GRA6612 list.

version April 17, 2023.

This is a list of things you should know after going through parts of the course. The list is not
exhaustive, it is meant as a constructive pedagogical device, not as a
you-do-not-need-to-know-stuff-that-is-not-on-this-list type of list. Chapter 1

in Mastering ‘metrics: Randomized trials

- You should be able to explain what a causal effect / treatment effect is (the difference
between two potential outcomes) and explain the difference between treatment heterogeneity
and treatment effect homogeneity.

Causal effect that is also known as the treatment effect, is the difference between two potential
outcomes that will arise when we have two groups where one gets some form for treatment or
intervention. The participants might be in a randomized trial (casual analysis) where they are randomly
assigned treatment or control status. For instance, in an experiment where they want to find the effect
of a new medication for stress relief, they will give the medication to the treatment group and give
placebo to the control group. They will after some analysis of the results look to see if there is causal
effect on the treated group, thus look for lower stress in treatment group. But it´s important that the
groups we control is a ceteris paribus comparison, we need to be certain we don’t have selection bias
or omitted variable bias that would explain the effect, not the treatment.

Difference between treatment effect heterogeneity and treatment effect homogeneity. Treatment effect
heterogeneity means that the effect of the treatment can differ depending on different characteristics or
states. hence, that the treatment affect varies throughout different subgroups of participants.
Heterogeneity in the treatment effect show consider these variances and analyze them. An example
can be differences in the participants stress levels throughout the day due to different gender or
different jobs. While treatment effect homogeneity refers to the situation where the effect of the
treatment is equal for all the participants. So, no matter participants states or characteristics the effect
is consistent. Its often used as an simplifying assumption when there is no evidence of heterogeneity in
the sample.

- You should be able to explain with a reasonable amount of rigor why we cannot generally
compare two sample means to get an unbiased estimate of a causal effect. The rigor amounts
to be able to derive the decomposition of the difference in sample means into the sum of the
treatment effect and the selection effect in the case of treatment effect homogeneity (from
page 15 in the textbook).

When comparing two sample means to estimate a causal effect, we cannot generally obtain an
unbiased estimate due to the presence of confounding factors and selection bias. Confounding factors
are variables that are related to both the treatment and the outcome, making it difficult to isolate the
true causal effect. Selection bias refers to the non-random assignment of individuals to treatment
groups, which can introduce systematic differences between the groups.
To derive the decomposition of the difference in sample means into the treatment effect and the
selection effect, let's consider the case of treatment effect homogeneity. In this scenario, we assume
that the treatment effect is the same for all individuals or groups receiving the treatment, which
simplifies the analysis. We can express the difference in sample means as follows:

ΔY = E[Y|T=1] - E[Y|T=0],
where ΔY represents the difference in the outcome variable (Y) between the treatment group (T=1)
and the control group (T=0). This difference can be decomposed into two components: the treatment
effect (TE) and the selection effect (SE):

ΔY = TE + SE.

The treatment effect (TE) represents the causal effect of the treatment on the outcome. It captures the
true impact of the treatment, assuming no confounding or selection bias. In an ideal randomized
controlled trial (RCT), where individuals are randomly assigned to treatment groups, the treatment
effect would be the unbiased estimate of the causal effect.

The selection effect (SE) accounts for the differences between the treatment and control groups due to
confounding and selection bias. It captures the portion of the difference in sample means that arises
from factors other than the treatment itself. These factors can include pre-existing differences in
characteristics, attitudes, or behaviors between the groups.
It's important to note that in the case of treatment effect homogeneity, the selection effect is the main
source of bias, as the treatment effect is assumed to be constant across individuals. By understanding
and accounting for the selection effect, we can aim to estimate the treatment effect more accurately.

In summary, comparing two sample means to estimate a causal effect is generally not sufficient due to
confounding factors and selection bias. Decomposing the difference in sample means into the
treatment effect and the selection effect helps us recognize the sources of bias and better understand
the true causal impact of the treatment. Rigorous analysis and techniques such as randomized
controlled trials, matching methods, or instrumental variables can help mitigate these biases and
provide more reliable estimates of causal effects.

- You should be able to explain the statistical methods required to make a t-test comparing
two sample means. You should also be able to interpret the output from a t-test in STATA
and to understand tables that exploit this method in the textbook and in papers.

To make a t-test comparing two sample means, you would typically use a statistical method called the
independent samples t-test. This method is used when you want to compare the means of two groups
that are independent of each other. Here's an explanation of the steps involved in conducting a t-test
and interpreting the output, along with its application in STATA and the presentation of results in
textbooks and papers.

1. Hypotheses formulation:
 Null hypothesis (H0): The means of the two groups are equal.
 Alternative hypothesis (HA): The means of the two groups are not equal.
2. Assumptions:
 The data in each group are independent.
 The populations from which the samples are drawn are normally distributed.
 The populations have equal variances (homogeneity of variances assumption) unless
you specify otherwise (such as using Welch's t-test when variances are unequal).
3. Calculate the t-statistic:
 The t-statistic measures the difference between the sample means relative to the

variation within the groups.
 The formula for the t-statistic is: t = (mean1 - mean2) / (sqrt((s1^2/n1) + (s2^2/n2))),
where mean1 and mean2 are the sample means, s1 and s2 are the sample standard
deviations, and n1 and n2 are the sample sizes of the two groups.
4. Determine the degrees of freedom:
1.
 The degrees of freedom (df) is calculated based on the sample sizes and the
assumption about equal variances.
 If the variances are assumed to be equal, df = n1 + n2 - 2.
 If the variances are not assumed to be equal (Welch's t-test), a more complex formula
is used to calculate df.
2. Calculate the p-value:
 The p-value is the probability of observing a t-statistic as extreme as the one
calculated if the null hypothesis is true.
 The p-value is compared to a significance level (e.g., α = 0.05) to determine if the
null hypothesis is rejected or not.
 If the p-value is smaller than the significance level, we reject the null hypothesis and
conclude that there is a statistically significant difference between the two means.

Interpreting the output in STATA:

 STATA provides a convenient way to conduct t-tests using the "ttest" command.
 The output will include the t-statistic, degrees of freedom, and p-value.
 STATA also provides additional information such as confidence intervals for the difference in
means.

Interpreting tables in textbooks and papers:

 Tables in textbooks and research papers often present t-test results in a standardized format.
 The tables typically include columns for the sample means, standard deviations, t-statistics,
degrees of freedom, and p-values.
 The format may vary slightly depending on the specific research context or the statistical
software used.
 It is important to pay attention to the reported p-values to determine the statistical significance
of the results.
In summary, the t-test is a commonly used statistical method to compare two sample means. It
involves formulating hypotheses, checking assumptions, calculating the t-statistic and p-
value, and interpreting the results. STATA provides a straightforward way to perform t-tests,
while tables in textbooks and papers present the results in a standardized format for easy
interpretation.

- You should be able to explain what a randomized controlled trial is and the standard
(minimalist) methodology for doing this: Randomize subjects into a treatment and control
group, testing for balance of treatment and control groups using pretest characteristics and
test for the effect of treatment on outcomes using a t-test.

A randomized controlled trial (RCT) is a research design that aims to determine the causal effect of a
treatment or intervention on an outcome of interest. It is considered the gold standard for evaluating
the effectiveness of interventions because it allows for rigorous comparisons and minimizes biases.
The standard methodology for conducting an RCT can be summarized in the following steps:
1. Randomization:
 Subjects are randomly assigned to either a treatment group or a control group.
 Randomization helps ensure that the groups are comparable at the outset, minimizing
the impact of confounding factors.
2. Testing for balance:
 Pretest characteristics of subjects in the treatment and control groups are assessed
and compared.
 Common pretest characteristics include demographic variables, baseline measures of
the outcome variable, and other relevant factors.
 Statistical tests (e.g., t-tests or chi-square tests) are conducted to check if there are
any significant differences between the treatment and control groups.
 Balance between the groups implies that any observed differences in outcomes can
be attributed to the treatment itself rather than preexisting differences between the
groups.
3. Treatment implementation:
 The treatment group receives the specific intervention or treatment being studied.
 The control group typically receives a placebo, no treatment, or an alternative
treatment (if applicable) that is believed to have no effect on the outcome.
4. Outcome evaluation:
 After the treatment has been administered, the outcome of interest is measured or
assessed for both the treatment and control groups.
 Common outcome measures can include health outcomes, behavioral changes,
quality of life, etc.
5. T-test for treatment effect:
 To estimate the treatment effect, a t-test is often used to compare the mean outcomes
between the treatment and control groups.
 The t-test calculates a statistic that measures the difference in means relative to the
variation within the groups.
 The resulting t-statistic, along with the degrees of freedom and p-value, is used to
determine if the treatment effect is statistically significant.
 If the p-value is smaller than the chosen significance level (e.g., α = 0.05), it suggests
a statistically significant treatment effect.

By following this minimalist methodology, an RCT provides a rigorous and unbiased assessment of
the causal effect of a treatment or intervention. It ensures randomization to minimize selection bias,
tests for balance between groups to address confounding factors, and employs statistical tests such as
the t-test to evaluate the treatment effect.

- You should know the basic relationship between t-test of sample means and linear
regression with a dummy right hand side variable - and know how one can apply a t-test
using a regression model, including the possibility of (and the motivation for) controlling
for background characteristics.

The relationship between a t-test of sample means and linear regression with a dummy right-hand
side variable is that they both allow for hypothesis testing regarding the difference in means
between two groups. The t-test compares the means directly, while regression with a dummy
variable allows for controlling for background characteristics and examining the treatment effect
while accounting for other factors.
In linear regression, a dummy right-hand side variable (often represented as a binary variable,
taking values of 0 or 1) is used to represent group membership (e.g., treatment group and control
group). The coefficient associated with this dummy variable captures the difference in means
between the two groups.

To apply a t-test using a regression model, including the possibility of controlling for background
characteristics, you can follow these steps:

1. Set up the regression model:

 Use a linear regression model where the outcome variable is regressed on the
dummy variable representing group membership and any relevant background
characteristics or control variables.
 The regression equation would typically be of the form: Y = β0 + β1X + β2Z + ... +
ε, where Y is the outcome variable, X is the dummy variable representing group
membership, Z represents additional background characteristics or control
variables, β0, β1, β2, etc., are the coefficients, and ε is the error term.
2. Interpret the coefficients:
 The coefficient associated with the dummy variable (β1) represents the difference
in means between the two groups.
 A positive coefficient suggests that the treatment group has a higher mean outcome
compared to the control group, while a negative coefficient indicates the opposite.
 The coefficient can be tested for statistical significance using a t-test, just like in
the case of comparing two sample means directly.
3. Assess the significance:
 The t-test is used to determine if the coefficient of the dummy variable is
statistically different from zero.
 The t-statistic is calculated as the estimated coefficient divided by its standard
error.
 The resulting t-statistic, along with the degrees of freedom and p-value, can be
used to determine the statistical significance of the treatment effect.
 If the p-value is smaller than the chosen significance level (e.g., α = 0.05), it
suggests a statistically significant treatment effect.

Controlling for background characteristics:

 By including additional background characteristics or control variables (Z) in the

regression model, you can account for their potential influence on the outcome variable.
 This helps to isolate and estimate the treatment effect more accurately by holding constant
the effects of these background characteristics.
 Controlling for background characteristics can enhance the precision of the treatment effect
estimation and address potential confounding factors.

In summary, the t-test of sample means and linear regression with a dummy right-hand side
variable both enable hypothesis testing for the difference in means between two groups. Regression
with a dummy variable allows for controlling for background characteristics and assessing the
treatment effect while considering other factors. By applying a t-test within the regression
framework, you can test the significance of the treatment effect and control for background
characteristics simultaneously, leading to a more comprehensive analysis.
 - You should have a fairly good idea (this goes beyond this chapter) of how randomized
controlled trials (lab experiments and field experiments) are currently used in economics
and be able to give a brief overview of the history of the use of experiments in
economics.

Currently, RCTs are widely used in economics to study causal relationships and evaluate the
impact of interventions or policies. RCTs can be conducted in controlled laboratory settings or in
real-world field environments, each serving different purposes and offering unique advantages.

Lab experiments:
 Lab experiments are conducted in controlled environments, such as university laboratories,
where researchers have high control over the experimental conditions.
 In lab experiments, participants are typically recruited and randomly assigned to different
treatment conditions.
 The advantage of lab experiments is the ability to precisely control variables and isolate
causal effects while minimizing external influences.
 Lab experiments are often used to study human behavior, decision-making, game theory,
and economic principles under controlled conditions.

Field experiments:
 Field experiments are conducted in real-world settings, such as schools, communities,
workplaces, or markets, where researchers observe participants in their natural
environment.
 In field experiments, participants are randomly assigned to different treatment conditions,
and the effects of the intervention or policy are measured in real-life contexts.
 The advantage of field experiments is their high external validity, as they provide insights
into how interventions or policies work in practice.
 Field experiments are used to evaluate the effectiveness of various interventions, including
education programs, health policies, behavioral interventions, and economic incentives.

Overview of the history of experiments in economics:

 Experimental methods have a long history in economics, but their widespread use gained
prominence in the latter half of the 20th century.
 The first notable experiments in economics were conducted by Vernon Smith in the 1950s
and 1960s. Smith's experiments focused on market behavior and revealed important
insights into price formation and trading mechanisms.
 In the 1990s, there was a surge in the use of lab experiments in economics, particularly in
the fields of behavioral economics and game theory.
 Nobel laureates Daniel Kahneman and Vernon Smith, among others, contributed
significantly to the development and popularization of experimental methods in
economics.
 Over time, the use of experiments in economics expanded beyond lab settings, with
researchers conducting field experiments to examine real-world phenomena and policy
interventions.
 The use of RCTs in economics has grown rapidly in recent decades, driven by
advancements in experimental design, statistical methods, and the recognition of the
 importance of rigorous causal inference.

Today, experiments, including RCTs, have become an integral part of the economist's toolkit. They
provide valuable insights into human behavior, market dynamics, policy effectiveness, and
decision-making processes. The ongoing advancements in experimental methods continue to shape
the field of economics and deepen our understanding of economic phenomena.

- You should be reasonably familiar with a small number of applications of field

experiments in economics that you can use as examples (e.g. the health insurance examples
from the textbooks and the papers we have been through in class).

1. Oregon Health Insurance Experiment:

 The Oregon Health Insurance Experiment (OHIE) is a landmark field experiment conducted
in the United States.
 The experiment aimed to assess the impacts of expanding Medicaid (government-funded
health insurance for low-income individuals) on various outcomes, such as healthcare
utilization, financial well-being, and health outcomes.
 The experiment utilized a randomized controlled design, where eligible individuals were
randomly assigned to either the treatment group (received Medicaid coverage) or the
control group (did not receive coverage).
 The study found that Medicaid coverage significantly increased healthcare utilization,
improved financial protection, and reduced rates of depression.
 The OHIE provided valuable insights into the impacts of health insurance expansion on
vulnerable populations.

2. Natural experiment – The Iceland natural experiment:

The Iceland natural experiment, often referred to as the "tax free year," was a unique field
experiment that occurred in Iceland in 1987 due to a major tax reform. During this period,
income taxes were eliminated for one year, resulting in a tax-free environment. This event
provided an opportunity for researchers to study the effects of temporary tax reductions on
various economic outcomes.

Econometric details of the field experiment in Iceland included the following:

Identification Strategy:

 Researchers used a difference-in-differences (DID) approach, a commonly used

econometric technique in observational studies, to compare outcomes before and after the
tax-free year.
 They compared the changes in outcomes for individuals affected by the tax reform (i.e.,
Iceland residents) with the changes in outcomes for a suitable control group (i.e., non-
residents and unaffected individuals).

Data Collection:

 Data on various economic outcomes, such as labor supply, consumption patterns, and
investment decisions, were collected both before and after the tax-free year.
 Researchers relied on surveys, administrative records, and other relevant sources to gather
 the necessary data for their analysis.

Analyzing Treatment Effects:

 The researchers estimated the treatment effect of the tax-free year by comparing the
changes in outcomes for individuals residing in Iceland (the treatment group) with the
changes for individuals not affected by the tax reform (the control group).
 By comparing the pre- and post-reform differences in outcomes between the treatment and
control groups, the researchers could isolate the effects of the temporary tax reduction.

Statistical Analysis:

 Econometric techniques, such as regression analysis, were used to estimate the treatment
effect while controlling for other potential factors that could influence the outcomes.
 The researchers employed appropriate statistical tests to assess the statistical significance of
the estimated treatment effects and the reliability of the findings.

The Iceland natural experiment provided valuable insights into the effects of temporary tax
reductions on economic behavior. The study's findings revealed several interesting outcomes,
including increased labor supply, changes in consumption patterns, and alterations in
investment decisions during the tax-free year. This natural experiment offered a unique
opportunity to study the short-term effects of tax policy changes in a real-world setting,
contributing to the understanding of the economic consequences of tax reforms.

Continue

Chapter 2 in Mastering ‘metrics: Regression

Going to private collage vs public when it comes to salary. How to compare equaly. Harvey vs Uma,
girl boy etc. replace Harvey with Harvard girl student. Same gender and SAT-scores. Key-summary
measure.

- You should be able to explain the assumption of “selection on observables”.

The assumption of "selection on observables" is an important concept in the context of causal inference,
particularly in observational studies or non-randomized experiments. It refers to the assumption that,
conditional on the observed variables or characteristics, the treatment assignment is independent of the
potential outcomes. In simpler terms, it means that any differences in outcomes between treated and
untreated individuals can be attributed solely to the treatment itself, rather than to unobserved
confounding factors.

Here's a breakdown of the assumption and its implications:

1. Observables: "Observables" refer to the variables or characteristics that can be measured and
observed in a study, such as demographic information, pre-treatment characteristics, or baseline
variables.
2. Treatment assignment: The assumption of "selection on observables" implies that the
assignment of individuals to treatment or control groups is determined only by the observed
variables. In other words, individuals with similar observable characteristics have an equal
chance of receiving treatment or remaining untreated.
3. Independence of treatment and potential outcomes: The assumption states that, within each level
of the observed variables, the assignment to treatment or control group is unrelated to the
potential outcomes (the outcomes that would be observed under different treatment conditions).
4. Causal interpretation: When the assumption of selection on observables holds, it allows for a
causal interpretation of the treatment effect estimate. In other words, any differences in
outcomes between treated and untreated individuals can be attributed to the causal impact of the
 treatment.

It is important to note that the assumption of selection on observables is an assumption and cannot be
directly tested. It relies on the researcher's understanding of the context and the available observed
variables. However, sensitivity analyses and robustness checks can be performed to assess the potential
impact of unobserved confounding on the estimated treatment effects.

In summary, the assumption of "selection on observables" in causal inference assumes that, given the
observed variables, the assignment to treatment is independent of the potential outcomes. This
assumption allows for making causal claims about the treatment effect based on observed data, but it
should be carefully considered and evaluated in any causal analysis.

- You should be able to know how one can estimate causal effects under the assumption of
“selection on observables” using matching estimators. (You are not required to know the
mechanics of matching estimators - only the idea.)

Estimating causal effects under the assumption of "selection on observables" can be done using
matching estimators. Matching estimators aim to create comparable treatment and control groups
by matching individuals with similar observed characteristics.

Here's an overview of how matching estimators work:

1. Preprocessing:

 Before estimating the causal effect, a preprocessing step is performed to create matched
pairs or groups.
 This involves identifying individuals who received the treatment (treated group) and
individuals who did not receive the treatment (control group).
 The individuals are then matched based on their observed characteristics, such as age,
gender, education level, or any other relevant variables.

2. Matching methods:

 Various matching methods can be used, such as nearest neighbor matching, propensity
score matching, or kernel matching.
 Nearest neighbor matching involves pairing each treated individual with the closest control
individual in terms of observed characteristics.
 Propensity score matching involves estimating the probability of treatment assignment
(propensity score) based on observed characteristics and matching individuals with similar
propensity scores.
 Kernel matching is a non-parametric method that assigns weights to control individuals
based on their similarity to treated individuals.
 3. Estimating the causal effect:

 Once the matched pairs or groups are created, the causal effect is estimated based on the
differences in outcomes between the treated and control individuals within each pair or
group.
 This estimation is typically done using statistical methods such as paired t-tests,
difference-in-differences estimators, or regression models that account for the matching
structure.

4. Assessing balance and sensitivity:

 After estimating the causal effect, it is important to assess the balance achieved through
matching.
 Balance refers to the similarity of observed characteristics between the treated and control
groups after matching.
 Sensitivity analyses can also be conducted to evaluate the robustness of the estimated
causal effect to potential hidden bias or unobserved confounding.

By using matching estimators, researchers aim to create comparable treatment and control groups,
reducing the influence of observed confounders and allowing for a more accurate estimation of the
causal effect. Matching estimators provide a way to address selection bias and make causal
inferences under the assumption of "selection on observables" by leveraging observed
characteristics to create comparable groups.

- You should be able to explain why regression works as a matching estimator. This is a bit
difficult, and rests on the Frisch-Waugh theorem

Explaining why regression works as a matching estimator requires understanding the Frisch-Waugh
theorem and its implications. The Frisch-Waugh theorem is a fundamental result in econometrics that
provides insights into the relationship between regression and matching.
Here's an explanation of why regression can serve as a matching estimator based on the Frisch-
Waugh theorem:

1. Background on the Frisch-Waugh theorem:

 The Frisch-Waugh theorem states that the coefficients estimated in a regression model can be
interpreted as the average treatment effect for individuals who are "as if" they were in the
treatment group or the control group, depending on the specification of the regression.
 It allows for the estimation of the treatment effect by controlling for observed covariates,
effectively removing the influence of those covariates on the estimated treatment effect.

2. Applying the Frisch-Waugh theorem to matching:

 In matching, the goal is to create comparable treatment and control groups by matching
individuals with similar observed characteristics.
 Regression can serve as a matching estimator by controlling for the observed characteristics
in the regression model.
 3. Steps to use regression as a matching estimator:

 Start with a binary treatment variable (1 for treated, 0 for control) and other observed
covariates.
 Perform a regression analysis, where the outcome variable is regressed on the treatment
variable and the observed covariates.
 By including the observed covariates in the regression model, the estimated treatment effect
represents the average treatment effect for individuals who have the same observed
characteristics as the ones included in the regression model.
 The regression coefficients associated with the treatment variable, along with their standard
errors, can be interpreted as the estimated treatment effect and its precision.

4. Interpreting the regression results:

 The coefficient associated with the treatment variable represents the average treatment effect,
adjusting for the observed covariates.
 The standard error of the coefficient provides an indication of the precision of the estimated
treatment effect.
 Hypothesis tests or confidence intervals can be used to assess the statistical significance of
the treatment effect estimate.

In summary, the Frisch-Waugh theorem allows regression to work as a matching estimator by

controlling for observed covariates. By including the observed covariates in the regression model, the
estimated treatment effect can be interpreted as the average treatment effect for individuals with
similar observed characteristics. The theorem provides a theoretical foundation for using regression to
estimate treatment effects in the context of matching.
 - You should be able to explain omitted variable bias and argue about the direction of omitted variable bias
- the direction of the bias depends on two factors.

Omitted variable bias is a common concern in econometric analysis that arises when a relevant variable is left out
of a regression model. It refers to the bias introduced in the estimated coefficients due to the omission of one or
more important explanatory variables.
Here's an explanation of omitted variable bias and the factors that determine its direction:
1. Omitted Variable Bias:
 Omitted variable bias occurs when a relevant variable, which is correlated with both the dependent variable
and the included explanatory variables, is excluded from the regression model.
 The omission of this variable leads to a violation of the ceteris paribus assumption (i.e., holding all other
factors constant) and can bias the estimated coefficients of the included variables.
2. Factors determining the direction of omitted variable bias: a) Correlation with the omitted variable:
 The direction of omitted variable bias depends on the correlation between the omitted variable and the
included explanatory variables.
 If the omitted variable is positively correlated with the included variables, the estimated coefficients tend
to be upwardly biased (overestimated).
 Conversely, if the omitted variable is negatively correlated with the included variables, the estimated
coefficients tend to be downwardly biased (underestimated).
b) Impact on the dependent variable:
 The direction of omitted variable bias also depends on how the omitted variable affects the dependent
variable.
 If the omitted variable positively affects the dependent variable, then the estimated coefficients tend to be
underestimated (biased towards zero).
 Conversely, if the omitted variable negatively affects the dependent variable, the estimated coefficients
tend to be overestimated.
It's important to note that the direction of omitted variable bias is not solely determined by the correlation between
the omitted variable and the included variables or its impact on the dependent variable. Other factors, such as the
magnitude of the omitted variable's effect and its correlation with the error term, can also influence the direction
and magnitude of the bias.
To mitigate omitted variable bias, researchers should carefully consider and include all relevant variables in their
regression models. Additionally, sensitivity analyses and robustness checks can be performed to assess the
potential impact of omitted variables on the estimated coefficients and draw more reliable conclusions.

- You should know how we try to assess whether we have successfully been able to control for the relevant
observables: We try to include even more covariates beyond the “selection controls” and see if they matter
for the effects.

Assessing whether we have successfully controlled for the relevant observables in a study typically involves
including additional covariates beyond the initial "selection controls" and examining their impact on the estimated
treatment effects. Here's an explanation of this process:
1. Initial control variables:

 In a study assuming "selection on observables," researchers typically include a set of control variables
(selection controls) that are believed to be correlated with both the treatment assignment and the outcome
of interest.
  These control variables aim to reduce confounding and establish balance between the treatment and control
groups.

2. Assessing the impact of additional covariates:

 To further assess the adequacy of controlling for relevant observables, researchers can introduce additional
covariates beyond the initial selection controls.
 These additional covariates may capture specific aspects of the treatment or outcome that were not captured
by the initial controls.
 By including these additional covariates in the analysis, researchers examine whether they have a
substantial impact on the estimated treatment effects.

3. Significance and interpretation of additional covariates:

 If the additional covariates are statistically significant and have a notable impact on the estimated treatment
effects, it suggests that the initial set of control variables may have been incomplete or insufficient in
capturing all relevant confounding factors.
 The significance and magnitude of the additional covariates provide evidence of their relevance in
explaining the treatment-outcome relationship.
 The inclusion of these additional covariates helps strengthen the case for controlling for relevant
observables and enhances the credibility of the estimated treatment effects.

4. Robustness checks:

 Assessing the impact of additional covariates can be seen as a robustness check, ensuring the robustness of
the estimated treatment effects to the inclusion of different sets of control variables.
 Researchers may also perform sensitivity analyses by adding or removing specific covariates to evaluate
the stability and consistency of the estimated treatment effects.

By including additional covariates beyond the initial selection controls and examining their impact on the estimated
treatment effects, researchers gain insights into whether the initial controls were sufficient in capturing relevant
confounding factors. The significance and interpretation of these additional covariates provide valuable evidence
for the adequacy of controlling for relevant observables in the study. This process helps strengthen the validity and
credibility of the estimated treatment effects by addressing concerns about omitted variable bias and uncontrolled
confounding.

- You should be able to explain why it is harder to argue why results are credible using the “selection on
observables”-assumption than e.g. with experiments. Specifically, since we do not know what drives the
remaining part of treatment assignment (beyond what is explained by the selection controls) it is very hard
to know why it should be unrelated to potential outcomes.

It can be more challenging to argue for the credibility of results based on the "selection on observables" assumption
compared to experimental studies. The key difficulty arises from the fact that, in non-experimental studies, we have
limited knowledge about what drives the remaining part of treatment assignment beyond what is captured by the
observed variables.
Here's an explanation of why it is harder to establish credibility using the "selection on observables" assumption:

1. Limited knowledge of unobserved factors:

  In non-experimental studies, the treatment assignment is typically based on real-world processes and
decisions, which can involve complex and unobservable factors.
 While the "selection on observables" assumption assumes that treatment assignment is independent of
potential outcomes, it does not guarantee independence from all unobserved factors that may influence both
treatment assignment and outcomes.
 The lack of knowledge about these unobserved factors makes it difficult to provide a strong argument for the
credibility of the results.

2. Potential for unmeasured confounding:

 The assumption of "selection on observables" relies on the idea that conditioning on observed variables
adequately captures the confounding factors and achieves balance between treatment and control groups.
 However, there may still exist unobserved confounding factors that are correlated with both treatment
assignment and potential outcomes.
 If unmeasured confounding exists, it can bias the estimated treatment effects and undermine the credibility
of the results.

3. Difficulty in establishing causal inference:

 Establishing causal inference based on the "selection on observables" assumption requires a convincing
argument that the observed variables fully capture the relevant factors driving treatment assignment.
 Without a clear understanding of the underlying mechanisms and determinants of treatment assignment, it
becomes challenging to assert that the remaining part of treatment assignment is unrelated to potential
outcomes.
 This lack of knowledge hampers the ability to confidently claim a causal relationship between the treatment
and the observed outcomes.

In contrast, experiments offer stronger credibility in establishing causal effects because treatment assignment is
typically randomized, minimizing the influence of both observed and unobserved confounding factors. Experimental
studies allow researchers to directly control and manipulate the treatment, leading to a more straightforward
interpretation of causal effects.
To address the challenges associated with the "selection on observables" assumption, researchers can employ
sensitivity analyses, robustness checks, and alternative estimation techniques (such as instrumental variable
methods) to assess the potential impact of unobserved confounding and strengthen the credibility of their findings.

- You should be able to explain the example case from the textbook.

Private vs public school will it affect salary. How to ceterius parebus. Matching strategy. Characteristics, gender, SAT
scores. But selection bias bc family background, ambition and ability. Got some significant results speaking for
private premium, but when exploring the selection bias, by selective schools etc, they turned out to be zero. What
about OVB? Like family background. It affects. No strong results for higher earnings for private school students, they
likely would earn more anyway due to smaller family, higher middleclass family, ambition and ability etc.

Chapter 3 in Mastering ‘metrics: Instrumental variables.

Example: charter schools vs public schools. Specific KIPP Lynn school vs public school on test scores.
Lottery if get in to KIPP or not. Will they go no matter what or do they not go no matter the outcome.
Looking at those who go if they win and don’t if they lose to compare apples to apples. Control variables, IV
estimator converts KIPP offer effects into KIPP attendance effects.
Dummy is indicating receiving offer to KIPP or not. Instrument needs to meet three requirements. Causal
effect on variable whose effect we want to study, instrument need to be randomly assigned and requires
exclusion restriction. LATE, TOT. KIPP recipe that indicate children might find easy to learn but might not
work for those that are bright to start with.
AND: Abuse, relapse. Page 115-. IV fixes experiment bc coddling was not delivered randomly. IV allowed
us to capture the casual effect of treatment on the treated in spite of the nonrandom compliance decisions
made by participants in experiment like the MDVE. Coddling was contaminated with selection bias, but IV
eliminates this source of selection bias. Asymmetry in the coddling- almost no always takers in the MDVE,
therefore all treated=compliers, in which case, LATE=TOT (p.121).
And: the population bomb – bad living standards (poverty) - smaller families, one child, sterilization.
Family size on child’s education. Twins vs non twin family. (ALS study), using twin birt as an instrument
for family size (p.128). What about sibling sex composition.Dummy 1 when same sex children and 0 when
not same sex simblings, but same sex composition showed no evidence of a quantity-quality trade off. The
exclusion restriction required for casual interpretation of sex-mix IV estimates asserts that sibling sex
composition matters for adult outcomes only insofar as it changes family size. Are more.

- You should be able to explain the key idea behind the instrumental variable method: Even if the variable
you want to measure the effect of is not exogenous: if you have access to a third variable that is exogenous
and that only affects the outcome through the variable that you want to measure the effect of - then you
can back out the effect of “x on y” by comparing the correlation between “z and x” and “z and y”.

The key idea behind the instrumental variable (IV) method is to address the issue of endogeneity when estimating
the causal effect between two variables, X and Y. Endogeneity occurs when the variable of interest, X, is correlated
with the error term in the regression equation, leading to biased estimates.

The IV method provides a way to obtain consistent and unbiased estimates by using an instrumental variable, Z. An
instrumental variable is a third variable that is exogenous (i.e., unrelated to the error term) and affects the outcome
variable, Y, only through its relationship with the variable of interest, X.

Here's how the IV method works:

1. Find an instrumental variable: Identify a variable, Z, that meets the criteria of being exogenous and
correlated with X. The key assumption is that Z affects Y only through its impact on X and is unrelated to
the error term in the regression equation.
2. Establish the relevance of the instrumental variable: Show that Z is correlated with X. This correlation
ensures that Z can help explain the variation in X.
3. Estimate the first-stage regression: Regress X on Z to quantify the effect of Z on X. This step provides the
causal relationship between the instrumental variable Z and the variable of interest X.
4. Estimate the second-stage regression: Regress Y on the predicted values of X obtained from the first stage.
Instead of using the original values of X, you use the predicted values to account for the potential
endogeneity. The coefficient on the predicted X variable in this regression gives you the causal effect of X
on Y.

By comparing the correlation between Z and X with the correlation between Z and Y, the instrumental variable
method allows you to obtain a consistent estimate of the causal effect of X on Y, even when X is not exogenous.
The underlying assumption is that the instrumental variable, Z, affects Y solely through its influence on X and not
through any other pathways.

It's important to note that the instrumental variable method relies on strong assumptions and careful consideration
of the instrument's validity. Violations of these assumptions can still lead to biased estimates. Therefore,
researchers must carefully select appropriate instrumental variables and assess the validity of the assumptions
before applying this method.

- You should be able to explain in detail how you can proceed to estimate a parameter using either indirect
 least squares or two stage least squares. This includes knowing how to take into account several
instruments in one analysis and knowing the role of control variables.

I'll explain in detail how you can proceed to estimate a parameter using both the indirect least squares (ILS) and
two-stage least squares (2SLS) methods. I'll also cover the consideration of multiple instruments and the role of
control variables in these estimation techniques.

1. Indirect Least Squares (ILS):

The ILS method is an instrumental variable approach that estimates a parameter of interest using a two-step
process. Here's how you can proceed:

Step 1: First Stage

 Identify a set of instrumental variables (Z) that are exogenous and correlated with the endogenous variable
(X) you want to estimate.
 Regress X on Z, using ordinary least squares (OLS).
 Obtain the predicted values of X (X-hat).
Step 2: Second Stage
 Regress the outcome variable (Y) on the predicted values of X obtained from the first stage, along with any
control variables you want to include.
 Use OLS to estimate the coefficient on the predicted X variable, which represents the causal effect of X on
Y.

If you have multiple instrumental variables, you can include them in the first stage regression equation as additional
predictors of X. This helps improve the precision of the estimation by accounting for more sources of exogenous
variation in X. In the second stage, you would still use the predicted values of X obtained from the first stage
regression.
Control variables can also be included in the second stage regression to account for other factors that may affect the
outcome variable Y. By controlling for these variables, you aim to isolate the causal effect of X on Y more
accurately.

2. Two-Stage Least Squares (2SLS):

The 2SLS method is another instrumental variable approach that addresses endogeneity and provides
consistent parameter estimates. Here's how you can proceed:

Step 1: First Stage

 Identify a set of instrumental variables (Z) that satisfy the requirements of being exogenous and correlated
with the endogenous variable (X).
 Regress X on Z, using OLS.
 Obtain the predicted values of X (X-hat).
Step 2: Second Stage
 Regress the outcome variable (Y) on the predicted values of X obtained from the first stage, along with any
control variables.
 However, instead of using OLS, use the predicted values of X as the explanatory variable and estimate the
coefficient using instrumental variable regression, which accounts for the endogeneity.
 The resulting coefficient on the predicted X variable gives you the consistent estimate of the causal effect of
X on Y.

In the case of multiple instrumental variables, you can include them in the first stage regression equation as
additional predictors of X. Similarly, control variables can be included in the second stage regression to improve the
accuracy of the estimation.
Overall, both ILS and 2SLS methods provide instrumental variable approaches to estimate parameters in the
presence of endogeneity. Multiple instrumental variables can be used to enhance the precision of the estimation,
while control variables help account for additional factors influencing the outcome variable.

- You should be able to explain, without going deeply, what the issues are with finite sample bias and weak
instruments. And explain how we deal with this in practice in applied economics (using the rule of thumb
F-test in the first stage).

I can explain the issues with finite sample bias and weak instruments and how we deal with them in practice in
applied economics, often using the rule of thumb F-test in the first stage.

1. Finite Sample Bias:

Finite sample bias refers to the bias that arises in parameter estimation when using a limited sample size. It occurs
when the estimated parameter consistently deviates from the true parameter value. In other words, the estimated
effect may not accurately represent the underlying causal relationship due to random variation.

In applied economics, finite sample bias can be problematic because it can lead to incorrect inferences and
conclusions. The bias arises from various factors such as measurement errors, omitted variables, or
misspecification of the econometric model. It becomes especially crucial to address this bias when estimating
causal effects or making policy recommendations based on empirical analysis.

2. Weak Instruments:
Weak instruments refer to instrumental variables that have weak or low explanatory power in explaining the
variation in the endogenous variable. In other words, the correlation between the instrumental variable and the
endogenous variable is weak. Weak instruments can lead to biased and inconsistent estimates in instrumental
variable regression.

When instrumental variables are weak, the standard errors of the estimated coefficients tend to be large, reducing
the precision of the estimation. This can result in inflated Type I errors (false positives) and compromised
statistical power. Weak instruments can arise due to measurement errors, low variability in the instrument, or weak
correlation with the endogenous variable.

3. Dealing with Finite Sample Bias and Weak Instruments:

In practice, applied economists employ several strategies to address finite sample bias and weak instruments. One
common approach is to use the rule of thumb F-test in the first stage. Here's how it works:

 In the first stage regression, where you regress the endogenous variable (X) on the instrumental variable(s)
(Z), you can compute the F-statistic to assess the strength of the instruments.
 The F-test compares the joint significance of the instrumental variables in explaining X. If the F-statistic is
greater than a predetermined critical value (commonly 10), it suggests that the instruments are sufficiently
strong.
 A significant F-test indicates that the instruments explain a substantial portion of the variation in the
endogenous variable, reducing the risk of weak instruments.
 If the F-test is not significant (i.e., the instruments are weak), it implies that the instruments may not
adequately address endogeneity concerns. In such cases, alternative instruments may need to be explored
or the interpretation of the instrumental variable results should be done cautiously.

Additionally, other techniques such as robust standard errors, heteroscedasticity-robust inference, and model
specification tests can be employed to mitigate the impact of finite sample bias and weak instruments.
It's important to note that addressing these issues requires careful consideration, and a thorough understanding of
the data, context, and assumptions underlying the applied econometric analysis. Researchers must use appropriate
diagnostics and rely on robust methodologies to ensure reliable and valid empirical results.

- You should be able to explain the interpretation of instrumental variable estimates under treatment
effect heterogeneity. What extra assumption is necessary (monotonicity - absence of “defiers”) - what
are compliers, always-takers and
never-takers, and that the IV estimate can be interpreted as a LATE (a local average treatment effect - the
average treatment effect among the compliers).

I can explain the interpretation of instrumental variable (IV) estimates under treatment effect heterogeneity, the
additional assumption of monotonicity, and the concepts of compliers, always-takers, and never-takers. Will
also explain how the IV estimate can be interpreted as a Local Average Treatment Effect (LATE) among the
compliers.

1. Interpretation of IV Estimates under Treatment Effect Heterogeneity:

In the context of treatment effect heterogeneity, individuals in a study population may respond differently to a
treatment, leading to variations in the estimated treatment effects. IV estimation allows us to address endogeneity
concerns and obtain causal estimates. However, the interpretation of these estimates becomes more nuanced when
treatment effects vary across individuals.

2. Monotonicity Assumption:
The monotonicity assumption is an additional requirement when estimating treatment effects using instrumental
variables. It states that everyone in the population belongs to one of two groups: those who always benefit from
treatment (always-takers) and those who never benefit from treatment (never-takers). In other words, individuals
cannot "defy" the instrument and receive the opposite treatment of what is indicated by the instrument.

3. Compliers, Always-Takers, and Never-Takers:

 Compliers: Compliers are individuals who would receive treatment only if they are assigned to the treatment
group and would not receive treatment if they are assigned to the control group. These individuals are
influenced by the instrumental variable and change their behavior based on the treatment assignment.

 Always-Takers: Always-takers are individuals who would receive treatment regardless of their assigned
group. They are not affected by the instrument and receive treatment irrespective of the treatment assignment.

 Never-Takers: Never-takers are individuals who would not receive treatment regardless of their assigned
group. They are not affected by the instrument and do not receive treatment regardless of the treatment
assignment.

4. Local Average Treatment Effect (LATE):

The IV estimate can be interpreted as a Local Average Treatment Effect (LATE), which represents the average
treatment effect among the compliers. Since compliers are the individuals who are influenced by the instrument
and change their treatment status based on the assignment, the LATE provides an estimate of the causal effect of
treatment on this subgroup.

The LATE is specific to the compliers and not necessarily representative of the entire population. It captures the
treatment effect for individuals whose treatment assignment is influenced by the instrument but does not capture
 the effects for always-takers or never-takers. Therefore, the interpretation of the IV estimate as a LATE highlights
the treatment effect among those individuals who are responsive to the instrument.

It is important to note that the assumptions of monotonicity and exclusion restriction (the instrument affects the
outcome only through its impact on the treatment) are crucial for interpreting the IV estimate as a LATE and
making valid causal inferences in the presence of treatment effect heterogeneity.

Overall, the interpretation of IV estimates under treatment effect heterogeneity involves identifying the compliers and
estimating the treatment effect specific to this subgroup, while recognizing the differences in treatment effects
among always-takers and never-takers.

- You should know and be able to explain the instrumental variable (validity) assumptions - including the
ability to come up with examples that would violate the assumptions.

I can explain the instrumental variable assumptions and provide examples of situations that would violate these
assumptions. The instrumental variable (IV) assumptions are necessary to ensure the validity of IV estimation and
obtain consistent and unbiased estimates of causal effects.
There are three main assumptions:

1. Relevance/Exogeneity: The instrumental variable should be relevant and exogenous.

 Relevance: The instrumental variable should be correlated with the endogenous variable of
interest.
 Example of violation: Suppose we want to estimate the effect of education on earnings using
years of education as the endogenous variable and parental income as the instrumental variable.
However, if parental income is not correlated with years of education, it violates the relevance
assumption.
 Violation example: Using a variable that is unrelated to the endogenous variable, such as a
random lottery number, as an instrumental variable.

2. Independence: The instrumental variable should be independent of the error term in the outcome equation.
 Independence: The instrumental variable should not be influenced by factors that affect the
outcome variable through channels other than the endogenous variable.
 Example of violation: Consider estimating the effect of smoking on lung cancer using a genetic
variant that affects both smoking behavior and lung cancer risk as an instrumental variable. If the
genetic variant influences lung cancer risk directly, violating independence, it would violate the
assumption.
 Violation example: Using an instrumental variable that is affected by unobserved confounders
that also affect the outcome variable.

3. Exclusion Restriction: The instrumental variable affects the outcome variable only through its impact on
the endogenous variable.
 Exclusion Restriction: The instrumental variable should have no direct effect on the outcome
variable, bypassing the endogenous variable.
 Example of violation: Suppose we want to estimate the effect of physical exercise on body
weight, using distance to the nearest gym as an instrumental variable. If the distance to the nearest
 gym has a direct effect on body weight (e.g., due to better access to healthy food near gyms), it
violates the exclusion restriction assumption.
 Violation example: Using an instrumental variable that has a direct effect on the outcome
variable, independent of the endogenous variable.

Violations of these instrumental variable assumptions can lead to biased and inconsistent estimates. It is crucial to
carefully select valid instrumental variables that satisfy these assumptions to ensure the credibility of the IV
estimation. Additionally, sensitivity analyses and robustness checks can be performed to assess the robustness of
the estimates to potential violations of the assumptions.

- You should be reasonably familiar with a small number of applications of instrumental variable studies that
you can use as examples (e.g. the KIPP lottery and the
quantity-quality model estimates from the textbook and the example papers that we have been through in
class presentations)

Here are a few examples of instrumental variable studies that have been commonly discussed in the literature:

1. The KIPP Lottery Study: The Knowledge Is Power Program (KIPP) is a network of charter schools in the
United States. The KIPP lottery study is a well-known instrumental variable analysis that examines the causal
impact of attending a KIPP school on students' educational outcomes. In this study, students who applied to
attend a KIPP school were randomly selected through a lottery system. The lottery provides an instrumental
variable that is exogenous and affects the treatment (attendance at a KIPP school) but is unrelated to
unobserved factors that may influence the outcome (educational achievement). Researchers use the
instrumental variable approach to estimate the causal effect of attending a KIPP school on academic
performance while addressing potential endogeneity issues.

2. Quantity-Quality Model Estimates: The quantity-quality model is a framework used to explore the relationship
between family size (quantity of children) and investment in child quality (such as education, health, and
nutrition). Estimating the causal effect of family size on child outcomes can be challenging due to endogeneity
concerns. Instrumental variable studies have been conducted to address this issue. For example, studies have
used instrumental variables such as twin births or sex composition of children to identify the exogenous
variation in family size. By leveraging these instrumental variables, researchers estimate the causal effect of
family size on various child outcomes, such as educational attainment, cognitive ability, or health status.

3. Angrist and Lavy (1999) - The Impact of Compulsory Schooling Laws: In their study, Angrist and Lavy
examined the causal impact of compulsory schooling laws on educational outcomes. The instrumental variable
they used was the changes in compulsory schooling laws across different regions and birth cohorts in Israel.
By exploiting these variations, they were able to estimate the causal effect of education on various outcomes,
such as earnings and employment. The instrumental variable approach helped address the endogeneity
concerns that could arise from self-selection into education or omitted variable bias.

These examples illustrate the application of instrumental variable analysis in various domains, including
education policy, family economics, and labor economics. They highlight the importance of instrumental
variables in obtaining causal estimates when randomized experiments are not feasible or ethical. By carefully
selecting valid instruments, researchers can address endogeneity concerns and make more reliable causal
inferences about the effects of treatments or policies.
 Chapter 4 from Mastering ‘metrics : Regression Discontinuity Design

You should be able to explain the idea behind the regression discontinuity design, that treatment
assignment is based on whether a “running” variable is above or below a threshold. Under the
assumption that it is not possible to manipulate the value of the running variable (on a fine scale), units
close to the threshold on either side are as good as randomly assigned.

Also an answer to what RDD is: In simple terms, a regression discontinuity design (RDD) is a research
method that allows us to estimate the causal effects of a treatment or intervention based on a threshold or
cutoff point. The idea behind RDD is that units just above and just below this threshold are expected to
be similar in all aspects except for their treatment status. By comparing the outcomes of units on either
side of the threshold, we can attribute any differences in outcomes to the treatment rather than other
factors.
To implement an RDD, we use a "running" variable that smoothly changes across the threshold. Units with
values above the threshold receive the treatment, while units with values below the threshold do not. By
examining the outcomes of units on either side of the threshold, we can determine the impact of the
treatment.
The key assumption in RDD is that the running variable is a valid indicator of the potential outcomes. In
other words, the running variable is closely related to the treatment status, and its values determine
whether units receive the treatment or not. By leveraging this natural assignment mechanism, RDD
allows us to estimate causal effects without the need for random assignment.
In essence, RDD provides a way to estimate causal effects by taking advantage of naturally occurring
thresholds or cutoff points. It allows researchers to study the impact of a treatment or intervention in
situations where random assignment is not feasible or ethical.

The regression discontinuity design (RDD) is a quasi-experimental design that utilizes a threshold in a "running"
variable to determine treatment assignment. The fundamental idea behind RDD is that individuals just above or below
the threshold are expected to be similar in all relevant aspects, except for their treatment status. This allows for the
estimation of causal effects by comparing outcomes for individuals on either side of the threshold.

The key assumption in RDD is that treatment assignment is based solely on whether the running variable exceeds the
threshold, and it is not possible to manipulate the value of the running variable on a fine scale. This assumption is
crucial for the identification of a causal effect. When treatment assignment is determined by a discontinuity in the
running variable, units close to the threshold on either side can be considered as quasi-randomly assigned to the
treatment and control groups.

The rationale behind this assumption is that individuals near the threshold share similar characteristics and are subject
to similar external factors that influence the running variable. Therefore, the treatment effect estimated by comparing
outcomes on either side of the threshold can be attributed to the treatment itself rather than confounding factors.

For example, suppose a study aims to evaluate the impact of a tutoring program on students' test scores. The running
variable could be students' pre-treatment scores, and the threshold might be set at a specific cutoff point. Students
scoring just above the threshold would receive the tutoring (treatment group), while those scoring just below the
threshold would not (control group). By comparing the test scores of students on either side of the threshold, the RDD
design allows for estimating the causal effect of tutoring on test scores.

The strength of RDD lies in the assumption that units close to the threshold are as good as randomly assigned,
meaning their treatment assignment is determined by a non-manipulable characteristic. However, it is essential to
carefully examine the validity of this assumption by assessing potential manipulation or discontinuities around the
threshold. Researchers should conduct sensitivity analyses and robustness checks to ensure the estimated treatment
effects are not driven by factors other than the running variable.

In summary, the RDD leverages a threshold in a running variable to determine treatment assignment and allows for
estimating causal effects by comparing outcomes on either side of the threshold. Under the assumption that the
running variable is not manipulable (on a fine scale), units close to the threshold are considered quasi-randomly
assigned, providing an opportunity to estimate causal effects in a quasi-experimental setting.
 - You should know how to implement a regression discontinuity design study using
regression analysis, including the ability to specify regression equations that take
different slopes and curves on either side of the threshold.

When implementing a regression discontinuity design (RDD) study using regression analysis, it is
essential to specify regression equations that capture different slopes and curves on either side of
the threshold.

Here's how you can do it:

1. Determine the functional form: Start by considering the functional form of the regression
equation. In an RDD study, you typically include a polynomial term or indicator variables
to allow for different slopes and curves on each side of the threshold.
2. Specify the regression equation: The general form of the regression equation in an RDD
study is as follows:

Y = β0 + β1X + β2X^2 + β3*T + ε

where Y is the outcome variable of interest, X is the running variable (e.g., pre-treatment scores),
T is the treatment indicator variable (1 for units above the threshold, 0 for units below), and ε
represents the error term.

3. Allow for different slopes and curves: To capture different slopes and curves on either
side of the threshold, include interaction terms between the running variable and the
treatment indicator (T). This allows for different effects of the running variable on the
outcome variable depending on the treatment status.
For example, you can modify the regression equation as follows:
Y = β0 + β1X + β2X^2 + β3T + β4(X * T) + β5*(X^2 * T) + ε
The coefficient β4 captures the difference in slopes between the treatment and control groups,
while β5 captures the difference in curves.

4. Estimate the regression model: Estimate the regression model using standard regression
techniques such as ordinary least squares (OLS). This will provide estimates of the
coefficients (β0, β1, β2, β3, β4, β5) and their corresponding standard errors.
5. Interpret the coefficients: Interpretation of the coefficients depends on the specific
research question and hypotheses. The coefficient of the treatment indicator variable (β3)
represents the average treatment effect at the threshold. The coefficients of the interaction
terms (β4 and β5) indicate the discontinuity in the treatment effect, capturing the
differences in slopes and curves on either side of the threshold.

By specifying regression equations that allow for different slopes and curves on each side of the
threshold, you can estimate the treatment effects and assess the presence of a discontinuity in the
outcomes around the threshold in an RDD study.
Note that the specific functional form and inclusion of additional covariates may vary depending
on the context and nature of the data. It's important to carefully consider the appropriate functional
form and potential confounders to obtain accurate and reliable estimates.
 - You should know the distinction between the sharp and the fuzzy RDD. In the sharp design,
treatment assignment is a function of whether the running variable is above or below the
threshold. In the fuzzy design, the probability of treatment is a discontinuous function of the
running variable at the threshold (but the probability of treatment does not jump from 0 to 1
as in the sharp design).

The distinction between the sharp and fuzzy regression discontinuity design (RDD) lies in how
treatment assignment is determined based on the running variable and the threshold. Here's an
explanation of each design:
1. Sharp RDD: In the sharp RDD, treatment assignment is a deterministic function of whether
the running variable is above or below the threshold. If the running variable exceeds the
threshold, the units are assigned to the treatment group; otherwise, they are assigned to the
control group. In this design, there is a clear discontinuity in treatment assignment at the
threshold, with a jump from 0 to 1 probability of treatment.
The sharp RDD design is characterized by a sharp change in treatment status at the threshold, making
it easier to identify the causal effect of the treatment on the outcome variable. The estimation of
treatment effects in the sharp RDD relies on comparing the outcomes of units just above and just
below the threshold.
2. Fuzzy RDD: In the fuzzy RDD, the probability of treatment is a discontinuous function of
the running variable at the threshold, but the probability does not jump from 0 to 1 as in the
sharp design. Instead, there is a more gradual change in the likelihood of treatment
assignment as the running variable approaches the threshold.
In the fuzzy RDD, treatment assignment is probabilistic and depends on the running variable's value
relative to the threshold. The treatment assignment mechanism might be based on a predetermined
probability function, such as a logistic regression model, that determines the probability of treatment
at different points in the running variable distribution.
The fuzzy RDD design introduces more complexity compared to the sharp design. Estimating the
causal effect of the treatment in the fuzzy RDD requires careful modeling of the treatment
assignment probabilities and may involve additional statistical techniques, such as instrumental
variable methods or propensity score matching, to account for the probabilistic nature of treatment
assignment.
Overall, the sharp RDD relies on a clear cutoff point in the running variable to determine treatment
assignment, while the fuzzy RDD involves a more gradual change in treatment probabilities as the
running variable approaches the threshold. Understanding this distinction is crucial for implementing
and interpreting RDD studies appropriately, as the design choice impacts the estimation strategy and
the assumptions underlying the causal inference.

- You should know the distinction (in principle) between parametric and non-parametric RDD
- and how the distinction becomes more blurry in practice. (RDD becomes “more non-
parametric” by zooming in on observations close to the threshold.)

I can help you answer that question. The distinction between parametric and non-parametric regression
discontinuity design (RDD) lies in the modeling approach used to estimate the treatment effects. Here's
an explanation of each approach and how the distinction becomes blurry in practice:
1. Parametric RDD: Parametric RDD involves specifying a functional form for the relationship
between the running variable and the outcome variable. This functional form is typically based
on prior theoretical knowledge or assumptions about the underlying data generating process.
Parametric RDD assumes a specific parametric model, such as linear regression or polynomial
regression, to estimate the treatment effects.
In parametric RDD, the regression equation includes predetermined functional forms and assumes a
specific distributional form for the error term. This allows for estimating the treatment effects using
standard parametric regression techniques. Parametric RDD may provide more precise and efficient
estimates if the chosen functional form accurately represents the relationship between the running
variable and the outcome variable.
2. Non-parametric RDD: Non-parametric RDD does not impose a specific functional form on the
relationship between the running variable and the outcome variable. Instead, it allows for more
flexible modeling without making strong assumptions about the data. Non-parametric RDD
methods use local polynomial regression or other flexible techniques to estimate the treatment
effects.
Non-parametric RDD estimates treatment effects by focusing on observations close to the threshold.
By zooming in on observations in the vicinity of the threshold, non-parametric methods capture the
local treatment effects without assuming a specific functional form for the relationship. This approach
accounts for the possibility of complex and non-linear relationships between the running variable and
the outcome variable.
In practice, the distinction between parametric and non-parametric RDD becomes blurry because
researchers often employ a combination of both approaches. While a parametric model may be
specified initially based on prior knowledge or assumptions, researchers may also examine the
sensitivity of the results to different functional forms or employ non-parametric techniques to test the
robustness of the findings.
Furthermore, researchers may use local polynomial regression or kernel-weighted regression to
estimate treatment effects in a way that focuses on the observations near the threshold, even within a
parametric framework. This allows for capturing the local treatment effects and addressing potential
concerns about the linearity of the relationship between the running variable and the outcome variable.
In summary, parametric RDD assumes a specific functional form and distributional assumptions, while
non-parametric RDD allows for more flexibility in modeling the relationship between the running
variable and the outcome variable. In practice, the distinction between these approaches becomes
blurry as researchers may employ a combination of both or use local regression techniques to capture
the local treatment effects within a parametric framework.

- You should know that the fuzzy design requires IV techniques, and you should know how to
implement a fuzzy regression discontinuity design graphically, by estimating the first stage
and the reduced form and taking their ratio (as the ILS method for IV).

The fuzzy regression discontinuity design (RDD) requires instrumental variable (IV) techniques
because treatment assignment in the fuzzy RDD is probabilistic rather than deterministic. In the fuzzy
RDD, the probability of treatment varies continuously as the running variable approaches the
threshold. This introduces endogeneity, meaning that the treatment assignment is not solely
determined by the running variable but may be influenced by other factors.

The use of IV techniques in the fuzzy RDD helps address the endogeneity issue by using an
instrument that is correlated with the treatment assignment but does not directly affect the outcome
variable, except through its influence on treatment assignment. The instrument is typically a variable
that affects the probability of treatment at the threshold but is unrelated to the outcome variable.

Here's how you can implement a fuzzy RDD graphically using IV techniques, estimating the first
stage and the reduced form, and taking their ratio (using the instrumental least squares, ILS, method):

1. Graphical representation: Start by plotting the running variable (X-axis) against the
probability of treatment (Y-axis) to visualize the fuzzy RDD. This plot should show a smooth
change in the probability of treatment as the running variable approaches the threshold. The
key is to identify a clear discontinuity or jump in the probability of treatment at the threshold.
 2. Estimate the first stage: In the first stage, you estimate the relationship between the running
variable and the instrument. The instrument should be a variable that affects the probability
of treatment but is unrelated to the outcome variable, except through its impact on treatment
assignment. The running variable itself can serve as the instrument in the fuzzy RDD.

Estimate a regression model where the instrument (running variable) is regressed on the treatment
indicator variable. The coefficient obtained from this regression represents the local average
treatment effect (LATE). Additionally, calculate the first-stage F-statistic to assess the strength of the
instrument.
3. Estimate the reduced form: Next, estimate the relationship between the instrument (running
variable) and the outcome variable. Regress the outcome variable on the instrument,
including any relevant control variables. The coefficient obtained from this regression
represents the reduced form estimate, which captures the overall relationship between the
instrument and the outcome variable.
4. Calculate the IV estimate: To obtain the IV estimate, take the ratio of the coefficient from the
reduced form to the coefficient from the first stage. This ratio represents the instrumental
variable estimate of the causal effect of treatment on the outcome variable. It provides an
estimate of the local average treatment effect (LATE) among compliers—individuals whose
treatment status changes based on the instrument (running variable) around the threshold.
By using IV techniques and estimating the first stage and reduced form, you account for the
endogeneity introduced by the fuzzy RDD and obtain a more reliable estimate of the causal effect.
The IV estimate allows you to attribute any observed differences in outcomes to the treatment, rather
than other confounding factors.
In summary, the fuzzy RDD requires IV techniques because treatment assignment is probabilistic,
and endogeneity needs to be addressed. By implementing the fuzzy RDD graphically and estimating
the first stage and reduced form, you can calculate the IV estimate, providing a more robust estimate
of the treatment effect. The ILS method, which involves taking the ratio of the coefficients from the
first stage and the reduced form, is a common approach to estimate the treatment effect in the fuzzy
RDD using IV techniques.

- You should be reasonably familiar with a small number of applications of RDD that you can
use as examples (e.g. the MLDA example and the Boston latin school example from the
textbook or the financial incentives to return to work from disability insurance or the effect
of paternal leave case from the student presentations.)

Let's elaborate on the Minimum Legal Drinking Age (MLDA) experiment and the Boston
Latin School experiment, which are both notable applications of the regression
discontinuity design (RDD).

1. Minimum Legal Drinking Age (MLDA) Experiment: The MLDA experiment focuses
on studying the effects of changes in the legal drinking age on various outcomes related
to alcohol consumption and related behaviors. In the United States, the MLDA has
varied across states over time, providing a natural experiment to evaluate the
consequences of different drinking age policies.

To implement an RDD in the MLDA experiment, researchers typically compare

individuals just above and just below the legal drinking age threshold. For example, if the
legal drinking age is 21, individuals who turn 21 just before a specific date would be
treated as the treatment group, while those who turn 21 just after that date would be treated
as the control group.
Researchers collect data on outcomes such as alcohol consumption, drinking patterns,
 alcohol-related accidents, educational attainment, and other related measures. By
comparing the outcomes between the treatment and control groups, they can estimate the
causal impact of the MLDA policy on these variables.

The RDD approach allows researchers to isolate the effects of the MLDA policy by
assuming that individuals near the threshold are similar in all aspects except for their legal
ability to drink. This helps address concerns about selection bias and confounding factors
that may affect the outcomes.

2. Boston Latin School Experiment: The Boston Latin School experiment focuses on
evaluating the effects of admission to the Boston Latin School, a prestigious exam-
based public high school in Boston, Massachusetts. The admission process for the
school relies on an entrance exam, and there is a cutoff score that determines admission.

Using RDD, researchers examine the outcomes of students who are just above and just
below the cutoff score. The students who score above the cutoff are admitted to the Boston
Latin School, while those who score below the cutoff are not admitted.
Researchers collect data on outcomes such as test scores, educational attainment, college
attendance, and future earnings. By comparing the outcomes between the admitted and
non-admitted students around the cutoff score, researchers can estimate the causal effects
of admission to the Boston Latin School.
The RDD approach allows researchers to identify the effects of attending the Boston Latin
School by leveraging the sharp discontinuity in admission based on the entrance exam
score. This design helps address concerns about self-selection bias and provides valuable
insights into the impact of attending a selective educational institution.

Both the MLDA experiment and the Boston Latin School experiment exemplify the use of
RDD to estimate causal effects by exploiting natural thresholds or cutoff points. These
applications highlight the power of RDD in providing rigorous evidence for policy
evaluation and decision-making.

Chapter 5 from Mastering ‘metrics: : Differences-in-Differences

- You should know the basics of the difference-in-differences methodology - we do not want to
simply compare outcomes before and after some treatment (or reform), called a first-
difference analysis, because time itself or something else that varies with time can affect
outcomes in addition to the treatment. We therefore want a control group that we can use to
construct a counterfactual: How would the outcomes in the treatment group have evolved
over time in the absence of treatment.

The difference-in-differences (DiD) methodology is a research design used to estimate the causal
impact of a treatment or intervention by comparing the changes in outcomes between a treatment
group and a control group over time. The goal of DiD is to identify the causal effect of the treatment
by isolating it from other factors that may influence the outcomes.
In a DiD analysis, researchers compare the differences in outcomes before and after the treatment
between the treatment group and the control group. The key idea is to utilize the control group as a
counterfactual, representing what would have happened to the treatment group in the absence of the
treatment.
Here's a breakdown of the DiD methodology:

1. Treatment group: This group consists of individuals or entities that are exposed to the
treatment or intervention. For example, it could be individuals who received a policy change, a
new program, or any other treatment of interest.
2. Control group: This group serves as a comparison or reference group and consists of
individuals or entities that are similar to the treatment group but did not receive the treatment.
The control group helps establish a counterfactual scenario, allowing researchers to estimate
the causal effect of the treatment.
3. Time periods: DiD requires observations before and after the treatment for both the treatment
and control groups. These time periods can vary depending on the research context.
4. Estimation: The DiD estimation involves comparing the changes in outcomes between the
treatment and control groups over time. By comparing the pre-treatment and post-treatment
differences in outcomes for both groups, researchers can estimate the causal effect of the
treatment.

The key advantage of the DiD methodology is that it helps address the issue of time-varying factors or
trends that may affect outcomes in addition to the treatment. By using the control group as a
counterfactual, researchers can control for time-invariant factors that affect both groups, while
focusing on the differential changes in outcomes associated with the treatment.
It's important to note that the validity of the DiD approach relies on the assumption of parallel trends.
This assumption implies that, in the absence of treatment, the treatment and control groups would have
followed similar trajectories over time. Deviations from this assumption can introduce bias into the
estimates.

In summary, the DiD methodology goes beyond simple before-and-after comparisons by utilizing a
control group to construct a counterfactual scenario. By comparing the changes in outcomes between
the treatment and control groups, DiD allows researchers to estimate the causal effect of a treatment or
intervention while addressing potential confounding factors associated with time.

- You should be able to show how to do a difference-in-difference analysis by taking the

differences between 4 sample means and how to do statistical inference (by adding the
sampling variance of those four sample means).

I can guide you through how to perform a difference-in-differences (DiD) analysis by taking
the differences between four sample means and conducting statistical inference using the
sampling variance. Here's a step-by-step explanation:

1. Define the groups: Identify the treatment group and the control group. These groups can
represent different entities, such as individuals, regions, or firms. Ensure that each group
has pre-treatment and post-treatment observations.
2. Calculate the differences in outcomes: Compute the differences in outcomes between the
pre-treatment and post-treatment periods for both the treatment and control groups. This
will result in four sample means: (1) treatment group pre-treatment mean, (2) treatment
group post-treatment mean, (3) control group pre-treatment mean, and (4) control group
post-treatment mean.
3. Calculate the DiD estimate: Compute the difference-in-differences estimate by taking the
difference between the changes in the treatment and control groups. This involves
subtracting the control group's pre-treatment mean from its post-treatment mean and
 subtracting the treatment group's pre-treatment mean from its post-treatment mean. The
formula is as follows:

DiD estimate = (Treatment Group Post - Treatment Group Pre) - (Control Group Post -
Control Group Pre)

4. Statistical inference: To conduct statistical inference and determine the significance of the
DiD estimate, you need to consider the sampling variance. This involves estimating the
standard error of the DiD estimate. You can use the following formula to calculate the
standard error:

Standard Error = sqrt((Var(Treatment Group Post) / n1) + (Var(Treatment Group Pre) / n1) +
(Var(Control Group Post) / n2) + (Var(Control Group Pre) / n2))

Here, n1 represents the number of observations in the treatment group, n2 represents the
number of observations in the control group, and Var represents the variance of each
respective group's outcomes.
5. Hypothesis testing: To test the null hypothesis that the DiD estimate is equal to zero
(indicating no treatment effect), you can calculate the t-statistic by dividing the DiD
estimate by the standard error. You can then compare the t-statistic to the critical values of
the t-distribution with appropriate degrees of freedom to determine statistical significance.

If the t-statistic is greater than the critical value (e.g., if it falls in the rejection region), you can
reject the null hypothesis and conclude that there is a significant treatment effect.
It's important to note that this explanation assumes a basic implementation of DiD analysis. In
practice, researchers often employ more sophisticated econometric techniques, such as
regression models, to account for additional covariates and control for potential confounding
factors.

Remember, performing a DiD analysis requires careful consideration of the underlying data,
assumptions, and potential limitations. Consulting relevant statistical software or resources
can help streamline the calculations and conduct more robust statistical inference.

- You should know how to do the corresponding analysis using a regression model and how
this can be used to expand the analysis with covariates.

When conducting a difference-in-differences (DiD) analysis, using a regression model is a common

and more comprehensive approach. Regression allows for the inclusion of covariates, which can help
control for additional factors that may influence the outcomes. Here's how you can expand the DiD
analysis using a regression model:

1. Define the variables: Identify the dependent variable, which represents the outcome of interest,
and the independent variables, including the treatment indicator variable, the time indicator
variable, and any covariates you want to include in the analysis.
2. Set up the regression model: Specify the regression model that captures the DiD framework.
The model should include the treatment indicator, the time indicator, and their interaction
term. The interaction term captures the differential effect of the treatment before and after the
treatment period. Additionally, include any covariates you want to control for in the model.
The general form of the regression model is as follows:
Y = β0 + β1 * Treatment + β2 * Time + β3 * (Treatment * Time) + β4 * Covariate + ε

Here, Y represents the dependent variable, Treatment is the treatment indicator (0 or 1), Time is the
time indicator (0 or 1), Covariate is the additional covariate(s), β0 to β4 are the coefficients to be
estimated, and ε represents the error term.

3. Estimate the coefficients: Use the regression model to estimate the coefficients, including the
treatment effect (β3). The coefficient β3 captures the difference-in-differences estimate, which
represents the average treatment effect.
4. Assess statistical significance: Perform statistical inference to assess the statistical significance
of the coefficients. You can calculate standard errors for the coefficients and use hypothesis
testing (such as t-tests) to determine whether the coefficients are significantly different from
zero. If the treatment coefficient (β3) is statistically significant, it indicates a significant
treatment effect.
5. Interpret the results: Interpret the coefficients and their significance in the context of the
research question. The coefficient for the interaction term (β3) represents the average treatment
effect, accounting for the time-varying factors and covariates included in the model. Positive
values indicate a positive treatment effect, while negative values indicate a negative treatment
effect.

By including covariates in the regression model, you can control for additional factors that may
influence the outcomes and improve the precision of the treatment effect estimate. Covariates can help
address confounding variables and provide more reliable and robust results.

Remember to check the assumptions of regression, such as linearity, independence, and

homoscedasticity, and consider potential issues like endogeneity or omitted variable bias. Addressing
these concerns can be achieved through various econometric techniques, such as using instrumental
variables or employing robust standard errors.

Overall, utilizing a regression model in the DiD analysis allows for a more comprehensive examination
of the treatment effect while incorporating relevant covariates to enhance the accuracy and
interpretation of the results.

- You should know how to specify a regression model to perform a

difference-in-difference analysis with many units and several points in time for reform
introduction (as the MLDA case in the textbook).

When performing a difference-in-differences (DiD) analysis with many units and several points in
time, such as the case of the Minimum Legal Drinking Age (MLDA) reform, you need to specify a
regression model that accounts for the panel structure of the data. Here's how you can specify a
regression model for such an analysis:

1. Define the variables: Identify the dependent variable, which represents the outcome of interest,
and the independent variables, including the treatment indicator variable, the time indicator
variable, and any additional covariates you want to include in the analysis. In the MLDA case,
the treatment indicator variable would indicate whether the reform was implemented or not.
2. Incorporate the panel structure: Since you have multiple units (e.g., states or countries) and
several points in time, you need to account for the panel structure of the data. This can be done
by including fixed effects for both units and time periods. Fixed effects capture the unit-specific
 and time-specific characteristics that may influence the outcome variable. The general form of
the regression model with fixed effects is as follows:

Y_it = β0 + β1 * Treatment + β2 * Time + β3 * (Treatment * Time) + β4 * Covariate + α_i + γ_t +

ε_it

Here, Y_it represents the dependent variable for unit i at time t, Treatment is the treatment
indicator (0 or 1), Time is the time indicator (0 or 1), Covariate is the additional covariate(s), β0 to
β4 are the coefficients to be estimated, α_i represents the fixed effect for unit i, γ_t represents the
fixed effect for time t, and ε_it represents the error term.

3. Estimate the coefficients: Use the regression model to estimate the coefficients, including the
treatment effect (β3). The coefficient β3 captures the difference-in-differences estimate, which
represents the average treatment effect after accounting for the panel structure and covariates.
4. Assess statistical significance: Perform statistical inference to assess the statistical significance
of the coefficients. Calculate standard errors for the coefficients and use hypothesis testing
(such as t-tests) to determine whether the coefficients are significantly different from zero. If the
treatment coefficient (β3) is statistically significant, it indicates a significant treatment effect.
5. Interpret the results: Interpret the coefficients and their significance in the context of the
research question. The coefficient for the interaction term (β3) represents the average treatment
effect, accounting for the panel-specific and time-specific characteristics, as well as any
covariates included in the model. Positive values indicate a positive treatment effect, while
negative values indicate a negative treatment effect.

By including fixed effects for units and time periods, you account for the unobserved heterogeneity
and time-varying factors that may influence the outcomes. This helps to control for unit-
specific and time-specific factors that are constant over the time period of analysis.

Remember to check the assumptions of the regression model, such as linearity, independence, and
homoscedasticity, and consider potential issues like endogeneity or omitted variable bias.
Addressing these concerns can be achieved through various econometric techniques, such as
using instrumental variables or employing robust standard errors.

Overall, specifying a regression model that incorporates fixed effects for units and time periods
allows for a robust analysis of the treatment effect in a DiD framework with many units and
several points in time, as in the MLDA case.

- You should be aware of issues with standard errors when doing

difference-in-difference analysis with more than two time units, and be able to explain how we
can solve this with “clustered standard errors”.

When conducting a difference-in-differences (DiD) analysis with more than two time units, there
are specific issues with estimating standard errors that need to be addressed. One common
solution is to use "clustered standard errors." Let's explore the issues with standard errors in
this context and how clustered standard errors can resolve them:
1. Autocorrelation of errors: In DiD analysis with multiple time units, there is often a correlation
 between the error terms for observations within the same unit over time. This violates the
assumption of independence of observations, which is required for valid standard error
estimation. Autocorrelation of errors can lead to underestimation or overestimation of standard
errors, affecting the accuracy of hypothesis testing.
2. Heteroscedasticity: Heteroscedasticity refers to the unequal variance of errors across different
units or time periods. In DiD analysis, the presence of heteroscedasticity violates the
assumption of homoscedasticity, which assumes constant error variance. Heteroscedasticity can
lead to biased standard error estimates and incorrect inference.
To address these issues, clustered standard errors can be used. Clustered standard errors allow for
the estimation of standard errors that are robust to both autocorrelation and heteroscedasticity.
The key idea is to group observations within the same unit together and treat them as a cluster.
Here's how clustered standard errors are implemented:
1. Define the clusters: In the context of DiD analysis, clusters are typically defined as the units or
groups of units. For example, if you are examining the effect of a policy change across different
states, each state would be considered a cluster.
2. Estimate the model: Run the regression model for the DiD analysis as usual, including the
treatment, time, and interaction terms, along with any covariates. Obtain the coefficient
estimates and the residual errors.
3. Compute clustered standard errors: To calculate clustered standard errors, you need to adjust the
standard errors for clustering. This is done by taking into account the correlation of errors
within each cluster while estimating the standard errors. The most commonly used approach is
the Huber-White sandwich estimator, also known as the Eicker-Huber-White or "robust"
estimator.
4. Interpret the results: Interpret the coefficient estimates and the corresponding clustered standard
errors. The clustered standard errors provide more reliable standard error estimates that account
for the correlation and heteroscedasticity within clusters. They allow for valid inference and
hypothesis testing in the DiD analysis with multiple time units.
By employing clustered standard errors, you can address the issues of autocorrelation and
heteroscedasticity when estimating standard errors in DiD analysis with more than two time
units. This approach ensures that the standard errors accurately reflect the uncertainty
associated with the estimated coefficients and improves the reliability of the statistical
inference.

- You should be aware of weighting issues when using US states, Norwegian

municipalities or similar units (of very different size) for difference-in-difference
analysis.

When conducting a difference-in-differences (DiD) analysis using units of different sizes, such
as US states or Norwegian municipalities, it is crucial to be aware of weighting issues. These
issues arise due to the potential influence of larger units on the overall analysis compared to
smaller units. Here's a well-rounded answer to address this:
1. Size disparity: In studies using units of different sizes, such as states or municipalities,
the impact or significance of a particular unit may depend on its size. Larger units tend
to contribute more observations or have a greater weight in the analysis due to their
larger population or sample size. As a result, the overall results may be influenced
more by the outcomes of the larger units compared to the smaller units.
2. Biased estimates: When conducting a DiD analysis with units of different sizes, the
estimates may be biased if the analysis is not appropriately weighted. Biases can occur
 when the influence of larger units dominates the analysis, leading to an inaccurate
representation of the treatment effect or average treatment effect.
3. Weighting approaches: To address the weighting issues in DiD analysis with units of
different sizes, researchers employ various weighting approaches. Two common
methods are population weighting and unit-level weighting:
a. Population weighting: In population weighting, the weight assigned to each unit is
proportional to its population size or sample size. By assigning higher weights to larger units
and lower weights to smaller units, population weighting accounts for the varying sizes and
ensures that the analysis reflects the overall population or sample accurately.
b. Unit-level weighting: Unit-level weighting assigns specific weights to each unit based on
predetermined criteria. These weights can be based on factors such as economic importance,
relevance to the research question, or specific characteristics of the units. Unit-level weighting
allows for a more nuanced approach in addressing the size disparity among units.
4. Interpretation of results: When employing weighting techniques in DiD analysis, it is
essential to consider the implications for interpreting the results. Weighted estimates
provide a more balanced representation of the treatment effect across units of different
sizes. However, it is crucial to acknowledge that the interpretation may differ based on
the weighting approach used.
In summary, weighting issues arise when conducting DiD analysis with units of different sizes.
The disproportionate influence of larger units can bias the estimates and affect the overall
results. To address this, researchers can employ weighting techniques such as population
weighting or unit-level weighting to account for the size disparity and ensure a more accurate
representation of the treatment effect. Careful consideration and appropriate weighting are
necessary to obtain reliable and unbiased estimates in such analyses.

- You should be reasonably familiar with a small number of applications of

difference-in-differences that you can use as examples (e.g. the monetary policy and
 MLDA examples from the textbook and the minimum wage case or the disability insurance case
from the student presentations.)
Here are a few examples of applications of difference-in-differences (DiD) analysis:
1. Monetary policy: In the context of monetary policy, researchers often employ DiD analysis to
evaluate the impact of interest rate changes on various economic outcomes. For example, they
may compare the economic performance of countries or regions before and after a specific
monetary policy intervention, such as a change in interest rates. By examining the differences in
outcomes between the treated (exposed to the policy change) and control groups (not exposed),
researchers can estimate the causal effect of the policy change on economic variables like GDP,
inflation, investment, or employment.
2. Minimum wage: DiD analysis is frequently used to assess the effects of minimum wage policies
on labor markets. Researchers compare the employment and wage levels in regions or industries
where minimum wage increases have occurred with those where minimum wage remains
unchanged. By comparing the differences in outcomes before and after the policy change
between the treated and control groups, the DiD approach helps estimate the impact of
minimum wage changes on employment, earnings, and other labor market indicators.
3. MLDA (Minimum Legal Drinking Age): The MLDA example you mentioned refers to studies
examining the effects of changes in the legal drinking age on alcohol consumption, traffic
accidents, and other related outcomes. Researchers employ DiD analysis by comparing the
outcomes of interest before and after the change in the MLDA between states with different
legal drinking ages. By comparing the changes in outcomes between states that raised or
lowered the MLDA and those that kept it constant, researchers can estimate the causal effects of
the MLDA on alcohol-related outcomes.
4. Disability insurance: DiD analysis is also utilized to study the effects of disability insurance
policies on labor force participation and other employment-related outcomes. Researchers
compare individuals who are eligible for disability benefits (treated group) with those who are
ineligible (control group). By comparing the changes in employment and labor force
participation before and after the policy change, the DiD approach helps estimate the causal
effects of disability insurance on individuals' work behavior and labor market outcomes.
These are just a few examples of how DiD analysis is applied in different contexts to evaluate the
causal effects of policy interventions or natural experiments. Each case involves comparing
outcomes before and after a treatment or policy change between treated and control groups,
allowing researchers to estimate the treatment effects by accounting for other factors that may
influence the outcomes

Computer problems and STATA (you should really expect to be tested on

this on the exam)

- You should be able to perform analyses on data. That includes the ability to perform
 analysis of all the “furious five” analysis methods covered by the textbook.
- You should be familiar with the “ttest”, “regression” and “ivregression” and “xtreg”
commands from STATA.
- In addition to the analysis commands, you should master simple commands to inspect and
manipulate data and be able to make simple plots, particularly plots that you need for the
“furious five”.
- You should know how to access the data from a research paper on the internet, find the
associated do-files and replicate the paper. (We have done this in class with the credit
supply paper.)
Notes from Adham:

 * Make histograms (regression discontinuity - se om tallene vi har er manipulerte)

 * hvis søylene rundt threshold er veldig varierende (veldig høy og
veldig lav) er det indikasjon på manipulasjon - kan variere, men ikke være syke
"sjokk"

Lab experiments, experimental/behavioral economics (Lecture 11)

- You should know in general terms how lab experiments are used in economic research, both
how they are conducted and what kind of research questions they are used for.
Lab experiments are commonly used in economic research as a powerful method for investigating
causal relationships and understanding human behavior in controlled settings. Here's an answer that
covers the general aspects of conducting lab experiments in economic research and the types of
research questions they are used for:
1. Conducting lab experiments: In a lab experiment, researchers create a controlled environment
where they manipulate specific variables of interest and observe participants' responses. Here
are some key steps involved in conducting lab experiments:
a. Designing the experiment: Researchers carefully design the experiment by defining the research
question, selecting the variables to manipulate (independent variables), determining the outcome
measures (dependent variables), and specifying the experimental conditions or treatments.
b. Random assignment: Participants are randomly assigned to different experimental groups or
conditions to ensure that any observed differences are due to the treatment and not pre-existing
differences between individuals.
c. Implementing the experiment: Researchers administer the experimental tasks or treatments to the
participants, carefully controlling the presentation of stimuli and instructions.
d. Data collection: Data on participants' behavior, decisions, or responses are collected using various
methods, such as surveys, behavioral tasks, or economic games.
e. Data analysis: Researchers analyze the collected data using statistical techniques to examine the
effects of the experimental manipulations and draw conclusions.
2. Research questions addressed by lab experiments: Lab experiments in economic research aim
to investigate various aspects of human behavior, decision-making, and economic phenomena.
They can address questions related to:
a. Behavioral economics: Lab experiments are used to explore deviations from rational behavior,
decision-making under risk and uncertainty, biases, and heuristics. For example, researchers may
investigate how individuals make choices in economic games, such as the Ultimatum Game or the
Prisoner's Dilemma, to understand social preferences, cooperation, or strategic behavior.
b. Market behavior: Lab experiments can simulate market settings and examine phenomena like price
formation, market efficiency, and the impact of policy interventions. Researchers may investigate the
effects of different market structures, pricing mechanisms, or information asymmetry on market
outcomes.
c. Public policy and incentives: Lab experiments can evaluate the effectiveness of public policies,
incentive schemes, or interventions aimed at promoting desirable behaviors or addressing market
failures. For instance, researchers may test the impact of financial incentives on pro-environmental
behavior or the effectiveness of nudges in promoting healthier choices.
d. Social interactions and networks: Lab experiments provide insights into social interactions, network
formation, and cooperation. Researchers may study how individuals form social networks, how
information spreads through networks, or how social norms influence behavior.
e. Experimental finance: Lab experiments are widely used in finance research to investigate topics
such as asset pricing, market anomalies, investor behavior, and risk preferences.
Lab experiments offer controlled settings where researchers can isolate specific factors and study their
causal effects. By manipulating variables and observing participants' behavior, lab experiments
provide valuable insights into economic decision-making, human behavior, and the underlying
mechanisms that drive economic outcomes.

- You should know the basic features that distinguish lab experiments in economics from
lab experiments in other social sciences such as psychology.

Lab experiments in economics and lab experiments in other social sciences, such as psychology,
have some similarities but also distinct features that set them apart. Here are some basic features
that distinguish lab experiments in economics from lab experiments in other social sciences:

1. Economic Decision-Making: Economics lab experiments typically focus on studying

individual and group decision-making in economic contexts. The primary objective is to
understand how individuals make choices and interact with others in economic situations.
The experiments often involve tasks related to resource allocation, incentives, pricing,
bargaining, auctions, and market behavior.
2. Rationality and Self-Interest: Economic experiments often assume that participants are
rational decision-makers who act in their self-interest. This assumption is central to
economic theory and is used to analyze and predict economic behavior. Participants are
typically incentivized to make choices that maximize their own well-being or financial
gain.
3. Incentives and Monetary Rewards: Economic lab experiments frequently use monetary
incentives to motivate participants' behavior. Monetary rewards are used to align
participants' actions with economic outcomes and create real-world incentives. For
example, participants may receive payments based on their performance or the outcomes
of market interactions.
4. Market Mechanisms: Economic experiments often involve simulated markets where
participants can trade goods, services, or assets. These markets can be designed to study
various economic phenomena like supply and demand, market efficiency, price
determination, and market power. The focus is on understanding how market forces and
individual decisions interact to shape outcomes.
5. Formalized Experimental Design: Economic lab experiments often employ rigorous
experimental design and procedures to ensure internal validity and control for confounding
factors. This includes random assignment of participants to treatment groups, pre- and
post-treatment measurements, manipulation of independent variables, and control groups
to establish causality.
6. Emphasis on Economic Theory: Economic lab experiments are closely tied to economic
theory, and researchers often aim to test or validate economic models and hypotheses. The
experiments provide empirical evidence to support or challenge economic theories and
help refine and develop new economic models.
In contrast, lab experiments in psychology and other social sciences may focus on different aspects
of human behavior, cognition, perception, or social interactions. They may explore topics like
memory, attention, perception, decision-making in non-economic contexts, social influence,
attitudes, emotions, and more. While some similarities exist, the emphasis and specific research
questions addressed in psychological lab experiments can differ from those in economics.

Here are some ways in which lab experiments in psychology may differ from those in economics:

1. Research Questions and Topics: Psychological lab experiments may investigate a wide
range of topics, such as memory processes, attentional mechanisms, perception, learning,
language acquisition, personality traits, social influence, motivation, emotions, and
psychological disorders. These topics are generally not directly tied to economic concepts
like markets, prices, or resource allocation.
2. Experimental Manipulations: Psychological lab experiments may involve different types of
manipulations compared to economic experiments. For example, psychologists may
manipulate variables related to stimulus presentation, cognitive load, emotional states,
social context, priming, or experimental tasks specifically designed to study psychological
phenomena.
3. Incentives and Rewards: While economic experiments often rely on monetary incentives
and rewards, psychological lab experiments may use alternative forms of incentives. These
can include non-monetary rewards like course credits, extra credit, tokens, or the
opportunity to contribute to scientific knowledge. The focus is more on understanding
psychological processes rather than economic outcomes.
4. Experimental Design: The experimental design in psychological lab experiments may
differ from economic experiments. Psychological experiments may employ various
designs, including within-subject or between-subject designs, counterbalancing, control
groups, and random assignment. However, the specific design choices are tailored to the
research question and phenomena being studied within psychology.
5. Theoretical Framework: Psychological lab experiments often draw on psychological
theories and frameworks to guide their research questions and interpretations. These
theories may come from cognitive psychology, social psychology, developmental
psychology, or other branches of psychology. The emphasis is on understanding human
behavior and mental processes from a psychological perspective, rather than economic
theory.
6. Data Analysis and Interpretation: Psychological lab experiments may utilize statistical
analyses and modeling techniques specific to psychological research, such as analysis of
variance (ANOVA), correlation analysis, regression analysis, factor analysis, or structural
equation modeling. The interpretation of results is based on psychological theories and
concepts, rather than economic models.
While there may be some overlap and interdisciplinary studies that bridge economics and
psychology, the primary focus and research questions addressed in lab experiments in each field
tend to differ due to the disciplinary orientations and theoretical frameworks of the respective
disciplines.

- You should know a variety of different games that are often used in lab experiments in
economics (dictator games, ultimatum games etc).

In lab experiments in economics, various games are commonly used to study decision-making,
social interactions, and economic behavior. Here are some of the most well-known games used
 in economics lab experiments:

1. Dictator Game: In the Dictator Game, one participant (the dictator) is given a sum of money
and has the power to unilaterally decide how to divide it between themselves and another
participant. The dictator can choose to keep the entire amount or allocate some portion to the
other participant.
2. Ultimatum Game: The Ultimatum Game involves two participants: a proposer and a
responder. The proposer is given a sum of money and must propose a division of the money
between themselves and the responder. The responder can accept or reject the proposal. If
the responder rejects the offer, both participants receive nothing.
3. Trust Game: The Trust Game examines trust and trustworthiness in economic interactions. It
involves two participants: a trustor and a trustee. The trustor is given an endowment and can
choose to send a portion of it to the trustee. The amount sent is usually multiplied by a factor
before it reaches the trustee, who then decides how much to send back to the trustor.
4. Public Goods Game: The Public Goods Game explores cooperation and free-riding behavior
in a group setting. Participants are given an endowment and can choose to contribute any
amount to a public pool. The total amount contributed is multiplied by a factor and then
equally distributed among all participants, regardless of their individual contributions.
5. Auctions: Auctions are commonly used in economics experiments to study bidding behavior
and market dynamics. Different types of auctions, such as English auctions, Dutch auctions,
sealed-bid auctions, and ascending-bid auctions, are used to examine price determination,
efficiency, and strategic bidding strategies.
6. Market Games: Market games simulate economic markets and allow participants to engage
in trading activities. Participants can buy and sell goods or assets, negotiate prices, and
observe market dynamics. These games provide insights into market behavior, price
formation, and the impact of various market mechanisms.
7. Bargaining Games: Bargaining games, such as the Nash Bargaining Game or the Rubinstein
Bargaining Game, involve two participants negotiating the division of a surplus. Participants
make offers and counteroffers until an agreement is reached or the negotiation breaks down.
These games study strategic decision-making and the emergence of fair or equitable
outcomes.
8. Risk and Uncertainty Games: Games like the Risky Choice Game or the Stag Hunt Game
are used to study decision-making under risk and uncertainty. Participants make choices
involving probabilities, payoffs, or potential gains and losses. These games help examine
risk preferences, risk aversion, and attitudes towards uncertainty.
These are just a few examples of the games commonly used in lab experiments in economics.
Each game is designed to explore different aspects of economic behavior, strategic thinking,
cooperation, fairness, and decision-making in controlled laboratory settings.

Course introduction, papers from Teele-textbook.

- You should be aware of the “credibility revolution in economics” including the pre-
revolution description in Leamer (1982). The revolution is summarized in the
papers by Angrist and Pischke and more quantitatively in Panhans and Singleton. What are the
key elements in the credibility revolution?
- You should be aware of the explosive increase in the use of randomized controlled trials in
economics and in particular in development economics. You should be able to give a brief
description, including some examples, of how randomized controlled trials have been used in
 development economics. (If this is vague, look at Duflo’s ted talk and afterwards explain
what she talked about to your spouse, flatmate or little sister.)
- You should understand the key part from Chapter 1 in Teele. Under the assumption that we
do not know anything about the bias in observational studies, we cannot learn anything
particularly useful from observational studies and should not waste resources on such
studies. The flip side is of course that if we can safely assume the bias to be small, then we
can learn from observational studies.
 - You should know and understand the seminal paper by Lalonde (1986). In this paper, he
compares observational best-practice routines for evaluating the effects of a welfare program
and compares with the results from a randomized controlled trial. He finds that best-practice
observational study methods are not able to recover the results from the randomized
controlled trial. Note how this relates to the point above.
- You should know the key argument from Chapter 2 in Teele. The viewpoint that we cannot
learn from observational studies is characterized as “radical skepticism”. The key argument
is that if the criticism from Chapter 1 (“there could always be omitted variable
bias”/”selection effects”) is applied to randomized controlled trials, then “there could always
be omitted interaction terms”. You need to understand that an omitted interaction term
would only lead us to find an average between the true average treatment effects, not to bias.
- You should be pretty well aware of the criticisms of quasi-experimental and experimental
approaches in Deaton’s paper in the Teele textbook and how we could respond to these
criticisms.

Research papers:
You should be aware of and know the main results / contributions from the research papers that have
been presented in class.

Pres. Topic 1: Famous Field Experiments in Economics

- Randomized trials: treatment and control group.

Article 1:

- Are Emily and Greg more employable than Lakisha and Jamal? Field experiment on labor
Market discrimination by Marianne Bertrand and Sendhil Mullainathan (2004).
- Fehr and Goette (2007): Do workers work more if wages are high? Evidence from a
randomized field experiment.

Summary article 1:
Marianne Bertrand and Sendhil Mullainathan (2004) presents the findings of a field experiment
that investigates potential racial discrimination in hiring practices. The study examines whether
individuals with distinctively African-American names, such as Lakisha and Jamal, face
different employment opportunities compared to individuals with distinctively white names,
such as Emily and Greg.
The researchers conducted a controlled field experiment by submitting fictitious resumes in
response to actual job advertisements in Boston and Chicago. The resumes were randomly
assigned either a white-sounding name or an African-American-sounding name, while other
qualifications and experiences remained identical. The goal was to evaluate the callback rates
from employers to determine if there were any disparities based on the perceived race
indicated by the names.
The results of the study indicate a significant racial bias in hiring practices. Resumes with white-
sounding names, Emily and Greg, received about 50% more callbacks for interviews compared
to resumes with African-American-sounding names, Lakisha and Jamal. The researchers found
that even when controlling for other factors, such as qualifications and previous work
experience, the racial disparities in callbacks persisted.
The study sheds light on the prevalence of racial discrimination in the labor market and highlights
 the potential biases that job applicants with African-American names face during the hiring
process. The findings suggest that individuals with African-American names may experience
lower employment opportunities due to the implicit biases of employers.
The authors discuss the implications of their findings and suggest that policies aimed at reducing
racial discrimination in hiring should be considered. They emphasize the importance of raising
awareness about such biases and promoting equal opportunities for all individuals, regardless
of their racial or ethnic background.
Overall, the article provides empirical evidence of racial discrimination in the labor market by
examining differences in callback rates for job applicants with different-sounding names. It
contributes to the understanding of the challenges faced by minority groups in accessing
employment opportunities and highlights the need for policies and interventions to address
these disparities.

Econometric methods/theory used:

n their study, Bertrand and Mullainathan (2004) used econometric methods and statistical analysis
to analyze the data and arrive at their results. Here are some of the econometric methods and
theories employed in their research:
1. Field Experiment Design: The study utilized a field experiment design, which is a research
method often used in economics to observe real-world behavior in controlled settings. The
researchers strategically manipulated the racial identifiers (names) on resumes while keeping
other qualifications constant. This allowed them to isolate the effect of perceived race on
callback rates.
2. Hypothesis Testing: The study aimed to test the hypothesis of racial discrimination in hiring by
comparing the callback rates of resumes with white-sounding names to those with African-
American-sounding names. The statistical analysis involved hypothesis testing to determine if
the observed differences in callback rates were statistically significant.
3. Regression Analysis: The researchers employed regression analysis, a commonly used
econometric method, to estimate the relationship between the dependent variable (callback
rates) and the independent variable (racial identifiers). They used regression models to control
for other relevant factors, such as qualifications and previous work experience, in order to
isolate the impact of perceived race on callback rates.
4. Control Variables: The study incorporated control variables in their regression models to
account for other potential factors that might affect callback rates. These control variables could
include education level, years of experience, industry, or other relevant characteristics that
might influence hiring decisions. By including these variables, the researchers aimed to isolate
the effect of perceived race on callback rates while controlling for other factors.
5. Statistical Inference: Bertrand and Mullainathan used statistical inference techniques to draw
conclusions from their findings. They examined the statistical significance of their results to
determine if the observed differences in callback rates were unlikely to have occurred by
chance. This helped them assess the robustness and reliability of their findings.
6. Descriptive Statistics: The study employed descriptive statistics to provide a summary of the
data and present key characteristics of the sample, such as the mean callback rates for resumes
with different racial identifiers. Descriptive statistics allowed the researchers to provide an
overview of the data and facilitate the comparison of different groups.
By employing these econometric methods and statistical techniques, Bertrand and Mullainathan
aimed to rigorously analyze the data and uncover evidence of racial discrimination in hiring
practices. These methods helped them control for confounding factors, isolate the impact of
perceived race, and provide statistical evidence to support their findings.

- Summary article 2: Do workers work more if wages are high?

 The article "Do Workers Work More if Wages Are High? Evidence from a Randomized Field
Experiment" by Ernst Fehr and Lorenz Goette investigates the relationship between wages and
work effort. The study utilizes a randomized field experiment to examine how individuals
respond to different wage levels in terms of their work hours.

Fehr and Goette conducted their experiment in collaboration with a large Swiss corporation. They
randomly assigned employees to different wage groups, with some individuals receiving
higher wages than others for performing the same tasks. The researchers then analyzed the
impact of these wage differences on work effort.
The key findings of the study indicate that higher wages lead to increased work effort. Workers
who received higher wages showed a significant increase in the number of hours they worked
compared to those who received lower wages. This finding supports the hypothesis that
financial incentives, in the form of higher wages, can motivate individuals to exert more effort
and work longer hours.
The authors also explored potential mechanisms underlying this relationship. They found that
workers who received higher wages experienced higher levels of job satisfaction and perceived
their pay as fair. These psychological factors likely contributed to the increased work effort
observed among those receiving higher wages.
The study provides empirical evidence from a real-world setting, highlighting the importance of
financial incentives in shaping work behavior. It suggests that higher wages can be an effective
tool for motivating employees to work more and improve their productivity.
The findings have implications for both employers and policymakers. Understanding the
relationship between wages and work effort can help inform decisions related to wage policies,
incentive structures, and employee motivation strategies.

In summary, the article by Fehr and Goette examines the impact of wage levels on work effort
using a randomized field experiment. The study finds that higher wages lead to increased work
hours, suggesting that financial incentives play a significant role in motivating individuals to
work more. The research contributes to our understanding of the relationship between wages
and work behavior in real-world settings.

Econometric Theory:

In their article "Do Workers Work More if Wages Are High? Evidence from a Randomized Field
Experiment," Ernst Fehr and Lorenz Goette (2007) investigate the relationship between wages
and work effort using a randomized field experiment. Here are some of the econometric
methods and approaches they employ:

1. Randomized Control Trial (RCT): Fehr and Goette utilize a randomized control trial design in
which they randomly assign participants to different wage groups. This random assignment
ensures that any differences observed in work effort between the groups can be attributed to the
wage variation and not to other confounding factors. RCTs are considered a gold standard in
causal inference as they help establish a clear cause-and-effect relationship.
2. Difference-in-Differences (DiD) Analysis: The authors employ a DiD approach to analyze the
impact of wage levels on work effort. They compare the work effort of individuals in the high-
wage group with the work effort of individuals in the control group, who receive lower wages.
By examining the differences in work effort before and after the wage manipulation between the
two groups, they can estimate the causal effect of higher wages on work effort.
3. Statistical Analysis: Fehr and Goette use statistical analysis to quantify the relationship between
 wages and work effort. They may employ various statistical techniques, such as t-tests or
regression analysis, to assess the significance and magnitude of the wage effect on work effort.
These analyses help determine whether the observed differences in work effort between the
high-wage and control groups are statistically significant.
4. Robustness Checks: The authors conduct robustness checks to ensure the validity of their
findings. They may test the sensitivity of their results to different specifications or control
variables to assess the robustness of the wage-work effort relationship.
5. Psychological Factors Analysis: Fehr and Goette also investigate the role of psychological
factors in mediating the relationship between wages and work effort. They may use additional
econometric methods, such as mediation analysis or structural equation modeling, to explore the
mechanisms through which wages influence work effort, such as job satisfaction or perceptions
of fairness.

By employing these econometric methods, Fehr and Goette aim to provide rigorous empirical
evidence on the relationship between wages and work effort. Their randomized field
experiment and econometric analysis help establish causal links and provide insights into the
behavioral responses of workers to variations in wage levels

Pres. Topic 3: Instrumental variables

- Randomized trials: treatment and control group.

Articles:

- Lifetime earnings and the Vietnam era draft lottery: Evidence from social security
administration records by Joshua D. Angrist (1990).
- The effects of class size on student achievement: Nex evidence from population variation by
Caroline M. Hoxby (2000)

Summary article 1, lifetime earnings:

In short:
Joshua D. Angrist's article "Lifetime Earnings and the Vietnam Era Draft Lottery: Evidence from
Social Security Administration Records" (1990) examines the impact of military service during the
Vietnam War on individuals' lifetime earnings. Using data from the Social Security
Administration, Angrist finds that being drafted into the military during the Vietnam era led to
lower lifetime earnings compared to those who were not drafted. The study highlights the long-
term economic consequences of military service during that period.

More detailed

The article titled "Lifetime Earnings and the Vietnam Era Draft Lottery: Evidence from Social
Security Administration Records" by Joshua D. Angrist (1990) examines the impact of military
service during the Vietnam War on individuals' lifetime earnings. The study utilizes data from
the Social Security Administration to investigate the long-term economic consequences of
being drafted into the military during the Vietnam era.
Angrist focuses on the draft lottery system implemented in 1970, which randomly assigned draft
numbers to eligible young men. By comparing the earnings of those who were drafted due to
receiving low lottery numbers with those who were not drafted due to receiving high lottery
numbers, the study aims to estimate the causal effect of military service on lifetime earnings.
The author analyzes data from the Social Security Administration, which provides detailed
 earnings information over individuals' entire working lives. Angrist employs econometric
techniques, including regression analysis and instrumental variables, to control for
confounding factors and isolate the effect of military service on earnings.
The findings of the study suggest that military service during the Vietnam era had a negative
impact on individuals' lifetime earnings. Specifically, those who were drafted experienced a
significant reduction in their earnings compared to those who were not drafted. The effect was
particularly pronounced for individuals with lower levels of education.
Angrist further explores the heterogeneity in the earnings effect based on characteristics such as
race and ethnicity, educational attainment, and marital status. The study finds that the negative
earnings impact of military service was more substantial for certain subgroups, such as African
Americans and individuals with lower educational attainment.
The article concludes that military service during the Vietnam era had long-lasting economic
consequences, leading to lower lifetime earnings for those who were drafted. These findings
contribute to the understanding of the economic implications of military service and the long-
term effects of historical events on individuals' economic outcomes.
In summary, Joshua D. Angrist's article investigates the relationship between military service
during the Vietnam era and lifetime earnings. Utilizing data from the Social Security
Administration and employing econometric methods, the study finds that being drafted into the
military had a negative impact on individuals' earnings, with certain subgroups experiencing
more substantial effects. The research sheds light on the long-term economic consequences of
military service during a specific historical period.

Summary article 2: The effects of class size on student achievement:

Caroline M. Hoxby's article titled "The Effects of Class Size on Student Achievement: New
Evidence from Population Variation" (2000) investigates the impact of class size on student
achievement using a novel approach that leverages population variation in class size.
Hoxby's study aimed to address the limitations of previous research on the subject, which often
relied on samples of schools or classrooms that may not be representative of the broader
population. Instead, she utilized a comprehensive dataset that included information on all
public school districts in the United States, covering a wide range of demographic and
socioeconomic characteristics.
The key innovation in Hoxby's research was the identification of naturally occurring variation in
class size across different schools and districts. By comparing schools within the same district,
she could examine the effects of class size while controlling for other factors that may
confound the relationship.
Hoxby employed a variety of statistical techniques to analyze the data and estimate the causal
effect of class size on student achievement. She used both cross-sectional and longitudinal
analyses to explore the relationship between class size and academic outcomes, such as test
scores and graduation rates.
The findings of the study indicated that class size had a limited effect on student achievement.
Hoxby did not find a strong relationship between smaller class sizes and improved academic
performance. The effect sizes she observed were generally small and statistically insignificant.
Additionally, Hoxby explored potential heterogeneity in the effects of class size based on student
characteristics, such as race, socioeconomic status, and prior achievement. Her analysis
revealed that the impact of class size was consistent across different student groups, suggesting
that class size did not disproportionately benefit or harm particular subgroups.
Hoxby's study challenged the prevailing belief at the time that reducing class size would
substantially improve student achievement. While acknowledging that her findings did not
 imply that class size was irrelevant, she emphasized that other factors, such as teacher quality
and instructional practices, might play more significant roles in determining student outcomes.
Overall, Hoxby's research provided valuable insights into the relationship between class size and
student achievement by utilizing population-level data and accounting for potential
confounding factors. Her findings suggested that the effects of class size on academic
performance were modest and called for a broader consideration of other educational factors
that might have a more substantial impact on student outcomes.

Econometric methods/theory used:

Certainly! In her article, "The Effects of Class Size on Student Achievement: New Evidence from
Population Variation," Caroline M. Hoxby (2000) employs various econometric methods to
examine the relationship between class size and student achievement. Here are some of the
econometric methods and approaches she utilizes:

1. Instrumental Variable (IV) Analysis: Hoxby employs an instrumental variable approach to

address potential endogeneity issues in estimating the causal effect of class size on student
achievement. Endogeneity arises when there is a two-way relationship between class size and
student achievement, making it challenging to establish causality. By using instrumental
variables that are correlated with class size but unrelated to student characteristics or
achievement, Hoxby can obtain unbiased estimates of the causal impact.
2. Fixed Effects Models: Hoxby utilizes fixed effects models to control for unobservable time-
invariant characteristics of students and schools that could potentially confound the class size
effect. Fixed effects models allow for within-group comparisons, capturing the variation in
achievement within students or schools over time. By including these fixed effects, Hoxby aims
to isolate the impact of class size on student achievement.
3. Difference-in-Differences (DiD) Analysis: Hoxby employs a difference-in-differences approach
to analyze the impact of changes in class size on student achievement. By comparing changes in
student achievement before and after changes in class size, and across different schools or
districts with varying class size policies, she aims to estimate the causal effect of class size.
4. Instrumental Variable Regression Discontinuity Design: In certain cases, Hoxby uses an
instrumental variable regression discontinuity design, which leverages sharp cutoff points in
class size policies to estimate the causal effect of class size. This approach allows for a
comparison between students who are just above or just below the threshold of the class size
policy change.

These are some of the econometric methods and approaches employed by Hoxby in her article to
examine the relationship between class size and student achievement. By using these methods,
she aims to address potential biases, endogeneity concerns, and confounding factors that could
arise when estimating the causal impact of class size on student outcomes. The combination of
these econometric techniques helps provide robust evidence on the effects of class size on
student achievement.

Article 1:

In the article, Joshua D. Angrist (1990) used econometric methods and statistical techniques to
analyze the data and arrive at the findings. Here are the econometric methods and theories
 employed in the research:

1. Regression Analysis: Angrist used regression analysis, a commonly used econometric method,
to estimate the relationship between military service (draft) and lifetime earnings. The
regression models allowed him to control for other factors that might influence earnings and
isolate the causal effect of military service.
2. Instrumental Variables: The study utilized instrumental variables to address potential
endogeneity issues and establish causal inference. Angrist used the draft lottery numbers as
instrumental variables, as they were randomly assigned and could be used to separate the effect
of military service from other factors.
3. Control Variables: The study included control variables in the regression models to account for
other factors that might affect lifetime earnings. These variables could include education level,
demographic characteristics, and other relevant variables that might influence earnings.
4. Heterogeneity Analysis: Angrist explored heterogeneity in the earnings effect by examining
how the impact of military service varied across different subgroups, such as race and ethnicity,
educational attainment, and marital status. This analysis allowed for a deeper understanding of
how the effect of military service differed across various groups.
5. Longitudinal Data Analysis: The study utilized longitudinal data from the Social Security
Administration, which provided information on individuals' earnings over their entire working
lives. This allowed for the examination of lifetime earnings trajectories and the long-term
economic impact of military service.
By employing these econometric methods and statistical techniques, Angrist aimed to rigorously
analyze the data and estimate the causal effect of military service on lifetime earnings. These
methods helped control for confounding factors, address endogeneity concerns, and provide
empirical evidence supporting the relationship between military service during the Vietnam era
and long-term earnings outcomes.

From PP about IV in article.

IV analysis estimates the causal effect of military service on lifetime earnings, addressing the
problem of endogeneity.
The draft lottery numbers were randomly assigned, providing an exogenous source of variation in
the treatment variable.
Comparing the earnings of individuals above and below the draft eligibility threshold helps
estimate the causal effect of military service on earnings.
IV analysis shows military service during the Vietnam War had a negative impact on lifetime
earnings, especially for combat veterans, and is a crucial methodological contribution to this study.

Econometric methods/theory used:

Article 2:
Caroline M. Hoxby's article on "The Effects of Class Size on Student Achievement: New Evidence
from Population Variation" (2000) employed various econometric techniques and theoretical
frameworks to investigate the relationship between class size and student achievement.

1. Identification Strategy: Hoxby utilized a quasi-experimental design that relied on naturally

occurring variation in class size across schools and districts. By comparing schools within the
same district, she aimed to isolate the causal effect of class size on student outcomes while
controlling for other factors that could confound the relationship. This approach helped address
endogeneity concerns and increase the internal validity of the analysis.
 2. Fixed Effects Models: Hoxby employed fixed effects models to account for unobservable time-
invariant characteristics of schools and districts that could influence student achievement. By
including fixed effects for each school or district, she controlled for time-invariant factors that
might be correlated with both class size and academic outcomes.
3. Instrumental Variables: In some analyses, Hoxby used instrumental variable (IV) estimation to
address potential endogeneity between class size and student achievement. She identified
instruments that were correlated with class size but unrelated to student characteristics or
unobserved factors that could affect academic performance. By using these instruments, she
aimed to obtain consistent estimates of the causal effect of class size.
4. Propensity Score Matching: Hoxby employed propensity score matching techniques to compare
outcomes between schools with different class sizes while ensuring balance in observable
characteristics. This allowed her to create comparison groups that were similar in observable
attributes, reducing the potential bias due to selection effects.
5. Subgroup Analysis: Hoxby conducted subgroup analyses to examine whether the effects of
class size varied across different student groups. She explored heterogeneity based on factors
such as race, socioeconomic status, and prior achievement. This allowed her to investigate
whether the relationship between class size and academic outcomes differed for different
student populations.
6. Theory and Mechanisms: While the article primarily focused on empirical analysis, Hoxby also
discussed theoretical frameworks that could explain the relationship between class size and
student achievement. She highlighted the importance of teacher quality and instructional
practices as potential mechanisms through which class size could influence academic outcomes.
By considering these theoretical pathways, she provided additional context for interpreting the
empirical results.

Hoxbys strategies to eliminate all sources of bias

1.Instrumental variable
-Natural randomness in population as IV
-Explains the random variation in class size
-How the random variation affects student achievement
-Eliminates a lot of potential bias problems-
Not directly correlated with achievement ➔Fulfils the requirements of an IV

2.Regression discontinuity
-Maximum and minimum class size triggers

Instrumental variable strategy

3 steps
1.Obtaining the instrumental variable
2.First-stage regression: Identifying the random variation in class size
3.Second-stage regression: Uses the predicted value of class size
3 different approaches
-School level
-District level
-Kindergarten Cohorts (“K5”)

Overall, Hoxby's study utilized econometric techniques such as fixed effects models, instrumental
 variables, propensity score matching, and subgroup analysis to examine the effects of class
size on student achievement. These methods helped address endogeneity concerns, control for
confounding factors, and explore heterogeneity across student groups, enhancing the
robustness and validity of the findings.

Pres. Topic 4: More on Instrumental variables

Articles:

- Estimating the effect of unearned income on labor earnings, savings, and consumption:
Evidence form a survey of lottery players. By Imbens, Rubin and Sacerdote (2001).
- Does compulsory school attendance affect schooling and earnings? By Angrist and Krueger
(1991).

Summary article 1, unearned income on labor earnings, savings and consumption.

Now, let's discuss the article "Estimating the Effect of Unearned Income on Labor Earnings,
Savings, and Consumption: Evidence from a Survey of Lottery Players" by Imbens, Rubin,
and Sacerdote (2001).
In this article, the authors investigate the impact of unearned income, specifically lottery winnings,
on various economic outcomes such as labor earnings, savings, and consumption. The study
focuses on utilizing survey data collected from lottery winners to estimate the causal effects of
receiving unearned income on individuals' economic behavior.
The authors employ a quasi-experimental approach, leveraging the exogenous variation in lottery
winnings as a source of identifying the causal effect. By comparing the economic outcomes of
lottery winners before and after their windfall, and also comparing them to a control group of
non-winners, the researchers aim to isolate the impact of unearned income.
The study utilizes a comprehensive survey of lottery winners in Massachusetts, providing detailed
information on their earnings, savings, consumption, and other relevant variables. The authors
analyze this data using various econometric techniques to estimate the causal effects
accurately.
The findings of the study reveal important insights into the effects of unearned income on
economic behavior. The authors find that the receipt of lottery winnings has a substantial
positive effect on savings. In particular, they observe a significant increase in savings rates
among lottery winners compared to the control group. This suggests that unearned income, in
the form of lottery winnings, motivates individuals to save more.
Additionally, the study examines the effects of unearned income on labor earnings and
consumption. The authors find mixed evidence regarding the impact on labor earnings. While
there is some indication of a positive effect, the results are not as conclusive as the findings on
savings. Similarly, the effects on consumption are more nuanced, with some evidence of
increased consumption but not as consistently observed as the effects on savings.
Overall, the article contributes to the understanding of the behavioral responses to unearned
income, specifically lottery winnings. The study highlights the positive effect of lottery
winnings on savings behavior, indicating that windfalls can lead individuals to save more. The
findings on labor earnings and consumption provide more mixed results, suggesting that the
impact of unearned income on these outcomes may be less straightforward.
It is worth noting that this summary provides a general overview of the article's key points based
on the information provided. For a more detailed understanding, it is recommended to refer
directly to the original article by Imbens, Rubin, and Sacerdote (2001).

Econometric methods:
 In their article "Estimating the Effect of Unearned Income on Labor Earnings, Savings, and
Consumption: Evidence from a Survey of Lottery Players" (Imbens, Rubin, and Sacerdote,
2001), the authors employ several econometric methods and theoretical frameworks to
estimate the causal effects of unearned income on economic outcomes.

1. Quasi-Experimental Design: The authors use a quasi-experimental design that relies on the
exogenous variation in lottery winnings as a source of identification. By comparing the
outcomes of lottery winners before and after their windfall and comparing them to a control
group of non-winners, they aim to establish a causal relationship between unearned income and
economic behavior.
2. Difference-in-Differences Analysis: To estimate the causal effects, the authors employ a
difference-in-differences approach. By comparing the changes in economic outcomes over time
for lottery winners and non-winners, they can isolate the effect of unearned income on the
outcomes of interest.
3. Matching Techniques: The authors utilize matching techniques to compare the characteristics of
lottery winners and non-winners and ensure that they are comparable. This helps address
potential selection biases and improves the validity of the estimated effects. Matching methods
such as propensity score matching may be used to create matched pairs or groups of individuals
with similar observable characteristics.
4. Regression Analysis: The authors employ regression analysis to estimate the causal effects of
unearned income on labor earnings, savings, and consumption. They specify regression models
that control for relevant covariates and potentially time-varying factors to isolate the effect of
lottery winnings on the outcomes of interest.
5. Instrumental Variables: In some cases, the authors may employ instrumental variables (IV)
estimation to address endogeneity concerns and obtain consistent estimates. Instrumental
variables are variables that are correlated with unearned income but not directly influenced by
other factors that could confound the relationship. By using these instruments, the authors can
obtain unbiased estimates of the causal effects.
6. Economic Theory: The authors discuss economic theories that help interpret the estimated
effects of unearned income on economic behavior. For example, they may draw upon theories
of wealth effects or income shocks to explain the observed changes in savings, labor earnings,
and consumption.

Overall, the authors use a combination of quasi-experimental designs, difference-in-differences

analysis, matching techniques, regression analysis, instrumental variables, and economic
theory to estimate the causal effects of unearned income on various economic outcomes. These
methods help address endogeneity concerns, control for confounding factors, and provide a
rigorous framework for interpreting the results.

Summary Article 2: Does compulsory school attendance affect schooling and earnings?

The article titled "Does Compulsory School Attendance Affect Schooling and Earnings?" by
Joshua D. Angrist and Alan B. Krueger (1991) explores the impact of compulsory schooling
laws on educational attainment and subsequent earnings.
Angrist and Krueger utilize a natural experiment created by changes in compulsory schooling laws
across different U.S. states and cohorts. They examine how variations in the timing and
stringency of these laws affect individuals' schooling outcomes and earnings.
The authors employ an instrumental variable (IV) strategy to address potential endogeneity issues
and identify the causal effect of compulsory schooling on educational attainment and earnings.
 They use exogenous variation in the years of schooling induced by compulsory schooling laws
as an instrument to estimate the impact on individuals' education and earnings.
Using data from the National Longitudinal Surveys (NLS) and the Current Population Surveys
(CPS), Angrist and Krueger analyze the educational and earnings outcomes of individuals who
were affected by different compulsory schooling laws during their formative years.
The findings of the study reveal that increasing the length of compulsory schooling leads to higher
educational attainment among individuals. The authors find that individuals who were exposed
to more compulsory schooling years were more likely to complete higher levels of education,
such as high school and college. This suggests that compulsory schooling laws have a positive
effect on educational attainment.

Moreover, Angrist and Krueger find a positive correlation between additional years of schooling
due to compulsory attendance laws and higher subsequent earnings. Individuals who were
required to attend school for longer periods tend to have higher earnings in their adult years.
This finding suggests that the impact of compulsory schooling extends beyond educational
attainment and has long-term economic benefits in terms of increased earnings.
The article provides valuable insights into the relationship between compulsory schooling laws,
educational attainment, and earnings. By utilizing instrumental variable techniques and
exploiting natural variations in compulsory schooling laws, the authors offer evidence of the
causal effects of compulsory schooling on individuals' education and subsequent earnings.

Please note that this is a general summary of the article based on the information provided. For a
more detailed understanding, it is recommended to refer directly to the original article by
Angrist and Krueger (1991).

Econometric methods:

In their article "Does Compulsory School Attendance Affect Schooling and Earnings?" (Angrist
and Krueger, 1991), the authors employ various econometric methods and theoretical
frameworks to estimate the causal effects of compulsory schooling on educational attainment
and earnings.

1. Instrumental Variable (IV) Analysis: Angrist and Krueger use an instrumental variable strategy
to address endogeneity and identify the causal effect of compulsory schooling. They utilize
exogenous variation in compulsory schooling induced by changes in state-level compulsory
schooling laws as an instrument. By using this instrument, they aim to isolate the effect of
compulsory schooling on educational attainment and subsequent earnings.
2. Difference-in-Differences (DD) Approach: The authors employ a difference-in-differences
design to analyze the impact of changes in compulsory schooling laws. They compare
individuals affected by different compulsory schooling laws and cohorts to identify the
differential effects of these policies on educational outcomes and earnings.
3. Data Sources: Angrist and Krueger utilize data from the National Longitudinal Surveys (NLS)
and the Current Population Surveys (CPS) to examine the relationship between compulsory
schooling, education, and earnings. These datasets provide information on individuals'
educational attainment, labor market outcomes, and other relevant variables.
4. Regression Analysis: The authors use regression analysis to estimate the effects of compulsory
schooling on educational attainment and earnings. They specify econometric models that
control for individual and family characteristics, as well as other potential confounding factors
that may influence educational outcomes and earnings.
5. Economic Theory: Angrist and Krueger discuss economic theories that help interpret the
estimated effects of compulsory schooling on education and earnings. They draw upon concepts
 such as human capital theory, signaling theory, and the role of education in increasing
productivity and enhancing individuals' skills and abilities.
6. Sensitivity Analysis: The authors conduct sensitivity analyses to examine the robustness of their
results. They explore alternative specifications and control for additional factors to ensure that
the estimated effects of compulsory schooling remain consistent and reliable.

Overall, Angrist and Krueger employ instrumental variable analysis, difference-in-differences

design, regression analysis, economic theory, and sensitivity analysis to estimate the causal
effects of compulsory schooling on educational attainment and subsequent earnings. These
methods and theoretical frameworks help address endogeneity concerns, control for
confounding factors, and provide a rigorous framework for interpreting the results.

Pres. Topic 5: Regression Discontinuity.

Articles:

- How to financial incentives induce Disability Insurance recipients to return to work. By

Andreas Ravdal Kostøl and Magne Mogstad (2014).
- A flying start? Maternity leave benefits and long-run outcomes of children. By Pedro Carnerio,
Katrine V. Løken and Kjell G. Salvanes (2015).

Summary article 1, How to financial incentives induce Disability Insurance recipients to return to
work.

The article titled "How Do Financial Incentives Induce Disability Insurance Recipients to Return
to Work?" by Andreas Ravdal Kostøl and Magne Mogstad (2014) explores the effectiveness of
financial incentives in encouraging individuals who receive disability insurance to reenter the
workforce.
The authors investigate whether changes in the structure of financial incentives provided to
disability insurance recipients can influence their decisions to return to work. They aim to
understand how variations in these incentives affect labor market outcomes and the likelihood
of individuals transitioning from disability insurance to employment.
Using administrative data from Norway, Kostøl and Mogstad analyze the impact of a policy reform
that introduced more generous financial incentives for disability insurance recipients who
engage in work-related activities or find employment. They compare the outcomes of
individuals before and after the reform and compare them to a control group to estimate the
causal effects.
The study employs a difference-in-differences approach to examine the changes in employment
rates and labor market outcomes following the introduction of the policy reform. By
comparing the outcomes of disability insurance recipients affected by the reform with those of
unaffected individuals, the authors can identify the impact of the financial incentives on
individuals' work behavior.
The findings of the study indicate that financial incentives can indeed induce disability insurance
recipients to return to work. The authors find a significant increase in the employment rates
among individuals who were eligible for the more generous incentives. The reform led to a
higher likelihood of transitioning from disability insurance to employment.
Moreover, Kostøl and Mogstad find that the financial incentives were particularly effective for
individuals with less severe disabilities. These individuals were more responsive to the
increased incentives and showed higher rates of employment after the reform.
The article provides important insights into the role of financial incentives in promoting labor
 market reintegration for disability insurance recipients. It highlights the potential effectiveness
of providing stronger financial incentives as a policy tool to encourage individuals to return to
work, especially for those with less severe disabilities.
Please note that this is a general summary of the article based on the information provided. For a
more detailed understanding, it is recommended to refer directly to the original article by
Kostøl and Mogstad (2014)

Econometric methods:

In their article "How Do Financial Incentives Induce Disability Insurance Recipients to Return to
Work?" (Kostøl and Mogstad, 2014), the authors employ various econometric methods and
theoretical frameworks to estimate the causal effects of financial incentives on the labor
market behavior of disability insurance recipients.

1. Difference-in-Differences (DD) Analysis: Kostøl and Mogstad use a difference-in-differences

approach to estimate the causal effects of the policy reform. They compare the outcomes of
disability insurance recipients before and after the introduction of the more generous financial
incentives, and they compare them to a control group of individuals unaffected by the reform.
By analyzing the differences in outcomes between these groups, the authors can identify the
impact of the policy change.
2. Regression Analysis: The authors utilize regression analysis to estimate the effects of the
financial incentives on employment outcomes. They specify econometric models that control
for individual characteristics, time trends, and other potential confounding factors. This helps
isolate the causal effects of the incentives on labor market behavior.
3. Administrative Data: Kostøl and Mogstad use administrative data from Norway, which provide
detailed information on disability insurance recipients, employment status, and other relevant
variables. The use of administrative data allows for a comprehensive analysis of the impact of
financial incentives on labor market outcomes, as it provides a large and representative sample
of individuals.
4. Propensity Score Matching: To address potential selection biases, the authors employ
propensity score matching techniques. They use observable characteristics of disability
insurance recipients to create matched comparison groups. This helps ensure that the groups
being compared are similar in terms of relevant characteristics, improving the validity of the
estimated effects.
5. Heterogeneity Analysis: The authors conduct heterogeneity analyses to explore the differential
effects of the financial incentives based on the severity of disabilities. By examining subgroups
with varying levels of disability severity, they can assess how the effects of the incentives vary
across different segments of the population.
6. Economic Theory: Kostøl and Mogstad discuss economic theories related to labor supply and
behavioral responses to financial incentives. They draw upon concepts such as labor market
participation, moral hazard, and the income-leisure trade-off to interpret their findings in the
context of economic theory.

By employing these econometric methods and theoretical frameworks, the authors are able to
estimate the causal effects of financial incentives on the labor market behavior of disability
insurance recipients. The use of difference-in-differences analysis, regression models,
propensity score matching, and heterogeneity analysis enhances the validity and robustness of
their findings, allowing for a comprehensive understanding of the impact of financial
incentives on employment outcomes.
Please note that this summary provides a general overview of the econometric methods and
theoretical frameworks used in the article based on the information provided. For a more
 detailed understanding, it is recommended to refer directly to the original article by Kostøl and
Mogstad (2014).

Summary article 2, A flying start? Maternity leave benefits and long-run outcomes of children.

The article titled "A Flying Start? Maternity Leave Benefits and Long-Run Outcomes of Children"
by Pedro Carneiro, Katrine V. Løken, and Kjell G. Salvanes (2015) examines the impact of
maternity leave benefits on the long-term outcomes of children. The authors investigate
whether longer and more generous maternity leave policies have positive effects on children's
educational attainment, earnings, and other outcomes later in life.
To conduct their analysis, Carneiro, Løken, and Salvanes utilize a natural experiment arising from
changes in maternity leave policies in Norway. They exploit variations in the duration and
generosity of maternity leave benefits across birth cohorts to estimate the causal effects on
children's outcomes.
The study employs a difference-in-differences approach, comparing outcomes of children born
before and after the policy changes, and it compares the outcomes of children whose mothers
were eligible for different lengths of maternity leave. By examining these variations, the
authors aim to identify the causal effects of maternity leave benefits on long-term child
outcomes.
The authors gather data from various sources, including administrative registers, school records,
and surveys, to obtain comprehensive information on educational attainment, earnings, and
other relevant variables for the children in the study.
The findings of the study suggest that longer and more generous maternity leave benefits have
positive effects on children's long-term outcomes. The authors find that children born to
mothers who had access to longer maternity leave benefits tend to have higher educational
attainment, including higher high school completion rates and higher college enrollment rates.
They also observe positive effects on earnings in adulthood, indicating that these children have
higher earning potential later in life.
The article highlights the potential benefits of longer and more generous maternity leave policies in
promoting positive long-term outcomes for children. It suggests that providing mothers with
adequate time and financial support during the early stages of a child's life can have a
significant impact on their educational and economic trajectories.
It is important to note that this is a general summary of the article based on the information
provided. For a more detailed understanding, it is recommended to refer directly to the original
article by Carneiro, Løken, and Salvanes (2015).

Econometric methods:

In their article "A Flying Start? Maternity Leave Benefits and Long-Run Outcomes of Children"
(Carneiro, Løken, and Salvanes, 2015), the authors employ various econometric methods and
theoretical frameworks to estimate the causal effects of maternity leave benefits on the long-
term outcomes of children.
1. Difference-in-Differences (DD) Analysis: Carneiro, Løken, and Salvanes use a difference-in-
differences approach to estimate the causal effects of maternity leave benefits. They compare
the outcomes of children born before and after changes in maternity leave policies, as well as
the outcomes of children whose mothers were eligible for different lengths of maternity leave.
By comparing these groups, they aim to identify the effects of the policy changes on children's
 long-term outcomes.
2. Regression Analysis: The authors utilize regression analysis to estimate the effects of maternity
leave benefits on educational attainment and earnings outcomes. They specify econometric
models that control for individual and family characteristics, as well as other potential
confounding factors. This helps isolate the causal effects of maternity leave on children's
outcomes.
3. Instrumental Variable (IV) Analysis: Carneiro, Løken, and Salvanes employ instrumental
variable techniques to address potential endogeneity issues and strengthen the causal
interpretation of their results. They utilize exogenous variation in maternity leave benefits
induced by changes in policy as instruments to identify the causal effects on children's
outcomes.
4. Longitudinal Data Analysis: The authors utilize longitudinal data from various sources,
including administrative registers, school records, and surveys. This allows them to track the
educational and earnings outcomes of children over time and examine the long-term effects of
maternity leave benefits.
5. Economic Theory: Carneiro, Løken, and Salvanes discuss economic theories related to human
capital formation and early childhood development to interpret their findings. They draw upon
concepts such as investment in human capital, the role of parental time and resources in child
development, and the potential long-term effects of early-life interventions.
6. Robustness Checks: The authors conduct various robustness checks to ensure the validity of
their results. They explore alternative specifications, control for additional factors, and examine
different subsamples to assess the robustness of the estimated effects of maternity leave
benefits.
By employing these econometric methods and theoretical frameworks, the authors are able to
estimate the causal effects of maternity leave benefits on the long-term outcomes of children.
The use of difference-in-differences analysis, regression models, instrumental variable
techniques, longitudinal data analysis, economic theory, and robustness checks enhances the
validity and reliability of their findings, providing valuable insights into the impact of
maternity leave policies on children's educational attainment and earnings outcomes.
Please note that this summary provides a general overview of the econometric methods and
theoretical frameworks used in the article based on the information provided. For a more
detailed understanding, it is recommended to refer directly to the original article by Carneiro,
Løken, and Salvanes (2015).

Pres. Topic 6: Classic difference in differences studies.

Articles:

- Card and Krueger (1994): Minimum wages and employment: A case study of
the fast-food industry in New Jersey and Pennsylvania.
- Jonathan Gruber (2000): Disability insurance benefits effect on labor supply.

Summary article 1, Minimum wages and employment:

The article titled "Minimum Wages and Employment: A Case Study of the Fast-Food Industry in
New Jersey and Pennsylvania" by David Card and Alan B. Krueger (1994) investigates the
effects of minimum wage changes on employment within the fast-food industry in New Jersey
and Pennsylvania. The authors conduct a case study to examine the impact of a minimum wage
increase in New Jersey and compare it to the neighboring state of Pennsylvania, which did not
 experience a similar increase during the same period.
Using survey data from fast-food restaurants in both states, Card and Krueger analyze employment
levels and other relevant variables before and after the minimum wage increase in New Jersey.
Their objective is to assess whether the wage hike led to significant changes in employment
levels and if those changes were different compared to Pennsylvania.
The study employs a quasi-experimental approach, utilizing the neighboring state as a control
group to estimate the causal effect of the minimum wage increase. By comparing the trends in
employment between the two states, the authors aim to determine whether the policy change
had a discernible impact on employment outcomes.
The findings of the study challenge the conventional belief that minimum wage increases lead to
significant reductions in employment. Contrary to the expected outcome, Card and Krueger
find that employment levels in the fast-food industry in New Jersey actually increased
following the minimum wage hike, compared to Pennsylvania. The authors attribute this
counterintuitive result to potential labor market adjustments and factors specific to the fast-
food industry, such as turnover rates and the competitive nature of the market.
The article also explores alternative explanations and conducts robustness checks to support the
validity of the findings. Card and Krueger argue that the observed increase in employment
could be attributed to factors such as improved worker productivity, reduced turnover costs,
and the ability of firms to adjust to higher wages through price increases or cost-saving
measures.
The study's conclusions have generated considerable discussion and debate within the field of labor
economics, challenging previous assumptions about the negative employment effects of
minimum wage hikes. It highlights the importance of empirical analysis and case studies in
understanding the nuanced impact of minimum wage policies on employment outcomes.
Please note that this is a general summary of the article based on the information provided. For a
more detailed understanding, it is recommended to refer directly to the original article by Card
and Krueger (1994).

Econometric methods:

In their article "Minimum Wages and Employment: A Case Study of the Fast-Food Industry in
New Jersey and Pennsylvania" (Card and Krueger, 1994), the authors employ various
econometric methods and theoretical frameworks to estimate the causal effects of minimum
wage changes on employment within the fast-food industry. The study challenges conventional
economic theory and empirical findings related to the employment effects of minimum wage
hikes.
1. Difference-in-Differences (DD) Analysis: Card and Krueger use a difference-in-differences
approach to estimate the causal effects of the minimum wage increase in New Jersey. They
compare employment levels in New Jersey before and after the wage hike and compare them to
employment levels in Pennsylvania, which did not experience a similar increase. By analyzing
the differences in employment trends between the two states, the authors can identify the impact
of the minimum wage change.
2. Survey Data: The authors collect survey data from fast-food restaurants in both New Jersey and
Pennsylvania to obtain information on employment levels and other relevant variables. The use
of survey data allows for a detailed analysis of the fast-food industry and its response to the
minimum wage increase.
3. Robustness Checks: Card and Krueger conduct robustness checks to ensure the validity of their
findings. They examine alternative specifications of the empirical model and assess the
sensitivity of their results to different sample restrictions and control variables. These
robustness checks strengthen the confidence in the estimated effects of the minimum wage
increase on employment.
 4. Theory of Monopsony Power: The authors discuss the theory of monopsony power, which
suggests that firms with market power in the labor market can pay lower wages and exert
control over employment levels. They argue that the fast-food industry exhibits characteristics
of monopsonistic competition, where individual firms have the ability to influence wages and
employment. This theory informs their interpretation of the empirical results.
5. Alternative Explanations: Card and Krueger explore alternative explanations for their findings.
They consider factors such as improved worker productivity, reduced turnover costs, and firms'
ability to adjust to higher wages through price increases or cost-saving measures. These
alternative explanations help explain why the minimum wage increase did not lead to the
expected negative employment effects.
By employing these econometric methods and theoretical frameworks, Card and Krueger provide
evidence that challenges traditional economic thinking regarding the employment effects of
minimum wage increases. The use of difference-in-differences analysis, survey data,
robustness checks, the theory of monopsony power, and consideration of alternative
explanations enhances the validity and robustness of their findings, providing insights into the
specific case of the fast-food industry and its response to minimum wage changes.

Summary article 2, Disability benefits and labor supply.

The article titled "Disability Insurance Benefits and Labor Supply" by Jonathan
Gruber (2000) examines the impact of disability insurance benefits on
individuals' labor supply decisions. Gruber investigates how the availability and
generosity of disability insurance programs affect the likelihood of individuals
leaving the labor force due to disability.
Gruber's analysis focuses on the United States and utilizes a combination of
empirical data and econometric methods to estimate the causal effects of
disability insurance benefits on labor supply.
The study begins by providing an overview of the disability insurance programs in
the United States and their eligibility criteria. Gruber discusses the potential
trade-off between providing financial support to individuals with legitimate
disabilities and the potential negative incentive effects on labor force
participation.
To estimate the effects of disability insurance benefits, Gruber utilizes various
econometric techniques, including instrumental variable (IV) analysis and
regression models. He leverages exogenous variation in the generosity of
disability benefits across different states and over time to identify the causal
effects on labor supply outcomes.
Gruber's findings suggest that disability insurance benefits have a significant
negative impact on labor force participation. He finds that the availability and
generosity of disability benefits lead to an increase in the likelihood of
individuals exiting the labor force due to disability. The results also indicate that
the effects are more pronounced among individuals with marginal attachments
to the labor market.
The article discusses potential mechanisms through which disability insurance
benefits affect labor supply decisions, such as moral hazard and adverse
selection. Gruber highlights the importance of carefully designing disability
 insurance programs to balance the needs of individuals with legitimate
disabilities while minimizing the negative incentive effects on labor force
participation.
The research has implications for policymakers and provides insights into the
labor supply effects of disability insurance programs. Gruber's analysis
contributes to the broader literature on social insurance programs and their
impact on labor market outcomes.
Please note that this is a general summary of the article based on the information
provided. For a more detailed understanding, it is recommended to refer
directly to the original article by Jonathan Gruber (2000).

Econometric methods:

In his article "Disability Insurance Benefits and Labor Supply," Jonathan Gruber
(2000) explores the relationship between disability insurance benefits and
individuals' labor supply decisions. Gruber employs various econometric methods
to analyze the impact of disability insurance programs on labor force participation.
Here are some of the econometric methods used in the article:

1. Instrumental Variable (IV) Analysis: Gruber uses instrumental variables to

address potential endogeneity issues in estimating the causal impact of
disability insurance benefits on labor supply. Endogeneity arises when there is a
two-way relationship between disability insurance benefits and labor supply,
making it challenging to establish causality. By identifying instrumental variables
that are correlated with disability insurance benefits but unrelated to other
factors affecting labor supply, Gruber obtains unbiased estimates of the causal
effect.
2. Difference-in-Differences (DiD) Analysis: Gruber employs a difference-in-
differences approach to examine how changes in disability insurance benefits
affect labor supply decisions. By comparing the labor force participation of
individuals affected by changes in disability insurance policies to those not
affected, he estimates the causal effect of the policy changes on labor supply
outcomes.
3. Panel Data Analysis: Gruber utilizes panel data, which involves observing the
same individuals over multiple time periods, to study labor supply dynamics.
This allows him to examine how changes in disability insurance benefits over
time are associated with changes in labor force participation among individuals.
4. Control Variables: Gruber includes control variables in his analysis to account
for other factors that may influence labor supply decisions. These control
variables could include individual characteristics, socioeconomic factors,
 regional differences, or other policy changes that may affect labor force
participation. Controlling for these factors helps isolate the specific impact of
disability insurance benefits on labor supply.
5. Sensitivity Analysis: Gruber may conduct sensitivity analyses to examine the
robustness of his results. These analyses involve testing the sensitivity of the
findings to alternative specifications, control variables, or assumptions,
providing a more comprehensive assessment of the relationship between
disability insurance benefits and labor supply.

By employing these econometric methods, Gruber aims to estimate the causal

impact of disability insurance benefits on labor supply decisions while accounting
for potential biases and confounding factors. These methods help enhance the
validity and reliability of his findings, contributing to our understanding of the
relationship between disability insurance programs and labor force participation.

Pres. Topic 7: More on Diff and Diff

Articles:

- Money for nothing? Universal childcare and maternal employment. By Tarjei

Havnes and Magne Mogstad. (2011)
- Labor supply response to the earned income tax credit. By Nada Eissa and
Jefferey B. Leibman. (QJE 1996)

Summary article 1, Money for nothing.

The article "Money for Nothing? Universal Childcare and Maternal Employment" by Tarjei
Havnes and Magne Mogstad (2009) examines the impact of universal childcare policies on
maternal employment rates. The study investigates whether providing subsidized and easily
accessible childcare options affects the labor force participation of mothers.
Havnes and Mogstad focus on the case of Norway, which implemented a universal childcare
program in the 1970s. They utilize a combination of empirical data and econometric methods
to analyze the causal relationship between childcare availability and maternal employment.
The study begins by describing the universal childcare policy in Norway, which provides
affordable and high-quality childcare services for children aged one to five. The authors
discuss how the policy was designed to facilitate maternal employment and promote gender
equality.
To estimate the effects of universal childcare on maternal employment, Havnes and Mogstad
employ various econometric techniques, including instrumental variable (IV) analysis and
difference-in-differences (DiD) analysis. They leverage exogenous variation in childcare
availability across municipalities and over time to identify the causal impact of the policy.
The findings of the study suggest that universal childcare has a positive effect on maternal
employment rates. The authors find a significant increase in the likelihood of mothers
participating in the labor force due to the availability of subsidized childcare services. The
effects are particularly pronounced for mothers with lower education levels and those with
younger children.
Havnes and Mogstad discuss potential mechanisms through which universal childcare influences
maternal employment, such as increased flexibility in work arrangements and reduced
 childcare costs. They also consider potential spillover effects on children's outcomes and the
overall economy.
The research has important implications for policymakers and provides insights into the
relationship between childcare policies and maternal employment. The authors argue that
universal childcare can be an effective tool for promoting gender equality and labor force
participation among mothers.

Econometric methods:

In the article "Money for Nothing? Universal Childcare and Maternal Employment" by Havnes and
Mogstad (2009), the authors employ several econometric methods to examine the relationship
between universal childcare and maternal employment. Here are some of the econometric
methods and theories they use:
1. Instrumental Variable (IV) Analysis: Havnes and Mogstad address the potential endogeneity
issue where the availability of universal childcare could be influenced by unobservable factors
affecting maternal employment. To tackle this, they use instrumental variables as a way to
establish causality. They identify instruments that are correlated with childcare availability but
not directly with maternal employment. By using these instruments, they can estimate the causal
effect of childcare on employment outcomes.
2. Difference-in-Differences (DiD) Analysis: The authors employ a DiD approach to analyze the
impact of universal childcare on maternal employment. They compare changes in employment
rates over time between women with children of different ages (treatment group: children
affected by the universal childcare policy, and control group: children unaffected by the policy).
By comparing the differences in employment changes between the two groups, they can isolate
the causal effect of the childcare policy.
3. Fixed Effects Regression: Havnes and Mogstad use fixed effects regression models to control
for time-invariant unobserved characteristics at the individual or household level that could be
correlated with both the availability of universal childcare and maternal employment. By
including fixed effects, they can account for these unobserved factors and focus on the variation
in childcare availability within individuals or households over time.
4. Propensity Score Matching: In addition to IV and DiD methods, the authors employ propensity
score matching techniques to compare similar individuals who have different levels of exposure
to universal childcare. This allows them to create more comparable groups and estimate the
causal effect of childcare on employment outcomes.
The underlying theory in the article is based on the economic principle that the availability and
affordability of childcare can affect the decision of mothers to participate in the labor force.
The authors hypothesize that universal childcare, by providing accessible and subsidized
childcare services, reduces the costs and constraints associated with childcare, thereby
increasing maternal employment rates.
By applying these econometric methods and theories, Havnes and Mogstad aim to provide
empirical evidence on the causal impact of universal childcare on maternal employment
outcomes, shedding light on the effectiveness of such policies in promoting women's labor
force participation.

Summary article 2, Labor supply response to the earned income tax credit.
Can you summarize this article?: Labor supply response to the earned income tax
credit. By Nada Eissa and Jefferey B. Leibman.

The article "Labor Supply Response to the Earned Income Tax Credit" by Nada
Eissa and Jefferey B. Liebman explores the impact of the Earned Income Tax
 Credit (EITC) on labor supply. The EITC is a refundable tax credit designed to
provide financial assistance to low-income working individuals and families.
The study aims to understand how the EITC affects individuals' decisions to
participate in the labor market and their hours of work. It examines the
behavioral responses of individuals to changes in the generosity of the EITC,
using econometric methods and empirical analysis.
The authors start by providing an overview of the EITC program, its history, and its
structure. They discuss how the credit amount varies based on income level,
number of children, and marital status, and how it is intended to incentivize and
support work among low-income individuals.
To estimate the labor supply response to the EITC, Eissa and Liebman employ
various econometric methods, including difference-in-differences analysis,
instrumental variable estimation, and simulation techniques. They use a
combination of microdata from household surveys and administrative tax
records to analyze the effects of the EITC on labor force participation and
hours worked.
The findings of the study suggest that the EITC has a positive impact on labor
supply. The authors observe an increase in both labor force participation and
hours worked in response to the EITC, particularly among unmarried women
and those with children. They also explore the heterogeneity in responses
based on income level and family characteristics.
Eissa and Liebman discuss the implications of their findings for policy design and
implementation. They highlight the importance of considering the potential
labor supply effects when evaluating the EITC and similar programs aimed at
supporting low-income individuals.

In summary, the article analyzes the labor supply response to the Earned Income
Tax Credit (EITC) using empirical data and econometric methods. It provides
insights into the impact of the EITC on individuals' decisions to work and their
hours of work, offering valuable information for policymakers and researchers
interested in understanding the effects of income support programs on labor
supply.

Econometric methods:

In the article "Labor Supply Response to the Earned Income Tax Credit" by Eissa
and Liebman, the authors employ several econometric methods to analyze the
labor supply response to the Earned Income Tax Credit (EITC). Here are some
of the econometric methods they use:

1. Difference-in-Differences (DiD) Analysis: Eissa and Liebman use a DiD

approach to estimate the causal impact of the EITC on labor supply. They
compare changes in labor force participation and hours worked for individuals
eligible for the EITC (treatment group) to changes in similar individuals who are
ineligible for the EITC (control group). By comparing the differences in labor
supply changes between the two groups before and after the EITC policy
 changes, they can isolate the causal effect of the EITC.
2. Instrumental Variable (IV) Estimation: The authors employ instrumental
variables to address potential endogeneity issues that could arise in estimating
the causal effect of the EITC. They identify instruments that are correlated with
the eligibility for and receipt of the EITC but not directly related to labor supply
decisions. By using these instruments, they can obtain unbiased estimates of
the causal effect of the EITC on labor supply.
3. Simulation Analysis: Eissa and Liebman use simulation techniques to predict
and quantify the labor supply response to changes in the generosity of the
EITC. They simulate the labor supply decisions of individuals under different
EITC scenarios and compare the outcomes to assess the impact of varying
EITC parameters on labor supply behavior.

The econometric methods employed in the article aim to address potential biases
and endogeneity concerns that could arise when estimating the causal impact
of the EITC on labor supply. The DiD analysis allows for comparison between
treatment and control groups, controlling for time-varying factors. The IV
estimation helps address potential omitted variable bias and endogeneity,
providing more robust estimates. The simulation analysis provides additional
insights into the potential effects of changes in the EITC policy on labor supply.

Overall, the combination of these econometric methods enables Eissa and

Liebman to provide rigorous and reliable estimates of the labor supply
response to the EITC, enhancing our understanding of the program's impact
on individuals' work decisions.

Pres. Topic 8: Classic difference in differences studies.

Articles:

- The impact of the Mariel boatlift on the Miami labor market. Card (1990).
- Iceland’s Natural experiment in supply-side economics. Bianchi,
Gudmundsson Zoega (2001).

Summary article 1, The impact of the Mariel boatlift on the Miami labor market

The article titled "The Impact of the Mariel Boatlift on the Miami Labor Market" by David Card
(1990) examines the effect of the Mariel boatlift, a sudden influx of Cuban immigrants to
Miami in 1980, on the local labor market. The study investigates how the increased supply of
low-skilled workers from the boatlift affected employment and wages for native workers in
Miami. Here is a summary of the article:
David Card analyzes the labor market outcomes in Miami following the arrival of approximately
125,000 Cuban immigrants during the Mariel boatlift. He focuses on the impact of this
unexpected and substantial increase in the low-skilled labor force on native workers. The study
 aims to assess whether the arrival of the Mariel immigrants led to adverse employment and
wage effects for native workers.
Card's analysis involves comparing the employment and wage trends of low-skilled workers in
Miami with those in other similar cities that did not experience the influx of immigrants. He
uses data from the Current Population Survey (CPS) and applies various econometric
techniques to estimate the causal impact of the Mariel boatlift.
One of the primary econometric methods employed by Card is a difference-in-differences
approach. He compares the labor market outcomes of Miami with a control group of cities,
both before and after the boatlift, to identify any differential changes that can be attributed to
the immigrant influx. This method helps isolate the effect of the boatlift on native workers'
employment and wages.
Card's findings challenge the conventional belief that the Mariel boatlift had a significant negative
impact on the Miami labor market. Contrary to expectations, he finds no significant adverse
effects on native employment or wages. The study suggests that the increased supply of low-
skilled workers did not lead to substantial displacement or wage reductions for native workers
in Miami.
The article concludes by discussing potential explanations for the unexpected findings, including
the flexibility of the Miami labor market, the specific characteristics of the Mariel immigrants,
and the ability of the local economy to absorb the influx of workers.
In summary, Card's article on the impact of the Mariel boatlift on the Miami labor market
challenges conventional wisdom by providing evidence that the influx of Cuban immigrants
did not have detrimental effects on native employment and wages. The study utilizes a
difference-in-differences approach and compares labor market outcomes between Miami and
other cities to estimate the causal impact of the boatlift.

Econometric methods:

Certainly! In his article, "The Impact of the Mariel Boatlift on the Miami Labor Market," David
Card (1990) utilizes several econometric methods to estimate the effects of the Mariel boatlift
on the local labor market. Here are some of the key econometric methods employed in the
study:

1. Difference-in-Differences (DiD) Analysis: Card utilizes a DiD approach to compare the labor
market outcomes in Miami (the treatment group) with those in a control group of similar cities
unaffected by the Mariel boatlift. By examining the differential changes in employment and
wages before and after the boatlift, Card estimates the causal effect of the immigrant influx on
native workers. This method helps control for time-invariant differences between Miami and the
control group and isolates the specific impact of the boatlift.
2. Synthetic Control Group: As part of the DiD analysis, Card constructs a synthetic control group
that closely matches the pre-boatlift characteristics of Miami. He combines data from multiple
cities to create a weighted average that serves as a counterfactual for Miami's labor market
outcomes in the absence of the boatlift. This allows him to compare the actual changes observed
in Miami with the changes that would have been expected in the absence of the boatlift.
3. Difference-in-Differences-in-Differences (DDD) Analysis: Card extends the DiD framework by
introducing a third dimension to the analysis. He compares the changes in labor market
outcomes in Miami and the control group before and after the boatlift, but he also considers
differences between high- and low-skilled workers within these areas. This additional
dimension helps account for potential heterogeneity in the effects of the boatlift across different
skill levels.
4. Control Variables: Card includes various control variables in his econometric models to account
for other factors that may influence labor market outcomes. These control variables can include
 demographic characteristics, industry composition, local economic conditions, and time trends.
Controlling for these factors helps isolate the specific impact of the Mariel boatlift on native
workers' employment and wages.
5. Robustness Checks: Card conducts several robustness checks to ensure the validity of his
findings. These checks involve testing the sensitivity of the results to different model
specifications, control variables, and assumptions. By performing various sensitivity analyses,
Card assesses the robustness of the estimated effects and examines the potential influence of
alternative approaches on the overall conclusions.

These econometric methods employed by Card help address the challenges of causal inference and
account for potential confounding factors in estimating the impact of the Mariel boatlift on the
Miami labor market. The DiD and DDD analyses, along with the use of control variables and
robustness checks, provide a rigorous framework for identifying the causal effects of the
immigrant influx on native workers' employment and wages.

Summary article 2, Iceland’s Natural experiment in supply-side economics.

Bianchi, Gudmundsson Zoega (2001).

Can you summarise this article?: Iceland’s Natural experiment in supply-side

economics. Bianchi, Gudmundsson Zoega (2001).
ChatGPT
The article titled "Iceland's Natural Experiment in Supply-Side Economics" by Marcello
Bianchi, Gylfi Zoega, and Gudmundur S. Gudmundsson (2001) examines the
economic policies and their impact on the Icelandic economy during the 1980s and
early 1990s. The authors refer to this period as a "natural experiment" in supply-side
economics due to the substantial changes implemented in tax and labor market
policies. Here is a summary of the article:
The study focuses on Iceland's experience with supply-side policies, which aimed to
stimulate economic growth and enhance labor market efficiency. The authors
investigate the effects of tax reforms, changes in social security contributions, and
labor market deregulation on various economic indicators.
The article provides an overview of the policy changes implemented during the period
under study, including reductions in income tax rates, shifts from progressive to
proportional taxation, and the introduction of a value-added tax (VAT). Additionally,
changes in social security contributions and labor market policies, such as the
deregulation of wage agreements, are analyzed.
Bianchi, Zoega, and Gudmundsson employ various econometric methods to assess
the impact of these policies on key economic variables. They analyze time series data
and utilize econometric models to estimate the causal effects of the supply-side
policies on outcomes such as economic growth, employment, wages, and investment.
The authors find that the supply-side policies had mixed effects on the Icelandic
economy. While some indicators, such as investment and productivity, improved as a
result of the reforms, other outcomes, such as employment and wages, showed less
significant improvements. The authors attribute the limited impact on employment and
wages to factors such as labor market rigidities and the slow adjustment of real wages.
The article highlights the importance of considering both the short-term and long-term
effects of supply-side policies. While immediate results may be modest, the authors
suggest that the long-term benefits of such policies may become more evident over
time as the economy adjusts to the changes.
In summary, "Iceland's Natural Experiment in Supply-Side Economics" examines the
impact of tax reforms, social security changes, and labor market deregulation on the
Icelandic economy. The authors employ econometric methods to analyze the effects of
these policies on various economic indicators. The study concludes that while the
supply-side policies had mixed effects, they offer insights into the complexities of
implementing and assessing supply-side measures in an economy.

Econometric methods:

Chat GPT didnt want to tell me -_- so se pp.