Statistical Treatment of Analytical Data Lecture

Statistical Treatment of
Analytical Data
ANALYSIS OF ERRORS
 Analytical chemistry is based on reliability,
reproducibility & accuracy of results.
 However, every measurement has some degree of

uncertainty which is called as error in analytical
chemistry.
 The error is nothing but the difference between the

experimental mean value & a ‘true value’.
2
ANALYSIS OF ERRORS
 It is impossible to perform a chemical analysis
that is error free or without uncertainty.
 no analysis is free of error or “uncertainty”!!
 Our goals are to minimize errors &..

• …….. to calculate the size of the errors.
3
ANALYSIS OF ERRORS
 ‘We can only hope to minimize errors &

estimate their size with acceptable
accuracy’ !!!
4
ANALYSIS OF ERRORS
 Precision & Accuracy
 are commonly associated with any discussion of error.
 Precision = reproducibility of a measurement

 Accuracy = measure of the closeness to true
value.
5
Accuracy and Precision
Illustration of accuracy and precision using the pattern of darts on a dartboard

ANALYSIS OF ERRORS
 Three general types of errors occur in lab
measurements:
 random error,
 systematic error, &
 gross errors.
7
ANALYSIS OF ERRORS
8
The Nature of Random Errors
• All measurements contain random
errors.!
– Random, or indeterminate, errors occur
whenever a measurement is made.
– Caused by many small but uncontrollable
variables.
– The errors are accumulative.
Random errors
• are produced by any one of a # of unpredictable &
unknown variations in the experiment.
• These changes may occur in the measuring

instruments or in the environmental conditions.
– eg fluctuations in room temperature, fluctuations in line

voltage, mechanical vibrations, etc
Systematic Errors
• Systematic errors or determinant error have a
definite value, an assignable cause,..
– & are of about the same magnitude for replicate
measurements made in the same way.
Systematic Errors
• are instrumental, methodological, or personal
mistakes.
• These cause"lopsided" data….

– which is consistently deviated in one direction from
the true value.
Examples of systematic errors:
– an instrumental error results when a

spectrometer drifts away from calibrated
settings;
– a methodological error is created by using the
wrong indicator for an acid-base titration;
– a personal error occurs when an experimenter
records only even numbers for the last digit of
burette volumes.
13
Instrumental Errors
• are caused by non ideal instrument behavior, by

faulty calibrations, or by use under inappropriate
conditions
• Pipets, burets, and volumetric flasks may hold or
deliver volumes slightly different from those
indicated by their graduations.
• Calibration eliminates most systematic errors of this
type.
Instrumental Errors
• Electronic instruments can be influenced by

– noise,
– temperature,
– pH and are also subject to systematic errors.
• Errors of these types usually are detectable

and correctable.
Method Errors
• The non ideal chemical or physical behavior of the

reagents and reactions on which an analysis is based
often introduce systematic method errors.
• Such sources of nonideality include
– the slowness of some reactions,
– the incompleteness of others,
– the instability of some species,
– the lack of specificity of most reagents,
– and the possible occurrence of side reactions that
interfere with the measurement process.
Personal Errors
• Result from the carelessness, inattention, or

personal limitations of the experimenter.
• Many measurements require personal
judgments.
• eg
– estimating the position of a pointer between two
scale divisions,
– The color of a solution at the end point in a
titration,
– the level of a liquid with respect to a graduation
in a pipet or buret.
Personal Errors
• Judgments of this type are often subject to

systematic, unidirectional errors.
A universal source of personal error is

prejudice, or bias.
Number bias is another source of personal error

that varies considerably from person to person.
Personal Errors
• The most frequent number bias encountered

in estimating the position of a needle on a
scale involves a preference for the digits 0 and
5.
• Also common is a prejudice favoring small
digits over large and even numbers over
odd.
Systematic errors
• Systematic errors can be identified &
eliminated by
– careful inspection of the experimental methods,
– cross-calibration of instruments,
– & examination of techniques.

Gross errors
 are caused by experimenter carelessness or equipment failure.
 These "outliers" are so far above or below the true value that they
are usually discarded when assessing data.
 The "Q-Test" (discussed later) is a systematic way to determine if a

data point should be discarded.
22
Gross errors
 These differ from indeterminate & determinate errors.
 They usually occur only occasionally, are often
large, and may cause a result to be either high or
low.
 They are often the product of human errors.
 Can lead to outliers, results that appear to differ

markedly from all other data in a set of replicate
measurements.
summary
 Errors are caused by
 faulty calibrations or
 standardizations or
 by random variations and uncertainties in results.
 Frequent calibrations, standardizations, and analyses

of known samples can sometimes be used to lessen
all but the random errors
24
Errors & Statistical Data in Chemical Analyses
 Impossible: The analytical results are free of

errors or uncertainties.
 Possible: Minimize these errors & estimate

their size with acceptable accuracy.
 Statistical calculationsused to judge the

quality of experimental measurements are
many.
Bias, Precision & Accuracy
 We consider both bias & precision with respect to
how well an estimator performs replicate
measurements.
 The average of these replicate measurements is

called the expected value of the estimator.
 Bias?? -measure of how far the expected value of

the estimate is from the true value of the parameter
being estimated.
26
Bias, Precision & Accuracy
 Precision is a measure of how similar the multiple estimates are
to each other, not how close they are to the true value
(which is bias).
 Precision & bias are 2 different components of Accuracy.
27
Precision of a Set of Measurements
 used in describing the agreement of a set of
replicate results among themselves.
 usually expressed in terms of the deviation of a set of

results from the arithmetic mean of the set (mean &
standard deviation to be discussed later).
Note:– good precision does not always mean

good accuracy.!!
28
 Why doesn't good precision mean we
have good accuracy???
We have systematic & random errors.
 total error is the sum of the systematic error

& random error.
29
 random error likely be -ve or +ve
 a measurement that has only random error is
accurate to within the precision of measurement.
 & the more precise the measurement,..the better

idea we have of the true value, provided there is no
bias in the data.
 In the case of random errors only…good precision

indicates good accuracy.
30
 Now lets add the possibility of systematic error.
 systematic error will produce a bias in the data from
the true value.
 This bias will be -ve or +ve depending upon the type

& there may be several systematic errors at work.
 Many systematic errors can be repeated to a

high degree of precision.
31
 it follows that systematic errors prevent us from
making the conclusion that good precision means
good accuracy.
 When we go about the task of determining the

accuracy of a method, ..
• we are focusing upon the identification &
elimination of systematic errors.
32
 Don't be misled by the statement that 'good precision
is an indication of good accuracy.'
 Too many systematic errors can be repeated to a

high degree of precision for this statement to be true.
33
 terms 'repeatability' & 'reproducibility' used instead
of the more general term 'precision.'
 Repeatability (of results of measurements) –

the closeness of the agreement between the
results of successive measurements carried out
under the same conditions of measurement.
34

1. These conditions are called repeatability
conditions.
2. Repeatability conditions include
 the same measurement procedure,
 the same observer,
 the same measuring instrument, used under the
same conditions, the same location, & repetition
over a short period of time.
35
 Reproducibility (of results of measurement) –
 the closeness of the agreement between the
results of carried out under changed
conditions of measurement.
36
1. A valid statement of reproducibility requires
specification of the conditions changed.
2. The changed conditions may include
 principle of measurement,
 method of measurement,
 observer,
 measuring instrument,
 reference standard,
 location,
 conditions of use, and time.
37
 When discussing the precision of measurement data,
 it is helpful for the analyst to define how the data are
collected & to use the term 'repeatability' when
applicable.
 It is equally important to specify the conditions

used for the collection of 'reproducibility' data.
38
Characterizing Measurements & Results
 What is the mass of an average coin in

circulation?
 This is a problem that we might expect to be

able to answer experimentally.!
 A good way to begin the analysis is to acquire

some preliminary data.
39
40
 Looking at these data, it is immediately apparent that our

question has no simple answer.
 ie, we cannot use the mass of a single coin to draw a

specific conclusion about the mass of any other
coin..
 … (although we might conclude that all coins weigh at
least 3 g).
41
 We can characterize these data by providing a

measure of the spread of the individual
measurements around a central value or central
tendency
Measures of Central Tendency

 One way to characterize the data is to assume that
the masses of individual coins are scattered around a
central value
 this provides the best estimate of a coin’s true mass.
42
Measures of Central Tendency
 2 common ways to report this estimate of

central tendency are the mean & the
median.
43
 A data set of repetitive measurements is often
expressed as a single representative # called the
mean (average) or median.
 The mean (x̄ ) is the sum of individual measurements

(xi) divided by the # of measurements (N).
44
Example
 What is the mean for the data in the previous
Table given?
45
 The mean is the most common estimator of central
tendency.
 It is not considered a robust estimator, because extreme
measurements,..
 ie.. those much larger or smaller than the remainder of the
data, strongly influence the mean’s value.
 eg, mistakenly recording the mass of the fourth coin as

31.07 g instead of 3.107 g, changes the mean from
3.117 g to 7.112 g!
46
Median Xmed
 is the middle value when data are ordered from
the smallest to the largest value.
 is the value separating the higher half of a data

sample, a population, or a probability distribution
, from the lower half!
47
 When the data include an odd # of measurements, the median

is the middle value.
 For an even number of measurements,..
 … the median is the average of the n/2 & the (n/2) + 1
measurements, where n is the # of measurements.
48
 As shown in the examples, mean & median provide
similar estimates of central tendency when all data are
similar in magnitude.
 The median, however, provides a more robust estimate

of central tendency since it is less sensitive to
measurements with extreme values.
 It is not skewed so much by extremely large or small

values, and so it may give a better idea of a "typical"
value.
49
Some important terms
 To improve the reliability & to obtain infor about the

variability of results, ..
 2 to 5 portions (replicates) of a sample are usually

carried through an entire analytical procedure.
 Replicates are samples of about the same size that are

carried through an analysis in exactly the same way. !!
 Individual results from a set of measurements
are seldom the same
 Usually, the “best” estimate is considered to

be the central value for the set.
 The central value of a set should be more

reliable than any of the individual results.
 Usually, the mean or the median is used as
the central value for a set of replicate
measurements.
 An analysis of the variation in the data allows

us to estimate the uncertainty associated with
the central value.
median & mean
 Measures of average such as the median &
mean represent the typical value for a
dataset.
 Within the dataset the actual values usually
differ from one another & from the average
value itself.
53
median & mean
 The extent to which the median & mean are good
representatives of the values in the original dataset
depends upon the variability or dispersion in the
original data.
 Datasets are said to have high

dispersion/spread/variability when they contain
values considerably higher & lower than the mean
value.
54
Measures of Spread
 AKA measure of dispersion
 used to describe the variability in a sample or
population.
 It is usually used in conjunction with a

measure of central tendency, such as the mean or
median, to provide an overall description of a set of
data.
55
Measures of Spread
 If the mean or median provides an estimate of a
coins’ true mass,..
 .. then the spread of the individual measurements

must provide an estimate of the variability in the
masses of individual coins.
 describe how similar or varied the set of

observed values are for a particular variable (
data item)
56
Why is it important to measure the spread of
data?
 A measure of spread gives us an idea of how well the x̄

, for example, represents the data.
 If the spread of values in the data set is large, the x̄ is not

as representative of the data as if the spread of data is
small.
 This is because a large spread indicates that there are

probably large differences between individual scores.
57
Measures of Spread
 Additionally, in research, it is often seen as positive if
there is little variation in each data group.
 3 common measures of spread are range, standard

deviation,& variance.
58
Measures of Spread
Range, w
 difference between the largest & smallest values in the
data set.
 The range provides infor about the total variability in the

data set,
 .. but does not provide any infor about the distribution of
individual measurements.
59
Standard deviation
 Effectively it indicates how tightly the values in the
dataset are bunched around the mean value.
 It is the most robust & widely used measure of

dispersion.
− ie it is the most widely used measure of spread
associated with the mean.
60
 Precision (reproducibility) is quantified by
calculating;
 the average deviation (for data sets with 4 or
fewer repetitive measurements)
 or the standard deviation (for data sets with

5 or more measurements).
61
62
 Widely scattered data results in a large average or
standard deviation indicating poor precision.
 calculations contain the deviation from the mean
 the difference between the individual experimental

value and the mean value.
63
 The average deviation, Δ x̄ , is used when a data set contains
less than 5 repetitive measurements.
 A small average deviation indicates data points clustered
closely around the mean & good precision.
64
65
66
67
Measures of Spread
 Variance
 Another common measure of spread is the square of the
s, or the variance.
 It is the average squared deviation from the mean.
 It’s usefulness is limited because the units are squared &

not the same as the original data.
68
Variance
Variance = s2
 Or the formula:
69
Measures of Spread
 The standard deviation, rather than the

variance, is usually reported because the units
for standard deviation are the same as that for
the mean value.
70
71
How About Accuracy?
• Accuracy indicates the closeness of the measurement to
its true or accepted value & is expressed by the error.
– Accuracy measures agreement between a result & its true

value.
– Precision describes the agreement among several results that

have been obtained in the same way.
Figure
Illustration of accuracy and precision using the pattern of darts on a dartboard

Accuracy?
• Absolute Error
– The absolute error E in the measurement of a
quantity xi is given by the equation
where xt is the true, or accepted, value of the

quantity.
– Note that we retain the sign in stating the error.
Accuracy?
• Relative error
– Often, the relative error Er is a more useful quantity
than the absolute error.
– The % relative error is given by the expression

Accuracy?
• Relative error
– Relative error is also expressed in parts per
thousand (ppt).
e.g.
Characterizing Experimental Errors
• Errors associated with central tendency

reflect the accuracy of the analysis, ..
• ..but the precision of the analysis is

determined by those errors associated with
the spread…
– ..and s is the measure of spread!
77
The Confidence Limit—How Sure Are You?
• standard deviation provides an indication of the

precision inherent in a particular procedure or
analysis.
• But unless there is a large amount of data, ..
– it does not by itself give any infor about how close

the experimentally determined mean might be to the true
mean value μ !!
• Statistical theory allows us to estimate the range

within which the true value might fall, ..
– ..determined within a given probability, defined

by the experimental mean & the standard
deviation !
• This range is called the confidence interval,

– & the limits of this range are called the confidence limit.
• The likelihood that the true value falls within the

range is called the probability, or confidence level,
usually expressed as a percent.
– The confidence limit, in terms of the standard deviation, s is

given by:
– where t is a statistical factor that depends on the number of degrees of

freedom, ν, & the confidence level desired.
– The number of degrees of freedom is one less than the number of

measurements
v = N - 1 = degrees of freedom.
• Values of t at different confidence levels & degrees of freedom ν are given
below
The true value falls within the confidence limit, estimated using t at the desired confidence
level.
Example
• A soda ash sample is analyzed in the analytical chemistry

laboratory by titration with standard hydrochloric acid.
• The analysis is performed in triplicate with the following
results: 93.50, 93.58, and 93.43 wt% Na2CO3. Within what
range are you 95% confident that the true value lies?
So you are 95% confident that, in the absence of a determinate

error, the true value falls within 93.31 to 93.69 wt%.
Tests of Significance— Is There a Difference?
 In developing a new analytical method,..

– it is often desirable to compare the results of that
method with those of an accepted (perhaps standard)
method.

– How, can one tell if there is a significant difference
between the new method & the accepted one?
– Again, we resort to statistics for the answer.

Tests of Significance— Is There a Difference?
– Deciding whether one set of results is significantly different from
another depends not only on the difference in the means..
 ..but also on the amount of data available & the spread .
– There are statistical tables available that show how large a difference
needs to be in order to be considered not to have occurred by
chance.!!
– The F-test evaluates differences between the spread of

results
– while the t-test looks at differences between means.
Tests of Significance— THE F-TEST
• The F-test is used to determine if two variances are statistically
different.
• It’s designed to indicate whether there is a significant difference
between two methods based on their standard deviations.
– F is defined in terms of the variances of the two methods, where the variance is
the square of the standard deviation:
– where s21 > s22.
– The variances are arranged so that the F value is > 1

There are two different degrees of freedom, v1 & v2, where degrees of freedom is
defined as N - 1 for each case
• Conclusion about the results
If the calculated F value exceeds a tabulated F value at

the selected confidence level, ….
 ..then there is a significant difference between the
variances of the two methods.
• You are developing a new colorimetric procedure for

determining the glucose content of blood serum. You have
chosen the standard Folin-Wu procedure with which to
compare your results. From the following two sets of replicate
analyses on the same sample, determine whether the
variance of your method differs significantly from that of the
standard method.
The tabulated F value for v1 = 6 and v2 = 5 is 4.95.
Since the calculated value of 1.73 is less than this, we conclude that there is
no significant difference in the precision of the two methods
• Conclusion is that…
The standard deviations are from random error alone & do not
depend on the sample.
In short, statistically, your method does as well as the

established procedure. !!
Rejection of a Result: The Q Test
When a series of replicate analyses is performed, it is not uncommon that

one of the results will appear to differ markedly from the others.
 A decision will have to be made whether to reject the result or to retain it.
Unfortunately, there are no uniform criteria that can be used to decide if a

suspect result can be ascribed to accidental error rather than chance
variation.
 It is tempting to delete extreme values from a data

set because they alter the calculated statistics in an
unfavorable way,
 that is, increase the standard deviation & variance

(measures of spread)
 and they may substantially alter the reported mean.

• The only reliable basis for rejection occurs when it can
be decided that some specific error may have been
made in obtaining the doubtful result.
• No result should be retained in cases where a known

error has occurred in its collection.
• The Q test is used to determine if an “outlier” is due to
a determinate error.
• If it is not, then it falls within the expected random

error & should be retained.
– It is one of the most statistically correct for a fairly small

number of observations & is recommended when a test is
necessary.
• The ratio Q is calculated by arranging the data in
increasing or decreasing order of numbers.
– If you have a large data set, use excel!
The difference between the suspect number & its nearest

neighbor (a) is divided by the range (w), compare Qcalc to Qtable
• If Q exceeds the tabulated value for a given number
of observations and a given confidence level, the
questionable measurement may be rejected with, for
example, 95% confidence that some definite error is
in this measurement.
• The following set of chloride determinations on
separate aliquots of a pooled serum were reported:
103, 106, 107, and 114 meq/L. One value appears
suspect. Determine if it can be ascribed to accidental
error, at the 95% confidence level
• For a small number of measurements (e.g., 3 - 5),
– the discrepancy of the measurement must be quite large before it can be
rejected by this criterion,
– and it is likely that erroneous results may be retained.
• This would cause a significant change in the arithmetic mean because

the mean is greatly influenced by a discordant value!
• Consider reporting the median when an outlier cannot quite be

rejected.!
– The median has the advantage of not being unduly influenced by an outlying value.
• The Q test should not be applied to three data points if two are
identical.
• In that case, the test always indicates rejection of the third
value, regardless of the magnitude of the deviation, because a
is equal to w and Qcalc is always equal to 1.
– The same obviously applies for three identical data points in
four measurements, and so forth.

Statistical Treatment of Analytical Data Lecture

Uploaded by

Copyright:

Available Formats

Statistical Treatment of Analytical Data Lecture

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Statistical Treatment of Analytical Data Lecture

Uploaded by

Copyright:

Available Formats

Statistical Treatment of

 However, every measurement has some degree of

 The error is nothing but the difference between the

 no analysis is free of error or “uncertainty”!!

 Our goals are to minimize errors &..

 ‘We can only hope to minimize errors &

 Precision = reproducibility of a measurement

Illustration of accuracy and precision using the pattern of darts on a dartboard

 systematic error, &

• These changes may occur in the measuring

– eg fluctuations in room temperature, fluctuations in line

• These cause"lopsided" data….

– an instrumental error results when a

• are caused by non ideal instrument behavior, by

• Electronic instruments can be influenced by

• Errors of these types usually are detectable

• The non ideal chemical or physical behavior of the

• Result from the carelessness, inattention, or

• Judgments of this type are often subject to

A universal source of personal error is

Number bias is another source of personal error

• The most frequent number bias encountered

– & examination of techniques.

 The "Q-Test" (discussed later) is a systematic way to determine if a

 They are often the product of human errors.

 Can lead to outliers, results that appear to differ

 Frequent calibrations, standardizations, and analyses

 Impossible: The analytical results are free of

 Possible: Minimize these errors & estimate

 Statistical calculationsused to judge the

 The average of these replicate measurements is

 Bias?? -measure of how far the expected value of

 usually expressed in terms of the deviation of a set of

Note:– good precision does not always mean

We have systematic & random errors.

 total error is the sum of the systematic error

 & the more precise the measurement,..the better

 In the case of random errors only…good precision

 This bias will be -ve or +ve depending upon the type

 Many systematic errors can be repeated to a

 When we go about the task of determining the

 Too many systematic errors can be repeated to a

 Repeatability (of results of measurements) –

 It is equally important to specify the conditions

 What is the mass of an average coin in

 This is a problem that we might expect to be

 A good way to begin the analysis is to acquire

 Looking at these data, it is immediately apparent that our

 ie, we cannot use the mass of a single coin to draw a

 We can characterize these data by providing a

Measures of Central Tendency

 2 common ways to report this estimate of

 The mean (x̄ ) is the sum of individual measurements

 eg, mistakenly recording the mass of the fourth coin as

 is the value separating the higher half of a data

 When the data include an odd # of measurements, the median

 The median, however, provides a more robust estimate

 It is not skewed so much by extremely large or small

 To improve the reliability & to obtain infor about the

 2 to 5 portions (replicates) of a sample are usually