Determinants of States Tax Rate

4 ),,+,. 1 3 2 / # ),,+ # # ! $% &' ( )*))*+),+-., '$$/// 0 * " DETERMINANTS OF STATE TAX RATES1 Kodrat Wibowo Department of Economics Padjadjaran University 2004 ABSTRACT The problem addressed mostly in tax and government spending study is the endogeneity problem between tax/spending and income or economic growth. Therefore, the efforts to find instruments for tax variable are very crucial. This paper investigates the factors that determine changes in state tax rates with the US dataset from 1960 to 1999. I use a time series and cross-sectional approach (panel data) complemented by the utilization of fixed effect and interaction variables technique in the OLS estimates. I find that demographic, economic, and political structure variables are important for the determination of the change in state tax rates. Special for political issues, this empirical information supports the common knowledge that Democratic legislatures favor higher tax rates compared to Republicans, both in state and federal levels. Last, this result will allow researchers to address endogeneity concerns about tax and spending policy by incorporating 2SLS estimation in the analysis of economic growth. 1 This paper is a short version of an essay in my PhD dissertation at the University of Oklahoma (2003). First, I would like to thank W. Robert Reed, Daniel Sutter for their guidance and useful comments. I also appreciate Rex J. Pjesky for his help providing me the preliminary set of the US demographic and economic data. Last, I thank Rina Indiastuti and the Dept. of Economics and Development Studies, Unpad for asking me to present this “not so quite” fancy paper. 2 A. Introduction There are only a few studies addressing the endogeneity problem on tax or government spending variable as economic growth determinant2. One argument whether tax or spending can also affect economic growth is the Wagner’s law that views government expenditure as endogenous to economic development. As development proceeds there would be a long-run tendency for the public sector to grow relative to national income3, and taxes would also be increase to finance it. Therefore, the efforts to find instruments for tax variable are very crucial. This paper investigates the factors that determine changes in state tax rates with the US dataset from 1960 to 1999. Besley and Case (2000) suggest that greater use should be made of political variables as instruments in empirical studies of state’s policies. The empirical work in this paper follows up this suggestion. Thus, the second motivation is to identify new political variables that may prove valuable as instruments in other studies. This paper proceeds as follows. Section B summarizes previous studies that have examined the determinants of state fiscal policy. Particular emphasis will be given to the role of political variables. Section C identifies the (i) demographic, economic, and (ii) political variables that are significantly related to tax and spending policy variables. Section D and E estimate the determinants of changes in state tax rates. Section F and G concludes. B. Summary of Previous Studies of the Determinants of State Fiscal Policy Poterba (1994). Poterba (1994) examines the factors that determine how states respond to fiscal crises in the short-run. Fiscal crises have greater force at the state level because deficit finance is prohibited in most US states. Once a state has a fiscal crisis, politicians are confronted with a dilemma; to raise taxes or reduce outlays to restore fiscal balance. Poterba’s findings suggest that states react to unexpected deficit shocks with real changes in fiscal position. Raising taxes within the fiscal year has a small contribution to deficit reduction, but raising taxes that take effect in the next fiscal year is a better option than cutting spending to correct unexpected deficits. With respect to political variables, Poterba (1994) estimates that states with a single-party government raise more taxes and cut more spendings in response to unexpected deficit shocks. He provides two interpretations for this finding: (i) reaching political consensus in single-party states is easier than that in divided-party state governments; and (ii) the governor and the state legislature are more politically vulnerable in the states with a divided-party government. Unpopular actions such as raising taxes or cutting spending will be a threat for control of legislative seats in the next election. Poterba also explores the effect of the governor’s position in the electoral cycle to the magnitude of tax increases and spending cuts. The indicator for this variable equals unity in fiscal years immediately prior to gubernatorial elections. With a 10 percent level of confidence, his paper suggests that spending cuts and tax increases are significantly smaller when the governors are up for reelection. 2 See Easterly and Rebelo (1993), Mendoza, Milesi-Feretti, and Asea (1997), and Bleaney, Gemmel, and Kneller (2001). 3 This happens because of a substitution of public for private sector activity, an increase in cultural and welfare expenditures by the state, and because of government intervention to manage and finance natural monopolies. 3 Alt and Lowry (1994). Alt and Lowry (1994) examine whether state fiscal and political institutions affect the level of state spending and taxing rules. Using a formal model of fiscal policy where the state’s revenue equals the state’s expenditure, the paper concludes that (i) Democrats set state spending at a higher percentage of state personal income than Republicans; and (ii) states with divided governments have smaller responses to budget deficits than states with unified governments. A shortcoming of the Alt and Lowry (1994) study is that the data set is decomposed into a number of sub-samples. The breakup of the total sample into these sub-samples precludes the use of state and time fixed effects. This method could produce better results of the effect of state fiscal and political institutions on taxes if these subsamples have different structural relationships but there would not be a general conclusion whether political institutions affect the level of spending and taxing. Besley and Case (AER, 1995). Besley and Case, (AER, 1995) examine whether a state’s tax-setting behavior is affected by the tax-setting behavior of neighboring states. This study makes assumptions that voters have fairly open information across the states and they are able to make comparisons between jurisdictions to overcome political agency problems. Another assumption is that there is “asymmetric information” between voters and politicians: voters know less about the cost of providing public good than politicians. There are two types of politicians: rent seekers who charge more than the cost of public goods and non-rent seekers who provide public goods and services at cost. Voters choose to reelect the incumbents by evaluating the incumbents’ performances and comparing them to neighboring states’ incumbents’ performances. If voters are skeptical about the need to increase a tax, even a small increase may force elected officials to lose their seats. However, if voters find that taxes are increasing everywhere, voters will not mind an increase in taxes, even with a large increase. These assumptions lead incumbents into yardstick competition in which they care about what incumbents in neighboring political jurisdictions are doing for the tax-setting policy. Besley and Case suggest that voters are very sensitive to tax changes relative to the ones they observe in neighboring states, and this leads to votes against an incumbent whose tax changes are relatively high in regional standards. The estimated coefficients of the state demographic variables show that change in sales, income and corporate taxes increase with an increase in the proportion of elderly and young within the population. The proportion of young appears to be more significant than the proportion of elderly. This study also includes state and year dummy variables to absorb the impact of changes in national economic conditions and changes in federal fiscal behavior. Besley and Case (QJE, 1995). Another study by Besley and Case (QJE, 1995) examines whether governors in their last term behave differently with respect to taxing and spending behavior. The authors used a reputation-building model of political behavior to analyze the issue. Their argument is that governors facing a binding term limit behave differently compared to those who are able to run for reelection. This fact provides a source of variation in discount rates that can be used to test a political agency model. Besley and Case start with the assumption of asymmetric information about the types of politicians. Voters judge and gauge the types of their incumbents’ performances by using the outcome measures from incumbents. If incumbents want to be reelected, 4 then the possibility of reelection will affect policy choices. Officials try to develop a reputation that enhances their reelection chances. The results of this study show that when a governor faces term limits, sales taxes per capita as well as income taxes will be higher in his/her final term than if he/she did not face the term limit. However, corporate taxes appear to be insignificantly affected. For the insignificant estimates on corporate taxes, this study finds only weak positive results on total taxes. The proportion of young (aged 5-17) is a positive and significant determinant of sales taxes, income taxes, corporate taxes, and total taxes. The proportion of elderly (aged 65 and above) is only a positive and significant determinant for sales taxes. Results estimated by the model also suggest that term limits significantly affect state spending per capita, as do state demographic variables. State spending rises when the proportions of young increases while it falls with an increase in the proportion of elderly. With respect to the effect of political party effects, Besley and Case (QJE, 1995) estimate that if the governor who faces the term limit is a Democrat, total per capita taxes and its components are higher by $10 to $15 on average. On the other hand, Republican governors in their last term reduce sales taxes, corporate taxes, and real minimum wage while raising income taxes and state spending per capita, though by a lesser amount than Democrats in their final terms of office. Poterba (1997). Poterba (1997) studies the impact of “demographic structure”, particularly the proportion of a state’s population that is elderly, on state education spending. This focus of the paper is motivated by the tension between family with children who mostly receive the return from tax-financed public education spending, and older households with owner-occupied homes who pay taxes that finance K-12 education. This generational difference is believed to lead to a tension in the political process in which educational budgets are set. The fraction of the young and the elderly in the population significantly affect per-child spending on education. With state and year dummies included, the proportion of elderly has a negative relationship. The results suggest that, other things being equal, states with more elderly voters spend less on public schools. Comparing this result to the one estimated by “control equation” in which the dependent variable is per capita noneducation direct spending also strengthens this finding. The estimates of the “control equation” suggest that a larger fraction of the elderly in a state leads to a higher spending on non-education programs. Crain and Crain (1998). Crain and Crain (1998) investigate whether “current service budget baselines” increase state spending policy. A current service budget baseline sets the default level of public spending at the amount necessary to maintain existing services. This is in contrast to a “dollar budget baseline” in which the current level of expenditures is used as the baseline. Current service is widely criticized as biased toward higher spending in the existing budget process. The estimated coefficients show that during the 1980s, a current service baseline procedure had a positive and significant coefficient on spending growth. The current service baseline procedure led to higher spending than the dollar budget baseline procedure. The results also suggest that spending growth rates are significantly higher in states with 4-year terms limit on governance, as compared to states with only 2-year terms limits. The coefficient of “Party Stability” index for state Senates appears to be 5 significant: more predictability in the continued majority control by the same party promotes higher spending growth. However, the “Party Stability” index for the state House of Representatives fails to have a significant effect on state per capita spending growth. By examining the estimated coefficients of the fiscal structure variables, I can see that states that concentrate greater fiscal responsibility at the state level had higher spending growth than those that concentrate greater fiscal responsibility at the local level. A heavier dependence on income tax than other state revenue sources results in higher state per capita spending growth. States that have no requirement on the Constitutional Budget Balance have significantly lower state per capita spending than those that have a requirement. Finally, state spending moves in a positive direction with personal income and the share of the young population but moves negatively with the share of the populations who live in urban areas. Vedder (1990). The last study reviewed in this paper is a work by Vedder (1990). I saved this study for last because it comes closest to the analysis that I will undertake in this paper. Although Vedder is primarily concerned with the effect of state taxes on economic growth,4 he includes an analysis of the effect of political structure variables on the change in state tax rates. This is the only study that directly studies the determinants of changes in state tax rates. Vedder finds evidence that the states more likely to vote for Republican candidates also had significantly lower taxes. While Vedder (1990) does find a significant link between party affiliation and changes in state tax rates, his study is limited by a number of shortcomings. The inclusion of the District of Columbia is arguable. The regression analysis does not include any demographic characteristics of the states. Further, a great deal of information is thrown away in collapsing the data down to a cross-sectional sample. There is no doubt that state tax rates have had periods of substantial increase and decline over the 1967 to 1987 period. C. Implication of Previous Studies for Variable Selection Previous studies suggest that both demographic and political variables may be important explanatory variables for changes in state tax rates. Tables 1 and 2 summarize the findings of previous studies with respect to the estimated effects of demographic and political variables. Columns (1)-(3) identify the study, the dependent variable, and the respective explanatory variables. Column (4) reports the estimated effects of the variables on taxes or expenditures. Column (5) reports the significance of the estimated coefficients. ****TABLE 1 HERE**** State demographic variables. What can we learn from previous studies with respect to the selection of demographic variables? As Table 1 reports, demographic variables are frequently found to have significant coefficients in state taxes and/or state expenditures regressions. The coefficients of the unemployment rate, the proportion of the population aged 5-17 years old, the proportion of the population aged 65 and above, total population, the fraction of the population who own homes, and the fraction of the population who live in urban areas are all significant in some specifications. 4 This study is considered to be the only one that uses the term of the change in state tax rates as dependent variable. 6 Unfortunately, there is little consistency. For example, while both Besley and Case studies find that the proportion of the population aged 5-17 years old is positively related to higher taxes, Poterba (1997) finds that this same variable is negatively related to school spending, and negatively (but insignificantly) related to non-school state spending. ****TABLE 2 HERE**** Political variables. A similar conclusion holds with respect to previous studies’ findings on the effects of political variables. Table 2 reports that the following political variables are determinants of state taxes and/or spending with significant coefficients: control of the governorship and the lower house of the state legislature by the same party; an imminent gubernatorial election; the governor’s age; a Democratic governor; a Democratic governor in his/her last term; a large share of state and local revenues; a gubernatorial term limit; a 4-year gubernatorial term limit; “party stability” in the state senate; and “party stability” in the state house. As before, however, these findings generally lack consistency. For example, it seems contradictory that “party stability” in the state senate should be associated with higher spending, while “party stability” in the state house is associated with lower spending. The finding that comes closest to being a consistent finding in the literature is that party affiliation variables matter. Generally, a state which is characterized by a greater degree of “Democratic-ness” is likely to have higher taxes and spending than a state which is more Republican in nature. Nevertheless, even this finding is complicated by the fact that “Democratic-ness/Republican-ness” is measured in different ways by different studies. In conclusion, while previous studies do not establish strong priors about the expected effects, they do establish the fact that demographic and political variables can be significant determinants of state tax policy. Amongst demographic variables, the (i) proportion of the population aged 5-17 years old, (ii) the proportion of the population aged 65 and above, and (iii) the fraction of the population who live in urban areas appear to be particularly important.5 Amongst political variables, the findings of previous studies suggest that party affiliation variables should be included in analyses of state fiscal policy. D. Empirical Analysis of the Determinants of Changes in State Tax Rates General Description of Study and Variables As mentioned earlier, the objective of this paper is to identify the empirical determinants of the change in state tax rate. Moreover, I try to identify instruments for the change in state tax rate that can be used in 2SLS estimation of economic growth6. My sample consist of a cross sectional/time series dataset on 45 states (Alaska, Hawaii, Nebraska, Minnesota, and Wyoming are excluded) from 1960 to 1999. Nebraska is excluded from the analysis because of missing information in political party variables (Nebraska has a non-partisan, or the unicameral system, in the state legislature). 5 While the unemployment rate is also frequently a significant determinant of state taxes and/or expenditures, I choose not to include this variable in the empirical analysis below because of the concern with endogeneity. 6 2SLS is considered to be the most effective econometrics technique to solve the endogeneity problem in economic modeling. 7 Minnesota is also excluded since it had a unicameral system in the state legislature from 1959 through 1970, and Wyoming is omitted because its state tax rate is heavily dependent on severance taxes, which caused it to change dramatically in the late 1970s and early 1980s when the state experienced oil booms, as shown by Figure 1. Figure 1 State and Local Tax Burden of Wyoming (1960-1999) WYOMING 99 96 19 93 19 90 19 87 19 84 19 81 19 78 19 19 72 75 19 69 19 66 19 63 19 19 19 60 21% 20% 19% 18% 17% 16% 15% 14% 12% 11% 10% 9% 8% Source: the US Bureau of Census, selected years Due to relatively little year-to-year variation in most of the variables, especially the change in state tax rate, I analyze the data using 5-year intervals to net out short-lived shocks and business cycles7. This is also to avoid the problem of data unavailability and further, to overcome the problem of “wipe out” and “fiscal cycle” effects mentioned in the study by Beasley and Case, QJE (1995). Consequently, this study has data for eight time periods for the total of 360 observations: (1) 1960-1964, (2) 1965-1969, (3) 19701974, (4) 1975-1979, (5) 1980-1984, (6) 1985-1989, (7) 1990-1994, and (8) 1995-1999. All variables in the tax rate change equation, except the political variables, take their values at the beginning of the period. By doing this, the problem of endogeneity is minimized. The dependent variable employed is the change in state tax rate (Change in Tax Rate). The value of this variable is calculated as the difference between the state tax rate in year t and the rate in the previous 5 years (t-4). State tax rate is commonly referred to as “Tax Burden” and is defined as the percentage of the level of state and local taxes to the state personal income. Since tax calculations involve fiscal years and personal income is based on calendar years, tax rates are calculated by dividing state and local taxes in period t by state personal income in period t-1. The smallest change in state tax rate during my 5-year interval observations appeared in North Dakota from 1970 to 1074 in which state tax rates decrease by 3.814 percentage points. While Wisconsin had the highest change in state tax rates by increasing 2.120 percentage points from 1960 to 1964. The earliest data (1950’s until 1980’s) for state and local taxes is downloaded from the US Census Bureau homepage. I hand-entered the latest data (1990’s) from the US Census Bureau Government Finance (selected years). In the next two sections, I will 7 See Grier and Tullock (1989). 8 describe two groups of the independent variables: economic and demographic variables and political variables based on the descriptive statistics in Table 3. ***TABLE 3 HERE**** Economic and Demographic Variables State tax rates at the beginning of the period (Initial Tax Rate) is included as a determinant of tax rates changes since I expect that competition among states encourages tax rate convergence. Beasley and Case (AER, 1995) argue that “yardstick competition” forces the incumbents to reduce the tax-burden close to their neighboring states’ tax rate. Accordingly, I expect the sign of this variable to be negative. In my sample, Virginia was the state that had the lowest tax burden. In 1960, 7.15 percent of its real state personal income was burdened by state and local taxes. On the other hand, New York had the highest tax rate in 1970 with 15.85 percent. Data on the percentage of the elderly aged 65 and above at the beginning of period (Elderly) was not available for 1961 through 1964, but I overcame this data unavailability problem by using 5-year interval data instead of yearly observations, as mentioned earlier. During the time periods I observed, the average of the percentage of the elderly in the US was 10.7 percent; the lowest percentage of elderly was in Nevada in 1965 at 5.2 percent. The highest percentage of elderly was 18.2 percent in Arkansas in 1990. Data on population density at the beginning of the period (Density) is derived from the ratio of total population over land area. In the annual observations, the state with the highest population density was New Jersey in 1999 with 1,046 persons per square mile. And the lowest population density was Nevada in 1964 with 4 persons per square mile. The last demographic variable included is education attainment at the beginning of the period (Education), which is defined as the fraction of the population aged 25 years old and above who completed college or a higher degree program. The state with the lowest fraction of person 25 years old and above who completed a college or higher degree program was Illinois in 1964, with only 4.22 percent. In contrast, 31.67 percent of Colorado’s population 25 years old and above completed college or higher degrees in 1999. Data of income earned in agricultural (Farm) and manufacturing (Manufacturing) sectors are calculated as the proportion of the state personal income at the beginning of period. On average, for all states in my time observations, farm income constituted only 3.84 percent of personal income. In 1960, South Dakota had the highest percentage of annual farm income to personal income with 24.23 percent. On the other hand, the lowest value was North Dakota with -6.55 percent in 1980. During the time periods I observed, in 1965 Michigan had the highest concentration of manufacturing with 36.55 percent of its state personal income coming from manufacturing. The variables Elderly and Density are included because they were found to be significant in previous studies. The other demographic variables are included because these variables may be measuring preferences for state spending and taxes policy amongst different special interests in the state. 9 Political Variables The political variables employed as the determinants for the change in state tax rates are divided into two categories: political variables from the states’ federal legislature and ones from state’s state legislature. The political variables from federal legislature are generally approximated by only one variable, which is the adjusted mean of the American for Democrat Action (ADA) score.8 ADA Average in this study measures the mean ADA score for the state’s federal politicians (mean ADA score in House of Representatives plus mean ADA score in the US Senate divided by 2), over the 5-year period. I employ the mean value over the five-year period lagged one year (t-5 to t-1 rather than t-4 to t) because legislative changes voted in one fiscal year typically do not go into effect until the next fiscal year. This variable is designed to measure the states’ federal legislators' preferences for spending and taxes. At the federal level, a higher ADA score is generally associated with support for higher federal spending and taxes.9 I believe that voters who support federal legislators with higher ADA scores will also support state legislators who support higher spending and taxes. Thus, the prediction is that this variable will be positive. The adjusted ADA score data is loaded from the homepage of Tim Groseclose, a Political Science Professor of Stanford University.10 According to this measure, the most liberal state was Massachusetts in 1979 with a mean ADA score of 85.44. In that year, Democrats controlled both the House of Representatives and the Senate in the Massachusetts’ federal legislature. On the other hand, Idaho was the most conservative state in 1984, with an average of only 1.89 real ADA scores, which was extremely conservative compared with other states in the same year. Again, I can also note that for Idaho, Republicans controlled both of the chambers in the federal legislature. The second category is the political variables in the state’s state legislature. Democrat Legislature refers to the percentage of years during the 5-year period, in which that Democrat controlled both chambers in the state legislature, and Republican Legislature refers to the percentage of years during the 5-year period, in which Republican controlled both chambers in the state legislature. Based upon the differences in spending and taxes preferences of Republicans and the Democrats at the federal level, my prediction is that Democrat Legislature will positively related, and Republican Legislature will be negatively related to changes in state tax rate. Only 20 out of 45 states experienced a full 5-year unified Republican, while 33 states experienced a full 5-year unified Democrat during the time periods I observed. In fact, 12 states experienced a full 40-years unified Democrat from 1959 through 1999. Of those states, 92 percent are southern states. There was only one state, New Hampshire, that was unified Republican for all observation years.11 8 This variable is commonly used as a measure of how liberal or conservative a member of congress is in their office. Democrats are known more liberal than Republicans. 9 It is possible to have a negative number for ADA score since I use Real (Inflation-adjusted) ADA score produced by Groseclose, Levitt, and Snyder Jr. (1999) 10 http://faculty-gsb.stanford.edu/groseclose/turboadas.webpage.update061302.xls 11 The reason of not including governor as political variable in my study is statistical. I do not find that governor showing significant coefficient in my preliminary tax model regressions in any specification 10 Many state and time-specific factors have important effects on the tax rates change (such as unemployment rate, level of average wages, and the rise of legislatureimposed special state spending and revenues). Those factors may affect the level of state income taxes and transfer payments; I need to recognize the potential influences of such effects by allowing for state and time-fixed effects in the equation. In this case, I try to identify the coefficients of interest from variation among states (over time) in other structures that cannot be explained by economy-wide shocks to demographics and political conditions. The inclusion of the state and time-fixed effects in the equation also help me to avoid the problem of “specification bias” in the model. E. Empirical Results In this section, I report the results of regressing the change in state tax rate on state economic and demographic conditions and political variables. The first subsection reports the basic political specification results with the inclusion of cross-state fixed effects and time effects in the model, while the second, third, and fourth models add demographic and economic variables and also interaction variables among them. By interacting one specific variable to others, I will be able to alter their real impacts on state tax rates. Last two equations include the interaction variables between time fixed effects and demographic variables to analyze the robustness of the model. Equation 1 I begin the analysis with a model of the change in state tax rate with political variables and Initial Tax Rate that controls for state and time fixed effects. The use of state and time fixed effect is intended to overcome the problem of bias from inadvertently omitting any variables that potentially affects the change in state tax rates model. Let the Change in Tax Rate be denoted by DTRst, the basic specification is: DTRst = 0 + 1 Initial Tax Ratest + 2 Republican_Legislaturest 48 + 3 Democrat_Legislaturest + 4ADA_Averagest + i s i 5 55 + i t+ s,t …………….… (equation.1) i 49 The equation has a good fit, explaining approximately 46.5 percent of the variation in the change in state tax rates. This coefficient of determination is resulted without any contemporaneous variables other than the political structure variables. Some may suspects that the inclusion of state and time fixed effects in the model is a waste since the R2 is not even close to 60 percent12. However, the state fixed and time fixed effects are jointly significant at a 0.01 level of significance in this equation. The estimations shown in the first column of Table 4 indicate that the coefficient on Initial Tax Rate is significant and has a negative sign, as expected. That initial levels of the state tax rate are negatively correlated with their changes reflects the process of convergence in state tax rates. The point estimate suggests that, ceteris paribus, a state having a tax rate that is one percentage points higher than other states at the beginning of 12 Kneller, Bleaney, and Gemmel (1999) show that much higher R2 is the artifact of the inclusion of state fixed effects and time effects in the equation. 11 a 5-year period will increase its tax rate 0.47 percentage points less than other states over that period. There are three political structure variables in this equation: adjusted mean ADA score (ADA Average), the percentage of years that Democrats controlled both of the chambers in the state legislature (Democrat Legislature), and the percentage of years that Republicans controlled the state legislature (Republican Legislature). As expected, the coefficient estimate of ADA Average is positive; a higher ADA score in the federal legislature tends to increase state tax rates. Democrat Legislature and Republican Legislature also have the expected signs. States in which Republicans (Democrats) have controlled both houses of the legislature are less (more) likely to raise taxes during that period. However, neither of the associated coefficients is significant at the 5% level, and only Democratic Legislature is significant at the 10% level. However, a test of the null hypothesis that the political structure variables corporately have no effect on the change in state tax rates is rejected at the 5% significance level. Equation 2 In consideration of public choice matters, I add 3 interactive variables: ADA Average Farm, Republican Legislature Farm, and Democratic Legislature Farm to equation 1. The main reason for adding these interaction terms is statistical. The equation that includes Farm interaction terms has the lowest Akaike Information Criterion (AIC) and Schwartz Information Criterion (SIC).13 Another reason is that it is well known that agricultural interests from farmer groups have a disproportionate impact on political outcomes in both federal and state legislatures. Regardless of the very-small shares of farm income to total state earning, farmer groups are still important voters that help politicians to get elected or incumbents to get reelected. Historically, the Democratic Party drew its followers from farmer groups. In 2000, Democratic budget resolutions favored farmer groups by providing increases in income assistance for farmers. In contrast, additional money for agriculture was not a sure thing in the Republican budget resolution. However, these farmers’ political alignments have changed because the Republican platform released at the 2000 Republican Convention is more in line with agriculture. For example, the platform put fourth specific goals to repeal the inheritance tax, and to grant a one-time exemption on the capital gains tax from the sale of farming products. Accordingly, the proportion of personal income earned from the agricultural sector (Farm) is interacted with the three political variables I have from equation 1. The regression results of equation 2 reported in the third column of Table 4 show a higher adjusted R2, 0.395 compared to 0.369 in equation 2. This helps to indicate the joint significance of Farm interaction terms. The formal testing also shows that Farm interaction terms’ coefficients are jointly significant with a p-value of 0.0008. With the inclusion of Farm interaction variables into the equation, now I have a total of six political variables in the equation. The inclusion of Farm interaction terms causes the coefficient of ADA Average to become significant. A test that both of the ADA (ADA Average and ADA Average Farm) coefficients are equal to zero is rejected with a pvalue of 0.0064. The coefficient of Democrat Legislature and Republican Legislature remain insignificant but in spite of this insignificancy, I will still include them since a test of the null hypothesis that all of the political variables’ coefficients (i.e., ADA Average, 13 AIC and SIC are commonly used in the issue of model selection in Econometrics. 12 ADA Average Farm, Republican Legislature, Republican Legislature Farm, Democrat Legislature, and Democratic Legislature Farm) are jointly insignificant is rejected at 99 percent confidence level. Due to the inclusion of interaction terms in the equation, I need to calculate the estimates of marginal impacts of the original political structures variables. By employing a simple differential rule, I use the following formulas to gather the marginal impacts of the original political variables: DTR ADA _ Average ADA _ Average DTR Republican _ Legislature DTR Democrat _ Legislature ADA _ Average*FARM Republican _ Legislature Democrat _ Legislature x FARM Republican _ Legislature*FARM Democrat _ Legislature*FARM x FARM x FARM 4 When evaluated at the mean value of Farm, the estimated marginal impacts for ADA Average, Democrat Legislature, and Republican Legislature are 0.0049, 0.0022, and -0.0023 respectively. These signs are consistent with what was expected and estimated in equation 1. However, none of these marginal impacts is significant at a 5% significance level when evaluated at the mean value of Farm. Equation 3 Equation 3 adds the economic and demographic variables into equation 1. DTRst = 0 + 1 ADA_Averagest + 2 Democrat_Legislaturest + 3 Democrat_Legislaturest + 4 Initial Tax Ratest + 5 Elderlyst 53 + 6 Densityst + 7 Farmst + 8 Manufacturingst + 9 Educationst i s i 10 60 + i t+ s,t ……………… (equation 3) i 54 The equation explains approximately 51 percent of the variation in the change in state tax rates. The estimations shown in the third column of Table 4, again suggest that the coefficient on the Initial Tax Rate is significant and has a negative sign as expected. The positive and significant estimate of the state’s population density at the beginning of period (Density) shows that states with higher population density are to be more likely to increase taxes than less densely populated states. The estimation results also suggest that all else equal, jurisdictions with more elderly populations are less likely to increase taxes than states with younger populations. Further, states whose economies that are more concentrated in the agricultural and manufacturing sectors, and whose populations are more educated, are individually estimated to be less likely to increase taxes than other states. The estimates of variable Farm, Manufacturing and Education are all shown to be significant at a 5% level of significance. The significances of state characteristic variables is also supported by the hypothesis testing which rejects the null hypothesis that 13 all of the State characteristic variables corporately have no effect on the change in state tax rates. The associated p-value is 0.000002. Again, the coefficient estimate of the political variables: ADA Average, Democrat Legislature, and Republican Legislature have the expected signs but insignificant coefficients. And, again, I reject the null hypothesis that all of the political structure variables jointly have no effect on the change in state tax rates (p-value is 0.0294). Equation 4 Equation 4 adds the farm interaction effects to the specification of equation 3. This equation has a higher adjusted R2, 0.437 compared to those in previous equations. This is consistent with the joint significance of the Farm interaction effects in the model. Formal testing also shows that the Farm interaction terms’ coefficients are jointly significant with a p-value of 0.0006. With the state characteristic variables in the equation, I find that the inclusion of the Farm interaction terms together with state characteristic variables causes each of the coefficients of political structure variables, except Republican Legislature to become significant. A test that both of the ADA, both of the Republican Party, and both of the Democratic Party coefficients are equal to zero, respectively is rejected with a p-value ranging from 0.0013 to 0.011. Moreover, a test with a null hypothesis that all six political variables corporately have no effect on the change in state tax rates is rejected at the 95 percent level of confidence. Using the marginal impact formulas presented above, I find that the signs of marginal impacts of the political variables confirm my expectations. ADA Average and Democrat Legislature have positive signs while Republican Legislature has a negative one. However, unlike the result from equation 2, the marginal impacts of the political variables when evaluated at the mean value of Farm are significant at the 5% of significance level Democrat Legislature and 10% for ADA Average. Equation 5 Next, I try to check the robustness of the political variables estimates by exploring the effects of time specific differences among states. I include interaction variables between time fixed effect and state characteristic variables (Density, Farm, and Manufacturing) into equation 3. The reason for the inclusion of time interaction variables is that states may differ from each other through time periods. For example, there were policies and regulations passed by Federal government on education, farm and manufacturing sectors throughout the observation periods. Those polices may induce different effects for state demographic variables over different time periods. With 21 additional variables, the results of equation 5 show a higher adjusted R2, 0.567 compared to previous equations. A test of the null hypothesis that all time and state characteristic interaction terms have no effect on the change in state tax rates is rejected with a p-value of 0.0001. If I compare the results of equation 5 to the results of equation 3, I see that all variables consistently have the same estimated signs. I decide to select this equation as the better equation compared to previous equations because it has the lowest AIC and SIC, as shown in Table 4. 14 Equation 6 Having knowledge that equation 5 is the most appropriate model to select concerning the AIC and SIC scores, I finally include the Farm and political variables interaction terms I have in equations 2 and 4 to analyze the robustness of the political variables’ effects on state tax rates. The result of this equation shows that about 70% of the variation in state tax rates can be explained by this model. A test of the null hypothesis that Farm and the political variables interaction terms have no effect on the state tax rates is rejected with a p-value of 0.06. The results demonstrate the consistent sign estimates for all major variables. The point estimate of Initial Tax Rate suggests that, ceteris paribus, a state having a tax rate that is one percentage point higher than other states at the beginning of a 5-year period will increase its tax rate 0.49 percentage points less than other states over that period. All political variables but Republican Legislature have significant estimated signs as expected. When evaluated at the mean value of Farm, the estimated marginal impacts for ADA Average, Republican Legislature, and Democrat Legislature are 0.00578, -0.00285, and 0.00092, respectively. These signs are consistent with what was expected and estimated in equations 2 and 4. Even so, only the marginal impact of Democrat Legislature is significant at the 5% significance level when evaluated at the mean value of Farm. The marginal impacts of ADA Average and Republican are significant only at the10% significance level. F. Implications and Discussion Comparing and analyzing the estimates of the six equations allow me to test whether there is any difference between the impacts of Republican legislatures and those of Democrats legislatures on the variability of the change in state tax rates. Table 5 reports the results of testing the difference in the marginal impacts of Democrats and Republicans. In each of the 3 equations containing political variables and their interaction terms, the null hypothesis that Democrats and Republicans have equal impacts on the change in state tax rates is rejected with associated p-values consistently ranging from 0.013 to 0.016. The results suggest that the impact of Republican legislatures is different than the impact of Democrats in determining changes in state tax rates. The test results in Table 5 also allow me to make some practical interpretations on the impact of changes in the partisan makeup of state As a practical matter, I ask what difference political party control of the state legislature means for the change in state tax rates. Using the value of the difference between the marginal impact of Democrats and the marginal impact of Republicans in equation 1 in Table 5, I calculate that if Democrats controlled both houses of legislature for a given 5-year period, then state tax rate would be 0.4 percentage points higher on average than if Republicans controlled both branches of the legislature for that period. This follows the conventional wisdom that Democratic legislatures favor higher tax rates compared to Republicans. Since the costs of passing regulations and policies are less in the single majority party, states in which the Democrats controlled both branches of the legislature had higher state tax rates. The 0.4 percentage points different are also consistent in each of the 3 equations. This fact gives me more confidence that I have estimated the true effect of the political variables on the change in state tax rates. 15 G. Conclusion The empirical evidence presented in this paper suggests that demographic, economic, and political structure variables are important for the determination of the change in state tax rates. Percentage of elderly at the beginning of the period (Elderly), population density at the beginning of the period (Density), income share from agricultural sector at the beginning of the period (Farm), income share from manufacturing sector at the beginning of the period (Manufacturing), and educational attainment at the beginning of the period (Education) appear to be significant determinants in at least 2 out of 6 equations at the 10 percent significance level. In general, states whose economies are more concentrated in the agricultural or manufacturing, and whose more elderly and whose populations are more educated are estimated to be less likely to increase taxes than other states, while states having higher population densities are estimated to be more likely to increase taxes than less densely populated states. In the matter of political structure variables, there are two major findings from the results. First, political variables are important for the determination of the change in state tax rates. ADA Average, Republican Legislature, and Democrat Legislature show significant and appropriate signs of coefficient estimates as expected with at least a 10 percent significance level. States whose federal legislators are characterized by higher ADA scores are more likely to increase taxes. Moreover, states in which Republicans control both houses of the state legislature are less likely to raise taxes during that period while Democrats are more likely to raise taxes when they control both houses of the state legislature. This finding provides prima facie evidence that these variables can serve as instruments in two-stage least squares estimations of the economic growth models. Second, the sign estimates and significances of political variables are shown to be robust to the inclusion of the set of the conditioning variables into the model. Reference: Alt, James E. and Robert C.Lowry (1994), “Divided Government, Fiscal Institution, and Budget Deficits: Evidence from the States,” American Political Science Review, 88(4), pp. 811-28 Besley, Timothy and Anne Case (1995), “Does Electoral Accountability Affect Policy Choices? Evidence from Gubernatorial Term Limits,” The Quarterly Journal of Economics, August, pp. 767-97 _________ (1995), “Incumbent Behavior: Vote-Seeking, Tax Setting, and Yardstick Competition,” The American Economic Review, 85(1) pp. 25-45 _________ (2000), “Unnatural Experiments? Estimating the incidence of Endogenous Policies,” The Economic Journal, 110, pp. 672-94 Bleaney, Michael, Norman Gemmel, and Richard Kneller (2001), “Testing the Endogenous Growth Model: Public Expenditure, Taxation, and Growth over the Long Run,” Canadian Journal of Economics, 34(1), pp. 36-57 Bound, John, David A. Jaeger, and Regina M. Baker (1995), "Problems with Instrumental Variables Estimation When the Correlation between the Instruments and the Endogenous Explanatory Variable is Weak.” Journal of the American Statistical Association, 90, pp. 443-450. 16 Crain, W. Mark (1999), “Districts, Diversity, and Fiscal Biases: Evidence from the American States,” Journal of Law and Economics, 42: 675-698. Crain, W. Mark and Nicole V. Crain (1998), “Fiscal Consequences of Budget Baselines,” Journal of Public Economics, 67, pp. 421-436 Crain, W. Mark and T.J. Muris (1995), “Legislative Organization of Fiscal Policy,” Journal of Law and Economics, 38, pp.1-18 Easterly, William and Sergio Rebelo (1993), “Fiscal Policy and Economic Growth, Journal of Monetary Economics, 32, pp. 417-458 Erikson, Robert S., Gerald C. Wright, Jr., and John P. McIver (1989) “Political Parties, Public Opinion, and State Policy in the United States,” American Political Science Review 83, pp. 729-750. Greene, William H. (1997), Econometrics Analysis, 3rd edition, Upper Saddle River, Prentice Hall, NJ Grier, Kevin B. and Gordon Tullock (1989), “An Empirical Analysis of Cross-National Economic Growth, 1951-80,” Journal of Monetary Economics, 24, pp. 259-276 Groseclose, Tim, James Levitt, and James R.Snyder, Jr. (1999), “Comparing Interest Group Scores across Times and Chambers: Adjusted ADA scores for the U.S. Congress,” American Political Science Review, 93, pp. 33-50 Higgs, R. (1989), “Do Legislators’ Votes Reflect Constituency Preference?: A Simple Way to Evaluate the Senate,” Public Choice, 63, pp. 175-181 Holtz-Eakin, Douglas (1988), “The Line Item Veto and Public Sector Budgets,” Journal of Public Economics, 36, pp. 269-292. Johnston, Jack and John DiNardo (1997), Econometrics Methods, 4th edition, The McGraw-Hill Companies, Inc. New York Kneller, Richard, Michael F. Bleaney,, and Norman Gemmel (1999), “Fiscal Policy and Growth: Evidence from OECD Countries,” Journal of Public Economics, 74, pp. 171-90 Mendoza, Enrique G., Gian Maria Milesi-Feretti, and Patrick Asea (1997), “On the Ineffectiveness of Tax Policy in Altering Long-Run Growth: Harberger’s Superneutrality Conjecture,” Journal of Public Economics, 66, pp. 99-126 Nickell, Stephen (1981), “Biases in Dynamic Models with Fixed Effects,” Econometrica, 49(6), pp. 1417-26 Peltzman, Sam (1985), “An Economic Interpretation of the History of Congressional Voting in the Twentieth Century,” American Economic Review, 75(4) pp. 656-675 _________ (1987), “Economic Conditions and Gubernatorial Elections,” American Economic Review, 77, pp. 293-97 Poterba, James (1994), “State Responses to Fiscal Crises: the Effect of Budgetary Institutions and Politics,” Journal of Political Economy, 102(4), pp. 799-821 ________ (1996), “Demographic Structure and the Political Economy of Public Education,” Journal of Policy Analysis and Management, 102(4), pp. 799-821 Vedder, Richard K. (1990), “Tiebout, Taxes, and Economic Growth,” Cato Journal, 10(1), pp.91-107 _________ (1996), “Taxation and Economic Growth: Lessons for Oklahoma”. State of Oklahoma, Office of State Finance, Unpublished manuscript. 17 Table 1: The Estimated Effect of Demographic Variables on State Taxes and Expenditures: Results from Previous Studies Study Besley and Case, (AER, 1995)b Besley and Case, (QJE, 1995)c Besley and Case, (QJE, 1995)d Poterba (1997)e Dependent Variable Taxes Taxes Expenditures Expenditures Variable 1) Unemployment rate 2) The proportion of population aged 5 – 17 years old 3) The proportion of population aged 65 and above 1) The proportion of population aged 5 – 17 years old 2) The proportion of population aged 65 and above 3) Population 1) The proportion of population aged 5 – 17 years old 2) The proportion of population aged 65 and above 3) Population 1) The proportion of the population aged 5 – 17 years old 2) The proportion of the population aged 65 and above 3) The fraction of population who own homes 4) The fraction of population who live in urban areas 5) The fraction of nonwhite population 6) The fraction of population below poverty line Estimated Effect Mixed Significant at 5% level? Sometimes Mixed Sometimes Positive Positive Sometimes Yes Positive Sometimes Mixed Positive Sometimes Yes Negative Yes Negative Negative Yes Sometimes Mixed Sometimes Positive Yes Negative Sometimes Positive No Negative No 18 Study Crain and Crain (1998)f Dependent Variable Expenditures Variable 1) The proportion of the population aged 5 – 17 years old 2) The fraction of population who live in urban areas Estimated Effect Positive Significant at 5% level? No Negative Yes NOTES: Estimates taken from Table 1, page 819 of Alt and Lowery (1994). b Estimates taken from Table 4, page 37 of Besley and Case (AER, 1995). c Estimates taken from Table IV, page 780, columns (1)-(4) of Besley and Case (QJE, 1995). d Estimates taken from Table IV, page 780, column (5) of Besley and Case (QJE, 1995). e Estimates taken from Tables 3 and 4, pages 57 and 58, columns (2)-(4) of Poterba (1997). f Estimates taken from Table 3, page 431, column (3) of Crain and Crain (1998). a 19 Table 2: The Estimated Effect of Political Variables on State Taxes and Expenditures: Results from Previous Studies Study Poterba (1994)a Poterba (1994)b Dependent Variable Expenditures Taxes Alt and See note below. Lowry (1994)c Besley and Taxes Case, (AER, 1995)d Besley and Taxes Case, (QJE, 1995)e Besley and Case, (QJE, 1995)f Expenditures Estimated Effect Positive Significant at 5% level? Sometimes Mixed Mixed Sometimes Sometimes Negative See note below. Yes See note below. 1) Governor’s age Mixed Sometimes 1) Republican governor is in his/her last term 2) Democratic governor is in his/her last term 3) Democratic governor 1) Republican governor is in his/her last term 2) Democratic governor is in his/her last term 3) Democratic governor Mixed No Positive Yes Mixed Positive Sometimes No Positive Yes Positive Yes Variable 1) Governor and the majority party of the lower house of the legislature are from the same party 2) Gubernatorial election is imminent 1) Governor and the majority party of the lower house of the legislature are from the same party 2) Gubernatorial election is imminent See note below. 20 Study Crain and Crain (1998)g Vedder (1990) Dependent Variable Expenditures Taxes Variable 1) Constitutional balanced budget requirement 2) State share of state + local revenues 3) Dependence on state income taxes 4) Gubernatorial term limit 5) 4-year Gubernatorial term limit 6) “Party Stability” in state senate 7) “Party Stability” in state house 1) Measure of support for Republican presidential candidates Estimated Effect Negative Significant at 5% level? No Positive Yes Positive No Positive Yes Positive Yes Positive Yes Negative Negative Yes Significant NOTES: Estimates taken from Table 5 on page 817, columns (1), (3), and (5) of Poterba (1994). b Estimates taken from Table 5 on page 817, columns (2), (4), and (6) of Poterba (1994). c Signs and significances are difficult to determine in Alt and Lowery (1994) because this study estimates separate regressions for each of eight different subgroups. d Estimates taken from Table 4, page 37 of Besley and Case (AER, 1995). e Estimates taken from Table V, page 782, columns (1)-(4) of Besley and Case (QJE, 1995). f Estimates taken from Table V, page 782, column (5) of Besley and Case (QJE, 1995). g Estimates taken from Table 3, page 431, column (3) of Crain and Crain (1998). h Estimates taken from Table 3, page 99 of Vedder (1990). a 21 Table 3: Descriptive Statistics Variable Mean Std Dev. Minimum Maximum Change in Tax Rate 0.2289 0.7102026 -3.8138378 2.1202783 Initial Tax Rate 10.5342 1.3424867 7.1526656 15.8320497 ADA Average 41.4217 18.1051594 1.8928442 85.4437721 Democrat Legislature 56.7778 45.7028742 0 100 Republican Legislature 26.0555 39.4106616 0 100 Elderly 10.6830 2.1125393 5.2 18.2 Density 158.7411 210.6379607 2.6320312 1022 Farm 2.7458 3.8446750 -6.5505760 24.2280664 Manufacturing 16.9434 7.3811766 2.9743608 36.5557554 Education 14.3782 5.8955821 4.220000 31.6742736