Do redistributive state taxes reduce inequality?

Inequality measures are calculated from the March CPS, using Stephen Jenkins' "ineqdeco" Stata routine. Person-weights were used, and hourly wages were not adjusted for family size. Since the CPS asks households about earnings in the previous year, the surveys from March 1978 to March 2003 provide data on household income in the years 1977-2002. The sample is further restricted to adults aged 16-55 with positive hours and earnings. Hourly wages are calculated by dividing annual earnings for the previous year by the total number of hours worked in the previous year (calculated by multiplying the number of weeks worked in the previous year by the usual number of hours worked per week in the previous year).

To avoid extreme values biasing the calculations, hourly wages below a minimum value are omitted and those above an upper threshold are truncated. In 2002, the minimum value was one dollar and the top-code was $500. In earlier years, these numbers are indexed to changes in average wages. For example in 1977, observations with hourly wages below $0.27 were dropped, while the top code was set at $134.11 per hour.

Since I calculate hourly wages as annual earnings divided by the total number of hours worked in the previous year, the number of hourly wage observations that are top coded in each year is affected by the top coding of annual earnings in the CPS. In income years 1977-1980, this is set at $50,000, in 1981-1983 at $75,000, and in 1984-1994 at $99,999. From 1995-2002, top coded values were given the mean value for all top coded observations (e.g., in 1995, all those who earned $150,000 or more were assigned earnings of $576,372). This change does not appear to have had a major impact on the number of hourly wage observations that I top coded, which ranged from 8-40 in income years 1977-1994, and from 27-58 in income years 1995-2002. The number of top coded hourly wage observations was 40 in 1994, and 48 in 1995.

Although the CPS is designed to be representative at a state level, the person-weights that are provided are calculated based on national demographics, rather than state demographics. However, this is unlikely to make a substantial difference. Using the CPS to calculate trends in inequality in California, a state whose demographic composition is very different to the nation as a whole, Reed, Haber, and Mameesh (1996, Appendix B) used census data to form new CPS weights for California, and found that it made virtually no difference to their estimates.

Tax Redistribution

To calculate redistribution measures, I use a national sample comprising a randomly selected ten percent of the March 1990 CPS (15,847 individuals). Income is indexed by multiplying each family's income by ([MedEarn.sub.st]/ [MedEarn.sub.1990]), where [MedEarn.sub.st] is median family income in a given state and year, and [MedEarn.sub.1990] is the median family income across the United States in 1990 ($38,640). This ensures that the distribution of earnings remains unchanged, but that incomes are at an appropriate level for the tax brackets in a given state and year.

For example, median family earnings in North Dakota in 1984 were $23,491, so in order to calculate tax redistribution, I take the 15,847 individuals from in the 1990 CPS sample, multiply their incomes by 0.607 ($23,491/$38,640), then assign them the state code for North Dakota, and the year 1984.

Each state-year sample is then fed through the National Bureau of Economic Research's Taxsim program (Feenberg and Coutts, 1993), version 5.1. To simplify calculations, I assume that all family income is wage income, that individuals file as singles, and couples file jointly (with two-thirds of the income assigned to the primary earner). Dependent child exemptions and age exemptions are taken into account. Post-tax income is net of state and federal taxes, but not net of FICA, which is regarded as akin to savings. Taxsim covers all 50 states plus the District of Columbia from 1977-2002. Therefore I feed the same sample (with incomes indexed according to the median income in that state and year) through the Taxsim program a total of 1,326 times (51 x 26). The ratio of post-tax income to pre-tax income gives (1 - ATR).

To calculate a measure of tax redistribution as it applies to hourly wages, I calculate pre-tax hourly earnings in the same manner as for the state inequality statistics, i.e., by dividing annual earnings for the previous year by the total number of hours worked in the previous year. As with the inequality measures, the sample is restricted to those aged 16-55, and the same bottom-coding and top-coding rules are applied to pre-tax hourly earnings. The pre-tax Gini coefficient for all states and years remains constant at 0.36, while the pre-tax S-Ginis are 0.15 ([delta] = 1.25), 0.24 ([delta] = 1.5), 0.43 ([delta] = 2.5), and 0.52 ([delta] = 3.5). Post-tax hourly earnings are then calculated by multiplying pre-tax earnings by (1 - ATR). The difference between the Gini (S-Gini) of pre-tax hourly earnings and the corresponding Gini (S-Gini) for post-tax hourly earnings is the measure of tax redistribution in a given state and year.

Other State Variables

Migration rates and hourly wages are calculated from March CPS data, applying the same sample restrictions as used in calculating the inequality measures (sample restricted to adults aged 16-55, hourly wages bottom and top-coded). Since the mobility question was only asked for the income years 1981-1984, 1986-1994, and 1996-2002, the sample for this specification is somewhat smaller. The migration question asks about mobility since March 1 in the previous year, and thus does not match up perfectly with the calendar year measures used for other statistics. For example, I match migration data from March 2002 to March 2003 with tax redistribution in tax year 2002. Note that the outgoing migration rate is smaller than the incoming migration rate, because some CPS respondents identify as interstate movers, but fail to identify the state from which they moved.

Real personal income and population are from the Bureau of Economic Analysis (http:// www.bea.gov/bea/regional/).

Unemployment rates are from the Bureau of Labor Statistics (http://data.bls.gov/).

Unionization rate is the percentage of each state's nonagricultural wage and salary employees who are union members. Estimates are based on the 1983-2002 CPS Outgoing Rotation Group (ORG) earnings files, the 1973-1981 May CPS earnings files, and the BLS publication, Directory of National Unions and Employee Associations, for various years. Details on data and methodology are provided in Hirsch, Macpherson, and Vroman (2001) (accompanying data online at http://www. unionstats.com/).

State sales taxes, state inheritance taxes, and state estate taxes are from the World Tax Database (http://www.bus.umich.edu/otpr/), downloaded December 10, 2007. Sales tax rates ignore exemptions (e.g., for food or prescription drugs). I combine state inheritance taxes and estate taxes into a single variable (no state has both), and also include a dummy variable to account for the possibility that the two types of taxes have different impacts.

Summary statistics for all variables are provided in Appendix Table 1. APPENDIX TABLE 1 SUMMARY STATISTICS Variable Mean SD N Current Period Variables (1983-2002)

Pre-Tax Gini 0.358 0.018 1,020

Post-Tax Gini 0.335 0.015 1,020

S-Gini ([delta] = 1.25) 0.144 0.010 1,020

S-Gini ([delta] = 1.5) 0.239 0.015 1,020

S-Gini ([delta] = 2.5) 0.432 0.019 1,020

S-Gini ([delta] = 3.5) 0.522 0.020 1,020 Incoming migration rate (from 0.048 0.019 918

interstate) Outgoing migration rate (to another 0.036 0.019 918

state) Wage ratio: incoming /nonmovers 0.968 0.194 918 Wage ratio: outgoing/non movers 0.980 0.310 917 Log population (non-institutional) 14.657 1.032 1,020 Sales tax rate 0.045 0.018 1,020 Maximum state inheritance / estate 0.032 0.051 1,020

tax rate Indicator for state estate tax 0.100 0.300 1,020 Unemployment rate 0.058 0.020 1,020 Log real state personal income per 9.894 0.307 1,020

capita Unionization rate 0.146 0.062 1,020 Current and Lagged Variables (1977-2002)

Redistribution (Gini) 0.025 0.003 1,326

Redistribution (S-Gini [delta] = 1.25) 0.012 0.002 1,326

Redistribution (S-Gini [delta] = 1.5) 0.018 0.002 1,326

Redistribution (S-Gini [delta] = 2.5) 0.027 0.003 1,326

Redistribution (S-Gini [delta] = 3.5) 0.027 0.003 1,326 Note: All specifications are restricted to dependent variables that are measured over the period 1983-2002. The maximum number of lags of the tax rate variables is six, so summary statistics for tax rates cover the years 1977-2002.

Acknowledgments