In addition, the general sampling and nonsampling errors associated with survey data, especially the underreporting of income, must always be kept in mind. Following is a brief discussion of some of the more important limitations on the estimates and the estimation process. The first limitation that should be mentioned is the difference between CPS and IRS income concepts. One phase of the tax estimation process is the calculation of adjusted gross income (AGI) based on the CPS income. The CPS excludes capital gains (or losses)

Amounts of capital gains were simulated for the CPS in the tax estimation procedure. (See details in appendix A of this report.) The computation of AGI on Federal individual income tax returns allows "adjustments" and various exclusions from total income. These include Individual Retirement Accounts, moving expenses, disability income exclusion, alimony paid, and employee business expenses. A simulation of the Individual Retirement Accounts was made using IRS statistics and data reported in the May 1983 CPS supplement. In addi

units in which both spouses had earnings. Simulations for the other adjustments were not made. Had these adjustments been simulated, the estimated AGI levels from the CPS would have been lower resulting in slightly higher after-tax incomes. While the overall CPS-estimated AGI was about the same as the IRS figure for 1983, the CPS and IRS amounts differ considerably by income type as discussed later.

Second, an initial step in the tax simulation process is the formation of tax filing units using the survey information on household relationship, marital status, and dependency rules based on income. The CPS records this information for each "permanent'' household member as of the time of interview in March. The simulation of tax filing units does not, therefore, account for differences in household composition that may have existed during the year for which taxes were simulated. Because of the CPS household definition, it was also not possible to simulate dependents living outside the household. The exact effect of these limitations is difficult to estimate since some simulated tax units will have too few dependents (exemptions) and some will have too many. It seems likely that, overall, too few exemptions would be simulated. This situation probably results in a slight underestimate of aftertax income levels because all exemptions have not been accounted for.

The combination of IRS tax return statistics with the March CPS income data may have also affected the final estimates to a small degree because the IRS returns include units which are not contained in the CPS universe. These include 1) prior year delinquent returns, 2) returns of Armed Forces members living overseas or on base without families, and 3) returns

The procedures for simulating Federal and State individual income taxes tend to underestimate the actual variation in taxes paid by AGI level and, therefore, may tend to underestimate the variation in after-tax incomes. This occurs because the simulation procedures used, in some cases, averages to assign statuses and amounts to CPS tax filling units. For example, the amount of deductions for units assigned itemizing status was simulated using a matrix showing the IRS ratio of itemized deductions to AGI for all tax units by AGI interval, type of return, number of dependents, and presence of a home mortgage. The true variation in deductions was not simulated since all units within a specified matrix cell were assigned the same proportion of their AGI as deductions. The net effect of this aspect of the simulation procedure on the final after-tax income estimates is not known.

Comparisons of the distribution of AGI derived from the March CPS with that based directly on tax returns indicate significant differences and year-to-year variation in these differences. These differences for 1983 can be examined in table A-4 of appendix A. Year-to-year variations can be examined by referring to similar tables in previous reports. Of note is the change in the relationship between simulated and IRS data for the "$75,000 and over" category. In 1980, the simulated estimate for number of taxable returns in this AGI interval was 17 percent below the IRS figure. For 1983 the simulated number is 6 percent higher. This change along with others that have occurred from year to year may affect comparisons of after-tax data between years. The effect of these changes on the after-tax income estimates is not known.

Finally, another important limitation is the underreporting of money income in the survey. This is a common problem encountered in household surveys that attempt to collect income data. Underreporting results in a downward bias in the estimates of income from the March CPS. While income underreporting is a serious problem in household surveys such as the March CPS, its effect on measures of year-to-year change in levels of income and poverty is much less important because year-to-year variations in underreporting are relatively small. Estimates of underreporting are contained in appendix D.

REVISIONS TO THE TAX SIMULATION METHODOLOGY

Appendix A contains a detailed description of the methodology used to simulate Federal income taxes in 1983. Included in the appendix are descriptions of several modifications to the 1983 Federal tax simulation procedures. These modifications include expanding the matrix used to impute capital gains and losses. In previous years, capital gains and losses were imputed to tax filing units according to their adjusted gross income level. Beginning in 1983, the matrix used to impute capital gains and losses was expanded to include adjusted gross income, type of return, and age of tax filer. The matrix used to assign amounts of income itemized

ning in 1983, the matrix used to assign amounts of income itemized was expanded to include adjusted gross income, type of return, number of dependents, and presence of a home mortgage.

In addition, in 1983 the simulation was modified to assure that the amount of income itemized was equal to or greater than the sum of property and State income taxes paid by the tax filing unit. Also, all tax filers with property and State income taxes exceeding the minimum needed to itemize were assigned to be itemizing tax filing units. Previously, itemization status and amounts were assigned independently of State and property taxes.

Finally, in 1983 the dependent child care credit was imputed to Federal tax filers for the first time. Previously, the only credit simulated was the earned income tax credit.

Federal income taxes paid, however, because comparisons of 1982 and 1983 data were important the 1982 data published previously were revised based on the methodology implemented in 1983 to provide comparability. In this report, all comparisons between 1983 and 1982 were made using the 1982 revised simulation file. Consequently, the 1982 tax data shown in this report differ slightly from those previously published.

Table G shows the effect of revisions on the 1982 after-tax income figures. Overall, mean after-tax income changed very slightly, from $18,906 to $18,926. There were some types of households, however, with somewhat larger changes. The after-tax income of elderly households declined by 0.5 percent, from $13,767 to $13,698. This change was caused mainly by the inclusion of age as a variable in the matrix used

1 Persons of Spanish origin may be of any race.

20,277 20,320 13,767 13,698

0.2

-0.5

21,594

21,659

0.3

13,928 13,863

-0.5

to impute capital gains. The use of a variable showing presence of a home mortgage in the matrix used to impute amounts of income itemized was the main cause of the differences in the revised figures for owner- and renter-occupied households. The mean after-tax income of owner-occupied households increased by 0.3 percent, from $21,594 to

$21,659, while the after-tax income of renters declined by 0.5 percent, from $13,928 to $13,863. The addition of the child care credit was a factor in the 0.5 percent increase in the after-tax income of married couples with children, from $23,307 to $23,434.