A Possible Key Reason for the Disparity in Income Statistics: ACS Income Estimates vs DC Individual Income Tax Data

In the previous post on this subject it was found that 1) ACS income statistics are systematically and consistently higher and ACS Gini coefficients are systematically and consistently lower than comparable statistics derived from the tax data; 2) some ACS and tax data reveal considerably different trends for the same phenomena; and  3) ACS Gini coefficients for the city show very little appreciable change over the 2006 to 2012 time period whereas tax data shows a 7.6 percent decline in the coefficients during that same time period but also a subsequent 3.6 percent increase between years 2011 and 2012.

The biggest distinction between the ACS and the tax data used in this analysis, full report here, is the basic socioeconomic unit of analysis. For this analysis, households are the basic socioeconomic unit for the ACS, and individual income tax returns are the basic unit of analysis for the tax data.  In the ACS, a household includes all the people who occupy a housing unit, and all income earned by household residents, regardless of their relation to the homeowner, is counted as income belonging to that household. On the other hand, the individual income tax return tends to represent an individual income earning resident (and their spouse and/or dependents, if applicable).

City income measures in terms of households (ACS) versus tax data returns (OTR) appears to be the most significant explanation in explaining the substantial and systematic difference in income statistics from the two data sources. The city’s mean and median income levels using the tax data are below the comparable ACS statistics.

jayron5

jayron6

In terms of the means, there was an average of 323,253 annual tax returns in the city for the years 2009-2013. But, ACS data tells us that there was an annual average of 263,649 occupied housing units in the city for the same time period. Dividing the city’s total income by the number of tax returns automatically yields a higher mean number than dividing the total income by the number of households because there were 22.6 percent more tax returns than households. In terms of the medians, the ACS tells us that the median income for households in the city was $65,830, but the tax data tells us that the median income for tax filers was $44,794.  But according to the ACS, the average household size in the city ranged between 2.11 and 2.31.  With 60.9 percent of the city’s tax filers being single residents, the tax data median of $44,794 most likely represents a single filer’s income. And indeed, the panel of tax data for this analysis reveals that there were 22 tax records (tax filers) with an exact income of $44,794 for the 2009-2013 time period, 59 percent of them were single filers, 23 percent were head of households, and 14 percent were married filers. Consequently, the ACS median household income, which represents a household with an average of slightly more than two residents (and possibly two income earners), is largely being compared to an income of an single individual tax filer.

jayron8

Given that the economic units being compared (households versus tax returns), may be a major cause for the discrepancy in the income statistics using the ACS and tax data, it is necessary to compare the overall income level of the city from the two data sources. The inflation–adjusted total income for the city for the five year period of 2009-2013 was $131.0 billion.  The comparable income amount for the same time period from the ACS was $150.8 billion. Notwithstanding the estimated $19.8 billion (13.1 percent) discrepancy between the two data sources, which may be an estimate primarily of the income earned by city residents that was not reported to tax authorities over the five year period, total income numbers for the time period under investigation from the two sources ($131 billion versus $151 billion) are fairly similar. Thus, controlling for the inability to classify income tax data by households, it appears that the tax data generally supports ACS income numbers for the District of Columbia in the aggregate.

 Differing Trends

The below table shows that while the mean income grew at an annual average rate of 3.27 percent over the study period, the tax data show mean income declined at an annual rate of 0.04 percent with the income in the final time period being less than the prior time period.  This may stem from the facts that over 60 percent of the city’s tax filers are single filers, and single filers have been the fastest growing cohort of individual income tax filers over the study period. According to the ACS, the number of occupied households grew at an annual average rate of 0.3 percent between years 2006 and 2013. According to the tax data, on the other hand, the number tax returns grew only at an annual average rate of 2.3 percent over the same time period. And more strikingly, single filers grew at an annual average rate of 2.4 percent, signifying that the number of single filers grew faster than all other tax filer types. Furthermore, 61 percent of the total growth in tax filers from 2006 to 2013 was accounted for by single filers, and there were 9,332 (4.7 percent) more single filers in 2012 than in 2011. All of this suggests that the robust net in-migration of single filers from 2006 to 2013 may have also been a major reason for the decline in average income per the tax data.

jayron9

jayron7

In sum, there has been a relatively large and continuous increase in new single filers migrating to the city.  But, with single filers having a mean income 69.4 percent lower than married filers (married filers and single filers account for over 90 percent of all income), the interplay of these dynamics may be the cause of the Gini coefficient per the tax data for 2012 beginning to trend upward while the ACS comparable statistic remained practically unchanged for years 2011 and 2012. For the study period, all of this suggests that while the out-migration of residents in the District of Columbia has been notable, the in-migration of new single residents has been slightly larger helping to cause the growth in the number of households to remain relatively unchanging. These important dynamics appear to have a significant bearing on the comparison of income statistics for the city when described in terms of tax filers and households.

This analysis finds great similarity in the citywide income level statistics as described by the ACS and DC tax data.  But means and median statistics from the two data sources differ substantially and systematically likely because the tax data is often explained in terms of tax filers/tax returns and the ACS often explains income, more appropriately, in terms of households. It also appears that the city’s demographic changes, and consequent changing tax filer profile, contributed to some of the differences in the levels and trends in income statistics from the two data sources.

What exactly is this data?

In the case of the District of Columbia, the ACS is designed to collect household data from about 4,000-5,000 city household respondents annually on their prevailing demographic and economic circumstances.

For the ACS household income comprises wage and salaries, military pay, commissions, tips, cash bonuses, social security payments, pensions, child support, public assistance, annuities, money derived from rental properties, interest and dividends. Earnings, as defined by the ACS, a narrower measure of income, are simply wage and salaries from employment and self-employment income.

The District of Columbia’s individual income tax data is collected and administered by the District of Columbia’s Office of Tax and Revenue. The analysis uses the Federal Adjusted Gross Income (FAGI) and the Wages, Salaries, and Tips variables from the individual income tax database as its primary measures of income variables. The FAGI of all tax filers is deemed the comparable income measure for the ACS’s “income”, and tax data “wages, salaries, and tips” is deemed the comparable income measure for the ACS’s “earnings”.

Income statistics for the District of Columbia: ACS Income Estimates vs. DC Individual Income Tax Data

The U.S. Census Bureau American Community Survey (ACS) is an important source of annual data on the economic, employment, housing, and demographic conditions of the nation and its subnational jurisdictions. Local jurisdictions also maintain a vast store of administrative data that captures some of the same socioeconomic measures included in the ACS. This raises the issue of how these measures from alternative data sources compare. In an attempt to address this issue, this analysis compares selected city income measures provided by both the ACS and the District of Columbia income tax returns database. This analysis also compares the Gini coefficient measure in the ACS to one derived from income measures in the District of Columbia income tax returns database.  See full report here.

Income Levels

For all time periods under investigation, median earnings for District workers in the ACS was at least $8,000 (20 percent) higher than the median wages and salaries in the DC tax returns database. The responses to the ACS survey indicated that the citywide median earnings in the 2009-2013 time period was $45,231, while the actual median wages and salaries of all tax filers was $36,288 for the same period.

2006-2010 2007-2011 2008-2012 2009-2013
Median Earnings for Workers (ACS) $41,171 $43,137 $44,423 $45,231
Median Wages, Salaries & Tips (DC tax data) $33,432 $34,679 $35,510 $36,288
Amount Difference $7,739 $8,458 $8,913 $8,943
% Difference 20.8% 21.7% 22.3% 21.9%

Examining household income, a slightly broader definition of income, the ACS data states that median household income for the city tended to be more than $16,000 (30 percent) higher than the median FAGI for all tax filers in the city.  The responses to the ACS survey indicated that the citywide estimated median household income in the 2009-2013 time period was $65,830 compared to the $44,794 actual median income of all tax filers.

  2006-2010 2007-2011 2008-2012 2009-2013
Median Household Income (ACS)  $58,526  $61,825  $64,267  $65,830
Median FAGI (DC tax data)  $42,017  $43,317  $44,124  $44,794
Amount Difference  $16,509  $18,508  $20,143  $21,036
% Difference 32.8% 35.2% 37.2% 38.0%

The above figures show persistent and fairly constant differences between the median measure of earnings in the ACS and the median measure of wages in the tax data.  But upon closer inspection, the above figure shows a mildly increasing difference between the median measures of income in the ACS versus the administrative data.  The figure and table below shows a more noticeable growing difference between measures of mean income in the two data sources. The ACS mean income in the 2006 – 2010 time period was $8,189 (9.3 percent) higher than the comparable statistic obtained from local income tax data.  By the 2009 – 2013 time period the ACS mean income was $17,589 (19.2 percent) higher than the mean income measure obtained from the tax data.  Over the period, the mean FAGI decreased by an estimated annual rate of 0.04 percent compared to a 3.27 percent estimated annual growth rate for mean household income in the ACS.

jayron3

  2006-2010 2007-2011 2008-2012 2009-2013
Mean Household Income (ACS)  $91,778  $96,183  $99,511  $101,076
Mean FAGI (DC tax data)  $83,589  $84,161  $84,448  $83,487
Amount Difference  $8,189  $12,022  $15,063  $17,589
% Difference 9.3% 13.3% 16.4% 19.2%
  2006-2010 2009-2013 Estimated Annual Average Growth
Mean Household Income (ACS)  $91,778  $101,076 3.27%
Mean FAGI (DC tax data)  $83,589  $83,487 -0.04%


The figure displaying the trends in mean income also displays the 90 percent confidence interval for the ACS means.  The ACS sample estimates and their statistical standard errors (the basis of the upper and lower bounds of the confidence interval) allow for the construction of confidence intervals.  The figure suggests that the process and method by which OTR obtains and processes its administrative data is considerably different from that of the ACS.

Income Inequality Levels

Gini coefficients reported by the two sources share a similar profile but differ in three important ways.  First, the Gini coefficients produced by the ACS are 12 to 18 percent lower than that produced using income tax data. This could significantly affect one’s view of inequality in the city. Second, the ACS coefficients stay in the very tight range of 0.53-0.54 whereas the tax data coefficients range from 0.61 to 0.66. These ACS coefficients suggest that the level of income inequality has been relatively unchanging whereas the tax data coefficients show an appreciable decline as the great recession began to take its toll on the city’s economy. Third, the tax data show income inequality began to increase in 2012 while the 2012 ACS coefficient appears practically indistinguishable from the coefficients in 2009 to 2011.

jayron4

In sum, this analysis presents two main findings. First, compared to comparable statistics derived from local tax data, ACS income statistics are systematically and consistently higher and ACS Gini coefficients are systematically and consistently lower than comparable statistics derived from the tax data. And second, and maybe more importantly, some ACS and tax data reveal considerably different trends for the same phenomena.

Possible Explanations

There are a few possible explanations for the large systematic differences in income statistics produced from ACS data and from the DC tax data.  First, ACS data is based on self-reported survey responses that may be less objective than income information provided on tax returns. Tax returns must be accompanied and supported by income documentation, and all tax returns are eligible for audit by OTR. Thus, data submitted on tax returns are subject to a greater degree of accountability and responsibility, making the quality of information in the ACS and tax data a possible issue.

Second, there may be a significant number of income earning residents that do not file tax returns or do not report all of their income. This pool of non-reported income by residents may represent a nontrivial number of households for the city and tax database. Consequently, the means, methods and processes of the ACS may better deal with this issue of capturing information from all households in the city regardless if they earn income, file tax returns, or report all of their income. ACS survey respondents are selected via a statistical methodology that makes them, in total, highly representative of the jurisdiction’s total population regardless of circumstances. The sum estimation of income characteristics of the city may be more comprehensively described by the ACS.

And third, for this analysis households are the basic socioeconomic unit for the ACS, and individual income tax returns are the basic unit of analysis for the tax data.  In the ACS, a household includes all the people who occupy a housing unit. The individual income tax return tends to represent an individual income earning resident (and their spouse and/or dependents, if applicable). It is not exceptional that different measures of the same variable that is quantified by different organizations using different data collection processes would be slightly different. But means and median statistics from the two data sources used in this analysis differ substantially and systematically. These differences may stem from the fact that the tax data is often explained in terms of tax filers/tax returns and the ACS often explains income in terms of households.

What exactly is this data?

In the case of the District of Columbia, the ACS is designed to annually collect household data from about 4,000-5,000 city household respondents on their prevailing demographic and economic circumstances.

For the ACS household income comprises wages and salaries, military pay, commissions, tips, cash bonuses, social security payments, pensions, child support, public assistance, annuities, money derived from rental properties, interest and dividends. Earnings, as defined by the ACS, a narrower measure of income, are simply wage and salaries from employment and self-employment income.

The District of Columbia’s individual income tax data is collected and administered by the District of Columbia’s Office of Tax and Revenue. The analysis uses the Federal Adjusted Gross Income (FAGI) and the Wages, Salaries, and Tips variables from the individual income tax database as its primary measures of income variables. The FAGI of all tax filers is deemed the comparable income measure for the ACS’s “household income”. Tax data “wages, salaries, and tips” is deemed the comparable income measure for the ACS’s “earnings”.