E-Mail This Article to a Friend Income, Poverty Data Flawed Middle-class incomes declined more sharply last year than the Census Bureau reported; a data error discovered by the Phoenix skewed the newly released figures. by David S. Bernstein
	Follow the Money Until 2001, the first $6000 of your income was always taxed at 15 percent. That year’s tax cut lowered the figure to 10 percent. Your savings, five percent of $6000, was exactly $300. Thus, in 2001, you paid $300 less in federal taxes than in previous years. To stimulate the economy, the Government mailed you a $300 " tax refund " check in the summer of 2001. In reality, that check was simply a loan, given to you that summer and taken back from you the next spring. The way the IRS took it back was by adding a five percent surcharge on the first $6000 of income on your 2001 tax form. To make it simple for you, the tax tables in your 2001 1040 instruction booklet combined that surcharge with the new 10 percent tax rate and presented it as 15 percent. In other words, the first $300 of your taxes that year were not taxes, but the return of the loan. The Census Bureau statisticians, it turns out, determine tax rates by looking in that very same 1040 instruction book. So, they calculated your 2001 taxes using the 15 percent rate -- concluding that you paid $300 more in taxes in 2001 than you actually did. They did not, however, add the refund check to your 2001 income. So where the government gave and then took away, leaving you even, the Census Bureau only saw them take the money away -- thus giving them the impression that you were $300 poorer, after taxes, than you really were that year. For couples, the figure is doubled, and the Bureau overestimated 2001 taxes by $600. For single heads of households, it was $500. Most households near the median in after-tax household income, (which is roughly $40,000, depending on which measurement is used), would have qualified for the entire amount of the tax cut. This means that the after-tax median income reported by the Census for 2001 is too high by somewhere between .75 percent and 1.5 percent.

The Census Bureau used the wrong marginal tax rates to calculate after-tax income data for 2001. As a result, the 2002 poverty report released a week ago contains numerous errors in its year-to-year comparisons. In particular, several measurements showed no change in median household income, when in fact there were significant declines. Other figures indicate that child poverty rates remained stable, when they actually may have risen.

The mistake, discovered by the Phoenix, has been verified by officials in the Census Bureau’s Housing and Household Economic Statistics Division, as well as by economists familiar with the data.

"The 2001 data is wrong," concludes David Betson, an economics professor at Notre Dame who served on a 1993 National Academy of Sciences panel on income and poverty measurement.

Bureau Division Chief Dan Weinberg says that if his statisticians verify the mistake "then over the next few months we will redo the data."

The data, released two weeks ago, included a variety of alternative measures of income and poverty, many of which depicted a less gloomy picture than the official numbers. Weinberg specifically made note of these measures when presenting the report. For example, when reporting that median household income had officially declined 1.1 percent, Weinberg added that "there was no statistically significant change for ... three comprehensive measures" of median household income.

However, those additional measures would have each shown an even higher rate of decline -- roughly 1.5 percent -- had the 2001 figures been correctly calculated.

Other alternative measures, of changes in statistics such as child poverty rates and income inequality, would also be affected. The Bureau reported that child poverty rates did not change significantly in 2002, but it is likely that the alternative measures will show an increase once the 2001 data is corrected.

The Census Bureau has reported the alternative measures for years, and many economists consider them better indicators of income and poverty than the official data. They account for several things the official numbers do not, including taxes.

The Census Bureau relies on surveys for its income information, and calculates taxes from those numbers.

In calculating taxes for 2001, the Bureau used 15 percent as the lowest marginal tax rate, rather than the correct rate of 10 percent, which went into effect that year. This resulted in the Bureau overestimating most taxes by $300 for individual taxpayers, $500 for heads of households, and $600 for couples. Accordingly, the estimates of after-tax income were low by the same amount.

The total amount of tax unaccounted for is roughly $39 billion, based on data from the Internal Revenue Service.

By one important definition cited by the Bureau in its report, median household income (adjusted for inflation) was $42,194 in 2001 and $42,061 in 2002, a statistically insignificant decline of 0.3 percent. However, adding back in the extra tax mistakenly taken out makes the correct 2001 figure approximately $42,700, meaning that median household income declined by 1.5 percent in 2002, by this measure.

The Phoenix obtained the portion of the Bureau’s computer program code for calculating 2001 alternative measures that contains the marginal tax rates; it shows .150 (15 percent) as the lowest marginal rate. Census Bureau statistician Ed Welniak of the Income Statistics Branch confirms that the 10 percent figure was incorporated only in 2002. The IRS confirms that the correct marginal rate for 2001 was 10 percent.

The Bureau’s computer code also shows that it failed to adjust three other marginal tax rates that were changed under the 2001 law. The code lists the additional rates as .280, .310, .360, and .391. The correct figures for that year are .275, .305, .355, and .391. "Those figures were incorrectly entered into that program for 2001," Welniak concedes.

As a result, the Bureau’s figures all use at least a $300-per-taxpayer greater tax burden for most people than they actually paid that year.

Arloc Sherman, poverty analyst with the Children’s Defense Fund in Washington, D.C., believes that the error explains some unexpected results in the Census data -- particularly, why certain alternative measures seemed to respond differently to the economy than the official figures did. He also suspects that close analysis might reveal further problems stemming from the extensive tax changes. "I know the [Census] division was confused about how to use the 2001 tax cut in their modeling."

The mistake may have resulted in part from the Bureau failing to realize how the Internal Revenue Service accounted for the $300-per-taxpayer "refund" sent out in the summer of 2001. This was actually a credit, or "advance payment check" against the taxpayer’s 2001 return, representing the difference between the old and new marginal tax rate. (See "Follow the Money.")

To recoup the advance payments, the IRS published the 2001 tax charts using the 15 percent rate. The first 10 percent is the tax, the other five percent is a "rate reduction credit," according to the IRS. (For further explanation, see box.)

Welniak says that the Bureau relied on the published IRS 2001 tax rate tables, taken from publication 17, the instruction booklet that accompanies 1040 tax forms.

"I think they just did what they always do, and it was the wrong thing to do that year," Sherman says.

The effect would have been offset if the "refunds" were accounted for by the Bureau as 2001 income. The Bureau now argues that this was why it used the 15 percent rate. "We noticed that people did report the tax credit as income. If we were to incorporate the 10 percent rate, it would be counted twice," says Welniak.

However, it appears that very little, if any, of that money was reported as income in the survey.

The Bureau collects income data through its Annual Demographic Survey, and did not specifically ask people about the "tax refund" in its 2001. Survey respondents would have had to include it in one of the Bureau’s 17 main income categories, but examination of the demographic survey data shows no category in which substantially more respondents reported income in 2001 than in 2000.

In fact, the Bureau’s survey takers are instructed not to include rebates and tax refunds as income.

Welniak suggests that some people included it as "other income" or as some form of government transfer payment. However, fewer people reported "other income" on the survey in 2001 than in the previous year. Although 7.5 percent more people reported government transfer payments in 2001, that increase came almost entirely in the categories of unemployment compensation and social security.

The federal government mailed out an estimated 91.6 million of these checks, worth a total of $39 billion, or an average of $426 per check, according to IRS figures. Some households received more than one check.

It is impossible to determine whether some respondents may have included the refund under "earnings." But that category specifically asks for wages, salary, and self-employment earnings. It is unlikely that most people would think to add the government check there.

The Bureau calculated all respondents at the mistakenly high rate, but probably accounted for very little of the "refund" money as income.

Estimating the exact effects of this error is difficult. The general effect, however, would be twofold, changing both the 2000-to-2001 and the 2001-2002 changes in income and poverty rates. It would show that the 2.2 percent decline in pre-tax income during the 2001 recession year was mostly offset by the tax cut. For the year following the recession, however, it would show a large drop in after-tax median household income, suggesting that the "jobless recovery" has had a greater negative effect on households’ ability to spend and save money.

The seasonally adjusted unemployment rate rose sharply in 2001, from 3.9 to 5.8 percent, and then remained at that level throughout 2002, ending at 6.0 percent. The latest figure, released today, is 6.1 percent for September 2003.

The effect on other figures, such as poverty rates, is more difficult to predict, Betson says. For example, it is hard to estimate how many households on the cusp of the poverty line paid enough taxes to be affected by the error.

The error may also have worked its way into many other databases that rely on the Census Bureau figures. "It probably affects our own analyses," says Sherman of the Children’s Defense Fund.

David S. Bernstein can be reached at dbernstein[a]phx.com