Harris School’s Dan Black finds flaws in how U.S. Census Bureau counts same–sex couples

A lack of clarity in how the government counts same–sex couples could lead to an inaccurate tally of such partnerships in the 2010 Census, according to University of Chicago economist Dan Black.

Gay and lesbian couples living in committed relationships across the country had two options to choose from when filling out the 2010 Census: “husband/wife” or “unmarried partner.” At a time when the policy stakes are at an all–time high—from challenging limitations on gay adoptions to nationwide legalization of gay marriage—which box they checked could have serious policy implications.

Although five states currently allow gay marriage, the U.S. government will not legally recognize same–sex spouses, allowing no clear way to categorize them in the census. The Census Bureau’s attempt to still count them hinges largely on what procedure it uses to code its data. According Black, deputy dean in the Harris School of Public Policy Studies, if it employs the same strategy it used in 2000, the 2010 results could be erroneous.

In 1990, any same–sex couples who identified as “husband/wife” were interpreted by the Census Bureau as heterosexual partners, which automatically edited the gender of one spouse.

In an attempt to collect more accurate data on same–sex couples, the Census Bureau modified its interpretation of this data for the 2000 Census. Instead of editing the “husband/wife” responses and counting them as male/female partners, the bureau edited same–sex couples’ marital status responses, changing them from married to “unmarried partner.”

This change was based on the bureau’s assumption that same–sex couples who identified as married were making a personal or political statement. By unofficial estimates, that coding change, in part, caused the number of unmarried same–sex couples to jump drastically from 145,000 in 1990 to almost 600,000 in 2000.

“The Census Bureau was responding to complaints that they were undercounting gay and lesbian couples,” said Black, who studies demography. “It was certainly a reasonable guess.” But when Black looked more closely at the data he found that this modification in procedure had only made the problem worse.

In his unpublished paper, “The Measurement of Same–Sex Unmarried Partner Couples in the 2000 U.S. Census,” Black and three colleagues found that more than 40 percent of respondents who identified as “unmarried partner” were not same–sex partners, but were heterosexual couples who incorrectly marked their gender.

To test this further, Black compared the census data to an alternative source of similar information and specifically looked at the number of couples with children, which are much lower in same–sex households. Using Current Population Survey data, Black estimated that 9 percent of same–sex male households and 19 percent of same–sex female households had children in 2000.

The edited 2000 Census data, however, indicated a much higher rate of households with children, with 23 percent for same–sex male couples and 35 percent for same–sex female couples. But when Black looked at the unedited responses of same–sex couples—those who checked the “unmarried partner” box to begin with—the tally was more in line with Current Population Survey predictions—8 percent for same–sex males and 20 percent for same–sex females.

The Census Bureau’s interpretations of and edits to same–sex couple responses suggest a drastic flaw in the process of tabulating data on this population. Among the edited responses, Black also found a disproportionate number of them were made by senior citizens and non–English speakers, two demographics that are more likely to commit mistakes on surveys.

“The problem is that the gay and lesbian population is relatively small when compared to the married population,” explains Black, who, since 2009 has been working on the Relationship Study Project offering recommendations to the Census Panel on Same–Sex Couples about reducing these mistakes. “The very small error rate in the number of married couples who made one error about sex on their census form turned out to cause a great deal of contamination of the sample.”

To be more precise, there were 2.7 million married couples who filled out a 2000 Census form. Given previous estimates that the sex–miscoding error rate on similar surveys is about 20 per 10,000, Black said it’s no surprise the number of same–sex couples are being misrepresented.

There’s no easy solution. Going back to the procedure used in 1990 is neither politically or statistically feasible. But in its attempt to reduce the likelihood of sex–coding errors in the 2010 data collection, the bureau retained its 2000 editing strategy. The bureau made only a few process and layout changes to the American Community Survey and 2010 Census. In the past, for example, if a form had multiple boxes checked (both “male” and “female” for the same person), the bureau would only count whichever box came first on the form. Today, ambiguous forms are now set aside for individual adjudication.

Understanding how exactly these changes will affect the eagerly awaited 2010 Census numbers is still unclear—they are not expected for release until 2013.

No matter how the 2010 census count adds up, for the first time in U.S. history, the number of same–sex couples who self–identified as married on their forms will be released to the public. High figures would be another boon to gay–rights efforts, which gained momentum recently when a federal judge overturned Proposition 8 allowing same–sex unions in California. Low numbers, on the other hand, would satiate opponents’ claims that the media has exaggerated the size of a nationwide movement.

“It depends on whether you think the number of gay and lesbian couples is important to whether you grant them rights to marriage,” Black says. “But it certainly could make either case stronger.”

—Steven Yaccino