As an integral part of the U.S. Census Bureau’s commitment to making changes to the design of the decennial census in order to meet the strategic goals and objectives for the 2020 Census, the 2015 National Content Test (NCT) served as the primary opportunity to test content on a nationally representative sample. The NCT used a 1.2 million housing unit sample to test content modifications, different contact strategies designed to optimize self-response, and different approaches to offering in-language materials. The focus of this report is on the relationship content testing that was conducted.

The NCT relationship content testing is the latest in the Census Bureau’s efforts to improve the estimates of coupled households. The 1990 Census first introduced “unmarried partner” as a response category to the relationship-to-the-householder question. The 2000 and 2010 Censuses built on this work by changing the processing of the relationship question to more accurately represent same-sex couples. However, the 2000 and 2010 Censuses tabulated counts of same-sex couple households by analyzing responses to the questions of sex and relationship rather than featuring a relationship question with response categories that explicitly distinguish between same-sex and opposite-sex spouses and partners. This collection method is prone to errors from opposite-sex couples mismarking their sex, and in 2010, the Census Bureau released a set of modified state-level same-sex household estimates. The revised relationship question tested in the NCT, one that features explicit same-sex and opposite-sex spouse and partner response categories, has the potential to improve same-sex couple estimates.

Two relationship question versions were featured in the NCT: a slightly modified version of the 2010 Census question (used in the control panel) and a new version where the two couple relationship categories were expanded to distinguish between opposite-sex and same-sex relationships (used in the experimental panel). Item nonresponse rates and relationship response distributions were the primary analytical measures used.

Replacing the relationship response options “husband or wife” and “unmarried partner” with “opposite-sex husband or wife,” “same-sex husband or wife,” “opposite-sex unmarried partner,” and “same-sex unmarried partner” did not result in statistically different item nonresponse rates for the relationship question overall. In addition, no significant differences were found within each mode.

The use of the new categories was not associated with a significant difference in the distribution of coupled household categories. No differences were found when looking at distributions by mode.