This article has been updated after a reader discovered and brought to the authors’ attention a coding error with a control variable. Add Health codes two responses for declining to report income as 9999998 and 9999996. In the original analyses, these responses were not treated as missing. This authors have corrected the error and rerun the analyses, which are presented in the current version.

The most consequential results of this correction are as follows:

• The number of respondents in the analytical sample changed from 10,721 to 10,281 (p. 674)
• The descriptive statistics in Table 1 have changed due to the change in sample size. In addition, due to the recoding of income for some respondents to missing, the descriptive statistics for income at the bottom of the table have changed.
• The correlations reported in Table 2 have changed.
• Point estimates for coefficients and standard errors for the models reported in Table 3-6 have changed. These changes generally do not involve changes in statistical significance, and are generally not consequential for the interpretations offered, with the following exceptions:
  • In Table 5, the effect of “GLB” in the “All other race” column is no longer significant at the p<.05 letter. As a result, the text on p. 679 has been changed to say "GLBs are estimated to select four [previously: five] of six friendship network composition categories over “all your race" at significantly higher rates than are heterosexuals."
  • In Table 6, the effect of “Less than High School” in the “Almost all other races” column is no longer statistically significant.

The originally published version of the paper is available here.