for race/ethnicity in the base model after excluding all cases missing on poverty. The new estimates based on this truncated sample did not affect our results for black-white differences. That is, poverty still appeared to partially account for the black-white gap in the risk for dependence symptoms. However, the Hispanic-white difference was no longer explained, in part, by poverty because the newly estimated Hispanic-white difference was smaller, and changed little when poverty was included in the model (new AOR=2.72 in base model, AOR=2.36 when poverty was included). Additional sensitivity analyses were performed to assess whether any other results were affected by missing data, and showed that no other results were affected.
