Marginal effects of question comprehension from weighted probit models of generalized trust for blacks and whites.

Notes: Marginal effects were estimated with all other covariates held at their means. Robust standard-errors are reported in parentheses. Significance levels: †10%, *5%, **1%, ***0.1%. Individual controls: gender, age, age-squared, a language dummy, education dummies, marital status dummies and log of real household income. Black respondents oversampled in 1982 and 1987 were excluded.