Demographics are presented in . The participants in our sample range in age from 14 to 43 years with an average age of 24.6 years. Many participants were of lower socioeconomic status, with nearly half the participants (44%) making <$25,000 annually, and a third (34.3%) with less than a high school education. The majority of our sample identified as Mexican or Mexican American.
We fit a regression model using EOD frequency, acculturative stress, Anglo marginality, Mexican marginality, Mexican American marginality, age, and SES variables (income and education) as predictors of depression (BDI). Prior to fitting the model, we visually inspected the distribution of the dependent variable and, due to a moderate positive skew, we transformed it with a natural log. We assessed the linearity of independent variables by fitting quadratic models for each variable individually. Log transformations were used to transform the education, EOD, and the pressure to acculturate variables to linearize the relationship between the independent and dependent variables.31
Because the log-transformed education variable was still curvilinear after the log transformation, we fit a polynomial education effect by first mean centering the variable to reduce collinearity with the quadratic term, and then squared the variable to form the quadratic value. We divided income by 1000 to aid interpretation of coefficients. We assessed multicollinearity of all variables in the model using the variance inflation factor (VIF). The largest VIF that we observed was 2.45 for Mexican American marginality, which was well below the recommended cutoff of 10.31
After fitting models, we assessed the normality of residuals using quantile-quantile (Q-Q) plots and assessed the homogeneity of variance by plotting fitted values and residuals in a scatter-plot. The Q-Q plots adhered to a straight line and the fitted values by residuals scatterplot showed that there was no relationship between the fitted values and residuals indicating homogeneity of variance of the residuals. We examined the number of cases that were missing under listwise deletion in ordinary least squares regression and determined that 43 cases would be excluded due to missing data for at least one of the variables in the model. Post hoc power for the entire model, assuming a two-tailed alpha of .05 is > .99; R-Squared = .10. The results of this model are displayed in . (R-squared is not available for models from multiply imputed data and we therefore used the model with listwise deletion.)
Regression model for predictors of depression N = 515a
illustrates the curvilinear nature of the relationship between education (measured in number of years) and BDI (measured in units) scores and demonstrates that BDI increases when the numbers of years of education are 10–12; however, BDI totals decrease after 12 years of education.
BDI is values are the logged value of the participants’ total BDI score. Education values are the number of years of education the participants had completed.
Age, EOD frequency, and Anglo marginalization were significant predictors of depression in the model. In the regression model, age exhibited a significant negative relationship with BDI (t (501) = −2.88, P = .004), EOD frequency exhibited a significant positive relationship with BDI (t (502) = 5.44, P <.001), and Anglo marginalization exhibited a significant positive relationship with BDI (t (503) = 3.56, P <.001) and as such were significant predictors of depression in the model. The quadratic education parameter was significant (t (499) = −2.12, P = .035), though the linear component was not (t (494) = −0.10, P = .920), indicating a curvilinear relationship between education and BDI. The relationship is displayed in fitted values in , which demonstrates that BDI increases until approximately 12 years of education, then decreases. We show that the relationship between age and depression is negative. There was not a significant relationship between income (t (452) = −0.48, P = .631), pressure to acculturate (t (500) = 0.06, P = .952), Mexican American marginality (t (503) = −0.78, p = .435), and Mexican marginality (t (503) = 0.18, P = .859). Discrimination, Anglo marginality and education had the overall greatest impact on depression in this model.