Statistical Analysis

We conducted structural equation modeling in Mplus 7.4. Prior to each analysis, the model’s assumptions were tested. The data met statistical assumptions for multivariate normality, linearity, and non-multicollinearity. Table 3 displays Pearson’s product moment correlations, means, and standard deviations for study variables. We first conducted a confirmatory factor analysis (CFA) to test the model fit of the anxiety and depression latent variables. A structural model was then examined that included the three sleep items as mediators, insomnia, unrestful sleep, and TST. The model included paths from anxiety at W1 predicting insomnia, unrestful sleep, and TST at W2, to depressive symptoms at W4 as the outcome variable. Anxiety was modeled as both a direct and indirect predictor of depressive symptoms through insomnia, unrestful sleep, and TST. W1 Depression was included as a predictor of W4 depression to control for the influence of prior depression. Gender, SES, age, and health were also included as controls. Robust maximum likelihood (MLR) estimation method was used as it provides standard errors robust to unmodeled nonnormality of data (Muthén & Muthén, 2007). Full-information maximum-likelihood (FIML) estimation was used to estimate missing values (Arbuckle, Marcoulides, & Schumacker, 1996).

Pearson’s product moment correlations, means, and standard deviations for relevant study variables.

Practical indices of goodness of fit were evaluated in place of the chi-square statistic, as chi-square values are highly influenced by large sample sizes. Model fit was evaluated based on the following criteria (Kline, 2015): Tucker-Lewis Index (TLI) ≥ .90, with > .95 preferred, Root Mean Squared Error of Approximation (RMSEA) and Standardized Root Mean Square Residual (SRMR) < .05. If model fit was less than acceptable, modification indices were examined for correlated error terms that were conceptually justified and would enhance model fit.