Computational protocol: Cross-sectional relationship of perceived familial protective factors with depressive symptoms in vulnerable youth

Similar protocols

Protocol publication

[…] We calculated frequencies and tested patterns of the missing values. The overall frequency of missing values was low (77.8% of respondents with less than 5% missing data). The missing values were either completely missing at random or missing at random.Descriptive statistics (mean and proportions) were calculated, and the data were checked for normality. We analysed the construct validity of the familial protective scales and depressive symptoms through confirmatory factor analysis (CFA). As advised by Cole (2007), we allowed correlations between residual terms of the attachment to the mother and father scales, that are implied by the measurement strategy, which are the equally expressed items of the mother and father [].We performed structural equation modelling (SEM) to investigate the relationship between the family protective factors and depressive symptoms in AIA using the SDM.First, we tested the measurement invariance of the SEM multigroup modelling with respect to sex. Configural invariance is present if the factor structure, loading pattern, and intercepts are similar in both groups []. For testing weak invariance, the factor loadings are set equal across groups. If this models proves the stage, structural relationship between groups, such as factor correlations, can be examined and compared across groups []. Strong invariance is tested by additionally constraining the intercepts to be equal across groups. Confirmation allows a comparison of latent means and regression parameters between groups [–]. The nested model is compared with the previous less restricted model by a χ2 difference test. As noted by Chen (2007), χ2 differences have the same problem as absolute χ2 tests by being highly sensitive to the sample size and violations of the normality assumption. Therefore, goodness-of-fit statistics are recommended. When the sample size is adequate (n > 300), a change of the Δ comparative fit index (CFI) ≤ − 0.010 supplemented by a change in the Δ root mean square error of approximation (RMSEA) ≤ 0.015 indicates invariance []. An analysis of partial measurement invariance is possible, if the nested model is worse than the previous model. Partial invariance is present if at any of the restrictions of the aforementioned stages are freed for some indicators to improve the model fit []. If at one stage, partial invariance is present, then this partial model is the basis for the next step of assessing measurement invariance [].To analyse the differences in path coefficients with respect to sex, we performed model comparisons by using the χ2 differences of the restricted model (equal constraint regressions) and the model without the restriction. For model comparisons, we used the Satorra–Bentler-scaled χ2 difference test using difference test scaling correction and the differences in the degree of freedom [].We applied the maximum likelihood estimator with robust standard errors (MLR) to obtain appropriate fit indices []. We used the CFI, Tucker–Lewis Index (TLI), and RMSEA to evaluate the model fit. Furthermore, we checked the standardised root mean square residual (SRMR < 0.1). The following parameter estimates and goodness-of-fit statistics describe an acceptable model fit; CFI and TLI ≥ 0.90, RMSEA ≤0.08. Moreover, a good model fit is represented by the following estimates; CFI and TLI ≥ 0.97 and RMSEA ≤0.05, which were used to evaluate the model data [].For the descriptive and multivariate analyses, we used SPSS Version 22 (IBM Corp. Released 2013. IBM SPSS Statistics for Windows, Version 22.0. Armonk, NY: IBM Corp). For the CFA, SEM, and measurement invariance tests, we used R (Version 3.2.4) with the missForest, semTools and lavaan packages (0.5–20) [, , ]. […]

Pipeline specifications

Software tools SPSS, lavaan
Application Miscellaneous
Organisms Homo sapiens
Diseases Alcoholism, Substance-Related Disorders
Chemicals Ethanol