Page 50 - Never Too Far Away? The Roles of Social Network Sites in Sojourners’ Adjustment

within waves to be equal. Then, for the non-lagged model, we removed the cross- lagged paths from the previous cross-lagged model, and replaced the correlations of the variables within waves with reciprocal causal paths. These reciprocal causal paths within waves (represented by two distinct reciprocal arrows in the model) disentangle the correlational association (single bidirectional arrow) within waves in the cross-lagged model. Thus, the directions of influence in the short-term are clarified (see Mathisen et al. (2007) for the application of similar procedures).
Evaluation of the cross-lagged model. Table 2 provides a summary of the effect sizes (unstandardized estimates of the paths based on z-scores) in all the models we tested. Only the values between T1 and T2 are presented because we imposed equality constraints across the waves (i.e., values between T1 and T2, T2 and T3 are the same). We first tested a cross-lagged model without the control variables (Model 1a). The results showed that Model 1a had a good fit to the data (χ2(80) = 96.49; p = .101; χ2/df = 1.21; RMSEA = .02; CFI = .98; TLI = .97). To check if the assumption of constancy of structural effects is warranted, we then tested Model 1a where we relaxed the equality constraints. We found that Model 1a without the equality constraints did not represent a statistically significant improvement in fit to the data (χ2(55) = 70.48; p = .078; χ2/df = 1.28; RMSEA = .03; CFI = .98; TLI = .97) compared to Model 1a with equality constraints (χ2diff = 26.01; dfdiff = 25; p = 0.407); thus, the assumption of constancy of structural effects was acceptable. The results showed that perceived social support predicted higher Facebook interaction with the host-country network at a later time point (RQ1b). Also, perceived social support predicted lower depressive symptoms at a later time point (b = -0.10; SE = 0.05; p = .030, one-tailed), supporting H2.
We then tested a cross-lagged model with the control variables gender and length of stay included (Model 1b). Model 1b showed an adequate fit to the data (χ2(90) = 113.99; p = .045; χ2/df = 1.27; RMSEA = .03; CFI = .97; TLI = .96). We found that Model 1b with and without equality constraints (χ2(65) = 87.64; p = .032; χ2/df = 1.35; RMSEA = .03; CFI = .97; TLI = .95) did not differ significantly in terms of model fit (χ2diff = 26.35; dfdiff = 25; p = .389); thus, the assumption of constancy was acceptable. When gender and length of stay were accounted for, the coefficients showed that Facebook interaction with the host-country network lowered perceived social support at a later time point (RQ1a); and higher perceived social support increased Facebook interaction with the host-country network at a later time point (RQ1b). We also found that greater perceived social support decreased depressive symptoms at a later time point, supporting H2. Regarding
48 Chapter 2