Page 74 - Never Too Far Away? The Roles of Social Network Sites in Sojourners’ Adjustment

Cross-lagged path analysis. Table 2 provides a summary of the unstandardized estimates of the paths in all the models we tested. We only presented the values between T1 and T2 since we imposed equality constraints across the waves (i.e., the values between T1 and T2, T2 and T3 are the same).3 Based on cross-lagged path analysis (Model 1a), the model had an adequate fit to the data (χ2(85) = 105.21; p = .031; χ2/df = 1.32; RMSEA = 0.02; CFI = 0.96; TLI = 0.95. The results showed that only the lagged causal path from Facebook interaction with the host-country network to homesickness was significant. Specifically, Facebook interaction with the host-country network decreased homesickness at a later time point (b = -0.14; SE = 0.06; p = .021, two-tailed) (RQ2). The cross-lagged paths between homesickness and sociocultural adjustment were not significant, thus H1 and H2 were rejected.
Non-lagged path analysis. We proceeded by testing the proposed model using non-lagged path analysis (Model 2). The results showed that Model 2a had a good fit to the data (χ2(84) = 106.56; p = .049; χ2/df = 1.27; RMSEA = 0.02; CFI = 0.96; TLI = 0.96). The results indicated that Facebook interaction with the host-country network decreased homesickness in the short-term (b = -0.24; SE = 0.11; p =.026, two-tailed). The reverse effect also held, homesickness increased Facebook interaction with the host-country network in the short-term (b = 0.16; SE = 0.08; p = .047, two-tailed) (RQ5). Moreover, homesickness decreased adjustment in the short-term (b = -0.09; SE = 0.05; p = .040, one-tailed) (H1). In this model, H1 was supported and H2 was rejected.
Model tests controlling for length of stay. We tested both cross-lagged and non-lagged models with the control variable length of stay included. For the cross- lagged analysis, the model (Model 1b) showed a good fit to the data (χ2(85) = 106.02; p = 0.061; χ2/df = 1.25; RMSEA = 0.02; CFI = 0.97; TLI = 0.96). For the non-lagged analysis, we also found that the model (Model 2b) had a good fit to the data (χ2(84) = 106.56; p = 0.049; χ2/df = 1.27; RMSEA = 0.02; CFI = 0.96; TLI = 0.96). The pattern of the main results was similar in both the cross-lagged and the non-lagged models with and without the control variable. Moreover, both in the cross-lagged and non-lagged models, length of stay did not have significant effects on the main variables. Figures 2 and 3 summarize the significant results
3 To ensure that the assumption of constancy of structural effects is justified, we ran all the models again without equality constraints. Using χ2diff test, the results showed that the models with equality constraints did not have a significant difference in fit to the data compared to the models without equality constraints.
72 Chapter 3