Page 72 - Never Too Far Away? The Roles of Social Network Sites in Sojourners’ Adjustment

at random (MCAR) test, which indicated that the missing data can be considered as completely at random , χ2(57, N = 414) = 57.24, p = .466. Thus, it can be assumed that there were no serious selection problems due to attrition.
Table 1 summarizes the means, standard deviations and correlations of the study variables. Across the three time points, the rank order of the four types of interaction in terms of frequency was the same, with FtF interaction with the host-country network being the most frequent, and Facebook interaction with the home-country network being the least frequent.
Testing the proposed model using SEM
We tested the proposed model using cross-lagged (for the long-term effects) and non-lagged (for the short-term effects) reciprocal causality path analyses. The non-lagged reciprocal causality analysis is used to investigate causal effects that occur within a short span of time (Finkel, 1995; Kline, 2016). The use of panel designs does not necessarily prove causality as conclusively as experimental designs. However, they are useful to estimate reciprocal effects and assess whether a set of results is consistent with a causal model (Finkel, 1995).
To test the models, we conducted structural equation modeling (SEM) using Stata 14 and the estimation method maximum likelihood for missing values (MLMV). We used the observed variables for the three types of interaction and homesickness, as well as the composite mean for sociocultural adjustment to reduce the number of parameters to be estimated and ensure statistical power. We also used z-score standardization to ensure the comparability of the coefficients across the variables. Since the panel waves were equally spaced and the participants were in varying stages of their sojourn in T1, we estimated the models by placing equality constraints on the parameters (Finkel, 1995). We assumed that “the changes in the underlying reciprocal causations have already manifested their effects and that the system is already in a steady state” (Kline, 2016, p. 137). Firstly, with the cross-lagged model, we put equality constraints on the corresponding autoregressive paths from T1 to T2 and from T2 to T3; the corresponding cross- lagged causal paths across the waves, and the corresponding residual variances between variables within waves. Then, with the non-lagged model, we removed the cross-lagged paths and replaced the correlations within waves with reciprocal causal paths (please see Mathisen et al., 2007 for the application of similar procedures). Thus, within each wave, the directions of influence are clarified.
70 Chapter 3