Ridderprint

Chapter 5 98 5.4.3 Coding We retrieved 852 effect sizes relating social support to success criteria, which the first author and another research assistant coded for several characteristics. Social support constructs were categorized on two levels – domains and agents – and a third level for organizational practices (family domain [spouses], community domain [HCNs, expatriates], work domain [peers, supervisors, mentors, organization]). For agents, we recorded their physical locations (home; host) and mobility status (HCN; expatriate). Additionally, for support by organizations, we distinguished between cross-cultural training, logistical support (i.e., assistance with previews, relocation, spousal employment, healthcare and education), repatriation support, and perceptions of overall organizational support and its dimensions (i.e., adjustment, career, and financial). Similarly, success criteria were categorized on two levels (adjustment [general, interaction, work], commitment [affective, normative, continuance], performance [task, extra-role], retention [in assignment, in country, in organization]). Moreover, we recorded measurement reliabilities and raters (self, spouse, supervisor, peer, hybrid, other). Several sample- and document-level characteristics were recorded as well but not reported consistently enough for analysis (e.g., sector, gender ratio, assignment duration). The estimated interrater reliability was 93.8% for the effect size characteristics, ranging between 83.3% (agent’s location) and 99.8% (success criteria [level 1]), and all inconsistencies were solved in tripartite with the second author. 7 5.4.4 Meta-analytical Procedure Because all effect sizes were correlation coefficients, we followed Hunter and Schmidt (2004) and estimated the mean true score correlation (ρ) by correcting for measurement error. If no measurement reliability (Cronbach’s α) was reported, we used the average reliability of that specific construct across all included samples. In case of reversely scored scales (e.g. family problems, turnover intentions), correlations were recoded so that a positive value indicates a positive relationship between support and success. One outlier, representing a perfect true correlation between family support and success, was found using schematic plot analysis and excluded from the final sample. Because not all correlations were statistically independent, we calculated one synthetic effect size per sample, using adjusted weighting to minimize bias in heterogeneity estimates (Cheung & Chan, 2004, 2008). Random effects models were conducted in R (R Core Team, 2017) using the metafor package (Viechtbauer, 2010). Random effects allow inferences regarding participants and contexts different from those included in the samples (Field, 2001; Hedges & Vevea, 1998) and therefore more appropriate if we seek to generalize results to the global business expatriate population. However, we do note that the statistical power to test moderator hypotheses is low in random effects models (Hedges & Pigott, 2004). For each support- success relationship, we estimated the mean effect size across studies, its confidence and credibility intervals, and several heterogeneity statistics. A 95% confidence interval (CI) 7 An example of the coding sheet can be provided upon request.