Ridderprint

Chapter 5 106 Table 5.11: Random effects models estimating correction for publication bias Before correction After correction k N r ρ se CI95 k' r ρ se CI95 Adjustment 60 12506 .21 .25 .02 [ .21, .29] 60 .21 .25 .02 [ .21, .29] Commitment 18 3549 .28 .33 .03 [ .27, .40] 18 .28 .33 .03 [ .27, .40] Performance 33 5703 .16 .19 .02 [ .15, .23] 39 .19 .22 .02 [ .18, .27] Retention 27 6167 .18 .21 .02 [ .16, .25] 32 .16 .17 .02 [ .13, .22] Note: k = number of independent studies; N = total independent sample size; r = estimated correlation in the population; ρ = estimated true correlation in the population; se = standard error of the true correlation; CI95 = 95% confidence interval of the true correlation; k’ = estimated number of independent studies after including estimated missing studies. 5.5.3 Results Summary Figure 5.3 and Table 5.11 summarize the results regarding the direct effects found in Tables 5.2 through 5.6. The network in Figure 5.3 was generated with Kamada and Kawai’s (1989) graph drawing algorithm. This algorithm seeks to minimize the total length of its edges (i.e., relationships) by cleverly placing nodes (i.e., constructs). Moreover, edges were weighted for the inverse effect sizes reported in Table 5.11, so that stronger relationships were resembled by shorter, broader edges. All this implies that, in Figure 5.3, social support constructs were located closer to success criteria they related to more strongly (and vice versa), and that social support constructs and success criteria that have similar relationships were automatically co-located. Figure 5.3. The agents/types of social support and their direct relationships to the success criteria as reported in Table 5.11, colored for their domain. No arrow indicates k < 2.