Users Beware: Variable Effects of Parenthood on Happiness Within and Across International Datasets

As a robustness check, we specified two other conversions of 4-point ISSP happiness variable to the 11-point ESS format (2.75, 5.5, 8.25, 11 and 0, 3.33, 6.67, 10). Doing so produced substantively identical findings for the random-effects analyses (with some differences in the lower and upper population variance and coefficient confidence limits, but all three variance components for parenthood remaining significant and overlapping). We therefore opted for the 2.5, 5.0, 7.5, 10.0 format because it provided even numerical spacing between choices while also creating a minor distinction between bottom categories (i.e., the bottom category for the ISSP refers to ‘not at all happy’ whereas the bottom category for the ESS refers to ‘extremely unhappy’—the ISSP bottom category does not seem as subjectively severe as the ESS bottom category).

Indeed, random-effects modeling using the original rather than the transformed happiness variable led to substantively identical conclusions. Thus the transformed variables are used in the interest of greatly improving overall model fit.

.

The ESS, like the ISSP, contains information on whether there are multiple children in the household. Thus we could have broken down children’s effects on happiness by number of children in the household. However, because the current argument rests on the issue of within- and between-dataset variance, we constrained the analysis simply to the presence or absence of children.

The ISSP and ESS both provide weights corresponding to the probability of selection. These weights were standardized within countries so that within-country means equaled 1. Analyses presented in this paper are unweighted, but weighted fixed- and random-effects analyses led to substantively identical conclusions in that weighted random-effects analyses (using xtmixed with Level-1 probability weighting in Stata 12) produced significant and overlapping effect variances at the cross-sectional, dataset and country levels, and weighted fixed-effects analyses revealed similar patterns of within- and between-method instability within countries. Mean country happiness fixed-effects rank-orders remained same for ISSP 2007 except for flipping of Belgium and UK; ISSP 2008 remained same; ESS 2006 remained same except for flipping of France and Germany and mutual interchanging between Ireland, Sweden and Germany; ESS 2008 remained same except for flipping of Belgium and Germany. Parental happiness fixed-effects rank-orders shifted somewhat due to a tendency for coefficients to cluster closer together in weighted analyses. All rank-order shifts (unweighted analysis → weighted analysis) were within three ranks except for Russia’s shift in ESS 2008:

Visual inspection of Table 3 reveals that the ESS 2006 estimate for Poland is highly aberrant. Exclusion of this cross-section from the multilevel random-effects model leads to some differences in the lower and upper population variance and coefficient confidence limits, but all three variance components for parenthood remain significant and overlapping.