Selective Data Analysis in Brown et al.'s Continued Critical Reanalysis

In their latest critique [1], Brown et al. verify the primary statistical results of our 2015 PLoS ONE report [2]. The results Brown et al. report for their mixed effect linear model analyses of our Confirmation study and pooled Discovery and Confirmation studies in their Table 3 [1] are nearly identical to the results we reported in our Tables 2 and 3 [2]. Nevertheless, Brown et al. continue to dispute the conclusions that follow from these results. They do so by selectively re-analyzing our Discovery study dataset (N = 76), which represents only 25% of the data presented in our 2015 report. Using this approach, Brown et al. argue that the relationship between eudaimonic well-being and gene expression is sensitive to (1) the inclusion vs. exclusion of a single data case (SOBC1-1293), and (2) the effects of a coding error in the originally posted covariate data for another data case (SOBC1-1299). However, analysis of the full set of data presented in our Discovery and Confirmation studies (N = 198) reveals that the association of eudaimonic well-being with gene expression is not materially affected by either of these factors (see Table 1 herein). The mixed effect linear model analyses reported in Table 1 account for correlation among the multiple indicator genes examined [3] and continue to indicate a significant inverse relationship between eudaimonic well-being and gene expression, regardless of SOBC1-1293 exclusion or the SOBC1-1299 coding error. (Because SOBC1-1293 and SOBC1-1299 come from the Discovery study sample, they have no effect on analyses of the Confirmation study dataset alone [N = 122] or the Generalization study dataset [N = 107].) The Discovery study sample alone is too small to provide a well-powered mixed effect linear model analysis. Thus, it is unsurprising that Brown et al.’s Table 4 [1] shows non-significant regression coefficients for eudaimonic well-being and point estimates that vary substantially from those of the better-powered analyses of the Confirmation study and the pooled Discovery and Confirmation studies (reported in our Tables 2 and 3, respectively [2], and Brown et al.’s Table 3 [1]). This discrepancy in statistical power between Brown et al.’s selective reanalyses (reported in their Table 4) and a more complete analysis (replicated in their Table 3) is evident in the larger Standard Errors (SE) in their Table 4 versus Table 3 [1]. In their previous critique of our 2013 report [4] on gene expression correlates of well-being, Brown et al. [5] argued for the replication of findings in additional samples using mixed effect linear model analyses. Such data are now available from two new samples with 229 new participants, and results continue to indicate a significant inverse relationship between eudaimonic well-being and gene expression. Brown et al.’s claims of statistical instability rely on selective