Statement of Context
A post-publication statistical review was initiated. The reported findings from the additional analysis and review remain unchanged from the original manuscript.
Additional Analysis Description
Upon initiation of the post-publication statistical review, an additional analysis using normality assessments, non-parametric hypothesis testing, and power analysis was conducted. The Shapiro-Wilk method and visual assessment of histograms were used to assess the normality of distributions for all five subscales of the Professional Quality of Life (ProQOL) measure. In our original analysis, paired t-tests were conducted; unlike an independent-sample t-test that assumes the observations themselves are normally distributed, paired t-tests assume that the pre- and post-score differences are normally distributed between pairs (McDonald, 2014). For each subscale, the differences demonstrated normal distributions (Compassion Satisfaction: p = 0.091; Perceived Support: p = 0.221; Burnout: p = 0.063; Secondary Trauma: p = 0.347; Moral Distress: p = 0.479). Although normality was suggested, some of the subscales trended towards non-normal distributions (Compassion Satisfaction and Burnout), so we continued the additional analysis to include non-parametric hypothesis testing for thorough evaluation.
Because tied values can affect rank-based tests, Wilcoxon signed-rank analyses were examined using standard normal approximation methods. Results were consistent across analytic approaches, and all subscales remained non-significant. For transparency and consistency with the original analyses, results from the unaltered dataset are reported here. Finally, we consulted with a post-doctoral researcher in our department to assist with conducting a power analysis for a paired t-test With a sample size of 22, a significance level of .05 and a power of .80, the study is powered to detect an effect size of approximately d =.63. Since we did not report any significant findings, this power analysis is acceptable for the current study. Additionally, we noted that an increased sample size would allow for us to control for potentially confounding variables, such as demographic or workplace factors, which may influence changes in professional quality of life over time.
Impact on Findings
Upon completion of the additional analysis, the results did not change the original conclusions of the study.
Considerations for Small Sample Sizes
Due to the relatively small nature of the child life profession, small sample sizes are common among child life related data sets. According to the Central Limit Theorem, larger samples (n ≥ 30) can use parametric hypothesis testing, such as paired t-tests, to approximate normal distribution in measurements. However, for smaller sample sizes, parametric tests must meet assumptions about normality to ensure accurate hypothesis testing that excludes false positive results (Type I errors) and about statistical power to exclude false negative results (Type II errors). Non-parametric testing, such as the Wilcoxon Signed-Rank test, is especially beneficial for ensuring results from non-normal distributed scales do not demonstrate Type I errors (Meek et al., 2007). As our initial results were non-significant, there should be no chance for false positive results, as reflected in our Wilcoxon Signed-Rank test results; we likewise did not see strong evidence for false negative results after evaluating statistical power for each subscale. Nonetheless, normality, non-parametric testing, and power analysis should be reported for smaller sample size, and conferring with a statistician for quantitative work is recommended. This ensures that the analyses are accurate and provide the audience with clear, robust information that progresses the field forward.