“… - Five additional Bayesian meta‐analyses using different weakly‐informative priors for μ (effect size) and for τ (heterogeneity), following the narrowing strategy suggested by Korner‐Nievergelt et al 48 Additionally, we also tested different scenarios for priors showing greater variance for the parameter τ (between‐study heterogeneity), to investigate the potential effect of having greater levels of variance across studies. Models with alternative weakly‐informative priors were, therefore, computed as (a) μ ~ N(0,1.5), τ ~ half‐Cauchy(0,0.5); (b) μ ~ N(0,1), τ ~ half‐Cauchy(0,0.5); (c) μ ~ N(0,1.81), τ ~ half‐Cauchy(0,1); (d) μ ~ N(0,1.5) and τ ~ half‐Cauchy(0,1); (e) μ ~ N(0,1) and τ ~ half‐Cauchy(0,1);
- Bayesian meta‐analysis using an uninformative flat prior for μ (a prior with the precision = 0 and variance = ∞), and a very weakly‐informative prior for τ (τ ~ half‐Cauchy(0,0.5)), to investigate a more non‐informative/improper scenario for priors;
- Additional meta‐analysis excluding a study identified as an outlier for reporting a higher prevalence of PD based on a small sample 29 ;
- Finally, a frequentist meta‐analysis of the same 20 studies included in our main meta‐analysis was also computed, using a random intercept logistic regression model. The analysis used the Clopper–Pearson confidence interval for individual studies, and the maximum‐likelihood estimator for the between‐study heterogeneity parameter tau‐squared (τ 2 ).
…”