Mp 16

Free-text comments on non-standard use of ROBINS-I were grouped into categories and summarized. For each review, we calculated the proportion of judgments in each RoB category and the range of categories used.
For reviews that did not report overall RoB judgments, inferred overall judgments were calculated by taking the highest judgment in an individual domain, in accordance with the ROBINS-I guidance.[4] (link) For reviews that reported both domain-specific and overall judgments, the inferred overall judgment was compared to the reported judgment to identify instances where they differed.
We considered the following explanatory variables as potential predictors of RoB judgments: methodological quality as assessed using AMSTAR 2, whether RoB assessment was performed in duplicate, whether the authors reported industry funding or competing interests, and whether the review included RCTs.
Since the RoB assessments within the same review were expected to be similar, multilevel regression was used. The overall (or inferred overall) RoB judgments were treated as an ordinal outcome, and a separate generalized ordered logit model was fitted for each predictor with a review-level random intercept. Results were presented as odds ratios and population-average marginal predicted probabilities. Analyses were carried out in Stata MP/16.1[23] using the gllamm command[24] .

Appropriate statistical tests were chosen to perform the response comparisons between groups; both parametric and non-parametric tests were employed. Due to the ordinal nature of the response variables, we perform a non-parametric pairwise multiple comparison^{50 (link)} and adjust the false discovery rate using the Benjamini–Hochberg stepwise adjustments. We also report the results of the Kruskal–Wallis rank test of the hypothesis that responses from different fields are from the same population. We employed both mean comparison t-test and the non-parametric Wilcoxon rank-sum for comparison between US and non-US responses. For US and non-US difference within fields, we implemented the Bonferroni correction to account for multiple comparison by inflating the significance cut-off by five-fold. To calculate the effect size for these comparisons, we follow the transformation of Cohen’s d for ordinal data proposed by^{51 (link)}, where d $\text{=}\text{2*z/}\sqrt{n}$ . Exact p-values (two-tailed) are reported. Statistical analyses were performed using Stata MP 16.1.

We used multiple imputations with chained equations to address any missing data between an individual's first and last appearance in UKHLS under a missing at random assumption (Desai, Esserman, Gammon, & Terry, 2011 (link); White, Royston, & Wood, 2011 (link)). Observations with high missingness (those missing >9 of the 22 variables required for analysis, totalling 12.7% of all observations) were dropped. Due to the need for lagged data on time-varying confounders, the first wave of data for each individual only contributed information about baseline confounding characteristics. Twenty imputed datasets were then created with all variables including exposure, mediator, outcome, and lagged time-varying confounders.
Gender, age, wave, and number of children were used as imputation variables; due to issues achieving model convergence missing observations for region (n = 211, 0.0007%) were dropped. Poverty and caseness variables were dichotomised after imputation. Online Table S1 in the supplementary appendix details the regression models used to impute each included variable.
Stata MP 16.1 was used for all analyses. Graphs were generated in R using ggplot2.