Public support for smoke-free policies in outdoor areas and (semi-) private places: a systematic review and meta-analysis 79 4 used, all answer options that were more positive than neutral were combined to indicate support. Policies were categorised by the places that they cover: 1) indoor private places, 2) indoor semi-private places, 3) outdoor private places,4), outdoor semi-private places, 5) outdoor hospitality places and 6) outdoor nonhospitality places. When multiple estimates per category were presented (e.g. for outdoor eating places and for outdoor café places within outdoor hospitality places), the average support across the category was calculated. For studies based on the same samples a hierarchy of criteria was used to include one of them; the included study was most representative of the general population had the lowest risk of bias, was based on the largest sample size. If relevant data were missing, corresponding authors were contacted. Data analysis All analyses were conducted with R V.3.6.5 (R Foundation for Statistical Computing, 2020). Support reported as proportions ranging between 0 and 1 did not meet the normality assumption, and therefore logit transformations were applied.17 If support was reported as the mean score of a Likert scale ranging between 1 and a maximum score, this was transformed to the proportion support using the following formula: If the Likert scale ranged between 0 and a maximum score, the mean score was divided by the highest scale value. For ease of interpretation, proportions were converted to percentage support. We assumed support was related to country of residence and thus violated the meta-analysis independence assumption if multiple studies from a single country were included. Hence, a three-level meta-analysis was conducted to account for within-study, between-study and country-level clustering.18 The metafor package version 3.0.2. was used, which applies inverse variance weighting and accounts for dependence between the estimates.19 Subgroup analyses were conducted by sex (men vs. women; none of the studies reported data according other categories), smoking status (smokers vs. nonsmokers, and smokers vs. former smokers), parental status (parents vs. others), and age group (youngest age group vs. oldest age group reported). Log odds ratios (ORs) were pooled; if ORs were not reported the following formula was used: Finally, pooled log ORs were back-transformed to ORs for easier interpretation.
RkJQdWJsaXNoZXIy MTk4NDMw