Variable selection strategy for PS model 

Balance diagnostics 

Matching
Ratio for matches Matching strategy Number of subjects and balance diagnostics prematch and post match Variance estimation

1:1 or 1:2 matching is sufficient Nearestneighbour with callipers strongly preferred 0.2 SD of the logit of the PS Untreated subjects chosen with or without replacement Without replacement—untreated matches chosen with greedy or optimal matching Account for matched pairs in outcome model with clustering, stratification or regression Account for matching with replacement

Inverse probability weighting
Application of weights Extraneous values Variance estimation
 Throughout; otherwise, if heterogeneity in treatment effect expected, apply subgroupspecific weights Use stabilisation, trimming and truncation, if appropriate Nonparametric bootstrap method preferred

Stratification
Number of strata Size of strata Combine estimates


Direct adjustment
Balance diagnostics 

Causal interpretations
1. Inclusion criteria 2. Target populationATE ATT ATU

Methods consistent with target population Describe the inclusion criteria and: Treatment effect in treated and untreated groups Treatment effect in the treated subgroup only Treatment effect in the untreated subgroup only
