Instrumental variable model average with applications in Mendelian randomization

Mendelian randomization is a technique used to examine the causal effect of modifiable exposure on trait using an observational study by utilizing genetic variants. The use many instruments can help improve estimation precision but may suffer bias when are weakly associated with exposure. To overcome difficulty high-dimensionality, we propose model average estimator which involves different subsets (single nucleotide polymorphisms, SNPs) predict in first stage, followed weighting submodels' predictions penalization common penalty functions such as least absolute shrinkage and selection operator (LASSO), smoothly clipped deviation (SCAD) minimax concave (MCP). averaged then genetically predicted obtain response second stage. novelty our also lies that it allows number submodels sizes grow sample size. practical performance examined series numerical studies. We apply proposed method real dataset investigating relationship between stature blood pressure.