DEVELOPMENT AND ANALYSIS OF NEW NONPARAMETRIC TECHNIQUES FOR CAUSAL INFERENCE IN OBSERVATIONAL STUDIES

DEVELOPMENT AND ANALYSIS OF NEW NONPARAMETRIC TECHNIQUES FOR CAUSAL INFERENCE IN OBSERVATIONAL STUDIES

S. R. Thanoon

[PDF]

Abstract

The lack of randomization methods in observational studies blocks researchers from reaching valid conclusions based on their data. The lack of randomisation techniques results in multiple experimental factors which appear in the research results. Research managers now use modern nonparametric analysis methods to reach superior causal results while gaining higher flexibility than traditional parametric procedures. This research develops a new analytical approach which merges matching techniques with instrument variables through kernel estimation methods for evaluation. The analytical procedures execute their functions without depending on specific distributional assumptions for discovering causal dependencies. Programmers who analyze complex observational data need to conduct theoretical evaluations and simulation tests to determine how specific causal data estimations are generated. The approaches make it possible to directly use them in epidemiology, economic research and social sciences to boost the estimated results from observed datasets.

Keywords

nonparametric causal inference, kernel-based estimators, instrumental variable techniques, marginal structural models (MSMs).