Effect Estimation Of Causal Mediation Model With Measurement Error

The causal mechanism is investigated to explain the effect of independent variables on outcome variables through mediating variables in causal mediation analysis.It is assumed that the independent variable influences the mediating variable,which in turn influences the outcome in the mediation models.Causal mediation analysis is widely used in behavioral,psychological,biomedical,and social sciences.The problem of independent variables with measurement errors is often encountered in both statistics and other fields,such as acceptance sampling,epidemic diseases,and instrumental measurements.Due to the influence of measuring instrument,external environment,observer’s condition and other factors,the generated deviation between the measured value and the true value is called measurement error.For the case of mediating var with measurement errors,we consider the effect estimations in linear regression models and logistic regression models in this article respectively.The main work of the article consists of the following two parts.The first part is mainly aimed at the linear causal mediation analysis model with measurement errors in this article,where a linear regression structure equation is established,and the natural indirect and direct effect estimation expressions are given based on the counterfactual framework.The instrumental variable method and the least squares method are used to obtain the model parameter estimations so as to obtain the effect estimations.Under a limited sample,the simulation studies are conducted for the model errors from two different distributions.The simulation results show that the proposed estimations perform well in this article.In the second part,we use odd radio to define the mediation effect for a binary outcome,establish the logistic and linear regression models for the two cases with no interaction and interaction respectively and obtain the analytical expressions of natural direct and indirect effects.In the simulation studies,the deviation,standard deviation and 95% confidence interval of natural direct and indirect effect estimators are given,respectively.The simulation results show that the proposed effect estimates perform well.Figure [3] Table[4] Reference [53]...