Abstract:
|
Marginal Structural Models (MSMs) quantify the marginal effect of time-varying treatments, which, however, are underdeveloped for correlated multinomial choice outcomes, e.g., consumer expenditures. We propose simple-to-interpret MSMs for multinomial choices taking into account correlations of expenditure categories, with an application to study the effect of lockdown on consumer shopping behavior in the COVID-19 pandemic. Also, most existing MSMs use a discrete-time setup, which requires all subjects to be followed at the same pre-fixed time points but can be restrictive in many practical situations with irregularly spaced observations. We propose semiparametric doubly robust estimators to address time-varying confounding and irregularly-spaced observation times, capitalizing on semiparametric efficiency theory and advanced machine learning methods. Simulation studies are employed to confirm the validity of the proposed estimator. We apply the proposed estimator to investigate the lockdown effect on an individual’s economic behavior through the market transaction data of distinct regions in India with different periods of lockdown implementation.
|