Causal Inference on Smokers' Online Friends
By: Vaidhervi Kansal
Department: Mathematics
Faculty Advisor: Dr. Luella Fu
A previous study found that smokers had more Facebook friends than nonsmokers. If this is true, what are the possible reasons that smokers would have more friends? Or are there covariates not yet accounted for that affect the number of friends that smokers have? We control for covariates, such as gender, for which a higher proportion of smokers are female and which affects number of Facebook friends. For data exploration, we use SQL and visualizations in Python. To control for covariates, we use matching and hypothesis testing. For estimating causal effects, we use inverse probability of treatment weighting and marginal structural models. We show directed a-cyclic graphs to show the relation between covariates, smoking, and Facebook friends; display matching outcomes; and provide model results.