Identification and estimation of effects with dynamic treatment assignments
Student No.:100
Time:Fri 09:50-11:25, Dec.21-Jan.4
Instructor:Per Johansson  
Place:Lecture Hall, Jin Chun Yuan West Bldg.
Starting Date:2018-12-21
Ending Date:2019-1-4

In the treatment effects literature (see Rosenbaum 2002, Lee 2005, Pearl 2009 and Imbens and Rubin 2015, among many others), typically there is a treatment unambiguously defining the treatment and control groups at a given time point, and its effect on a response variable is usually obtained by comparing the two groups' mean responses. But there are many cases where the treatment timing is chosen, which complicates the analysis.

For instance a medical treatment is taken by the ill, but when to take it is up to the individual or the treatment administrator and will only be taken as long as not recovered. Ignoring the treatment timing issue in these cases will lead to false inferences. Another example is the situation of estimating the effect of the chosen treatment timing effect on other outcomes related to the duration. For example, what is the effect of having a child on future income or life time earnings.

I will first discuss the problem when estimating the effect of the chosen treatment timing on outcomes not measured within the duration in which the treatment was given. To this end I will exemplify the problem by using a thematic large research field: `what is the effect of fertility on female labor supply?'. Then I will discuss the problem with estimating effects of a treatment given/taken in a state on the exit to another state, that the situation when the outcome is a function of the waiting time to treatment.

Undergraduate or master level in statistics

[1] Fredriksson, P. and P. Johansson (2008). Dynamic Treatment Assignment -- The Consequences for Evaluations using Observational Data", Journal of Business and Economic and Statistics, 26, 435-445;
[2] de Luna, X. and P. Johansson (2008). Non-parametric inference for the effect of a treatment on survival times with application in the health and social sciences, Journal of Statistical Planning and Inference 140, 2122-2137, 2010;
[3] Hernán, M., Brumback, B., and Robins J. M. (2001). Marginal structural models to estimate the joint causal effect of nonrandomized treatments, Journal of the American Statistical Association 96, 440-448.