Research software

Age-based policy changes that can be anticipated may lead to delayed behaviors. We develop a simple linear RDD specification with data-driven donut hole selection that enables us to compare the choices of delay and shift and assess whether there is a net change. Our programs can be found on GitHub. GitHub.

Details can be found in the paper Reductions in Out-of-Pocket Prices and Forward-Looking Moral Hazard.

This program estimates bias-reduced glm models with dummy-variable fixed effects and substantially improves upon BRGLM, below, much faster, omits fixed effects from output but saves them directly in a variable for further assessments. Finally, it also works with the margins command. Detailed descriptions and codes can be found on Github. 

Details can be found in paper Predicting Individual Effects in Fixed Effects in Panel Probit Models. 

Estimation of fixed effects in non-linear models, such as probit, logit, cloglog, and poisson is difficult to estimated due to (1) incidental parameter bias, (2) perfect prediction, and (3) small sample bias. The Kosmidis-Firth Bias-Reduced Generalized Linear Model estimator, developed for cross-sections can be used to estimate these models with favorable properties; the model is particularly useful if researchers are interested in the distribution of the fixed effects.  

Stata program to estimate bias-reduced glm models can be installed using ssc install brglm in Stata, or downloaded via brglm, see citation information here. 

Details can be found in Predicting Fixed Effects in Panel Probit Models. 

Contracts often involve non-linear pricing schedules. For example, health care insurance often contains time-limited deductibles that reset after a certain period of time. This phenomenon leads to forward-looking moral hazards. We develop a new approach based on a structural model of healthcare demand. In our model, individuals are exposed to random health shocks arriving according to a homogeneous Poisson process. Individuals are myopic and decide at each instance whether or not to visit a doctor. In the random utility decision model, the out-of-pocket costs for seeing a doctor drop nonlinearly after the first visit, leading to an increased hazard rate for subsequent visits. The model introduces a dynamic aspect absent in econometric models used in prior work.

Stata program and simulation files to estimate the stochastic dynamic hurdle models can be downloaded via sdhurdle. 

Details can be found in An Econometric Model of Health Care Demand.