A1015
Title: Penalized synthetic control method for truncated data with latent clustering
Authors: Gourab Mukherjee - University of Southern California (United States)
Wreetabrata Kar - Purdue University (United States)
Bikram Karmakar - University of Florida (United States) [presenting]
Abstract: A novel penalized synthetic control (SC) method is developed that accommodates latent structures often inherent in panel data structures. Using the proposed method, the effects of the passage of a medical marijuana law (MML) are studied by a state on direct payments to physicians. As an example of this latent structure, consider a physician who receives payments periodically, every six months, while another physician receives payments every five months. Direct use of an SC method while comparing these two physicians ignores these distinct latent patterns and thus will result in interpolation bias in the estimated effect. Under a truncated flexible additive mixture model, it is theoretically established that the SC method has uncontrolled maximal risk without a penalty; by contrast, the proposed penalized method provides efficient estimates. The analysis also estimates heterogeneous causal effects. Using the proposed method, a significant decrease in direct payments from opioid manufacturers is found to pain medicine physicians as an effect of MML passage. Evidence is provided that this decrease is due to the availability of medical marijuana as a substitute. Finally, the substitution effect is comparatively higher for female physicians and in localities with higher white, less affluent, and more working-age populations.