[Informs conference Rehearsal]

정우형의 informs rehearsal

 

문제 : overpopulated hostials in seoul metro area(sma) & underpopulated local hospitals in non-SMA due to financial, aceesibility

 

Korean government aims to alleviate this issue by building new local hosptial, modifying insurance coverage system, and enhancing credibility in local health care. To facilitate such action, he proposed context driven hospital choice method, which utilizes novel structure of choice model, named FS-LCCM. 

 

This make sense because even though policy is implemented it is up to the decision makers to drive the change. If patient's choice does not change, effect won't be manifested.

 

There are many covariates including, but not limited to, temporal effect, socio-demographic, medical experience, cancer characteristic, and regional properties. Using the multinomial logit model, we were able to infer the coefficients of the variables. If we take one step further, we could utilize lccm to divide people into classes and better capture heterogeneity. It was possible to observe the inference values of coefficients and corresponding confidence interval for each covariates. 

 

However, the problem is that some features have high level of heterogeneity while some others does not have high level of heterogeneity, which may not requrire specific one.

 

fs-lccm infer different paramter with respect to each covariates. Flexibility of heterogeneity come with bettter fitness, and thus mcfadden't r square value increased. EM-algorithm was used to estimate coefficients. 

 

Also it's possible to gain more insight such as high income, young, patients living far from seoul are less likely to choose sca.