Hi all
I am working on a model to identify if there are economies of scale at hospital department level. The dataset is patient level costs for all patients at all hospitals and across all types of hospital activity. The ‘department’ size is the total number of patients treated at the same hospital with the same activity type.
The cost of treating a patient is largely determined by the type of the activity itself, so we would like to absorb these as fixed effects. There are also many hospital level unobserved heterogeneity which influence costs (such as quality of senior management) and which we would like to absorb as a fixed effect. Given the vast size of our dataset (10s of millions), the large number of activity types (100s), the large number of hospitals (100s), and limited processing capacity (16GB RAM), we thought reghdfe would be a good solution.
The idea is that the coefficient on log_department_size would identify any economies of scale. My question is this: If there are economies of scale at department level, and department size is correlated (imperfectly) with hospital size, would absorbing hospital fixed effects bias our estimate of department economies of scale?
Unfortunately we only have 1 year of data, so cannot exploit longitudinal variation to deal with time invariant hospital heterogeneity.
Sergio Correia - I would be very appreciative if you were able to have a look at this question.
Thanks!
I am working on a model to identify if there are economies of scale at hospital department level. The dataset is patient level costs for all patients at all hospitals and across all types of hospital activity. The ‘department’ size is the total number of patients treated at the same hospital with the same activity type.
The cost of treating a patient is largely determined by the type of the activity itself, so we would like to absorb these as fixed effects. There are also many hospital level unobserved heterogeneity which influence costs (such as quality of senior management) and which we would like to absorb as a fixed effect. Given the vast size of our dataset (10s of millions), the large number of activity types (100s), the large number of hospitals (100s), and limited processing capacity (16GB RAM), we thought reghdfe would be a good solution.
Code:
reghdfe log_patient_cost log_department_size `patient_controls’, absorb(hospital_ID activity_type_ID)
Unfortunately we only have 1 year of data, so cannot exploit longitudinal variation to deal with time invariant hospital heterogeneity.
Sergio Correia - I would be very appreciative if you were able to have a look at this question.
Thanks!
Comment