Choice set construction

[imccart]

We’ve tried these versions of choice set constructions so far: All specialists in the beneficiary’s HRR as a potential option Specialists that a PCP has referred to in the past, plus an “other” outside option Same as above but limited to DRGs 469 and 470 (major hip and knee joint replacement) Specialists that a PCP has/will ever refer to. Note, can’t have outside option here because it’s never selected by construction.

Takeaways regarding choice sets:

Choice sets with outside options are changing over time, potentially endogenously. Any time a PCP experiments with a new specialist, that specialist is an outside option at time t and becomes part of the choice set for future periods.

Ideally we do not include a generic outside option, but estimation is very slow with the full set of possible specialists in each market Also, most operations are isolated among few specialists. Specialists with very few operations may not be true “referrals” and instead introduce noise into our estimate on the effect of failures

[imccart]

current setup: choice set consists of all specialists with at least 10 operations in that year. We lose less than 5% of all operations from 2013 to 2018 with this restriction.

[imccart]

Update as of Feb 2023: choice set consists of all specialists with at least 20 operations that year. We lose about 10% of all operations from 2013 to 2018 with this restriction.

[imccart]

How to restrict choice sets based on distance? Look at 95th percentile of distance across markets and consider dropping all specialists located beyond the 95th percentile distance from a patient's choice set. Is this more or less restrictive than a fixed 75 mile threshold?