I want to dig a bit deeper into Gold uptake on Healthcare.gov as there was wild variability this year. Some counties saw 50% of the on-Exchange market share go Gold while other counties barely saw any Gold sold. One of the first metrics that I’ve used is the gap between the Benchmark Silver and the the least expensive plans of any other metal. My logic has been that the marginal buyers are reasonably healthy and they are mainly shopping on prices.
Andrew Sprung did a deeper dive into the Philadelphia regional Gold purchasing pattern as he had noticed that Philly looked odd compared to the rest of Pennsylvania. He goes into benefit design but he begins his search for knowledge by looking at the least expensive plan in each metal band.
Gold plan enrollment north of 30% in populous counties such as Allegheny (e.g., Pittsburgh), Lancaster and York made sense, as the cheapest gold plans in these counties, offered by dominant insurers UPMC (in Allegheny) and Geisinger (in York and Lancaster) were cheaper than cheapest silver. I was somewhat mystified, however, by high gold takeup compared to prior years in Philadelphia, Montgomery and Bucks counties, ranging from 9%-14%. In these regions,the cheapest silver plan offered by sole insurer Independence Blue Cross was considerably cheaper than the cheapest gold. In Philadelphia, gold selection was 9% — compared to 3% in 2017.
This is a really good idea and I think it provides a bit more illumination. I cleaned up the data and looking at only the counties where there was non-suppressed Gold data, I ran a simple regression that sought to predict Gold market share by the size of the gap between the least expensive Silver and the least expensive Gold.
Gold MS = .000966(Price Gap) +.1769
R^2 = .48 P<.001 f=1099
This is way too quick and dirty to do anything beyond go “Umm… this is interesting” but it is very interesting and confirms the prior that as Gold gets comparatively cheaper, more people buy Gold.
A visualization of the relative prices of the cheapest plans is below the fold: