Obamacare, politics, and previously insureds

11/10/2014 09:37:00 AM
They did the survey here.
A recent working paper sheds some light on a couple questions regarding the ACA implementation. The researchers sent out mail surveys to a representative sample of 40,000 Washington (the state) registered voters, receiving about 4,200 responses.

The first question is that of "previously insured" people. The data from the first year of ACA exchange enrollment suggests that a large majority of enrollees had previously had insurance before purchasing Obamacare. This led many commentators to claim that the ACA was failing to reduce the uninsurance rate, which as I explained here, is simply invalid reasoning. This new study helps shed a bit more light on why. They found that people who reported being previously uninsured were 7.8 times as likely to say they would enroll in an ACA exchange plan than people who reported being previously insured. Yet despite the fact that uninsured people are vastly more likely to enroll than insured people, a majority of those who enrolled were previously insured. That's just how probabilities work: most of the population was previously insured, so that a large proportion of previously uninsured is still smaller than a small proportion of previously insured.

Since they didn't really include this in the paper (...why I hate odds ratios), I crunched the numbers on this. 89.7 percent of the survey respondents had insurance of some type, so that 9.3 percent were uninsured. 22.4 percent of the uninsured indicated that they would enroll in the ACA exchange, compared to 5.5 percent of those who had some type of insurance. So you can see that the uninsured were way more likely to indicate that they would enroll. Yet, because 89.7 percent were previously insured, that works out to 206 enrollees who were insured, compared to just 87 enrollees who were uninsured (in the sample of 4,200 respondents). It is both true that the uninsured were 7.8 times more likely to enroll, and that enrollees were 2.3 times as likely to have been previously insured. Neither datum is sufficient to infer whether the uninsurance rate is rising or falling.

The second question is whether some people chose not to buy insurance as a result of the political controversy surrounding the ACA. One way to test this is to see whether individuals with views of the ACA more closely aligned to the GOP's position were less likely to buy insurance, after controlling for things like demographics. As a proxy for whether the respondent shared the GOP's views of the ACA, the authors used whether they blamed democrats, republicans, or both for the 2013 government shutdown (which was largely about a GOP effort to repeal the ACA). Among those who blamed both parties equally, the uninsured were 1.99 times more likely to enroll than the insured, compared to 3.6 times more likely to enroll among those who blamed just the Republicans, and a statistically insignificant odds ratio of 1.1 among those who blamed only Democrats. (note, these OR are demographically adjusted, whereas the raw 7.8 figure above was not. For comparison purposes, the adjusted baseline for all individuals was 2.0.)

This requires a bit of interpretation. The demographically-adjusted odds ratio gives a measure of the incentive effect of the mandate on the probability of buying insurance on the exchange. That is, the uninsured are more likely to buy insurance than the insured for obvious reasons, but after controlling for all those reasons, any additional difference in the probability of enrolling through an ACA exchange is probably a result of the incentives in the ACA, such as the mandate requiring everyone to have insurance or pay a fee. For most people, especially those who shared the Democrats views of the ACA, these incentives substantially increased their likelihood of buying an insurance plan. Yet, the results show that for those who shared the GOP's view of the ACA, these incentives had essentially no effects--that is, the GOP successfully convinced some individuals to defy the ACA mandate.

One critique I have for the authors is that they didn't specifically test the difference in odds ratios between the different political groups. While the ratio for democrat-leaning individuals was statistically significant and the ratio for republican-leaning individuals was not, in general the difference in statistical significance needn't be significant. The confidence interval for the former was 1.6 to 8.2, while the interval for the latter was 0.5 to 2.8 (an odds ratio of 1 means the mandate had no incentive effect on whether they would enroll in an exchange plan, greater than 1 implies more likely to enroll, less than 1 means less likely). As you can see, the two confidence intervals overlap, so working with the statistics they reported we can't really conclude that the odds ratio for those agreeing with the GOP was lower, statistically speaking, than for those agreeing with the Democrats.