Sunday, August 23, 2015

Does having a state-run exchange improve health insurance access under the ACA?



With the ruling in King vs. Burwell behind us, focus on the differences between state sponsored health exchanges vs. the federal exchange has fallen away. But as state-based data has been rolling in from Gallup, the CDC, and the Urban Institute on declines in people without health insurance, I began wondering whether providing a state-based exchange has any advantages over a the federal marketplace. Gallup’s comments and tables in particular seem to push the idea that states with state-based exchanges seem to have had more success with reducing the uninsured rate.

Of course, the real reason that state-based exchanges exist is political. The original House of Representatives bill had a national marketplace, while the Senate Bill incorporated state-based marketplaces. This particular breakdown shouldn’t surprise anyone, since Senators represent entire states, and all states are equally represented. The general state-based structure of the Senate bill won out. The final ACA incorporated a mechanism that defaulted to a federal backstop, but the markets themselves were still based on state boundaries.

However, despite that structural political reason, there might be some practical reasons why state exchanges might have superior performance to a unified federal exchange. First, commentators often refer to states as “laboratories of democracies” that can innovate and try numerous different ideas. Over time, the theory goes, good ideas from some states will diffuse across other states naturally and more quickly than if the federal government had installed and tried to improve a clunky national idea. Second, there’s the idea that differing conditions and preferences across states mean that state-based exchanges will allow individual states to customize their exchanges to best fit the needs of their state.

We would have to hold these potential advantages against some very real drawbacks. First, there’s administrative complexity and cost of constructing and running an exchange for a state-level. With federal grants to construct exchanges running out, several smaller states are already transitioning back to the federal marketplace, while others are having trouble paying the upkeep costs. Also note that there’s a question of whether several states even have enough potential subscribers to form a healthy individual insurance market to begin with.

With these ideas in mind, I used Gallup’s state-based data to build an extremely simple statistical model to predict the effects of a state-based exchange on improvements in health insurance coverage.  Follow me below the fold for more details.


The next four paragraphs and the chart describe the model in some detail. If you want to cut to the executive summary written in English, skip the next four paragraphs. 

This Section Has the Wonky Methods Explanation and Statistics Stuff.

The dependent variable (what we’re trying to predict) is the percentage decrease in the uninsured population in each state according to Gallup’s polling between 2013 and the first half of 2015. Note that this measure is the percentage decrease, not the percentage point decrease. So Massachusetts, whose uninsured rate declined from 4.9 percent to 3.0 percent, would show a reduction of 39 percent , not 1.9 percentage points. The reason to employ this measure is to  avoid the bias that accrues by some states have the ability to have a much larger percentage point decrease because they started out with a far greater proportion of their population uninsured. 

My model includes two independent variables. The first represents whether or not the state implemented the Medicaid expansion by the Jan. 1 2015, with states implementing it scoring 1, and those that didn’t scoring 0.  For the purposes of the exercise, I counted Wisconsin as expanding Medicaid, since they extended Medicaid coverage to all people underneath 100 percent of the poverty line (while cutting it for those over, but they still have access to heavily subsidized insurance).  The other variable, the one of interest here, is what type of exchange a state had in 2014 (according to the Kaiser Foundation). States with a full-state exchange scored a 2, states with a partnership scored a 1, and states relying on the federal marketplace were a zero. 

I regress the reduction in state percentage of the uninsured on the Medicaid expansion and State Exchange variables using Ordinary Least Squares Regression (the simplest form of regression). If having a state exchange helps reduce the uninsured rate while controlling for the effects of the Medicaid expansion, we should see a positive coefficient on the exchange variable. Basically, this is a three-dimensional version of your basic y=m(x)+b slope-intercept formula you might remember from your first-year algebra class (So it’s really y=m(x1) +m(x2) +b). Regression takes all the points on the graph (which are the Gallup results for the states) and finds the best-fit line through the data. The slope of that line estimates the size of the effect that each independent variable (the presence or non-presence of Medicaid expansion and the State-based exchanges) on the dependent variable (the percent change in the uninsured rate). Where the line “misses” each point, the regression notes the size of the miss and records the distance of the "miss" as error.

Table 1 shows the results. Overall, the two independent variables explain about 28 percent of the variation in the Gallup data, which is quite a high for only a two-variable model in cross-sectional data. 

Both expanding Medicaid and having a state exchange seem to have a positive effect. The Medicaid expansion has the largest effect, increasing the percent decline by 15.5 points. Having a state exchange would increase the percent decrease by 5.1 points (having a partnership exchange would increase it about 2.5 points.) Think of the constant (which is the Y-intercept) as the estimated “baseline” decrease before calculating the effect of the  I do not report standard errors here, because this analysis incorporates every state, not a subsample of them. 

Finally note while I report on all 50 states, there is a source of error from the Gallup data while derives its estimates from a sample people in each state. Most of these samples are quite large, holding the margin of error down (assuming the sampling was done correctly), but some of the smaller states have samples of several hundred people, which leaves a larger margin of error (See Charles Gabba, aka "Brainwrap" at ACA, for more on this.) Having said this, the Gallup estimates seem to track very similar to other data points.



Table 1: Influence of Having a State-Based Exchange
and Medicaid Expansion on % Decline in the Uninsured Rate
Variable Name
Coefficient
Medicaid Expansion:
15.51
State-based Exchange
2.55
Constant
26.46
N= 50

Adjusted R-squared =.2811

 
 Plain English Explanation:

The regression results in table 1 show that having a state exchange decreases the uninsured rate by a modest amount (about 5.1 percent) In contrast, expanding Medicaid has a large effect on decreasing the uninsured rate (about 15.5 percent).  Both together have a large effect on the effectiveness of reducing the uninsured.

Here’s a practical illustration using the Gallup data: The median state had a 34.6 percent reduction in its uninsured rate between the end of 2013 and 2015.  A state in the 25th percentile had a 26.9 percent reduction, while a state in the 75th percentile had a reduction of 48.4 percent.  The upshot is that a state performing in the 25th percentile without having its own exchange nor expanding Medicaid would have performed in the 75 percentile if it had done both of those things (Or think of it as the equivalent of increasing your ACT score by eight points).

Again, however, note that the lion’s share of the effect comes from expanding Medicaid, not having a state-run exchange.  We also have a pretty good idea of how the Medicaid expansion increases coverage (it provides more money to get people covered) In contrast, the small effect that having state exchange has might be a statistical artifact: After all, states that deployed a state-based exchange were also aggressively implementing Obamacare in other ways as well, like more and better navigators, public-awareness campaigns and like.  Perhaps the improvement that the model shows in the state-run exchange really is a reflection of these factors.

Bottom line: does having a state-exchange help improve the ACA’s performance?  The answer: It might help a little, but the effect is probably overstated in this simple model, and it is far overshadowed the the effect of Medicaid.

Next time, I'll build a slightly different model to explore how much accepting the Medicaid expansion might help reduce the uninsured rate.

No comments:

Post a Comment