The Mystery of the Million "Missing" Enrollees (or: Why my projection could still be accurate even if it's "wrong")

Until this year, most of the ACA exchanges, including HealthCare.Gov, would simply report how many people selected QHPs through the exchange, whether paid up or not. There's nothing wrong with this as long as it's made clear at some point how many people actually paid their premiums and had their policies effectuated; the argument over this issue was the entire basis of the infamous "But how many have PAID???" fuss back in 2014. It was such a Big Deal that the Republicans on the House Energy & Commerce Committee even published the results on a laughably garbage-filled "survey" they had sent out to a portion of the insurance carriers.

(Note In the end, it turned out to be roughly 85-90% depending on the state/carrier in 2014; for 2015 the payment rate nudged up to around 90% overall, which is where it will likely stand going forward).

For the most part, once I figured out the overall payment rate, this simplified things for me when it came to The Graph. I could simply have 2 different numbers for private policy enrollment tracking: The total number of QHP selections, and the estimated paid number, which has been either 88% or 90% of the total depending on the year.

The caveat to this is that a few state exchanges handled the paid/unpaid issue differently. Massachusetts, for instance, doesn't even report QHP enrollments at all until they're paid for. That means that when they reported 180,000 people "enrolled/paid" or "new/paid" way back on 12/10, the "full" number of QHP selections was likely around 20K higher than that. However, since the lower number is included in the official ASPE report as well, I never really bothered "correcting" for this number. Plus, it's just one mid-size state.

Washington State did something similar for the first two years: They handled payments for exchange enrollees as Massachusetts does, and thus "pre-purged" most of their "scrubbed" enrollees before including them in the tally. This year, however, they've dropped the payment process, making enrollees pay directly through the carrier, and have therefore joined just about every other state in reporting the "full" QHP selection number.

On Tuesday, I posted the latest weekly update from the Rhode Island exchange. HealthSourceRI, as I noted the other day, is doing something sort of in between: They were reporting the total number of QHP selections up until this week, but were also reporting how many of those people voluntarily chose to cancel their policies. In other words, the reports were something like this:

  • 30,000 Enrollees plus 500 new enrollees minus 100 cancellations = 30,400 net enrollees.

Bear in mind that until the end of December, the January payment deadline for those 30K or so people hadn't yet passed, so there were still some people who were "unpaid" simply because they hadn't sent in the check yet. This is common; most people don't pay their cable or utility bills until the last minute either. The point is that there are different people at different "stages", and it's a rolling number, since anyone who enrolls today doesn't have to pay until two weeks from now, while someone who enrolls Saturday doesn't have to pay until mid-February (since their policy won't kick in until March 1st anyway).

Still, since we're well into January now, anyone who signed up for coverage starting on January 1st (which is to say, the vast majority of them) is well past their payment deadline, so it makes sense that a solid number of Rhode Island residents would be dropped by now...which is exactly what HealthSource RI did in this week's report:


30,015 individuals are enrolled in 2016 coverage through HSRI, paid and unpaid.*

4,333 of the 30,015 individuals have selected a plan for 2016 coverage and are new to HSRI this year or returning after being enrolled with HSRI at some point during a prior year.

28,756 of the 30,015 individuals are enrolled in 2016 coverage through HSRI, and have paid their first month’s premium.

*Individuals who selected plans for January 2016 coverage but did not make a payment were cancelled after the passing of the payment deadline, causing a decrease to the unpaid enrollment count last week.

The only problem with this is that they didn't include the grand total number of QHP selections in the report, so I had no idea what the "purged" number was down from. I asked about this...and they responded:

As of 1/9/16, total gross plan selections numbered 39,246 (this includes plan selections that were since cancelled or terminated), compared to the 30,015 total active plan selections in the system as of 1/9/16.

The 9,231 variance between those figures represents individuals who voluntarily cancelled or terminated their 2016 coverage (ie. autorenewals), as well as individuals who were enrolled in plans for January coverage and were cancelled for non-payment last week (approx. 5,500 individuals).

Bingo. That's what I was looking for; now I can do the math for Rhode Island:

  • TOTAL QHP selections: 39,246
  • minus 3,731 people who were auto-renewed but didn't want to be (9.5% of the total)
  • minus 5,500 people who were cancelled for non-payment by the deadline (14.0% of the total)
  • net of 30,015, of whom 28,756 have paid up already.

So what about the remaining 1,259 (3.2%) who are still in the system but haven't actually paid/been effectuated yet? I presume most of these came in after 12/29 and aren't actually due to start coverage until February 1st, thus giving them another coule of weeks to pay up.

The main point here is that Rhode Island's official enrollment number, which had been around 34.6K the week before, "dropped" by nearly 4,500...even though they actually added nearly 900 new enrollees last week.

Now, Rhode Island is a tiny state, so the numbers here are nothing more than a rounding error nationally...but what happens when 38 states start doing this "partial live purging"? Here's what makes this important:

Plan Selections: The weekly and cumulative metrics provide a preliminary total of those who have submitted an application and selected a plan. Each week’s plan selections reflect the total number of plan selections for the week and cumulatively from the beginning of Open Enrollment to the end of the reporting period, net of any cancellations from a consumer or cancellations from an insurer during that time.

Because of further automation in communication with issuers, the number of net plan selections reported this year account for issuer-initiated plan cancellations that occur before the end of Open Enrollment for reasons such as non-payment of premiums. This change will result in alarger number of cancellations being accounted for during Open Enrollment than last year.Last year, these cancellations were reflected only in reports on effectuated enrollment after the end of Open Enrollment. As a result, there may also be a smaller difference this year between plan selections at the end of Open Enrollment and subsequent effectuated enrollment, although some difference will remain because plan cancellations related to non-payment of premium will frequently occur after the end of Open Enrollment.

As far as I can tell, this year, unlike the last two years, HealthCare.Gov is doing the same thing as Rhode Island in their weekly "snapshot" reports: they're culling some of the "cruft" enrollees...people who voluntarily cancel their policies, get refused enrollment for some legal/financial reason (failure to prove their identity, bounced check, whatever)...or, presumably, failure to pay their first premium by the deadline.

Do you begin to see what I'm talking about here? The 8.68 million QHP selections officially reported via as of 1/09 has already been "culled" to some degree, with some unknown number of cancelled accounts (voluntarily or not) already subtracted from the totals. this is a good thing, because it means the numbers are "cleaner"...but it's also different from their reporting methodology the first two years.

When I originally projected 14.7 million QHP selections, that was based on the assumption that CMS/ wouldn't subtract any cancellations until after Open Enrollment had ended.

Instead, at least some portion of these have already been pulled from the numbers during OE3...making the official "QHP selection" numbers smaller than they otherwise would have been. Again, it's a good thing...but this can make a huge difference across over 2/3 of the states!

How big of a difference? Well, let's suppose that 5% of the "selection" total has been "purged" already. That means that instead of 8.68 million,'s "unpurged" total may be as high as 9.14 million...or a whopping 460,000 people!

If that's accurate, the national total "QHP selections" number is actually more along the lines of 12 million instead of 11.5 million.

Now, unlike Rhode Island, I have no idea what that "unpurged" grand total actually is for; it might only be 8.7 or 8.8 million...but it's definitely higher than the 8,682,471 listed in the offical Snapshot report. I just don't know how much higher.

So, what proof do I have that CMS is doing this, beyond them mentioning it in the footnote?

Well, that's where Montana and Pennsylvania come into play.

Take a look at the weekly Snapshot reports for Week Nine and Week Ten. Since the net increase this week was only around 74,000 across all 38 states, most of the state numbers are only a few hundred higher, or a couple thousand for the larger states.

However, take a look at both Montana and Pennsylvania:

  • Montana: Week Nine: 55,552; Week Ten: 55,474
  • Pennsylvania: Week Nine: 412,914; Week Ten: 411,675

Both states saw their "QHP selection" number decrease week over week: 78 in Montana, over 1,200 in Pennsylvania.

This is proof that yes, CMS is indeed purging/providing cleaner numbers this year. The only question is by how much?

It's possible that Montana added 100 people but cancelled 178. Or perhaps they added 500 but cancelled 578, and so on. The same holds with Pennsylvania; they might have added 2,000 but cancelled 3,200...or added 10,000 but cancelled 11,200.

As for the other states, you can't see the cancellations because their numbers did go up...just not by much. Alabama did add about 2,000 people...but it's possible that they actually added 5,000 while cancelling 3,000 for a net gain of 2K.

So, what kind of a difference could this mean nationally?

Well, last spring, the official QHP selection total was 11,688,074 as of 2/22/15...but this had "dropped" to just 10,187,197 still effectuated as of 3/31/15, only 5 weeks later. Is it plausible that over 1.5 million people--nearly 13% of the total--all decided to drop their coverage in the same 5 week period? Answer: Of course not. A chunk of them were forcibly kicked off due to legal residency verification problems, but most of them likely never paid their January invoice in the first place and thus were never actually just took until the 3/31 report for CMS to actually document this.

Now, let's suppose that it turns out that 14.7 million people really do "select a plan" by the end of January. Let's further assume the same 13% "drop" between January and the end of March. That would be a reduction of nearly 1.9 million people this year, to around 12.8 million as of 3/31/16.

In that scenario, around 11.2 million of the 14.7 million would be run through 13% of that would be a "drop" of up to 1.4 million via the federal exchange.

Again, I have no idea just how many people have been "purged" via so far, but it seems reasonable that at least half of them could be removed from the official reports before Open Enrollment even ends.

THIS is the main reason why I officially dropped my OE3 projection down from 14.7 million to around 14.0 million (and why I might drop it further still).


UPDATE: Nice catch from commenter farmbellpsu: In addition to Montana and Pennsylvania "losing" enrollees last week, Alaska "lost" 37 enrollees between Week 8 and Week 9 as well.

Again, you're only going to see an actual negative number in cases where the overall increase is very small, since the gross additions will overbalance them in other weeks...but this is proof that there's definitely some amount of culling/purging going on this year, and not just this week.

UPDATE: Thanks to Jed Graham for reminding me that CMS actually did do a substantial amount of "pre-purging" last year after all...which I even wrote about at the time!

On the surface, this appears to only be about 40,714 people higher than the last official report (8,797,577 QHP selections), which ran through 2/15, right?

Wrong. As you'll recall, while the HHS Dept. included all 8.8 million people in the actual "snapshot report" itself, they also deliberately subtracted 200,000 people in the accompanying text and subsequent marketing/promotion of the "11.4 Million" number (which was actually over 11.6 million). They estimated that around 200,000 renewed enrollees from 2014 would have to be kicked off of their policies due to ongoing problems with documenting their legal residency.

Officially, these people were to be given the boot on 2/28, and therefore, in my opinion, really should have been kept on the official tally until March 1st. Instead, HHS decided to nip an obvious "cooking the books" GOP attack point in the bud by pre-purging these 200K people last week.

In other words, the "unpurged" QHP selection number last year was actually more like 11.9 million than 11.7 million.

Ironically, in the end only about 90K actually were cancelled, making it even messier, but the point is that there's ample precedent for CMS to do this, and even more evidence that yes, the "unpurged" number actually could be several hundred thousand people higher than the official reports indicate.

To reiterate: It's a good thing that CMS has improved/speeded up the process of reporting cancelled enrollees; I just wish that they would provide the raw numbers alongside the culled ones to make it easier to keep track of what number/percentage are being purged, that's all.