# Taunter shreds health insurers' claims that recission is rare

A big tip'o the hat to okanogen at CorrenteWire for picking up this one:

Death by Math (that should get Oaks' attention!). Taunter analyzes the statement by Assurant CEO Don Hamm’s that "Rescission is rare." Rescission is when a health insurer cancels a policy, usually because the insured "lied" on the original application, usually by failing to disclose a previous condition. If you haven't heard the horror stories of people who required costly medical care only to be dropped by their health insurer, you haven't been paying attention to the world around you. Here is what Hamm told Congress in his prepared remarks:

Rescission is rare. It affects less than one-half of one percent of people we cover. Yet, it is one of many protections supporting the affordability and viability of individual health insurance in the United States under our current system.

Late last week, James Kwak of Baseline Scenario suggested that the numbers put forward by Hamm were misleading because rescissions are targeted only at those who fall ill. [NOTE - this corrects my previous, erroneous statement that Kwak had accepted Hamm's argument; I apologize for that major error.] Taunter, who has a mathematics background, came up with the hard numbers:

To understand why 0.5% of the people Assurant covers is a lot of people – a jarring, terrifying, probably criminal lot – you need to understand a little bit of math.  You need to understand just enough math to understand what Don and his legal team are not telling you.  You need to understand conditional probability.  And the folks at Assurant are counting on the fact that you don’t. . . .

Half of the insured population uses virtually no health care at all.  The 80th percentile uses only \$3,000 (2002 dollars, adjust a bit up for today). You have to hit the 95th percentile to get anywhere interesting, and even there you have only \$11,487 in costs.  It’s the 99th percentile, the people with over \$35,000 of medical costs, who represent fully 22% of the entire nation’s medical costs.  These people have chronic, expensive conditions.  They are, to use a technical term, sick.

An individual adult insurance plan is roughly \$7,000 (varies dramatically by age and somewhat by sex and location).

It should be fairly clear that the people who do not file insurance claims do not face rescission.  The insurance companies will happily deposit their checks.  Indeed, even for someone in the 95th percentile, it doesn’t make a lot of sense for the insurance company to take the nuclear option of blowing up the policy.  \$11,487 in claims is less than two years’ premium; less than one if the individual has family coverage in the \$12,000 price range.  But that top one percent, the folks responsible for more than \$35,000 of costs – sometimes far, far more – well there, ladies and gentlemen, is where the money comes in.  Once an insurance company knows that Sally has breast cancer, it has already seen the goat; it knows it wants nothing to do with Sally.

If the top 5% is the absolute largest population for whom rescission would make sense, the probability of having your policy cancelled given that you have filed a claim is fully 10% (0.5% rescission/5.0% of the population).  If you take the LA Times estimate that \$300mm was saved by abrogating 20,000 policies in California (\$15,000/policy), you are somewhere in the 15% zone, depending on the convexity of the top section of population.  If, as I suspect, rescission is targeted toward the truly bankrupting cases – the top 1%, the folks with over \$35,000 of annual claims who could never be profitable for the carrier – then the probability of having your policy torn up given a massively expensive condition is pushing 50%. One in two.  You have three times better odds playing Russian Roulette.

And that's just the beginning of Taunter's article. There are some other great zingers in the rest, such as this:

It is in the health insurer’s interest to have application fraud, not only because it saves time and expense on the front end, but also because it lets them get out of any policy that isn’t going well for them. If the health insurer had to verify the information – if, in essence the insurance company had to behave as an accredited investor with adequate expertise to make a decision without reliance – it wouldn’t have the opportunity to bail out.  It would catch more genuine liars, but many of these liars would have turned out to be healthy, profitable customers, and what the carrier really wants is a population devoid of expensive claims, not devoid of liars.

Just go read it: Unconscionable Math. And get it someone in your Congresscritter's office to read it.

## Forum Categories:

### nothing lies better

than conditional probability. I saw this and it's a great post, trying to expose one of the common areas of confusion, conditional probability which is used to lie with statistics.

You must have Javascript enabled to use this form.

### Actually ...

This is what I said: "And remember, insurers only try to rescind policies if you turn out to need them; so the percentage of people who lose their policies when they need them is even higher. (The denominator should exclude all those people who never need expensive medical care, at least not before 65 when they go onto the single-payer system.)"

I'm generally not this sensitive, but I am proud of my statistical chops, so I don't want people to think I missed this one.

That said, Taunter dealt with the issue in much more depth and much more clearly than I did, and I'm glad that everyone is picking up his post.

You must have Javascript enabled to use this form.

### Welcome to EP Sir Kwak

I didn't catch this, Tony, he did catch it.

The original post from Baseline Scenario, Health Insurance "Innovation".

The issue is conditional probability is very difficult to comprehend. People see "0.5%" and can't track where the rest of the numbers are coming from and Kwak didn't spell out the procedure on how he got the ballooning rescission numbers. Probably (pun intended) because most people, including a lot of PhDs, do not have graduate probability/statistics course work or a lot of experience with conditional probability, Bayes' Thm. and Actuarial Science.

The insurers claim that rescission is very rare; at the Congressional hearing, two of three industry representatives said it happens to less than 0.5% of policies per year. But that is a deeply misleading number. That means that if you are in the individual market for twenty years, you have a 10% chance of your policy being rescinded; 30 years, and it goes up to 14%.

Even myself, despite passing the "quiz", walked into Las Vegas and simply "forgot" the concept of independent events, much to the consternation of my bank account. (went the opposite way, maybe blowing on the dice will help!)

You must have Javascript enabled to use this form.

### Tony

I saw your corrected the miscategorization of Kwak over on that evil orange satan (:)), might correct it on the original for while we don't get the participation/comments on EP, we often beat the actual number of reads per post than DK....

Also, in your title, should be rescission.

Folks, I've done this myself, spell checker doesn't work in the titles and those damn programs...

often they just don't like them big fancy words and will get you to put in recession as an example. (I just did it, easy to do!).

but another big favor, regulars on EP, add EP to your DK blogroll.

You must have Javascript enabled to use this form.

### Maybe someone should do an intro

to conditional probability. Go beyond Markovian and talk about non-Markovian dynamics. Which would be a good lead in to a genus of Very Dark Grey Swans.

You must have Javascript enabled to use this form.

### stochastic out!

I just added some LaTeX and graphing tools, which are straight out of Academia style to write complex mathematical formulas.

Frankly, while I think one can get away with writing up a few formulas and even venture into Calculus.....

This should give capabilities like econbrowser and Krugman have on their blogs (I believe they are using something similar).

but I think if you're going into Markov models you'll have an audience of 0.5.

You must have Javascript enabled to use this form.