How I Used Science to Fight Back and Win my Insurance Company.
This post is dedicated to new students starting their graduate school studies today. Basic science skills can also be quite useful in the ‘real world’.
My cellphone buzzed as I was about to leave my house for work. My wife was on the line, her voice shaking: “Yaniv, someone hit us”. She had just dropped off our son at daycare and picked up two daughters of a family friend on her way to school, where she taught. While she was waiting at the traffic light in the left lane, a large pick-up truck with a dolly came in the right lane and made an overly ambitious wide turn (see figure below). Luckily, neither she nor the kids were hurt. But the dolly scraped the right part of the car, all the way from the back wheel fender to the front passenger door of our 2003 Toyota Matrix. The pickup driver did not even bother to stop. He just continued quickly driving and my wife did the right thing by not chasing him with two kids in the back.
I was not too worried: no casualties, which is the most important thing. The car also looked totally functional, including the two doors on the right side (see cover photo of this story). In fact, my wife could even drive the car to her school no problem. The only issues were the scratches and two dents that were just cosmetic. After she calmed down, I asked my wife to give our insurance company a call and report the accident. I thought we would end up paying the deductible and that the car would spend a few days in the garage, which is not a big deal since our policy covered reimbursement for car rental after an accident.
Part 1: Questioning authority
I was stunned when J., the adjuster told me that they are going to total the car. “You must be kidding me”, I said in disbelief over the phone, “This car can drive. It only needs some body work”. He calmly replied that the cost of repairing the car is very similar to the amount they would return if they total the car and there is nothing he can do about it. The car must be totaled. “How much are you going to give me?,” I asked with concern. At that point, he provided a long list of numbers and the method they used to calculate the value of the car. I have to admit that it sounded impressive — a solid system to precisely measure the value of the car. The result was much less impressive to me: ~$2800.
It never occurred to me that insurance companies can profit from their clients’ losses.
What am I going to do with $2800? Buying a similar car was going to cost about 10 grand and I really did not need the ‘fun’ of car shopping. “Can you please undo everything?”, I asked him, “We just want to get our car back. Please cancel the claim and we will take the car to our own mechanic”. “I cannot do that”, he replied, “Once I totaled your car, you are technically driving a salvage”. I was about to declare a total defeat, but thought to give it one last chance. “Is there any way that I can keep the car even as a salvage?”, I asked. “Well, you are the first one to ask me that in 14 years”, he replied confused. “We can give you the car, but then we need to deduct the cost”. “How much is that?”, I asked impatiently. Checking on the computer for a few minutes, he replied “The value of the car as a salvage is around $3700, so basically you will get nothing”.
I could not believe his answer. In probability classes, we teach that insurance works because of the “central limit theorem”. When risk is pooled among many insured, the expected losses approach the actual losses, meaning that the premium is a fair measure of the expected loss plus some profit for the insurer that underwrites the risk. It never occurred to me that insurance companies can profit from their clients’ losses.
“This is ridiculous!” I cried. “You value my car higher as a salvage than a drivable vehicle. Of course you want to total it — you are going to make money this way! Does it sound reasonable that you will profit from my accident?” There was a long silence on the other side of the line. “Let me check what I can do, Mr. Erlich”, he answered.
That evening, I found that J. salvage value was wrong. Checking a few salvage purchasing websites showed that at best my salvaged car would be worth a bit over $1000. How can I trust this person?
A few days later, J. sent me an email that contained a 19-page report about the appraisal of my car. This report also looked super-impressive. It had numbers, charts, and references on how exactly the value of my car was calculated. But at that point, I already developed mental antibodies to this quantitative analysis theater. If they can claim that the value of my car as a salvage is higher than an operating vehicle, anything is possible.
There was also the first sign that being stubborn pays — they increased the return to $3312. On the phone, J. said that he was playing nice with me and decided to not include a few scratches that existed prior to the accident. I thought to myself: Wow! Fifteen minutes on the phone can really save you 15% or more. Just not the fifteen minutes that you think.
Part 2: Address confounders
Feeling motivated by this quick victory, it was tempting to keep poking holes in their analysis. The costs of the prior damages were too complicated and they already gave me something there. Instead, I thought to focus on the Base Vehicle Value that describes the Fair Market Value of similar cars without any prior damage.
The report included a list of 23 car advertisements of similar models that used to appraise my car. In addition to the ad price, it also reported the distance of the seller from my house. My insurance company simply averaged all ads to estimate the value of the car. However, I found it a bit odd that some of the ads were for cars in Connecticut, over 80 miles from my house in Cambridge, in towns whose poverty rates are doubled compared to the Boston metropolitan. You do not need to have a Nobel Prize in Economics to suspect that the same model is sold a bit cheaper in places where income is lower.
“Don’t mess with me. I am a nerd. I actually enjoy doing this and have all the time in the world to keep at it”
Twenty minutes later, I created a graph that presented the price as a function of the distance of the ad from my house. Using linear regression to remove the effect of distance, the graph showed that my car should be worth about $200 more than the value they estimated. In fact, for every mile away from Cambridge, the ads were on average $4 lower. That does not sound like “Fair Market Value” to me.
At that point, J. handed me over to his supervisor, F. I emailed F. the graph and explained the matter over the phone. My line was that as a scientist, I need to reveal the true Fair Market Value of the car before we can proceed forward. The subtext was “Don’t mess with me. I am a nerd. I actually enjoy doing this and have all the time in the world to keep at it”.
A day later, F. sent an updated report. He filtered all of the long distance ads, which raised my return to $3495.
Part 3: Always ask for the raw data
So far, I got six hundred dollars over the initial offer for a net of 2–3 hours of my time. Not so bad, if you consider that this money is not taxable. The Fair Market Value suddenly looked much more vulnerable to the rigors of analysis, so I decided to keep hammering at it. The new report by the insurance company listed now only five ads that were used to estimate the value of my car.
I love raw data. There are too many examples in the recent history of genomics where the raw data and the presented story in a manuscript did not agree. So I decided to call to one of the sellers of the ads. The car was very similar to mine and the report claimed that it would cost $6,979 to buy the car.
Surprise. Surprise. The sale representative told me that this car was just sold for $7,887, almost a thousand dollars above what the report claimed. I took the name of the sales representative just in case.
I sent the data to F. who immediately replied that an external company makes the reports for them and I should speak with them.
I called the external company, but I wanted to record the conversation in order to use it as evidence if needed. Massachusetts is part of the eleven states that require a two-party consent to record phone conversations. No problem — I started the conversation by asking the representative if I can record our phone call. She agreed without hesitation. After presenting the situation, I asked her why there is a discrepancy between the ad and the actual price. Her reply was stunning:
“[T]hat happens quite often. They will list it lower online because it is their Internet sales pricing, which is cheaper than what you are going to find on a window sticker or on the vehicle on the lot”.
Bingo! She just admitted on record that their online ad scanning system is likely to be downward biased compared to the actual selling prices of the cars. I called F. again and played the phone conversation with the report company to him. Again, there was a long silence on the other side of the phone. Then, he told me to wait a few days until he can speak with his bosses.
Four days later, he sent me an email. No more long reports with fancy charts and numbers. He probably realized that this will just make the situation worse. His “final offer” was described in just one sentence and amounted $4561.78.
By exercising grad school skills — questioning the authority of the adjusters, removing confounders from the analysis, and looking at the raw data — I was able to win my insurance company. Their tactics are well-known in science: they cherry picked their data (ads, distance); they tried to overwhelm me with information to make their case; and they insisted on communicating nonsensical results (salvage value that is higher than an operable vehicle value). These are the tricks that bad scientists use. Nothing new — just a different domain.
After I texted the new offer to my wife, her reply had one word: “genius”. I have a PhD, some nice publications, and won a few awards in my life, but she has never called me a genius before. I guess this is the difference between grad school and the real world.