Tuesday, March 29, 2016

Car Shopping, Part 1

Let's start with the shameless plug: I'm a data scientist at a zero-inventory car dealership called carlypso. I spend a lot of time thinking about how people buy cars.

I also spend a lot of time thinking about behavioral economics and cognition. So when I think about how people buy cars, I start by completely ignoring what people SAY goes into car buying, and instead pay attention to their behavior and the underlying processes this behavior suggests. This post is about those underlying processes.

Let's imagine someone buying a car. Car buying is a big purchase, so most buyers do a lot of research. Presumably they compare different makes, models, trim packages, option packages, colors, new vs used, mileage, warranty, safety, trunk space, seating, gas mileage, and of course price. They make a short list, compare trade-offs, test drive, and eventually make a decision. Right?

Well... not really. Yes, all of that stuff happens. But that's just the most visible part of the process. One of the big lessons of the psychology of decision-making is that you usually need to look past the surface. Tversky (one of the two fathers of behavioral economics) provides a good example.
"Over the past few years, we have discreetly approached colleagues faced with a choice between job offers, and asked them to estimate the probability that they will choose one job over another.  The average confidence in the predicted choice was a modest 66%, but only 1 of the 24 respondents chose the option to which he or she initially assigned a lower probability, yielding an overall accuracy rate of 96%."
       —Dale Griffin and Amos Tversky, "The Weighing of Evidence and the Determinants of Confidence."  (Cognitive Psychology, 24, pp. 411-435.)
Think about that for a moment. A decision as big as a new job, everybody spends some time researching and carefully considering the decision. But for all that careful consideration... 23 out of 24 people went with the option they leaned toward originally. Apparently all that "research" and "careful consideration" didn't actually have much impact on the final decision. The real decision was actually made much earlier.

To explain this, we need a bit of jargon. Decision-making psychology talks about "system 1", which makes fast intuitive judgements, and "system 2", which makes slow, rational judgements. Within any individual person, both systems operate in parallel. In examples like Tversky's, system 1 has already decided. Once system 1 has decided, a host of cognitive biases steer system 2 toward the same conclusion. Even though it looks and even feels like the decision hasn't been made yet, most people will spend more time looking for evidence which supports their system 1 decision, and be more skeptical of evidence which opposes their system 1 decision. As a result, the real decision is usually made by system 1. System 2's de-facto job is to rationalize it. (This process is a major theme in a 2011 best-seller by Kahneman, the other father of behavioral economics.)

Time to take it back to cars. Tversky's job-decision example seems like a pretty direct analogy. Find 24 friends who are "deciding" on a car, and ask them what they're currently leaning toward. I bet that at least 22 times out of 24, they'll go with the same car they were still "deciding" on, from the same dealer they were still "deciding" on. Seem like a good bet?

There's a weird thing with SUVs. Ask people why they like SUVs, and one of the most common answers is "safety". Statistically, this is hogwash. The risk of rollover in an SUV is so dramatically higher than other vehicles that it makes them much more dangerous. But statistics are system 2 thinking. What about system 1? Do SUVs *feel* safer? Yes. They feel big, and tall, and durable. They feel like if there's an accident, it's the SUV that'll survive. Whether or not that's true is irrelevant. It feels safe, which is what appeals to system 1.

That doesn't mean system 1 is always wrong, though. Quite the opposite. The whole point of system 1 is that it uses heuristics which usually work well for quick judgements. The point is that people don't update those judgements much when new information comes along.

Let's look at some pictures.


Stealth IAT

Assume these two jeeps are the same year/make/model/trim, same options, same mileage, etc. They're the same price. Which one would you want? Go ahead and think about it for a second. But don't think too hard.

Ok, now what if I tell you that the car on the top needs a $3000 timing chain replacement? Think about that for a minute. How does that update your estimate? First car, or second? How much money difference is there?

Some of you are probably thinking "hmm, $3000 is a lot, but that second car obviously needs 4 new tires and probably a battery and maybe a tow and who knows what else...". Maybe you came up with a whole list of things that might be wrong with the second car, considering what's visible. But I bet you didn't stop to think what else might be wrong with a car that needs a timing chain replacement.

This is classic confirmation bias - system 1 has decided, so system 2 goes looking for reasons to support the original decision... but it doesn't do a good job looking for reasons NOT to support the decision. This is the biggest single mechanism which makes system 1 decisions so important.

At carlypso, I hear the struggle between system 1 and system 2 all the time as salespeople talk to customers. Sometimes, a customer's system 1 doesn't want to buy from us. The purchase makes sense rationally, but the customer's gut says no. The poor salesperson sits there spouting all sorts of facts and arguments and explaining why it really does make sense... but they're fighting an uphill battle. System 1 doesn't like it, so system 2 discounts all these nice reasons and generates counter-arguments, and it just goes into a drawn-out "grind", as the salespeople call it.

But other times, a customer's system 1 does want to buy from us. Those are the easy sales. Even if they want a really specific car and it's tough to get exactly the right one at a good price, those are the good cases. The customer wants to buy the car; all we need to do is help their system 2 to rationalize it.

This post has mostly explained that system 1 is the real decider, and system 2 usually rationalizes system 1's decision. Next post will lay out my current thinking on how system 1 chooses a car.

No comments:

Post a Comment