How to buy a car using data: Part II — the final decision and epilogue of our adventure

Guestpost by Dev Nambi on Feb 29th

In our previous blog post we identified 27 cars based on a list of features, and then narrowed our list down to 3 based on Internet data and test drives. Now, it is time for more data!

Near death experience, Wales 1964 © by PhillipC, used under Creative Commons license.


With a list of cars this small, we can do more in-depth research. We found out the cost of car insurance, average maintenance costs, vehicle crash ratings, accident data, and insurance data. We also tried to estimate how much each model would cost to own over 5 and 10 years.

However, the most interesting data was about crash test ratings and accident statistics. Vehicle crash-test ratings are designed to be predictive, which means they try to imitate real-world conditions. Accident data is far more interesting, because it shows what actually happened.

You're OUT, Honda Civic!

The Honda Civic accident data suggests it is less safe than a Fit or Prius. We eliminated it from our list. You can't argue with data.

We are left with two options: the Prius and the Fit. It is time to look at specific cars for sale.


The Internet makes it easy to find data. In our case, we used and to get a list of cars within 200 miles of Seattle. We wanted to find any Prius or Fit for less than $20,000 and with under 60K miles. We found 105 cars. Now that we had data, it was time for analysis!


Our biggest question was how to consider several variables. Which is better: a $15,000 car with 32,000 miles or a $12,000 car with 46,000 miles? What if one is a year older than the other?

The way we handled this is by focusing on the variables that mattered the most to us: price, mileage, and age. We created a "score" for each car's variable, from 0% to 100%. 100% meant it was the best deal. 0 mean it was the worst. For example, the car with the lowest mileage had a "mileage score" of 100%.

To find a "Good Deal Score," we weighted the different scores, and then added them. We said that price matters 50%, mileage 16.6%, age 16.6%, and warranties 16.6%.

You can see the results below. The best cars had a good deal score of over 5. You can see how the best scores are often given to cars with low mileage and a low price.

We looked at the top 2 Priuses and Fits, sorting by their Good Deal Score. We realized that the 2009 Prius for $15,000 and with 25,000 miles was what we wanted. We called the car dealer, had them email us the final price ($16.5K with sales tax and registration), and we bought the car that day. No pressure, no hassle, and we knew we got a great deal. Success!


Lessons Learned:

  • The Internet levels the playing field. A few days' research can make you a much savvier car shopper, giving car salesmen less of an edge.
  • Remember that the buyer has all of the leverage. I can choose not to buy a car from someone at any time.
  • Make car dealers bid for your business. Use phone or email, so they can't pressure you.


  • Discussing features first was a brilliant idea. We both compromised to make that list of features, but in a low pressure situation. Later on, Kate & I never disagreed about whether a car was a good fit for us, because we both were looking for the same thing.
  • Decide which model(s) to buy before deciding which specific car to buy.
  • The amount of money you can save by comparison shopping is incredible. We could immediately tell whether a specific car was a good deal or not based on the data.
  • A modest amount of time = massive savings. We spent ~40 hours doing research, analysis, and test drives that may save us $5,000 to $15,000 over the life of the car. That comes out to $125 to $375 saved per hour.
  • Using data = fewer disagreements. Kate & I always agreed which car was safer, because the data told us. We knew which specific car was a better deal, because our analysis said so.


  • Self awareness. Why are certain features and options important to you?
  • Emotional control. It is hard to walk away from a nice car because you are not intellectually ready to buy it.
  • Takes time. Those 40 hours were not spent on sleep, reading, or blogging.

What We Wish We Had:

  • Car maintenance/reliability data. We still don't know which car models are more reliable than others.
  • Car price predictor data. We didn't know whether car prices were going up or down.
  • A service to do this for us. I would gladly pay $100 for some company to do all of this work, and deliver the car to my front door.

How did you make your last car purchasing decision?

Read more posts about: , ,

About Dev Nambi

My name's Dev Nambi, and I'm a guy who loves numbers, bicycles, and Legos. I ended up (inevitably?) becoming a computer geek. I spend the daytime perfecting my craft and the rest of the time forgetting it. I can usually be found dreaming up impossible worlds, or reading about them in books.