Note: Code for this analysis available on GitHub.
Last March I purchased Automatic, an adapter that plugs into my car’s diagnostic port and syncs with an app on my iPhone to report and record data about my driving. Among other things, Automatic provides:
During the first few months, I had trouble getting the adapter to kick in reliably when a trip started, and it still alerts me regularly that it hasn’t connected to the smartphone app. Reliability has certainly improved, but I wanted to get an idea of accuracy – is the cost and gas mileage being reported actually close to what I’m experiencing? To get some independent data, I started recording odometer readings, gallons and price at each fill-up in August.
So… How well do the two datasets match up?
I have two datasets. One downloaded from the Automatic Dashboard. Each row is a trip. As mentioned above, there are missing trips in this dataset.
The second is a set of data recorded at each fill-up.
Generally, Automatic appears to be accurate when it actually registers a trip.
|Measure||Automatic Data||Mileage Data|
|Fuel cost||$ 1,011||$ 1,095|
|Miles per gallon||27.3||27.1|
|Price per gallon||$ 3.105||$ 3.174|
Here you can see that, while total mileage and fuel consumption are off, the price per gallon and gas mileage are pretty close to actuals.
Even over time, the Automatic rate data seems accurate. The total monthly mileage tracks well between actuals and Automatic’s calculations. I suspect the major month-to-month differences are a result of missed trips and timing mismatches between the trips data and my fill-up mileage data.
Rate data (mpg and cost per gallon) show even better accuracy.
It’s a little disappointing to see the once-plummeting gasoline price level off in February, but it had to end eventually…
Since trips and fill-ups don’t line up perfectly with months, I’d like to compare the trips data directly with each tank of gas. This also gives me a chance to play with time intervals in the lubridate package.
The big swing you see at the end of November in my mileage records is the result of stopping at a gas station that had some technical problems and I was able to fill my tank only partially. Since accurate estimates of fuel consumption require me to fill the tank each time, this skewed the numbers for this fill-up and the following one.
As you can see in the graphs above, cost per gallon, gas mileage and even total miles driven tracked pretty well with my manually recorded data – the exceptions coming mostly when AUtomatic missed trips. There are probably other systematic sources of error, too. For instance, I usually pay a slightly higher credit price at the pump, and Automatic might be assuming the lower cash price.