I know the owner's manual recommends premium only. But being the skeptic that I am, I decided to test it to see what difference it really makes. With rising gas costs and all, I wanted to find out which was cheaper in the long run, regular (87) or premium (93). I find it difficult to imagine that subaru would sell a car that can't run on 87, even though it does run better on 93.
What if you're stuck in the middle of nowhere, run out of gas, and the only thing available is 87? Well somewhere along the lines subaru made a decision to tune the car for 93 (or 91?), and put safeguards in place that retard timing if it detonates, which the ECU assumes is due to crappy gas.
Call me cheap, but I wanted to find out if it was cheaper to use bad gas, let it detonate and retard timing. I wanted to see what effect it had on fuel mileage.
I decided to run 87 for three tanks of gas, and then run 93 for three tanks of gas. I was concerned about the ECU learning period interfering with test results, so I built in a buffer tank of gas between the two. I.e., I actually have results for 7 tanks, but I'm throwing out number 4 because it was the "relearn tank" going from 87 to 93. I had been running 87 prior to starting the test, so there was no need to use a buffer tank initially.
The test had some (subjective) rules. I planned to drive in a similar manner on each tank, with a similar amount of spirited driving. Obviously things like traffic were beyond my control. I also planned to calculate mileage in the same way each time. When I filled up, I divide the number of miles on the trip odometer by the number of gallons of gas I put in. When pumping, as soon as the nozzle stopped pumping the first time, I stopped, i.e. I wasn't topping it off.
Ok, I'll admit it. I'm an engineer. This is way beyond the amount of analysis you would need to do to get results from this test, but here goes.
A simple average of the mileage shows that there is a difference in using 87 and 93. The mean of the mileage for 87 was 25.57, and the mean for 93 was 27.80. I achieved 8.7% better mileage using 93, so in order for it to be cheaper, 93 would have to be less than 8.7% more than expensive than 87. Just for kicks, let's say 87 is $3.50 today, as long as 93 is less than $3.80 it is cheaper to use 93.
From what I've seen during the test, 93 seems to be on average about $0.25 more per gallon and 87. This means that as long as regular is more than $2.87 per gallon ($0.25/8.7%), it will be cheaper to use 93.
Just for kicks I did a statistical analysis of the data, to determine if they are different. The test I used is called a 2-sample t-test. This test is used to determine if two sets of data come from the same population (i.e., that they have the same mean and variance).
The results of this test are shown below.
This test just tells us that we can say with 99% confidence that there is a different in the two sets of data, i.e. that they have a different mean and/or variance. Not particularly useful in an obvious case like this, but interesting anyway.
After the test I can confidently say that it is cheaper to use 93 in a stock 05 5-speed wrx driven by me in the traffic conditions I see. What does this mean for you? I'm not sure. But it probably holds true.
How much cheaper are we talking here? Let's use the $3.50 and $3.75 numbers for 87 and 93, and the means of the fuel mileage above. Let's assume you drive 15,000 miles per year. Fuel cost for 87 would be $2053, fuel cost for 93 would be $2023. The average driver would save a whopping $30 per year.
There are, of course, added benefits of 93. The car runs better. I had problems starting my car in the rain/cold using 87, which I have never had using 93. (I actually took it to the dealer to see if they could diagnose the problem, and they couldn't). There is more power due to proper timing using 93, though the butt dyno doesn't really register it. I also feel better giving the car 93, since that's what I know it wants. And I get a smug sense of satisfaction knowing that I am screwing the oil companies out of $30 per year.