About 18 months ago, I reached my main conclusion from my test of acetone:
The most important thing that can be taken from this is how hard it is to accurately measure the change caused by a variation of just one of the many factors that affect fuel mileage. There are just too many uncontrollable factors involved in road-testing to be definitive.
I went into that test with “my own fairly good understanding of my highway mileage,” which means I didn’t have any actual data. The test started with 2, ~90 mile stretches which were intended to establish a baseline. Now that I’ve collected my actual mileage data for over a year, I can better demonstrate just how hard it is to draw correct conclusions from road-testing, especially when just regular driving is involved because of the high levels of natural variability. It will also demonstrate just how easy it can be to get false positives from acetone. Furthermore, it will emphasize the need for laboratory testing, or, short of that, a very controlled road-testing program.
Let’s take a look at my actual data.
There is a lot of variability there. That variability raises an important question. If I were to use this data as a baseline for future comparisons, just what is the baseline? The best answer would probably be the average.
“The average” isn’t as simple as it might seem. Some people would want to average all the values in the chart, but that is mistaken. Each data point is itself an average (miles traveled divided by gallons of gasoline consumed), and it is nothing but bad math to try to draw a conclusion from an average of averages. To calculate the real average mileage, I have to divide the total miles driven by the total gallons consumed. That average is 22.1 MPG.
Just before I analyzed this data, I was asked what kind of mileage I had been getting. My best guess was 24 MPG. I had been recording my mileage regularly, and I was off by almost 2 MPG, about 9% too high. Unless there is considerable data used to establish a baseline, then the baseline it at best highly suspect.
We can learn another lesson from this data. The plot contains 53 data points. As will frequently happen with averages, about half the data points are above average. In this case, 28 points are above average. This is very important. If someone has taken the time to really establish a proper baseline, there’s about a 50% chance, based on nothing but pure randomness, that their next tank will be above average. So if you were to put something in your tank in hopes of improving your fuel mileage, as long as that something isn’t specifically detrimental, there’s a pretty good chance that it will appear to cause increased mileage.
Also, 26 points are higher than the point before them. Similar to the point made in the above paragraph, if you happen to use a single tank as a baseline, you’ve got a 50% chance of increasing your mileage on the next tank regardless of what additives you put in the tank.