Thanks, InsightfulRay! Now, don't quote me on this to any statisticians; I asked one I know whether I can in good conscience give a confidence interval for an experiment with just three data points in each condition of the variable, and he laughed at me.
But according to the standard deviation calculator at
Standard Deviation Calculator - Calculate mean, variance of the numbers
The pre-tab mpg has a mean of 45.09 and a stdev of 0.18
The post-tab mpg has a mean of 47.05 and a stdev of 1.10
So, the airtabs dramatically increased the variance in your mpg
(I can't think of a mechanism by which that would occur, unless they increased your susceptibility to crosswinds and one of those tanks was during a week of heavy winds).
Because of that big variance in the post-tab mpg, the distribution for the post-tab values is smeared out over a big range. If you could do three more fillups and post those numbers, we could probably get a much higher confidence.
*However*, according to
Student's T-Test (which I picked because it's the easiest to use, not because it's the most appropriate here), there's only an 8.9% chance that nothing changed between these two distributions of numbers.
That's not good enough for journal publication, but if nothing changed besides the airtabs, that's pretty damned convincing to me.