[stupac2]

I'll have to think a bit about what I think these numbers really mean.
One issue that comes to mind is whether standard deviation is a good measure of the kind of occasional face-plants I am trying to analyze; particularly if the underlying sample doesn't follow a normal distribution and in the presence of very small samples.
So, how does a bimodal distribution with two small spreads show up versus a single broad distribution?
Not sure if that is significant or not.

Yeah, that's reasonable. I would expect that a bimodal distribution would significantly increase stdev compared to a normal gaussian, but it obviously all depends on the parameters. I suppose you could also ask "how many games were worse than N deviations from mean?", I think I can run that in excel but at some point might want to bust out an actual scripting language on this. And, of course, you're right that these numbers are adjusted for opponent while our feelings aren't, so some of the games that feel like total clunkers really weren't.

Also, my plan to look at previous years was scuttled by the pre-2017 data not actually breaking out offensive and defensive efficiencies per game. Rats.

At any rate, if analyses like this are interesting to people I can try to [have my friend who knows how to do this stuff] scrape cfbstats or something.

[stupac2]

I'll have to think a bit about what I think these numbers really mean.
One issue that comes to mind is whether standard deviation is a good measure of the kind of occasional face-plants I am trying to analyze; particularly if the underlying sample doesn't follow a normal distribution and in the presence of very small samples.
So, how does a bimodal distribution with two small spreads show up versus a single broad distribution?
Not sure if that is significant or not.

Yeah, that's reasonable. I would expect that a bimodal distribution would significantly increase stdev compared to a normal gaussian, but it obviously all depends on the parameters. I suppose you could also ask "how many games were worse than N deviations from mean?", I think I can run that in excel but at some point might want to bust out an actual scripting language on this. And, of course, you're right that these numbers are adjusted for opponent while our feelings aren't, so some of the games that feel like total clunkers really weren't.

Also, my plan to look at previous years was scuttled by the pre-2017 data not actually breaking out offensive and defensive efficiencies per game. Rats.

At any rate, if analyses like this are interesting to people I can try to [have my friend who knows how to do this stuff] scrape cfbstats or something.