Yes, Comiket is the most mainstream, but perhaps for this reason (countersignaling involved?) I've read various comments that point towards: Don't go there if you're looking for good ero-doujin.

You're just trying to diagnose a trend, right? I think a bias like that would only be important if you were trying to estimate the absolute amount or if the bias itself were changing over time so the early figures were more/less biased toward ero-doujin; also, the bias sounds like it would be to decrease any increases in ero-doujin ratios so the increases would be underestimates: if you wound up seeing a statistically significant trend upwards anyway, then you wouldn't have to worry about that one-way bias.

First and obvious thing to do, however, would be to check whether someone else has already done part of the work on something like this that I could go steal data from.

Well, don't look at me! My hafu data, while occasionally involving porn stories (mostly yaoi, for some reason...), keeps me busy enough and I haven't even learned the fancier statistics that will be involved like capture-recapture.