Yeah, using Oracle here. The DATE 'YYYY-MM-DD' should be compatible, and the date part extraction was separated into a second CTE, for ease of modification. Although i used Dual, a CTE named Dual can be added on top to include one record and make it all work elsewhere (and in Oracle too.)

ROLLUP is great for reporting or (less frequently) generating records. I tried CUBE, but without a serious study, the month totals seemed better done in the second query, in line. FWIW, i was too lazy to (re-learn and) write a PIVOT query. :)

Unfortunately, i did not have enough polls with rep to check the correlation between rep and posts. Though, that avenue looks promising.

It'd be a lot of fun to write these queries against the actual database. That'd not only remove clerical errors, but be on all the data. By year, by month, and so on.

There is one other effect, perhaps (it's just a guess), worth looking into. That is, if polls are changed often, does that cause subsequent polls to have a higher rating? That is, if they are changed often, people might anticipate them, and thus be more excited when one shows up. That excitement could translate into more posts, or so goes my hypothesis. If true, the quality of the poll would be of lesser significance than the frequency, vis-a-vis the popularity of a poll.