Read on Twitter
I'm re-examining MI Oakland County from first principles, since I agree that similarity of 2016 and 2020 doesn't preclude manipulation in both elections, tho I think that it counts against it.
2/ Technically, Shiva's comparison of total margins to straight ticket margins is (at best) "exploratory" - a term of art in statistics. We don't know anything about the properties of this statistic. So Shiva's claim to have PROVED manipulation is unjustified armwaving.
3/ Having said that, merely showing (as I did last night) that there were similar patterns in 2016 Oakland Co doesn't end the discussion, as commenters rightly observed, since we still don't really understand the result.
4/ with that in mind and with a clean slate this morning, to get to a more robust statistical analysis, we need something more surgical than a comparison of Straight Ticket subset to Total (of which Straight Ticket is a subset). I.e. back out the "MixedR" and "MixedD" and analyse
5/ leaving aside Other votes (at least for now), this leaves 4 columns: StrR, StrD, MixR and MixD - with the MixR and MixD having Trump and Biden respectively at top of ballot. B/c precinct sizes vary a lot, as first cut, I calculated % in each precinct.
6/ I then calculated principal components (ordinary, not Mannian), ordered by PC1 and plotted (without any presupposition). It's pretty interesting. (StrR on left; StrD on right).
7/ Some obvious points. In intense D precincts, straight D share reached 80%, whereas straight R share never reached 50% in any precinct. The proportion of mixed tickets with Biden presidential vote was ~20% in all precincts.
8/ this graph contains ALL the information in the Shiva graph plus considerably more. The downward slope in Shiva graph at right arises because of larger increase in StrR share than MixR share (as shown in replot of graph in X-order Dem to R left to right).
9/ one observation can already be made from this graph on Shiva claims: the spread between StrD and MixD (and thus between StrD and TotalD) increases even more dramatically on left than spread between StrR and MixR (and thus TotalR) on right.
10/ I.e. the increasing spread between Str and Total with greater intensity is likely to be a property of Straight Ticket voting and supports my original criticism that an exploratory analysis with novel statistical comparison cannot PROVE manipulation (or even begin to prove it)
11/ all of the underlying data comes from https://results.enr.clarityelections.com//MI/Oakland/105840/269402/reports/detailxls.zip
12/ wrangled data is online in csv http://www.climateaudit.info/data/election/MI_Oakland-2020.csv
Column headings self-explanatory
Column headings self-explanatory
13/ there are only 6 states which offer Straight Ticket option. SC is one of them. Here is County-level graph in same format as MI-Oakland precinct graph above. Increased spread between StrR and MixR in more R counties.
