A second issue with this data is that people seem to be calculating the % of vote share and calling it the % of batch. batch size = votes - prior votes. batch breakdown by party = batch size * party share. % of batch for party = batch by party / (batch by rep + batch by dem). Substitution and simplification turns that last equation into vote_share_dem / (vote_share_dem + vote_share_rep) which has nothing to do with the batch. It's just recalculating the overall vote share while excluding 3rdparty or writein votes.
When people are including the batch size in their calculation, they are introducing rounding errors. These rounding errors are what makes the %batch for <party> jump around some (it jumps whenever vote_share_dem or vote_share_rep change).
Here's the rules for multiplying and significant digits - http://scientifictutor.org/1998/chem-multiplying-and-dividing-significant-figures/
Be careful of charts based on NYT data. When you multiply millions of votes by xx.x%, you get a potential rounding error of thousands of votes. 4 mil votes times a 0.1% rounding range is 4,000 votes. The nyt share percentage is rounded to 0.1%. .
Clearly, this doesn't apply to the big spike. BUT, be careful of saying 134,886 because it still has a potential swing of thousands of votes due to the error. the likelihood it's exactly that number of votes is very small.