A Case for Recounts
Hillary Clinton was positioned to fend off Donald Trump’s surge if not for several low probability breaks in exactly the right places that would typically require the sort of campaign skill set that the Trump campaign never previously displayed. These breaks were not heavily weighted towards white working class voters or Trump’s message.
I examined the reported performance of both candidates in the states where Jill Stein’s campaign has called for recounts: Wisconsin, Michigan and Pennsylvania. Trump flipped all of these states from Democrat to Republican. In each state and county, the reported results are mathematically possible and explicable. At no point have statistical methods showed a clear and obvious case of fraud or manipulation. However when the results are scrutinized, several county level results in each state appear improbable.
In these states Trump underperformed Romney’s vote totals in each state’s largest GOP base counties, while Clinton outperformed Obama’s vote totals in several of the largest GOP base counties. This should have put Trump at a disadvantage in these states overall. However in the smaller held Republican counties (won consecutively by Romney and Trump), Clinton’s support collapsed in all states.
Trump managed to flip critical counties of significant size that Obama won in these states. In some cases these flips did not conflict with underlying demographic data. But several of the larger flipped county outcomes were not generally supported by demographic data. Several critical flipped counties were generally less white, more educated, more affluent and more densely populated than held Republican counties.
In held Democratic counties (won consecutively by Obama and Clinton), there were significant drops of support in certain core urban counties. Some of this was offset by stronger Democratic support in suburban counties. However Democratic support tended to drop off beyond suburban counties.
In these states Clinton underperformed Obama’s 2012 vote totals and Trump outperformed Romney to some extent.
Conventional political wisdom suggests that flipping multiple core states of the opposition is exceedingly difficult. It typically requires the incumbent political party to be involved in a major scandal (1976) or massively unpopular (1980). Furthermore accomplishing this kind of performance requires a unifying message and a resource intensive field campaign effort.
None of these factors appeared in this election. Trump’s message and disapproval numbers speak for themselves. Trump put forward one of the more anemic field efforts in recent Presidential campaign history. He had fewer field offices than Clinton in any of the contested states below and lacked much of the Republican party’s most experienced operatives. He was outspent and outworked in each state.
Field office background
http://fivethirtyeight.com/features/trump-clinton-field-offices/
Wisconsin
Counties either used paper ballots or a combination of paper ballots and DRE (Direct Recording Electronic) machine voting. Paper ballots were a mix of optical scan, marked ballots and hand counted paper ballots. Nearly all smaller counties in Wisconsin used a combination of paper ballots and DRE machine voting.
The disparity in performance for the two candidates based on whether DRE machine voting was used in a given county was the first signal that raised concerns. Once census demographic data was applied, most of the disparity was explained. However continued investigation into the results raised other questions.
The largest traditionally Republican counties in Wisconsin saw tepid support for Trump. The largest Republican county is Waukesha, where Clinton had a flat performance (+0.53%) versus Obama’s 2012 vote totals, while Trump underperformed significantly (-10.61%) versus Romney’s 2012 vote totals. The two other famously Republican counties in suburban Milwaukee were similar. In Washington county, Clinton underperformed (-9.98%) and Trump underperformed (-5.54%) as well. In Ozaukee county, Clinton outperformed (+5.26%) and Trump underperformed (-15.58%). This kind of performance in the Milwaukee suburbs would normally indicated disaster for a statewide Republican candidate. Furthermore total votes in the larger held Republican counties were down across the board. All three of these counties used only paper ballots for voting.
The smaller held Republican counties (won consecutively by Romney and Trump) that accounted for less than 1% of all votes statewide told a different story. In these counties Clinton underperforms Obama massively (range -16.15% to -36.82%). Trump outperforms Romney in most smaller Republican counties (range +5.42% to +24.16%). The total difference in votes versus 2012 for these counties wasn’t very different (range -4.27% to +0.99%). This implies little demographic change or difference in campaign efforts in these counties. Furthermore these are very white and low population density counties. These are farmers, not factory workers. Agriculture was one of the major beneficiaries of the Fed’s Quantitative Easing that followed the recession. This makes little economic sense. Yet Clinton lost 41,000 votes from Obama’s totals in small counties that neither won.
Looking at both held Republican and flipped counties, the larger counties look nothing like the smaller counties in performance change versus 2012 and in demographics. All of the smaller held Republican counties are paper and DRE machine counties. Most of the smaller flipped counties are paper and DRE machine counties.
The larger flipped counties Racine, Winnebago and Kenosha are all densely populated. All are fairly well educated. And all are much less white than the large held Republican counties. All three had drops in total votes: Racine (-9.37%), Winnebago (-3.68%) and Kenosha (-6.27%). All three also had underperformance for Clinton: Racine (-19.81%), Winnebago (-18.47%) and Kenosha (-20.28%). Trump outperformed mildly in two counties: Racine (-5.53%), Winnebago (+3.15%) and Kenosha (+3.0%). Racine and Winnebago were paper only. Kenosha was paper and DRE machine counties.
These larger flipped counties would likely appear as outliers if not for Clinton’s underperformance in held Democratic counties like Rock (-20.08%), La Crosse (-11.69) and Eau Claire (-11.07%). Trump modestly outperformed in all three counties (range +0.19% to 3.17%). These three counties were generally more white than the three larger flipped counties.
Another major factor in Wisconsin is Democratic underperformance in Milwaukee county. Clinton underperformed (-13.07%) as did Trump (-18.61%) while county votes were down massively (-11.69%). Ultimately Clinton earned 43,000 votes fewer than Obama 2012. She even fell about 9,000 votes short of Kerry in 2004. Milwaukee county is easily the least white county in the state. Milwaukee was a paper only county.
The opposite was true in Dane county. Clinton was essentially flat (+0.66%) while Trump underperformed (-14.79%) and total votes were flat in the county (+0.18%). These are the two bastions of the Democratic party and accounted for 25.12% of all votes cast statewide. Dane was a paper only county.
Between the massive disparity between Milwaukee and Dane county as well as the three largest flipped counties that were less white than held Republican counties and also saw total drops in total votes, the effect of the voter ID law is obvious.
Trump’s poor showing in the largest traditionally Republican counties should have led to an easy win for Clinton. But a combination of the low turnout in Milwaukee county and the more diverse counties that flipped along with Clinton’s collapse in smaller held Republican counties led to a narrow win for Trump.
Wisconsin was one of Trump’s bigger states in field office investments. However it was still smaller than Clinton, Romney 2012 or Obama 2012.
File
https://docs.google.com/spreadsheets/d/1zrG0capcUtgbENEPLH7QkPk4j4eXpYobiUq5n_sy_So/edit?usp=sharing
Michigan
All voting statewide was by optical scan.
Similar to the pattern in Wisconsin, the two largest held Republican counties, Kent and Ottawa trended more toward Hillary Clinton. She outperformed Barack Obama’s 2012 vote totals in Kent (+3.87%) and Ottawa (+8.28%) counties while Donald Trump underperformed Mitt Romney in Kent (-5.11%) and showed a slight improvement in Ottawa (+2.6%). This is Gerald Ford country. A weak GOP performance in these counties would normally signal disaster for a statewide Republican candidate.
Macomb county is the third largest county in the state and has been the bellwether of statewide elections. It was the big shift from Democrat to Republican in this election. Clinton underperformed Obama (-15.28%) and Trump outperformed (+17.03%). Just taking the margin of loss in Macomb out, Clinton would win the state by over 30,000 votes. Macomb county was famous for its Reagan Democrats. However census data of the county’s educational and racial demographics do not generally support this swing. Of the counties that flipped (all Democrat to Republican), only Saginaw and Calhoun were less white than Macomb county. Furthermore Macomb, Saginaw and Calhoun counties were also less white than any held Republican county.
The pattern for held Democratic counties was similarly strange. In Wayne county Clinton underperformed (-13.09%) and Trump outperformed (+7.06%) while total votes declined (-5.34%). Wayne county had the lowest total votes for a Democratic Presidential candidate since 1996 (504,466) and the lowest Democratic vote share since 1992 (60.4%). Clinton lost 77,000 votes from Obama’s 2012 result.
Oakland was a bit more conventional with Clinton underperformance (-1.73%) and Trump underperformance (-2.49%) versus their predecessors 2012 vote totals. However Genesee county bucks all conventional wisdom. Clinton underperformed (-20.34%) and Trump outperformed (+17.22%). Obama won this county with a margin of about 57,000 votes in 2012. Clinton still held on by a margin of 18,000 votes. However Genesee county is home to Flint. The Republican Governor is public enemy number one in Genesee county, yet the county trended heavily against Clinton. Obama won Genesee county with 143,927 votes in 2008 and 128,978 votes in 2012. Clinton only garnered 102,744 votes.
Given Trump’s tepid performance in GOP strongholds, Clinton should’ve sailed to a win in Michigan. Instead a series of improbable outcomes in heavily Democratic counties along with flipped counties that defied demographics pushed Trump to a narrow win.
In terms of field offices, Trump barely had a presence in Michigan and was absolutely dwarfed by Clinton, Romney 2012 and Obama 2012.
File
https://docs.google.com/spreadsheets/d/19D62OF1rgcjG0KKSCxxQAAT6iKnnl_obiqK8dXyYW4Y/edit?usp=sharing
Pennsylvania
The majority of counties used DRE machines exclusively that lack a verified paper trail. A few counties used a combination of paper ballots and DRE machines. Several more only used paper ballots.
Unlike Wisconsin and Michigan, Pennsylvania offers party registration data for each county.
In larger held Republican counties, Clinton underperforms Obama by reasonable margins (+1.79% to -18.43%) while Trump outperformed Romney by similar margins (14.55% to 1.3%). However in the smaller held Republican counties, the same collapse in support for Clinton repeats as seen in other states. In most of these counties, the scale of Clinton’s underperformance is greater than the scale of Trump’s outperformance.
The two largest Republican registration counties are actually in held Democratic counties, Allegheny and Montgomery. In Allegheny county Clinton modestly outperformed (+2.93%) and Trump slightly underperformed (-1.74%). The trend in Montgomery county was stronger with outperformance by Clinton (+7.59%) and underperformance by Trump (-7.79%). Despite the large population of GOP registered voters in these counties, Democrat registration totals are even higher. Allegheny and Montgomery counties are DRE machine only counties.
The same trend in smaller held Republican counties applies in Pennsylvania as well. Clinton’s support collapses here with generally lesser gains for Trump.
The flipped counties were few overall. Clinton flipped suburban Philadelphia Chester county. Obama lost here by less than 1,000 votes in 2012. Clinton won by a margin of 24,000 votes. However Trump’s three flipped counties were decidedly more ex-urban. In the Delaware Valley, Trump flipped Northampton and Luzerne counties. Trump also flipped Erie county.
Demographics on race, education levels, income and population density are somewhat supportive of the flips in Luzerne and Erie counties. But there is nothing in the census data that would support a flip in Northampton county. All three of these counties had Democratic voter registration advantages. However Northampton is the most educated flipped Republican county. It is home to both Lafayette College and Lehigh University. Northampton is more educated than all held Republican counties but two, which both have Republican voter registration advantages. Northampton is also less white than any flipped or held Republican county, except Berks which was a held Republican county. Northampton has a higher median income level than the other flipped counties and is within $1000 of the highest median income level in any held Republican county. Overall Clinton slightly underperformed (-2.47%) in Northampton while Trump outperformed significantly (+16.17%). Northampton last supported a Republican Presidential candidate in 1988. There is virtually nothing in Trump’s message that should have resonated here. Northampton, Luzerne and Erie counties are DRE machine only counties.
On the Democratic side, Clinton modestly underperformed (-4.34%) Obama in Philadelphia county while Trump outperformed (+9.66%) Romney. Total votes were essentially flat (-1%). Philadelphia is DRE machine only county.
In the eastern portion of the state, Clinton underperformed significantly in counties that she still held. Monroe and Lackawanna counties neighbor Northampton and Luzerne in the Delaware Valley. In 2012 Obama won Monroe county by 9,000 votes. Clinton underperformed (-9.34%) and Trump outperformed (+18.01) in Monroe county, leaving Clinton with only a 224 vote margin of victory. Monroe is a DRE machine only county.
In Lackawanna county, Clinton underperformed (-16.57%) Obama and Trump outperformed (+37.1%) Romney. Obama won 61,838 to 35,085. Clinton won Lackawanna county 51,593 to 48,102. Trump garnered 124.59% support of registered Republicans. Even if independent voters who did not vote for a third party candidate are included, Trump earned the support of 101.34% of this larger base. This implies major Democratic defections. Meanwhile Clinton votes were equal to 62.62% of all registered Democrats. That isn’t the strongest showing, but it would require an incredible effort by Trump to flip that many Democrats. Even without winning, this was the best percentage performance for a Republican Presidential candidate in Lackawanna county since 1988. Lackawanna is less educated, more white and less densely populated than most held Democratic counties. But the opposite is true when comparing Lackawanna to most Republican counties. Lackawanna is a paper only county.
Trump’s weak performance with larger GOP populations didn’t help his cause. However his major increases in Delaware Valley counties with Democratic voter registration advantages coupled with a meager Democratic showing in Philadelphia pushed Trump to a narrow victory. If the 2012 results in Philadelphia, Northampton, Lackawanna and Monroe counties were replicated in 2016, Clinton wins.
Pennsylvania was Trump’s largest field office footprint. But it was still smaller than Clinton, Romney 2012 or Obama 2012.
File
https://docs.google.com/spreadsheets/d/1FCTrCDpS8oG0GenkFWkgLSS0g_4SxeU_gkcmDTz7E7o/edit?usp=sharing