The Forces of Darkness

 Not As Many Friends in Pennsylvania As Clinton Might Have Hoped

‘I have a deep-seated and, hopefully, irrational fear that somehow I am under-estimating the role vitriol, class and racial hostility, basic human stupidity, and cupidity play in elections and that this map might be not generous enough to the forces of darkness, but I am sticking to my guns and to what the demographic data says to me.’

This is a quote from my recent article ‘Election Predictions’ in which I presented maps with my opinion of what the results of the election might look like based on my attempts to understand the demography and the relationship between demographic data and voting history. It is virtually the first thing I said about the election, and was said deliberately because I felt it was possible I was being too optimistic about the chances of a surge in turnout for Trump merely on the basis of his racially tinged bluster. As it happens, as I will try to show in this article, the evidence was laying there for me to see, I saw it and consciously chose to downplay it in a vain hope that perhaps, in a post-Obama world, we had moved beyond the dog whistles of racial rhetoric.

I was wrong. My calculations were in fact quite accurate and precise for the most part, as I will show, but I chose to believe that the clear evidence in the data of interest in Trump in the smaller counties of Pennsylvania could not be true as the population was not large enough to overcome the lead Clinton would amass in the large counties of metro-Philadelphia. So I made a misjudgment in my calculations that was enough to conclude that the ‘worst-case scenario’ in Pennsylvania had little chance to become a reality in Pennsylvania. It did become reality and was in fact quite predictable had I been my more misanthropic, or at least more objective, self instead of assuming the more positive alter ego I normally present to the world. I had published an article on my findings about Pennsylvania, with the premise that if Pennsylvania was going to stay blue, then it was likely that other states with similar demography like Michigan and Wisconsin, and perhaps even Ohio, would also remain in the Democratic column. As a result of an optimistic attitude about voter behavior I made a mathematical adjustment in one small set of data which was amplified as a result by the conclusions that were logical from the final results I obtained about one state, which I then applied to states that were less likely to turn red than Pennsylvania. They too voted for the Republican Party, by smaller margins than that of Pennsylvania, consistent with my hypothesis that Pennsylvania was the one to examine as the others might be close behind. So what exactly happened in Pennsylvania? I will go through the logic of my prediction to examine what was accurate and what did not work and why.

Philadelphia

In my article on Pennsylvania I started with Philadelphia and discussed the fact that the Democratic party had in recent Presidential elections run up the score in Philadelphia and then usually lost the rest of Pennsylvania and still managed to win. Obviously this did not happen this time. Clinton as of Monday November 14 has 2,841,702 votes to Trump’s 2,905,959 votes a difference of 64,678 votes in favor of the Republican.  Two possibilities exist to explain this event: first, that Philadelphia gave Clinton a weaker margin than it did to Obama, either because Trump proved much more popular than expected in Philadelphia or because voter turnout was down in the city. The second possibility is that the vote in the rest of the state was higher than normal relative to that of Philadelphia, helping to overcome the large margin achieved in the the ‘City of Brotherly Love.’ The answer is, of course, a combination of both.

Clinton received 563,275 votes in Philadelphia, 25,531 votes lower than Obama’s 588,806 in 2012. Where did those 26,000 votes go? Trump got 105,786 votes, 9319 more than Romney’s 96,467 votes, indicating at least some potential new Trump voters or Democrats switching to the Republican candidate. Third Party candidates received 14,333 votes, 9279 more than they did in 2012, indicating that another sizeable chunk went in that direction. Overall, turnout in Philadelphia was down by 6,933 voters, at 683,394 votes, 1% lower than the 690,327 votes cast in 2012. So, it appears that about a quarter of the loss was in lower voter turnout, a little more than a third went to third party candidates and the remainder went in Trump’s direction.

Overall, however Clinton left Philadelphia with a very healthy margin of 457,489 votes. How did that fit into my prediction? Compared to Obama’s margin of 492,339 votes, she gave back 35,000 votes.

Here is what I wrote in my previous article before the election:

Thus turnout is the only variable in this equation, which is clearly why the campaign has become so nasty as the Republicans attempt to make things so ugly that turnout is suppressed, which they believe will help them win. My analysis of voting trends indicates that a likely Trump maximum vote in Philadelphia would be about 98,000 votes while a ‘low’ turnout result for Clinton would net about 559,000 votes, resulting in a margin of 461,000 votes. On the other hand, if turnout is high or is as one might anticipate from normal demographic change, Clinton stands to receive as many as 590,000 votes, while if Trump fares as trends indicate for Republicans, he is likely to receive under 90,000 votes. Here the gap between the two candidates reaches 500,000 votes. So, the effective range of vote margin is between about 460,000 and 500,000 votes, a number that is almost entirely dependent on turnout.

Personally I believe that turnout will be on the higher end rather than the lower end and thus I believe Clinton will have closer to a 500,000 vote lead in Pennsylvania before any other counties are factored in. If, as expected, Philadelphia represents about 11.6-11.7% of the electorate, then the electorate will total about 5.8-5.9 million voters, of which 700,000 are from Philadelphia. Of the remaining 5.1-5.2 million voters, Trump would need to win at least 2.8-2.9 million (about 54% of the remaining vote) to Clinton’s 2.3-2.4 million (about 46% of the remaining vote; also I am ignoring third-party numbers for the moment) to win the state. Is this possible?

So, Clinton left Philadelphia with about a 2,500 fewer vote margin than the lower end of my prediction, with Trump exceeding his potential high end of 98,000 by about 8,000 votes, and Hillary receiving 563,000 votes, 4,000 votes above the low end of her potential range. Not a bad performance at all for Clinton, but worryingly low on her part if you are a Clinton supporter and surprisingly high for a Trump supporter. Turnout in Philadelphia was a fraction under 62% (61.98%), compared to 62.80% in 2012, 63.66% in 2008, and 63.41% in 2004. I had predicted turnout to be between 670,000 and 693,000 so it actually fell in the middle of my prediction. However, I had anticipated that Philadelphia would make up about 11.75% of the Pennsylvania electorate, in the general range for recent elections (12.00% in 2012, 11.94% in 2008, and 11.68% in 2004). In the event Philadelphia turnout was low relative to the rest of the state at 11.47%. Turnout in Philadelphia did not match the city’s share of either the state’s registered voters (12.64%), or the state’s population (12.24% as of July 1, 2015). It is fair to say there was an enthusiasm gap: if Philadelphia had matched the rest of the state in voter turnout percentage, one might expect closer to 735,000 votes to have been cast in the city. If the share of extra votes broke as the city’s vote broke, one might expect an additional 42,000 votes for Clinton, and an extra 8,000 votes for Trump, increasing the Clinton margin in Philadelphia to about 490,000 votes. That did not happen and represents a failure of the Democratic party to retain enthusiasm in its base: if you can’t get your vote out in the places you are popular, it is clearly going to be worse in places where your message is less well-received. I submit my hypothesis that turnout of the base is the single most important factor in elections, not getting people to switch, is supported by the evidence in this case and, as we shall see, in the rest of the state as well.

Clinton received 2,278,427 votes in the rest of the state (about 22,000 fewer than the number I predicted she needed to win, while Trump received 2,800,172 votes in the rest of the state, exactly what I predicted he needed. She lost by 65,000 votes: 35,000 votes alone were lost in Philadelphia, so not an auspicious start, but definitely not far (2500 votes) off the range of the prediction I made, which had her carrying between 460,000 and 500,000 vote lead into the rest of the state.

The Philadelphia Five County Metro Area

Things were brighter for Clinton when the entire five county metropolitan area is taken into account.

Year      D Vote               R Vote          D Margin     % of PA total Vote     4 County margin

2016     1,286,471        649,598         +636,873           33.54%           +179,384

2012      1,278,876         663,120          +615,756             34.26%             +123,417

2008      1,345,107         662,715          +682,392             33.85%             +203,633

2004      1,200,000         700,770          +499,230              33.16%            +87,124

Prediction below

2016 (Low)  1,224,000          693,000     +531,000           33.25%         +71,000

2016 (exp)  1,283,000       650,000     +633,000         33.45%       +153,000

2016 (High)  1,300,000         651,000     +649,000           34.00%        +175,000

The data above show that even with the lag in the Philadelphia vote, the overall result in the five counties (Montgomery, Bucks, Delaware, and Chester, plus Philadelphia) was quite successful for Clinton, increasing the margin from 2012 although not matching the high of 2008, and massively surpassing Kerry’s margin over Bush in 2004. I had anticipated something like this happening. Here is what I said before the election:

To summarize: The five counties of the Philadelphia Metropolitan area will likely cast one-third of the vote in the upcoming election and there is little prospect of Trump closing the margin of 600,000 votes that has been the rule the last two elections. If anything, the trend seems to favor the margin widening. Thus the remaining two-thirds of the state must provide Trump with a similar margin in order to have a chance at victory.

I have to say, I nailed that right on the head. The area represents about 32% of the population of Pennsylvania, but votes at a slightly higher rate than the rest of the state, and this election was the same as usual. The demographic headwinds are strongest here: Clinton increased the gap over Trump in the four counties surrounding Philadelphia compared to Obama in 2012. The state’s population growth is predominantly here as well, and thus it seems that, regardless of the results of this particular election as I said previously, the future of Pennsylvania looks more like this area and the counties bordering this area than it does the rest of the state.

And yet Clinton still lost the election in Pennsylvania. Where did I go wrong? Let’s go to Pittsburgh and see what happened there.

Allegheny County

Pittsburgh always plays an out-sized role in the election, owing to it’s large size but even more to its relatively high turnout. This pattern prevailed again as the chart below shows: Allegheny County, representing 9.61% of the population of Pennsylvania, contributed 10.79% of the vote on Election Day.

Year           D Vote           R vote         Other Vote   D Margin     Total Vote   Reg Voter   Turnout    % PA Vote

2016          363,017          257,488        22,668         +105,529      643,173      924,573       69.56%      10.79%

2012         352,687          262,039          7,358          +90,658        622,081      924,351       67.30%      10.81%

2008          373,153         272,347           5,936         +100,806      651,436       956,096      68.13%      10.84%

2004          368,912         271,925           4,124          +96,987        644,961      918,877        70.19%      11.18%

2016(pre)  330-355K      260-270K      20-34K      +60 to +95    613-634K       67.7-70.0%     10.75%

Interestingly, Allegheny County proved to be a bright spot for the Clinton campaign, with the Democrats winning by the largest margin in the last four elections. This seems unusual on the surface, as it seems to go against the ‘story’ of the election, that her support among white working class voters collapsed and Allegheny is 79.1% White Not-Hispanic, a much higher number than Philadelphia (35.4%). However, Allegheny does have a higher college educated population (36.9%) than the state average (28.1%), and perhaps this plays a role. I will come back to this later in the article. Clinton in fact out-performed my prediction by over 8,000 votes while Trump under-performed by a little under 3,000 votes. Turnout was also close to 10,000 voters higher than I predicted.

After 6 counties have been tallied, accounting for 41.7% of the population and 2,641,783 voters, 44.3% of the total vote, Clinton had a lead 724,402 votes, well within my predictions. As you can see from the chart below, which is a summary of the results and my predictions for each county, I had predicted a low Clinton vote for the six counties at 1,569,000  votes, while a high Trump turnout would net 962,000 votes and a high Clinton turnout would net 1,655,000 votes and a low Trump turnout resulting in about 908,000 votes. The Democratic margin after these six counties I predicted would range from a 607,000 vote to a 747,000 vote lead.  The actual results were 1,649,488 votes for Clinton and 907,086 votes for Trump, a lead of 724,402 votes. In fact, Clinton was at the high end of her scenario while Trump was slightly under-performing his ‘low’ scenario. In fact, Clinton exceeded Obama’s vote total from 2012 (1,631,473), while Trump’s number’s were lower than Romney’s (907,086) and the net margin for Clinton was larger than the Obama margin (706,314).  And yet, Clinton lost. Clearly the answer will lie in the smaller counties remaining.

Other Democratic bright spots

The brightest trends for Democratic candidates have been in places in Pennsylvania that are growing, and have higher levels of educated citizens or increasing minority populations. Centre County, the home of Penn State, is one of the fastest growing counties in Pennsylvania and represents a promising area for future improvement. In the last five years alone, the population has increased by more than 4% and the number of Democratic voters has increased dramatically from 32,867 in 2004 to 51,155 voters in 2016, with Republicans increasing at a slower rate from 38,367 to 47,653 voters. Similarly election results have tilted increasingly towards the Democratic party. In 2004 Bush won by 33,133 votes to 30,733 for Kerry, but by 2012 Obama had forged a small lead over Romney of 34,176 to 34,001 votes. In 2016, Clinton widened that margin from 175 votes to 1,499 votes, winning 36,555 votes to Trump’s 35,099. Centre County, with a high, but rapidly changing White Not-Hispanic Population of 86.9%, unsurprisingly has an educated population, with over 40% of the 25+ population possessing at least a bachelor’s degree. Hence, the pattern of counties with educated voters breaking for Clinton continues.

In addition to the aforementioned counties, in 2012 Obama won Dauphin County, the location Harrisburg, the capital of Pennsylvania. This county has been moving in the Democratic direction for the last 12 years, with Democratic registration increasing from 67,345 in 2004 to 86,930 this year while Republican registration has declined from 83,699 in 2004 to 76,867 registered voters in 2016, a net change of over 26,000 voters. The actual vote has mirrored this change: Bush defeated Kerry in 2004 by 65,926 votes to 55,299 while in 2012 Obama beat Romney 64,595 to 57,450 votes. Clinton also was victorious in Dauphin County, receiving 64,287 votes almost exactly the same number of votes as Obama got in 2012. Trump, however, closed the margin slightly by earning 60,620 votes, 3,170 more than Romney and more than McCain in 2008 (58,238), but fewer than Bush in 2004. The margin of victory for the Democrat was reduced from 7145 votes to 3,667 votes. In my prediction I had Clinton winning between 64,000 to 68,000 votes, while I predicted that Trump would get between 57,000 and 60,000 votes. As should be obvious, although both final vote counts were inside the margins I predicted, Trump’s vote was at the high end while Clinton’s was at the low end. This is a pattern in most of the remaining counties that were won in the past by the Democrats and is the start of a pattern of vote surge that would be enough for Trump to overcome the large deficit accumulated by the large vote centers of Philadelphia and the city of Pittsburgh.

Lehigh County shows a similar pattern. The long term trend is Democratic, with voter registration increasing from 88,149 registered Democrats in 2004 to 115,733 in 2016, while the Republican registration numbers have increased only very slowly from 79,364 registered Republicans in 2004 to 80,618 in this election. However, the Democratic margin of victory in 2016 was much smaller than the margin in 2012: Clinton winning by 74,777 votes to 67,677 (+7,100 votes) in 2016, while Obama won by a more comfortable 78,263 to 66,874 votes (+11,389). Again, like Philadelphia, the Trump numbers were up slightly (+803 votes) while the Clinton numbers were down more substantially (-3,486 votes). Interestingly, third party votes were up substantially from 1,845 votes in 2012 to 5,501 votes in 2016, but unlike Philadelphia, the third part of the equation, lower turnout was not a factor: Lehigh County turnout was up from 146,982 to 147,955 voters. In fact, it could be argued that the total increase in turnout almost matches the increase in Trump support compared to Romney. Similarly, Dauphin County turnout was up, from 123,608 voters to 130,160 voters (6,552), Trump support was up by 3,170 votes, third party votes were up substantially from 1,193 votes to 5,253 votes (+4060 votes), and Clinton support was down slightly (-678 votes, about a 1% drop in support relative to Obama in 2012). Incidentally, both of these counties are near the statewide average for college educated voters 25+ (28.1%): Dauphin at 28.2% and Lehigh at 28.1 %. Both counties also have a lower White population than the state average (77.4%): Dauphin at 67.5% and Lehigh at 66.9%.

One common denominator in all nine counties I have discussed thus far is that the population in ALL these counties has grown in the last five years. Only 17 counties grew by more than 1,000 people between 2010 and 2015 and ALL nine of these are members of that select club. eight other counties grew by over one thousand and will be addressed shortly. Five counties had a small increase in population, ranging from +5 to +838, while the remaining FORTY counties declined in population in the same period. Clearly this decline had an impact on the election, as all but two of these counties voted for Trump, and the two counties Clinton won went from being strong Democratic counties to closely fought counties.

The nine counties discussed so far accounted for slightly more than half the vote in Pennsylvania in the 2016 election:

Year            Total Votes        D Votes          R Votes         D Margin   Other Votes    % Total  % Pop

2016          2,995,292           1,825,107         1,070,482    +754,625      99,703      50.26%     47.78%(2015)

2012          2,923,377           1,808,877         1,083,484    +725,393      31,016       50.80        47.27%(2010)

2016 (pre) 2.9-3.0 Mil      1.745-1.845 M   1.067-1.133M  +612 to +778K  92K    50.8-50.9%

In summary, I predicted quite accurately this segment of the Pennsylvania vote. The total vote was almost 3 million, the predicted Democratic vote was 1.825 million, near the high end of my potential scenario, while the Republican vote was closer to the low end, with the result that the margin for the Democrats at this point was actually closer to the best case scenario. The only fly in the ointment is the last set of predictions, that the percentage this segment of the vote would be closer to 2012 at 50.8%, when in fact it turned out to be a bit less than that. The obvious conclusion is that voter turnout in the other half of the state was up substantially. This turned out to be the case. I predicted the turnout for the election would be between 5.7 and 5.9 million voters. The total at this point unofficially is 5,959,727 voters, almost 60,000 votes higher than the high end. Thus my prediction was for turnout in the other half of the state to be between 2.7 and 2.9 million voters. In the end the actual number of voters was 2,964,435, 64,435 voters more than the high end prediction. Clinton lost by 64,678 votes.

My worst case scenario numbers were designed to calculate the absolute low Clinton could get and the absolute high that Trump could get. The idea was that if Trump reached his peak he would beat a weak Clinton by one vote. In the event he out did his peak by 65,000 votes and more than made up for his initial deficit with a surge of support in the rest of the state after getting off to a worse start than Romney in the nine counties already discussed.

The Trump Effect

So how is it that Clinton did so poorly in the rest of the state?

1. Clinton support was down 14% in the rest of the state compared to Obama in 2012 (1,016,595 Clinton votes to 1,181397 votes for Obama, a difference of -164,802 votes).

2. Third Party Votes were up substantially, from 51,946 votes in 2012 to 112,363 votes in 2016, a difference of 60,417 votes.

3. Total turnout was up in the rest of the state relative to 2012. In 2012 2,830,293 votes were cast in the rest of Pennsylvania, a number that surged to 2,964,435 in 2016, an increase of 134,132 votes.

4. Trump support was up dramatically in the rest of the state compared to support for Romney in 2012. Trump received 1,835,477 votes to Romney’s 1,596,950 votes, an increase of 238,527 votes.

One can parse these numbers in various ways, but the central fact is that Trump was up almost 15% over Romney while Clinton was down almost 15% from Obama. Looking at the data, it seems to me that most of the new vote (134,132 votes) went to Trump, as did about 100,000 votes from Obama, while the remaining 60,000 went to third party candidates. Of course it is possible that most of the 60,000 increase in third party votes came from Romney voters, in which case Clinton lost closer to 165,000 votes from 2012. This is hard to gauge.

In the final analysis, there are three possible reasons for Clinton’s loss: Loss of support through decreased turnout or voter switching, increased turnout for Trump, and increased third party support. Based on the evidence, one would have to conclude that decreased turnout was NOT the major problem: turnout was actually up in this election over 2012 by over 200,000 votes, Clinton actually outperformed Obama in her strong areas, and Trump’s numbers were higher than even George Bush in 2004 (2,793,847 votes for Bush, 2,905,959 for Trump, 112,112 more votes). It is not clear if the third party vote increase broke from Clinton or from Trump, so the logical conclusion is that Trump took voters from Clinton and increased turnout in these areas of Pennsylvania. 

I want to break these votes into four separate categories:

1. Counties which Democrats have traditionally won that moved in Trump’s direction. These comprise a series of counties in Northeastern Pennsylvania, including Berks, Monroe, Northampton, Lackawanna, and Luzerne, as well as Erie County in Northwestern Pennsylvania. These are part of the strategy of the Republicans I termed the Scranton Strategy in my article about Pennsylvania before the election.

2. The fast-growing area of South-central Pennsylvania including Lancaster, York, Adams, Cumberland, Lebanon, and Franklin Counties, which have been strongly Republican in the past and continued to be in 2016.

3. The counties in South-western Pennsylvania surrounding Pittsburgh that were once Democratic strongholds but have been increasingly trending Republican for many years.

4. The smaller counties in between these counties, which make up a large land area, but hold only about 19% of the population of the state. These counties have also been traditionally Republican and were more so in this election.

The Scranton Strategy

This three pronged attack on support for the Democratic nominee I discussed above in Dauphin and Lehigh Counties, decreased support for Clinton, increased third party support, and increased Republican support is seen in some of the other counties where the Democrats have traditionally fared well, only the changes are more dramatic. In fact, it is in counties where the traditional support for Democrats has come from the white working class that the vote changed the most and although I predicted a dropoff in support for the Democratic nominee in this election, the actual amount of the drop was much larger than I had predicted. For instance, there are four contiguous counties in Northeastern Pennsylvania, Northampton, Monroe, Lackawanna, and Luzerne Counties which were won by Obama in 2012. Although Clinton managed to win two of the four (Lackawanna and Monroe), the margins were down dramatically in all four counties. Monroe went from an 8,000 vote Obama victory to a 200 vote victory; Lackawanna (Scranton) went from a 27,000 vote victory to a little more than 3,000 vote victory; Northampton went from 6,000 votes ahead to 5,000 votes behind for the Democrats from 2012 to 2016; And the largest shift of all occurred in Luzerne County (Wilkes-Barre), where a 6,000 vote Obama lead turned into a 26,000 vote Clinton loss!

Another county Berks County, although narrowly lost by the Democrats in 2012 went in the same direction as the above four counties, shifting strongly in Trump’s direction, as Democratic support dropped by 8,000 votes while support for the Republican candidate increased by 9,000 votes. The same is true of Erie County in the far northwest corner of the state, where Obama’s 19,000 vote lead turned into a 2,000 vote Trump victory. The following chart shows the results and provides a little demographic data which sheds a little light on the  features that these counties have in common.

County     D-vote change  R-vote change    Net Vote +R    College %     % White   % NonWhite  %C+%NW*

Monroe             35-32   -3         27-32      +5          +8                   23.2               77.0            23.0           46.2

Northampton   67-66  -1        61-71      +10        +11                  27.2               77.9            22.1            49.3

Berks                  83-75  -8         84-93     +9          +17                  22.7               73.4            26.6            49.3

Lackawanna      62-52   -10     35-48      +13        +23                  25.7              87.0            13.0            38.7

Luzerne              64-52   -12      58-78     +20       +32                   21.4              84.1             15.9            37.3

Erie                      68-55   -13      49-57      +8         +21                   25.6              85.0            15.0            40.6

Total                 379-332  -47    314-379   +65      +112                  24.3             80.7              19.3           43.6

* % College Degree + %Not White. Note College Degree is Bachelor’s or higher, 25 years plus, from 2010-2014 American Community Survey.

The average vote total percentage for Clinton in these counties was 45.5%, which is not far off from the average sum of the % College Degrees Plus % Non-White Population, 43.6%. I have noticed that merely adding these two figures is a pretty accurate predictor of the potential Democratic vote in a county. In the nine counties where Clinton won discussed earlier this figure was always above 50% and Clinton won every one of those counties. These nine counties are the only counties where the sum of the % of the population with college degrees plus the sum of the non-white population exceeds 50%.  Three counties in the above chart, Monroe, Berks, and Northampton counties have a figure between 45 and 50%: Clinton won Monroe, and held the net change in votes to under 20,000. Although she did win Lackawanna, it must be pointed out that Scranton is Joe Biden’s hometown and he campaigned heavily there as well as being Hillary Clinton’s grandparents hometown. If I were a betting man I would predict larger losses in Lackawanna in the next election.

Here is what I said about the Scranton Strategy, specifically about the four counties of Lehigh, Northampton, Lackawanna, and Luzerne in the pre-election article:  Should Clinton lose 10% of the vote total from 2012, she would still earn about 245,000 votes, so even if Trump were to match Bush’s vote total from 2004, the Democratic margin would be about 4,000 votes, a gain of about 47,000 votes. Again, this is essentially a strategy of hoping Democratic numbers fall off and trying to get the maximum turnout which, in all honesty, is unlikely to exceed George Bush’s 2004 vote total of about 2.8 million votes statewide.

In the event, Clinton got 244,398 votes in these counties, just about at the low end of my prediction. However, Trump exceeded Bush’s 2004 total of 242,000 votes by a large margin, receiving 265,466 votes. Instead of coming out with a 4,000 vote lead, Clinton trailed by 21,000 votes, a 25,000 vote underestimate. This is a clear example of a surge in support for Trump coming at least in part from registered Democrats, much like the pattern already occurring in Western Pennsylvania that I will examine shortly. I did not believe this would happen, even though I had read an article by Sean Trende in 2013 which essentially stated that white working class voter turnout was down. I just did not believe the demographics were pointing in that direction. This is the big mistake I made in my analysis, despite my belief that charismatic politicians increase voter turnout.  In conclusion, Clinton and the Democratic Party seem to have lost the White Not-College Educated vote, and in these counties, which were all won (except Berks which was very close) by Obama, Clinton gave back 112,000 votes of Obama’s 310,000 lead. Whether this means they need to go and win them back next time is something I will address at the end of this article.

*Note- one other county I did not discuss in the previous article seems to fit into this category as well. Schuylkill County, between Berks and Luzerne County, also saw a higher than expected turnout, a drop in support for Clinton versus Obama and a large surge in Trump support. Schuylkill County is, however, over 91% White and fewer than 15% of residents 25 and older have a Bachelor’s degree, for a %CNW of only 23.3. Although Obama lost the county by 56%-43% in 2012, Clinton’s loss was of a much greater magnitude at 70%-27%, one of the largest drops of any single county, and a clear example of the Trump effect.

Lancaster County and the southern border counties.

Outside the five county Philadelphia metro area, the major source of population growth in Pennsylvania is in the counties to the west of Philadelphia and south of Harrisburg, the southern counties if you will, as they mostly border Maryland and Gettysburg is in the center of this area, in Adams County. Traditionally this has been very fertile territory for Republicans and proved to be again in 2016.

Lancaster County is the outlier in this group of counties. It is very large, the sixth largest county in Pennsylvania at 536,624 residents in 2015. it has been very faviorable to Republicans in the past but, owing to its relative proximity to Philadelphia, has become increasingly Democratic, both in registration and in voting outcomes. In 2004, registered Republicans outnumbered Democrats by 184,852 to 82,172 and Bush took 66% of the vote in the 2004 election. By 2012 the Republican margin had dropped by 35,000 to 167,063 Republicans and 100,056 Democrats, and Romney took 58.74% of the vote in the 2012 election. In 2016 Democrats had increased the registration figures by 8,000 and slightly cut into the Republican edge. In the 2016 election, Clinton increased the democratic vote total from 88,841 to 90,066, but there was an increased turnout of 17,000, almost an 8% increase, and most of that went to Trump and to third party support. Trump’s margin over Clinton was 137,145 to 90,066 votes, a difference of 47,079 votes versus 2012, where the margin was 42188 votes, a difference of 5,000 votes.

Here is what I wrote about Lancaster County: Bush’s winning margin of over 71,000 votes is unlikely to be matched. Even the 2012 margin of Romney over Obama, 42,000 votes, is unlikely to be much expanded and is likely to be slightly reduced by Clinton.

I was right that Bush’s 71,000 vote margin would not be matched but Trump did pick up 5,000 votes despite Clinton surpassing Obama’s 2012 numbers. Another example of the surge in turnout for Trump. The Trump Effect is real. Ironically, the results in Lancaster hold out hope for the Democrats in the future, as the result is qualitatively different from that of counties like Luzerne. Here, Clinton support increased and Democratic support in general is increasing, in a county that is one of the few growing counties in Pennsylvania. Using my rubric of %College +% Non-White(%CNW), Lancaster was at 39.8 in 2010 and was in 2014 at 42.1. In another six years, the population growth and the increased %CNW will definitely tilt the county in favor of the Democratic party.

The other counties in this area have been heavily Republican and the trend continued in this election. Of these counties I had expected her position to worsen more by population increases coupled with turnout and this is what happened. In York, Franklin, Cumberland, Adams and Lebanon counties, Obama fared extremely poorly.

County           2012 D       2016 D           2012 R        2016 R       +R 2012   +R 2016    Turnout 2012    2016

York                73,191         67,428            113,304       126,933       +40,113     +59,505      189,229       203,153

Cumberland 44,367         44,282             64,809       65,649       +20,442    +21.367      110,814        115,068

Franklin         18,995        17,322              43,260       49,554         +24,265   +32,232       63,078         69,345

Adams           15,091           14,077             26,767        31,249         +11,676      +17,172       42,457        47,138

Lebanon         19,900        17,860             35,872       38,804         +15,972     +20,944      56,580        58,906

Total               171,544       160,969          284,012     312,189         +112,468   151,220      462,158     493,610

Here is what I said about York, the only county I specifically discussed in my previous article: Even if Trump does very well in York, Clinton is likely to get about the same number of votes as Obama, diminishing any presumed success of the Republicans in this, their largest single source of remaining votes, by a large margin.

WRONG. This county and ones like them were the largest source of error in my calculations. First of all, I underestimated turnout, which was almost 7% higher than in 2012. Second I overestimated Clinton’s support not by much but she did lose 10,000 votes. Third, I totally underestimated the Trump Effect here, where most if not all the increased turnout went to Trump. The above counties, including Lancaster County gave Trump 50,000 more votes than I had anticipated.

Together with counties like Lackawanna, I underestimated Trump support by about 60,000 votes, while Clinton lost about 10,000 more than I had anticipated. The net result was a difference of 70,000 votes and was the difference in the election. There was one scenario I discussed in my previous article on Pennsylvania where Trump could win, by increasing turnout in counties like these and the ones in Western Pennsylvania, as well as the smaller counties, to overcome the deficit from 2012. Unfortunately, in my opinion, this is exactly what happened.

This already conservative area seems to have had the largest increase in turnout, unsurprising considering these counties are all growing. Although Clinton lost only about 10,000 votes from Obama’s total in 2012, Trump increased his total by 28,000 over Romney’s total, while turnout was up a little over 30,000 voters, again an indication of a surge in support for Trump. As this is the only region in Pennsylvania that is growing and is conservative, it seems to be the key to continued Republican success in Pennsylvania. With the exception of Cumberland County, which seems to straddle the Franklin and York type populations as well as the Dauphin County type demographic (all of which Cumberland has common borders with), these counties seem to be growing as well as increasing the registration gap between Democratic and Republican, and so will likely become the focal point in future of any sustained effort to retain Pennsylvania by the Republicans in 202, a task that will be very difficult as I will explain below.

Western Pennsylvania

In this part of the state, my data was more consistent with the result. I had discussed how the likelihood was that the Republican surge in voter registration reflected switching of Democrats who already voted Republican, rather than a true rise in new Republican registration. In Westmoreland County, for example, Democratic registration declined from 122,432 in 2012 to 113,584 in 2016, a drop of almost 10,000 voters, while Clinton’s vote total was 59,506, compared to Obama’s 63,722, adrop of only 4,000 votes. Clearly some people were voting for the Republican Party already and just officially switched. On the other hand, I was surprised that the turnout in these areas was as high as it was, considering that these are some of the fastest declining counties in the entire country, not just in Pennsylvania. This is consistent with my “last stand hypothesis,” that angry, white, less well-educated men (and women) in areas like Westmoreland County are declining in number and in influence, as well as in material well being and that they are lashing out at their perceived enemies and relying on an unlikely ‘saviour’ to restore them to what they perceive as their rightful place. Only, as these voters are older and die at a higher rate than the average old person in more well off counties, coupled with the fact that these counties are decreasing in population while other parts of the state grow, it seems that this would have to represent a dubious ‘high-water’ mark.

County                D 2012     D  2016    R  2012    R  2016   +R 2012  +R 2016   Turn 2012(%)  Turn 2016 (%)

Westmoreland   63,722     59,506      103,932   116,427    +40,210   +56,921        71.3              75.6!!  181,740

Washington        40,345     34,436      53,320      58,941     +12,975    +24,505       66.6              69.9      96,945

Butler                   28,550     28,560      59761        64398      +31,211     +35,838      72.9              74.8      97,014

Beaver                  37,055     30,225       42,344      46,081     +5289       +15,856      69.9              69.6      79,036

Total                    169,672    152,727     259,357     285,847   +89685     +133,120    70.2             73.1      454,735

Predicted            148,000-163,000        261,000-286,000          98,000-138,000    70.6-72.8%     434-448K

These counties were predicted to break even more heavily for Trump than they had for Romney and this is true. Again, however, I underestimated turnout slightly, predicting at most 448,000 voters in these four counties, while in the end there were 454,735. Also, while Trump’s final total of 285,847 votes, was within but at the high end of my prediction, Clinton’s total of 152,727 was at the low end of my prediction for her. So, the maximum differential I predicted would be 138,000 and it ended up at 133,120. Thus, despite the drop in support for the Democratic candidate, I had folded this into my model based on changing demographics, changing voter registration, and changing voting patterns. My model worked in this part of the state, with the possible exception of slightly underestimating turnout.

The Rest of the State

The above counties are the 25 largest counties in Pennsylvania (a 26th county, Adams, is the exception, as it is 32nd in population, but is growing rapidly). These counties represented a little more than 84% of the total vote in 2012. The remaining 42 counties account for 2,410,238 people, 18.83% of the population of Pennsylvania, a drop of 44,584 residents in 5 years. However, registration totals were down only about 7,585 voters, and the Republican voter registration advantage increased by over 112,000 voters.  The result was a very large win for Trump, by a much larger margin than that of Romney over Obama. Trump defeated Clinton by about 710,000 votes to 280,000 votes in the 42 smaller counties, compared to Romney’s 602,000 to 362,000 vote victory in 2012. Their are two main differences: the first is that voter turnout was up by more than 30,000 votes in these counties, despite the decline in voter registration and population. The second is the large shift from the Democratic Party to the Republican Party in 2016 compared to 2012. Not only did Trump receive new votes, he also got a large share of previously Democratic voters. Overall, 1,027,000 votes were cast in these counties in 2016, 17.23% of the total vote, higher than the 991,143 votes cast in 2012, which also represented 17.23% of the vote.

Here is what I wrote about these counties before the election: Clinton, according to the type of calculations I have described above for the counties I have discussed, will at the absolute worst likely have over 2.5 million votes after 25 counties. Even if she performs 10% worse than Obama did in 2012 in the remaining pool, Clinton will still receive over 2.8 million votes in Pennsylvania.  

In fact Clinton did have over 2.5 million votes after 25 counties (2.561) and she did receive over 2.8 million votes (2.841). The problem for her was the surge in Trump voters who did not vote in 2012, which boosted his totals by over 120,000 votes over the previous Republican high vote, set by George Bush in 2004.  In these smaller counties alone, despite a drop in voter registration and in population, the share of the total vote cast in 2016 did not change at all compared to 2012, despite the overall increase in turnout. This strongly implies that there was a voter surge for Trump, once again the so-called Trump Effect.

Here is what I wrote about the likely Trump vote: Trump would need to have 120,000 more votes than Romney to match the total of George Bush, and assume the impact of third-party candidates is negligible, in order to perhaps match Clinton’s worst possible vote total. Based on the evidence presented, I think that these are all unrealistic situations. In the best of circumstances, Trump will perform slightly better than Romney, perhaps even matching Bush, although the presence of that small increase in third-party voters will make even that task very difficult.

This is the part I missed, the increased turnout in smaller, more white counties. I suppose I noticed that voter turnout in previous elections had been lower than the turnout in more urban and suburban counties, but I assumed this was a constant phenomenon and not a variable one. In the end, Trump won not so much by taking potential Clinton voters, which he did do in certain parts of the state although notably, not in the largest counties in Pennsylvania, but by getting a lot of voters to turnout for the election.

Conclusions

The first conclusion I can make is that my methodology worked for the most part but was colored by my own prejudice. I noticed the rise in Republican registration, I noticed the general rise in voter registration, and I noticed the rise was occurring even in places where the population was dropping. I should have been able to say that these voters were likely to turnout in larger numbers than in previous elections, but I refused to believe the evidence out of, quite literally, a lack of faith in those very voters to turn up on election day. In the event it is fair to say that the combination of extra turnout and a drop in support for the Democratic party in more white and less-well educated regions of Pennsylvania was responsible for Trump’s victory.

Is this something that is reproducible? I have serious doubts about that. First of all, turnout in some of these counties, such as Westmoreland County, was in excess of 75%, which is a remarkably high number. Even if the enthusiasm those numbers reflect were sustained until 2020, the simple fact is that there will be fewer voters in counties like Westmoreland in four years as the aging population dies at an increasing rate as the Baby Boomers reach their 70s. This election represented a ‘perfect storm’ of a large pool of older white voters with sufficient enthusiasm for the charismatic candidate that turnout was generally higher than in previous elections. This pool will be much diminished in four years time and the enthusiasm gap will likely not be sustained as it is unlikely that these once bustling counties will recover anything like the industrial capacity they had fifty years ago, despite the promises of their ‘saviour’.

Another factor leading to the same conclusion is that the counties with the highest growth increasingly have the type of demographic that the Democratic Party excelled at attracting in 2016; educated, young, Black, Hispanic, Asian, LGBTQ. These voters will increase in 2020, particularly if the expected crackdown on civil liberties ensues, which will succeed mainly in bringing out the type of voter Republicans would like to stay home. Of course, history has a funny way of going in unexpected directions, so all of this assumes the conditions of today’s election exist in much the same way in 2020 which may not be the case. Wars, abrogation of voting rights, voter fraud or intimidation, selective pots of government money thrown at certain groups to try to peel them away from the Democrats: all is possible. My essential conclusion, however, is that the method used to win in 2016 by the Republican Party will fail in Pennsylvania in 2020 as a result of continued demographic change and once Pennsylvania has returned to the Democratic fold, expect other similar states to follow suit, especially as the margin of victory in those states was even smaller than the margin in Pennsylvania.

As I mentioned earlier, the combination of %College Degree and % Non-White (%C+NW) seems to be a decent predictor of voting outcome, with higher %C+NW correlating with Democratic success. The Five County Philadelphia Area as well as Allegheny County, Dauphin County, Centre County, and Lehigh County all have C+NW numbers greater than 50. Monroe, Berks, Northampton Counties are all between 45 and 50 and were the tightest contests in the state, with Clinton winning Monroe and losing the other two by smaller margins than the typical county. Lancaster is moving in the same direction. Hence, if trends continue, the eight largest counties in the state, and half of the next dozen largest counties will have a %C+NW number of 50 or larger, which would seem to indicate a difficult path for the divisive campaign which worked so well in this election for the Republican Party.

Should the next Democratic candidate for president go after these white working-class voters? I say no. Better to have progressive policies that will help everybody and hope some come along for the ride. The politics of division are not a recipe for long term success in Pennsylvania, nor in the country as a whole. It might have a short term payoff but is a long term disaster for the Republican Party. Apparently they have decided to hitch their wagons to it, and they will inevitably pay. In the short term we will all pay, perhaps more than we might want to pay.

As for other states, North Carolina and Florida disappointed, but both saw large increases in turnout for both sides that, if continued in the future, will eventually overwhelmingly favor the Democratic Party if current trends continue. Three other states are trending inexorably towards the Democrats as well: Arizona, Georgia, and Texas. If the current dynamic holds, there is little reason to doubt that the inevitable result of the politics of division and fear will be control of the Electoral College by the Democratic Party. California, Texas, Florida, New York, and Illinois are the five largest states and North Carolina, Virginia, Georgia, Florida, Arizona, and Nevada are the fastest growing and are all trending Democratic, while the large northeastern states that Democrats lost in this election are growing much more slowly. Any growth that does happen will likely involve an increase in the non-White population, such as Pennsylvania, which will inevitably swing the pendulum back into the Democratic column. My argument still holds, albeit my prediction was a bit premature. If, by some miracle, Trump actually succeeds in achieving what he claims to be able to do, bring manufacturing back, and provide better jobs for those in the ‘forgotten’ places, he will have a shot at re-election, but by 2024, regardless of the results, the demographic swing will be too great to resist except by immoral or illegal means if the same rhetoric continues to define the Republican Party. The ‘bullets of blue’ are continuing to spread and the red sea is parting, whatever the result of this particular election.