Investigating the Narrative: Trump's Claims of Fraud

On November 27, 2016 Donald Trump tweeted, “In addition to winning the Electoral College in a landslide, I won the popular vote if you deduct the millions of people who voted illegally.” Despite winning the Electoral College and the Presidency in the 2016 election Trump claimed that he actually won the popular vote as well, if not for voter fraud. A serious claim in a democratic republic, so let us take the claim seriously for a moment. 

In his tweet Trump claimed there were “millions” who voted illegally, later he would place the actual level of suspected fraudulent votes between 3 and 5 million. However, as Hillary Clinton won the popular vote in the 2016 election by 2,868,686 votes, we will provide the most generous position to Trump and assume that only 2,868,686 + 1 votes were illegally cast for Clinton to provide him with the narrowest popular vote margin. (You will see why this is generous in a moment).

This represents 2% of the total 136,669,276 votes cast in that election. However, it is reasonable to interpret Trump’s tweet as claiming that illegal votes were cast for Clinton only. Taking that into account, this 2.86 million vote margin represents 4% of the total 65,853,514 votes cast for Clinton. Further, if the Clinton campaign were to commit voter fraud, one would assume it would be to the extent that would result in her campaign winning the popular vote in that state. With that assumption, as well, the margin of victory in states Clinton won was 11,226,356, or 9% of the of the 33,399,673 votes cast for her in those 21 states she won.

However, maybe this last assumption is not so sound as Clinton failed to effectively campaign in the “Blue Wall” states of Wisconsin, Michigan, and Pennsylvania; assuming these historically Democratic states would continue to vote Democratic—she lost these three states by a combined 77,744 votes. 

Perhaps it was this hubris which led to the miscalculation of where voter fraud needed to be recorded. So, let’s look at the margin of victory of states Clinton won. Her average margin of victory was 535 thousand votes. But if we are to look more closely, we would see that she won four states by more than 900,000 votes—California, New York, Illinois, and Massachusetts. The Clinton Campaign may have been arrogant enough to ignore Midwestern states, however, they were aware of the Electoral College and understood its mechanics. Winning California by 4.26 million or New York by 1.73 million votes would not advantage Clinton in the Electoral College in any way. In other words, Clinton running up the score in states like California and New York did not matter, as Trump’s victory shows, and she knew this.   

With that said, we will continue under the assumption that roughly 4% of the total votes cast for Clinton in the 2016 election were fraudulent. Though it should be noted that Trump specifically identified California, along with New Hampshire and Virginia, as specific states where he believed fraud occurred. Remember, winning California by 4.26 million votes did not win Clinton bonus Electoral College votes from California. Put another way, almost 1 in 20 of the 65.85 million votes cast for Clinton were fraudulent if Trump is correct in his assertion.

Now, at this point Donald Trump is President; he has the resources of the Federal Government to investigate instances of voter fraud should he believe they exist. Which is exactly what he did. On May 11, 2017 Trump issued Executive Order 13799, establishing the Commission on Election Integrity to be headed by Vice President Mike Pence and Kris Kobach. Kobach at this time was the Secretary of State of Kansas—the chief elections officer of that state. Without getting into the details of Kobach, he already had a national profile for his efforts within his own state against alleged voter fraud. With the full backing of the National Executive Branch, Kobach was tasked with identifying voter fraud perpetrated during the 2016 election. But remember, Kobach did not have to investigate the more than 136 million votes cast, he only had to investigate the 65 million votes cast for Clinton. And he had a 1 in 20 chance at finding evidence of fraud. His odds increase to almost 1 in 10 if the upper limit of Trump’s claim of 5 million fraudulent votes is assumed to be the actual rate of fraud.   

Less than a year later—January 3,2018—the commission was quietly disbanded without issuing a report. Tasked with finding voter fraud, the commission was unable to identify any credible instances, which is consistent with studies that show that voter fraudis incredibly rare.

So then, what happened? Why did Trump lose the popular vote?

Well, one explanation could be that both of these candidates were historically unpopular. According to a favorability poll conducted by Gallup from November 2^nd to November 5^th 2016, Trump had a net favorability of -25% (favorable 36% to unfavorable 61%) to Clinton’s net favorability of -5% (favorable 47% to unfavorable of 52%). Dating back to 1956, these two candidates were the first in the history of Gallup’s favorability polling to both have a net negative favorability. (Barry Goldwater in 1964 is the only other candidate during this period to have a net negative favorability at -4%).  

Now I’m sure there are some out there who are going to claim that this poll is skewed, just like all the other of the polls in 2016 that underestimated Trump. That is reasonable to point out, so let’s compare these favorability rates to the total vote shares. 

Clinton’s results were reasonably consistent—a 47% favorable response and she won 48% of the popular vote.

Trump’s results were not so consistent—a 36% favorable response but he won 46% of the popular vote. Now, this 10% differential could be explained by biases in polling, however, FiveThirtyEight forecasted Trump would win 44.9% of the popular vote; much closer to the actual 46%. FiveThirtyEight—a poll aggregator—averages many polls, so this could be an issue with Gallup’s polling specifically, but this suggests that polling was not as far off in 2016 as some perceive. There is another explanation as well; the Gallup poll is correct, but voters cast their ballot for Trump despite having an unfavorable opinion of him. This would be consistent with broader trends of increasing partisanship and decreasing ticket splitting in elections. 

We can dig even further into the results of the 2016 election to show that this was an election between two deeply unpopular candidates. Between 2000 and 2012 third-party candidates received an average 2% of the total vote—between 1% and 4% or 1 million to 4 million votes. In each of these elections both candidates had a net positive favorability rating from Gallup. In 2016 the percentage of voters who voted for a third-party candidate increased to 6%, or 7,830,934 of the 136 million votes cast. This is a significant increase, indicating that in 2016 American’s were dissatisfied with either of the two major party candidates.

We can look at a third metric. The generic congressional ballot, which asks survey respondents a question along the lines of, “if the election were to be held today would you vote for a Democrat or a Republican?” This is understood by pollsters to be a good metric of how friendly the national environment is to either of the major parties. Returning to FiveThrityEight, we will this time look at their polling average of the Generic Ballot as of November 8, 2016—election day. At the time Democrats held a slight lead in the generic ballot, 45.4% to 44.2%. This shows a slimmer margin than the actual popular vote margin, but it reveals a slightly more favorable environment for the Democrats than Republicans. As such, it would be reasonable to assume that Democrats would win more votes in the 2016 election based on the generic ballot results alone. 

There might be some that again are claiming that polls had a clear liberal bias running up to the 2016 election, but if you follow the link to the FiveThirtyEight site you can see that Pollsters like Fox News and the Wall Street Journal were included and highly rated in this average. 

Regardless, all of these data points are directionally consistent; Clinton, while unpopular, was more widely popular than Trump in 2016. Unfortunately for her, that popularity was inefficiently spread geographically, such that it resulted in her losing the Electoral College. 

So, looking back at the 2016 election we see that Trump cried voter fraud and was unable to provide any evidence. Further, this claim did not benefit him in 2016 as he had won the necessary number of Electoral College votes and the Presidency, yet he made the claim anyway. 

While it did not benefit him in 2016, it planted a seed. That voter fraud was occurring, perpetrated by Democrats, and on a large scale. This allowed him to claim voter fraud during the 2020 election even before the first votes were cast.

We could run through a similar exercise with the 2020 election as we did with the 2016 election, but with a larger margin of victory the share of each vote breakdown is going to be larger, so in theory would be easier to identify instances of fraud. It is worth noting that Biden won all of the same states as Clinton, plus Michigan, Pennsylvania, Wisconsin, Georgia, and Arizona. So, let’s instead focus on those five states.

After Clinton’s defeat it was unlikely that a Democratic candidate was going to again ignore those three “Blue Wall” states, mentioned above, that cost Clinton the Presidency in 2016. As for Georgia and Arizona, Mark Kelly (D) from Arizona, Raphael Warnock (D) from Georgia, and Jon Ossoff (D) from Georgia were all elected in statewide elections in 2020 as well. This at least suggests that the statewide environment was friendly to Democrats in Arizona and Georgia in 2020. For what it’s worth, of the Blue Wall states only Michigan had a second statewide election; Senator Gary Peters, Democrat, won that election. Again, indicating a favorable environment for Democrats in Michigan.

We can look at these states in another way. The 2020 election saw historic turnout. Given this was conducted during a global pandemic no one had a clear sense of how turnout would be affected. The table below shows the 2016 and 2020 vote totals for the five disputed states and the percentage increase in votes. If we assume that Trump will again win a similar percentage of the vote in 2020 as he did in 2016, then we can apply this total increase to his vote share in 2016 to predict his 2020 vote total.         

2016 – 2020 Percent Change:

	2016	2020	Percent Change
Arizona	2,573,165	3,387,326	32%
Georgia	4,114,732	4,999,960	22%
Michigan	4,799,284	5,539,302	15%
Pennsylvania	6,165,478	6,915,283	12%
Wisconsin	2,976,150	3,298,041	11%

Trump 2020 Expectation v 2020 Actual:

	Trump ’16	x Percent Change	Trump ’20 Forecast	Trump ’20 Actual	Difference
Arizona	1,252,401	32%	1,648,666	1,661,686	13,020
Georgia	2,089,104	22%	2,538,546	2,461,854	-76,692
Michigan	2,279,543	15%	2,631,034	2,649,852	18,818
Pennsylvania	2,970,733	12%	3,332,014	3,377,674	45,660
Wisconsin	1,405,284	11%	1,557,275	1,610,184	52,909

As this table shows, Trump’s vote totals trend closely with the total increase in votes, and actually outperformed 2016, with one notable exception—Georgia. This is as expected. Increasing voter turnout would make it difficult to know beforehand the number of fraudulent votes necessary to win in any given state.

So then, if voter turnout increased, and if Trump outperformed his 2016 results, how did he lose?

Well, we could turn again to Gallup’s favorability polling, this time from October 16^th through October 27^th. Trump’s favorability actually increased by 9%, from 36% in 2016 to 45% in 2020. However, he still remained net unfavorable by 9% (45% favorable to 54% unfavorable).

While Joe Biden was not viewed by a majority of American’s favorably, he had a 2% advantage over Hillary in 2016 (47%), and at 49% had a 4% advantage over Trump. Further, Biden had a net favorability of 1% (49% favorable to 48% unfavorable).

As we examined before, the favorability of a candidate tracks closely with the results of an election. Trump improved his overall favorability, and his net favorability, and so was able to turn out more voters. However, in 2020, and in Joe Biden, he did not face the third most unfavorable candidate in recent history. (Trump was the most unfavorable in 2016 with a net favorability of -25% and second most unfavorable in 2020 with a net favorability of -9%). As such, Biden was able to turn out more voters for himself.

Looking again to the generic ballot, we can see that FiveThirtyEight’s average of generic ballot polling shows Democrats with a lead, however it is much more significant in 2020. Respondents favored a Democrat, 49.9%, to Republicans, 42.6%, as of November 3, 2020—election day. I again invite you to look at the FiveThrityEight site, Nate Silver and company have made all their data and methodology available.

Let’s now turn and look at the share of votes cast for third-party candidates. In 2020 third-party voting reverted back to the mean, taking 1.82% of the total vote share or 2,884,246 of the 158,383,403 votes cast. This is well below the 7,830,934 votes cast for third-parties in 2016, despite 21,714,127 more people voting in 2020 than in 2016, a 16% increase. 

So, what if we were to perform a similar analysis for Biden that we did for Trump, with one minor tweak? We will perform the same calculation, but assume that the difference between third-party votes in 2016 and 2020 went to Biden.

2016 – 2020 Third-Party Difference:

	Third-Party ’16	Third-Party ’20	Difference
Arizona	159,597	53,497	106,100
Georgia	147,665	64,473	83,192
Michigan	250,902	85,410	165,492
Pennsylvania	268,304	79,380	188,924
Wisconsin	188,330	56,991	131,339

2016 Third-Party Adjustment to Democrat Results:

	Democrat ’16	Third-Party ’16-’20 Difference	Democrat ’16 (Adjusted)
Arizona	1,252,401	106,100	1,267,267
Georgia	2,089,104	83,192	1,961,155
Michigan	2,279,543	165,492	2,434,331
Pennsylvania	2,970,733	188,924	3,115,365
Wisconsin	1,405,284	131,339	1,513,875

Democrat 2020 Expectation v 2020 Actual:

	Democrat ’16 (Adjusted)	x Percent Change	Democrat ’20 Forecast	Democrat ’20 Actual	Difference
Arizona	1,267,267	32%	1,668,236	1,672,143	3,907
Georgia	1,961,155	22%	2,383,071	2,473,633	90,562
Michigan	2,434,331	15%	2,809,689	2,804,040	-5,649
Pennsylvania	3,115,365	12%	3,494,235	3,458,299	-36,006
Wisconsin	1,513,875	11%	1,677,611	1,630,866	-46,745

In performing this analysis, we are able to see that not all those that voted for a third-party candidate in 2016 ended up voting for Biden in 2020, particularly in Michigan, Pennsylvania, or Wisconsin. But, again, we can see that our expectation is relatively accurate based on our assumptions made. Now, why are those assumptions reasonable? Well, as we first showed, Trump maintained his share of votes case in 2020, so we could assume the same of Democrats. But, that’s not really what we did for Biden here. The real assumption is that all third-party voters ended up voting for Biden instead of Trump. I will point back to the favorability polling and the generic ballot polling as to why this is a reasonable assumption. Assuming independent voters are not going to vote for a third-party candidate, which we can see they did not from the raw voter data, and if Biden has a higher favorability and Democrats lead in the generic ballot then one can assume independent voters would favor Biden over Trump. Obviously, this is not the reality of the matter, I have shown as much above with Biden underperforming and Trump overperforming our expected vote totals in these states.

Similar to 2016, Trump and his allies have not been able to provide credible instances of voter fraud. When incidents of fraud from the 2020 election have been identified they have largely been perpetrated by Republicans. I am not going to refute the specific claims of voter fraud, many others who have far more expertise in these areas than me have already done so. It should not even be necessary as Trump and his legal team have repeatedly failed to provide any evidence of widespread, pervasive and systematic voter fraud. Despite this, Republican lawmakers in statehouses across the country are using phantom fraud as pretense for instituting stricter voting laws, when the voting structure already favors Republicans in the Senate and Electoral College. Instead, my goal here, was to investigate the narrative put forth by Trump, beginning in 2016, that he was cheated out of both the 2016 and 2020 elections.

All the data presented here is directionally consistent. What’s more, Trump made these claims of fraud before it was clearly beneficial. This could have been due to his own ego—not believing that he could be less popular than Clinton—or this could have been a deliberate strategy; knowing that he won by a thin margin in 2016 and that next time he could lose by a thin margin. If he planted the seed of the lie in 2016, it would grow. By the time of the next election, should he need it, he could harvest the fruits of that lie. After the 2020 election Trump tried to harvest the fruits of that lie, culminating with the attack on the Congressional proceedings to certify the election on January 6, 2021. That lie continues to grow, ready to be harvested again in 2024.