Have the historical quirks of the Electoral College been weighing on your mind lately? Mine too. And so I thought I’d see whether I could find anything worth sharing on the subject here at Griffonage-Dot-Com that hasn’t already been covered many times elsewhere. For example, there’s the fact that in the United States of the 1780s and 1790s, “Electoral College” meant one thing and one thing only: the undemocratic assembly of prince-electors that chose the Emperors of the Holy Roman Empire. And it’s neat to see a British magazine imagining a U. S. Electoral College victory at odds with a popular vote in 1872, four years before this (possibly) happened for the first time:
But I’ve ended up devoting this blog post to a few graphs displaying statistics about United States presidential elections over the past 120 years. I was curious to see how much of a difference the Electoral College has actually been making in the outcome of elections, and how much of its impact has been due to specific components of the system such as “winner takes all” at the state level and the differential weighting of votes in different places. Since I couldn’t easily find any historical graphs of the kind I was looking for out there in the interwebs, I went ahead and created some myself.
Opponents of the Electoral College typically argue that it violates the “one person, one vote” principle: an individual vote in Wyoming, for example, weighs much more heavily than an individual vote in Florida. Proponents of the Electoral College like to counter that it protects the interests of smaller states against those of larger ones, or those of peripheries against those of population centers. However you might feel about either of those arguments, the fact of the matter is that rescaling the vote in and of itself—by which I mean the boosting of Wyoming votes relative to Florida votes, and so on—now hardly affects the outcome at all in practice. Below is a graph comparing the raw popular vote from 1896 to 2016 with the sum of popular votes in the states rescaled to match their proportional representation in the Electoral College (the whole scale runs from 0 to 1).
During the earlier part of the time span shown here, rescaling could have had a significant impact on the relative representation of political preferences by region, since it consistently boosted votes in Democratic-leaning areas and depressed votes in Republican-leaning ones. But the impact of such rescaling would have diminished greatly after the middle of the twentieth century. To show things a little more clearly, here’s a graph of differences between the raw popular vote and a rescaled popular vote for Democrats (blue) and Republicans (red) over the period 1896-2016, showing how much rescaling alone would have added or subtracted from their respective portions of the vote (on a scale running, once again, from 0 to 1):
The lines cross for the first time between the elections of 1960 and 1964, and for the last time between the elections of 1980 and 1984. Since then, rescaling has consistently favored Republicans, but only very slightly—and certainly not enough to account in itself for any conspicuous discrepancies between the popular and Electoral College vote. Whether we see it as a valid mechanism for protecting the concerns of less populated regions or an abhorrent vestige of slavery, it just hasn’t been making much of a difference. With rescaling alone, Hillary Clinton would still have won in 2016, notwithstanding statements such as this—
Some Democrats have argued that the Electoral College is undemocratic because it gives more weight to less populated states. That is how Hillary Clinton, who got more than 2.6 million more votes nationwide, lost the election to Trump.
And when I read remarks like these—
“Our Founding Fathers established the Electoral College because those larger states, those larger areas, don’t necessarily need to be the ones that rule,” said Mary Sue McClurkin, a Republican elector from Alabama….
“I feel like the Electoral College gives a very fair perspective, so that those who are in the rural areas are able to have an equal voice with those who are in the urban areas,” said Oklahoma elector Lauree Elizabeth Marshall.
—I can’t help but think many people are under the mistaken impression that rescaling is still having the sort of dramatic effect it had during the first half of the twentieth century.
A far more impactful feature of the Electoral College has been state-level rounding: the practice of determining the winner in each individual state—or, more rarely, in districts within states—and then throwing the weight of the whole regional electorate behind that candidate, regardless of the narrowness of the victory. If we think of an election not as a contest among champions with particular rules, but as a measurement of the will of the governed that aims at accuracy, this results in a compounded error equivalent to taking a bunch of fractional measurements (0.17+0.39+ 0.88+0.22+0.91+0.47=3.04) and then rounding each of them to 0 or 1 for convenience before adding them (0+0+1+0+1+0=2) rather than doing the rounding afterwards (0.17+0.39+0.88+0.22+0.91+0.47=3). The localized “winner takes all” approach originated with the self-serving desire of majority parties in state legislatures to maximize their impact, and once some states were doing this, other states followed suit in order not to yield an advantage. Maine and Nebraska round differently (two electors to the statewide victor, plus one each per congressional district), but they still round. No state apportions its electors according to the actual spread of its popular vote. Even if one did, an integer between 3 and 55—depending on the available number of electors—couldn’t reflect the real decimal precision of the spread in the popular vote, although the only reason a state couldn’t award, say, 3.14 “electors” to a particular candidate (much as the average American family notoriously consists of 3.14 “persons”) is that these are supposed to be real individuals capable in theory of intelligent deliberation, each of whom will in the end support a single candidate. In other contexts, the consequence would be regarded as a mathematical error, not just as the way the process is set up to work. As it stands, the impact of razor-thin wins is boosted, while the impact of decisive wins is diminished; and this only within designated geographic units (states or, rarely, congressional districts) rather than others (particular cities, say, or larger regions such as “the Midwest”). Arbitrary, to be sure. But to what actual effect? How has this played out in practice from election to election?
One way to think about this is that there’s only one rather narrow path towards casting an ultimately successful vote for president. You first need to vote for the person who wins your state (and/or your district, if you’re in Maine or Nebraska), and then that person needs to go on to win at the Electoral College as well. Otherwise your vote fails at either the state or federal level. A vote for Barack Obama in Indiana in 2012 was thus just as unsuccessful as a vote for Mitt Romney in Illinois that same year: it played no part in his victory, even though he ultimately won. Discounted votes for the winner aren’t generally separated out in considerations of the relationship between popular and electoral votes, but they arguably should be. Below is a graph comparing the percentage of successful state-level vote to the straightforward popular vote in favor of the winning candidate for each presidential election back to 1896:
The darker blue-green shaded area represents the proportion of the vote that actually ended up electing the president each time around, showing that as few as a quarter of the votes cast can suffice. For Donald Trump in 2016, the figure happens to have been 30%. The last time a majority of presidential votes cast were ultimately successful was in 1984, with the reelection of Ronald Reagan (57.6%). The average going back to 1896 is 42.6%, with a range running from 25.2% in 1916 to 60.4% in 1936.
Next, here’s a graph showing the proportions of total popular vote that have been successful at the state level (dark red and blue) as well the proportions that have been excluded from federal consideration through state-level rounding (light red and blue). The black bars at the edges indicate the rare instances when a third party has received electoral votes.
The previous election with the profile that comes closest here to 2016 is, curiously enough, 1916. In both years, the Democrat won the popular vote but not the “successful” popular vote, the only two times such a discrepancy arose during the whole period under investigation.
Wilson (D, 1916): 9,126,868 total, 4,672,669 successful
Hughes (R, 1916): 8,548,728 total, 5,237,688 successful
Clinton (D, 2016): 65,791,939 total, 33,289,110 successful
Trump (R, 2016): 62,941,178 total, 40,879,440 successful
In the case of Wilson versus Hughes, the rescaling of the votes of individual states threw the balance back in Wilson’s favor—remember, this was back when rescaling significantly favored Democrats—such that he narrowly won the electoral vote in addition to the popular vote; but of course that didn’t happen in the case of Clinton versus Trump. Compare the results of the 2000 election, which were plenty anomalous, but not in the same way:
Gore (D, 2000): 50,996,582; 29,160,500 successful
Bush (R, 2000): 50,456,062; 27,773,748 successful
In that case, there was no discrepancy between the popular vote and the “successful” popular vote, and the much-discussed reversal of the electoral result came only with the superimposition of rescaling.
Finally, here’s a pair of graphs that places the various tabulations of the historical vote we’ve been considering on common axes for comparison. The yellow line shows the straightforward national popular vote. The gray line shows the sum of popular votes within each state or district rescaled according to proportional representation in the Electoral College. The orange line shows the sum of all votes that were successful at the state level (i.e., factoring out any votes for candidates that didn’t win their states or districts). And the blue line shows the actual Electoral College result.
The gray line has almost exactly followed the yellow line since the middle of the twentieth century, confirming that the much-touted rescaling in favor of smaller states has virtually no effect on the outcome nowadays. Before that time, it had tended to favor Democrats, as I observed earlier. In fact, one oddity I turned up in the course of my analysis was the 1896 election, in which William Jennings Bryan lost both the popular and electoral votes to William McKinley but would have won the rescaled popular vote 49.3% to 48.3%.
Meanwhile, the blue line follows the orange line pretty closely, showing that variations between the Electoral College result and the popular vote have been due mostly to compounded state-level rounding—and not to any rescaling that protects the interests of small states against those of large ones. The Electoral College result has sometimes tugged the result rounded at the state level a little bit back towards the popular vote, but not by very much since the early twentieth century.
Supporters of the Electoral College in its current form should realize that what they’re defending is mainly the practice of state-level rounding, which erases cumulative differences between wins of different magnitudes. The case I’ve seen made in favor of this arrangement runs something like this: (1) the votes in battleground states, more than in other states, are accurate bellwethers of typical national sentiment; (2) if candidates can focus their general-election campaigning on just those few states, that’s less costly than running a truly national campaign would be; (3) the distinctive local interests of battleground states encourage candidates to tailor their messages more “inclusively” than they would for a national popular vote; (4) in the case of a close election, it’s easier to carry out recounts in a handful of battleground states than it would be to do the same thing at the national level; and finally (5) the United States as a whole “is a republic, not a democracy,” such that appeals based on democratic ideals miss the point. Note that I’m only summarizing a set of arguments here which I don’t personally find very compelling.
Meanwhile, opponents of the Electoral College should understand that the rescaling which seems to violate the “one person, one vote” principle so egregiously hasn’t actually been making much of a difference. In my opinion, the more meaningful violation of this principle is state-level rounding, which ends up denying a “voice” altogether to quite a lot of people at the federal level, both urban and rural, rather than simply weighting their votes differently.
Ultimately, I find the real effect of the Electoral College for the past fifty years or so has been to amplify overwhelming victories and to randomize the outcome when things have been close. It has probably also played a part in shaping the messages candidates have put forward during the general-election campaign, but that doesn’t lend itself to quantitative analysis.
Notes on the calculations that went into my graphs
- Two states—Maine (since 1972) and Nebraska (since 1992)—split electoral votes between two electors based on the statewide winner and individual electors (two in Maine, and three in Nebraska) based on the winners of congressional districts. In cases of a discrepancy, I’ve evaluated the success of each individual vote as based 50% on the statewide-vote elector and 50% on the congressional-district elector.
- I’ve counted Unpledged Electors as “Other,” even when they were for the winning party, since they generally opposed the official party nominees. In Alabama during the 1960 election, voters choosing among individual electors selected five pledged to the Kennedy/Johnson ticket and six unpledged, so I’ve counted the Kennedy/Johnson votes there as 5/11 successful and the unpledged elector votes as 6/11 successful.
- In 1916, West Virginia voters chose seven electors for Hughes (R) and one for Wilson (D), and I haven’t easily been able to find a relevant breakdown of popular votes by district. I’ve counted Wilson votes there as 1/8 successful, and Hughes votes as 7/8 successful. This is also how I’ve handled earlier splits of the same kind: California (1896, 1912), Maryland (1904, 1908), and Kentucky (1896).
- I’ve extended my analysis only back to 1896, since the 1892 election was unusually complicated (with numerous votes split at the state level plus the partial “fusion” of the Democrats and Populists).
- My sources of raw data have been this for 2000-2012 popular vote totals by whole state; Wikipedia for other data; and this for cases where the Wikipedia data didn’t add up to the stated totals. For 2016, I used the figures at Wikipedia at the time of writing, but with certified results substituted for unofficial ones whenever I could easily find them.