Wednesday, August 17, 2016

The Spoiler Effect--Is it Real?

With this year's Presidential candidates not exactly inspiring confidence in the voters, the typical talk of third party candidates tends to come up in political discussions. This of course inevitably leads to topics like the spoiler effect and first-past-the-post-voting. The common wisdom is that you are throwing your vote away if you vote third party. Worse yet, you are actually helping the mainstream candidate you are the most ideologically opposed to because you are taking a vote away from the candidate you would have voted for had the third party candidate not run. This sounds scary, but is it true?

One of my favorite political podcasts is Abe Lincoln's Top Hat, hosted by comedians Ben Kissel and Marcus Parks. I was listening to an episode the other day where Kissel claimed that if Perot hadn't run in the 1992 election, Clinton would have still won because Bush would have needed 100% of Perot's votes to beat him. Since this statement was against the conventional wisdom, I decided to investigate the claim.

First, let's talk about what actually happened in 1992. Perot had dropped out of the race, but returned. After election day, Clinton was the winner with 370 Electoral votes. Bush won only 168 Electoral votes, and Perot won no Electoral votes. Here's the strange thing, though: Clinton only won 43% of the popular vote. Yet, he beat Bush in the Electoral college by more than a 2-to-1 margin. How was this possible?

As I'm sure the reader is well aware, the American people don't directly vote for the President. The states do. The states vote for the Electors who actually cast the ballot for the President. The problem with the Electoral College is that the 538 votes are not distributed evenly by population. Every state (and the District of Columbia) gets 3 votes, right off the bat, regardless of its population. The rest are distributed proportionally by population. The result is that states with lower populations, like Montana, are actually over represented in the Electoral College, while highly populous states like California and Texas are under represented.

Additionally, the vast majority of states (with the exception of Nebraska and Maine) are "winner-take-all" states. That means all a candidate has to do is get the largest share of voters in any given state and they get all of that state's electoral votes. In a two-party race, it sort of represents the will of the people most of the time because a candidate has to get at least 50% of the popular vote to carry that state. Even so, there are several incidences where the winner of the popular vote was not elected President because the results of the Electoral College differed. But as my data will show, this problem is greatly exacerbated when a third party does unprecedentedly well, as we saw in 1992.

In 1992: Bill Clinton won a number of states with less than 50% of the popular vote. Those included:

New Hampshire
New Jersey
New Mexico
New York
Rhode Island

That's a lot of states, and many of them weren't states typically won by a Democrat, not even in 1992.  So the question is, would Clinton have won those states had Perot stayed out of the race? Which would he have won and lost?

The problem with answering questions like these is quite simply that we don't know what would have happened because it didn't happen. The best we can do is guess, based on the available information. So what do we know about Perot voters?

Perot was running as an economic conservative. In much the same vein as Donald Trump, he was vehemently against Free Trade Agreements, NAFTA being the hot-button topic of the time. I think it goes without saying that Perot's supporters had more in common ideologically with Bush than with Clinton.

Thus, it stands to reason that most of them would have voted for Bush. The question of how many is difficult, because again, there's no way to know that for sure. I am unaware of any polls indicating what Perot voters' "second choice" would have been. There's also a good reason to believe many of them would have stayed home rather than vote for either Bush or Clinton. And yet again, there's no way to know how many would have.

So, I decided to make a spreadsheet, listing all of the state's Clinton won, how many electoral votes each of those states had in 1992, and how many actual votes each of the three main candidates received. Then, I calculated a number of scenarios. Which were as follows:

1. Clinton and Bush share Perot's voters equally: 50/50
2. Bush gets 50% of Perot's voters, Clinton gets 25%, and the remaining 25% stay home on election day. This, I believe, is the most realistic scenario.
3. Bush gets 75% of Perot's voters, Clinton gets 25%. This is another believable scenario in my opinion.
4. Bush gets 100% of Perot's voters. Unrealistic in my opinion, but not impossible.

So what were the results?

Scenario 1: Bush and Clinton Share Perot Voters 50/50

Map of Scenario 1. Also the actual results of the 1992 Election 
In this scenario, nothing changes. Clinton still wins all of the states he won, and Bush wins all of the states he won. Clinton still wins.

Scenario 2: Bush gets 50% of Perot Voters, Clinton gets 25%, 25% Stay Home

In this scenario, Clinton still wins, but it's close. Bush gains Colorado, Kentucky, Montana, Nevada, New Hampshire, New Jersey, Ohio, and Wisconsin. But that only gets him to 255 electoral votes, 15 shy of the 270 needed to win.

Scenario 3: Bush gets 75% of Perot Voters, Clinton gets 25%

Map of Scenario 3
This is the turning point. If Bush had gotten 75% of Perot's voters--a very realistic possibility in my opinion--he would have won the 1992 election in a landslide. In addition to the states he picked up in Scenario 2, he would have gained Connecticut, Delaware, Iowa, Louisiana, Maine, Michigan, Minnesota, Missouri, Oregon, Pennsylvania, Tennessee, Washington and Wisconsin. That would have put him at 377 electoral votes to Clinton's 161, easily winning his second term.

Scenario 4: Bush gets 100% of Perot Voters

Map of Scenario 4

Though this isn't a very realistic scenario, it's certainly possible. Had all of Perot's voters voted for Bush in 1992, Bush would have won re-election with Reagan-level margins. He would have gained a whopping 486 electoral votes to Clinton's paltry 52. Clinton would have only retained Arkansas (his home state), D.C., Maryland, and New York. He would likely had been nothing but a footnote in history, regarded as another Walter Mondale.

Of course, none of these scenarios would have been exactly what happened. No matter how many scenarios I could come up with, I could only get close to what would have actually happened. These scenarios are snapshots.

In my opinion, scenarios 2 and 3 are the closest to what actually would have happened. This, of course, is unhelpful in deciding whether or not Bush would have definitively won the election had Perot remained out of the race, since scenario 2 has him losing by a narrow margin and scenario 3 has him winning by a large one.

I continued playing around with scenarios, but there are a number of variables that cannot be accounted for. I did discover that Bush would have needed 67% of Perot's votes in every state across the board at the minimum, if the rest of them had voted for Clinton.  Bush's chances improved the more Perot's voters stayed home.

So would Perot dropping out have changed the results of the election? We just don't know, and there is no way to know for sure without speculating. There is a good case for both sides, but my opinion is that it did. I base this on Scenario 2 being such a close race, and Scenario 3 being a landslide for Bush. If you average the two, Bush comes out on top.

If you'd like to view my spreadsheet, click here.

UPDATE: I was recently made aware of some peer-reviewed research in this matter. According to this research (Economics 737), 49.5% of Perot voters would have voted for Bush, and 50.5% of Perot voters would have voted for Clinton. This distribution means the election results would have been unchanged and Clinton still would have won.



Economics, Issues and the Perot Candidacy: Voter Choice in the 1992 Presidential Election. R. Michael Alvarez and Jonathan Nagler American Journal of Political Science Vol. 39, No. 3 (Aug., 1995), pp. 714-744

Friday, February 19, 2016

Why Iowa is Irrelevant

There is a massive media machine built to convince all of us that the Iowa Caucus is the holiest of holy holidays in the world of a political junkie like myself. This is mostly because the Caucus is the first time anybody gets to actually cast a vote for President of the United States in each election cycle. Admittedly, that's kind of a big deal.

Ever since I learned about the Iowa Caucus, the one burning question in my mind has always been thus: "Why does Iowa get to go first?" The answers are varied somewhat but they all essentially boil down to this: "Iowa is the most accurate cross-section of America and is a great predictor of Party Nomination winners."

The first part of this answer doesn't pass the smell test with me. Iowa an accurate cross-section of America? Really? Iowa is almost completely Rural. It doesn't have a particularly large population. It doesn't ever seem to matter in the general election. It is historically a swing state, yes, but its 6 electoral votes are rarely an attractive prize when there are much more populous swing states like Ohio and Indiana nearby. In fact, if there were any state that provided the best cross-section of America, it'd be Ohio. It bristles me to say that as a Michigan native, but the fact that no President has won without Ohio in over half a century is hard to argue with.

The second part of the pro-Iowa claim (that it predicts primary winners) takes some examination. So let's do that. Since the current arrangement with Iowa starting first began in 1972, the election results have been thus.

First, for the Democrats.

Year Winner Margin Race Type Predicted Winner
1972 Edmund Muskie +13 Contested No
1976 Jimmy Carter +15 Contested Yes
1980 Jimmy Carter +29 Uncontested Yes
1984 Walter Mondale +32 Contested Yes
1988 Dick Gebhardt +4 Contested No
1992 Tom Harkin +72 Contested No
1996 Bill Clinton +97 Uncontested Yes
2000 Al Gore +26 Contested Yes
2004 John Kerry +6 Contested Yes
2008 Barack Obama +8 Contested Yes
2012 Barack Obama +96 Uncontested Yes

At first glance, it looks like Iowa does pretty well. Of the 11 Iowa Caucuses held since 1972, 8 have accurately predicted the winner of the Democratic Nomination. That's a 72% success rate, significantly better than chance. 

But I still argue that Iowa is not a good predictor of nomination winners. Why? Because not all Caucuses are the same. Let's start with the 1980 Iowa Caucus. Jimmy Carter wins the nomination by a large margin. But he's the sitting President in 1980 seeking re-election. Not since the Iowa caucus started has a sitting President or Vice President lost his party's nomination. The same is true for Clinton in 1996. He was the sitting President. These contests weren't really the same as the contests we're seeing this year in 2016. Thus, I argue that they should not count toward Iowa's record, since those races are not contested within the Party. This leaves 1972, 1976, 1984, 1988, 1992, 2000, 2004, and 2008 as the only Contested races and the only true tests of Iowa's predictive power.

That is 8 total contests and 5 where Iowa predicted the winner. That is 5/8, or a 63% success rate, almost within the margin of error of chance with such a low sample size. Essentially, you could get the same results by flipping a coin 8 times, easily. But what about the Republican matches?

Year Winner Margin Race Type Predicted Winner
1976 Gerald Ford +2 Contested Yes
1980 George H.W. Bush +2 Contested No
1984 Ronald Reagan +100 (unopposed) Uncontested Yes
1988 Bob Dole +12 Contested No
1992 George H.W. Bush +100 (unopposed) Uncontested Yes
1996 Bob Dole +3 Contested Yes
2000 George W. Bush +10 Contested Yes
2004 George W. Bush +100 (unopposed) Uncontested Yes
2008 Mike Huckabee +9 Contested No
2012 Rick Santorum < 1 Contested No

Vice Presidents have a slightly more difficult time gaining the nomination than President's do. They are almost always successful, but not always. Determining if a race is uncontested or contested in that situation is a bit more nuanced. The 1976 GOP Nomination saw Gerald Ford, Richard Nixon's Vice President and sitting President win by only 2 percentage points in Iowa. Ford did win his party's nomination, but barely eeked a win against Reagan. The results of that nomination process were close enough for me to judge it as Contested, even though a sitting President was seeking his own party's nomination. Additionally, in 1988, George H. W. Bush was the sitting Vice President, but initially struggled in the early part of the nomination process. Though he ended up obtaining the nomination easily, the early part of the contest was close enough to consider it Contested as well, especially since we're discussing the Iowa Caucus in particular.

There were three uncontested GOP nominations in Iowa's record: 1984, 1992, and 2004, all with popular sitting Presidents running unopposed. Therefore, these contests are not applicable to 2016. Out of 7 Contested Primaries since 1976, the Iowa Caucus has predicted the eventual GOP nominee accurately 3 times, a 43% success rate.

From the data, it appears that Iowa's claims of predictive power over the nomination process are overblown, at best. Dismal at worst.