Primary turnout doesn't predict general turnout (a la the "monster vote" model)

Continuing the series on tempering expectations for what is going to be a close race, there's a major misconception we have to clear up about using the primary turnout to predict the turnout of the general election.

Shown below is the relationship between primary vs. general turnout for both parties back to 1976, when the 50-state primary system began. I've shown all years, not only those when both parties held primaries, in order to see how the general vote has changed each step along the way. Turnout is in millions, and the "multiplier" means how many times the general vote was compared to the primary vote. Blank entries mean no primary was held, and therefore no multiplier could be calculated either. Click to enlarge.

As I discussed here, primary and general elections are separate and independent from each other. The primary turnout reflects how motivated voters are to leave at an early stage, so whichever party has the more engaging primary contest will have higher primary turnout -- regardless of who will eventually have more on their side when it's the two parties vs. each other in the general. That's why knowing who had higher primary turnout tells you nothing about who won the general -- half the time it favored the primary winner, half the time it favored the primary loser.

Usually, Democrats have higher primary turnout, although in 2000 the Republicans did -- and still went on to lose the general turnout. In 2016, the Republicans have had a slightly higher primary turnout. Since there's only one other time when that happened (2000), there's no pattern there to guide us today.

Now, what if we looked to an earlier year and compared how a party's primary turnout compared to its ultimate general turnout, then applied that "multiplier" from the past to the current primary turnout? We'd have predictions for each one's general turnout, and hence a prediction of who would win and by how much.

This is the idea behind the "monster vote" model that was proposed at the Conservative Treehouse, first in a guest post and periodically discussed afterward, most recently here. I'm addressing this idea since a lot of folks have started to read TCT this election cycle, and may be relying on this model to predict what will happen.

Sadly, the model is fatally flawed. It only looks at 2008 and 2012 to calculate the "primary-to-general multipliers," even though there are data going back to 1976. Based on 2008 and 2012, the Republican primary turnout roughly tripled by the general stage. Assuming that same multiplier will hold this time, would predict a Republican general turnout of around 90 million -- 30 million more people than voted for McCain or Romney, an increase of 50%. The recent TCT post allows the multiplier to go down to just 2, predicting a general turnout of 62 million for Trump.

I don't see much of a problem with assuming the multiplier for the Republicans will be somewhere between 2 and 3, though probably closer to 2. If we look across all years, their multiplier ranges from 2.8 to 4.0.

However, we have to remember the relationship between primary excitement and general turnout. The more exciting and engaging the primary is, the more regular voters will be captured during this early stage -- and fewer additional ones left to turn out in the general. In short, the more engaging the primary, the lower the multiplier (so many have already turned out during the motivating primary), and the more pointless the primary feels, the greater the multiplier (everyone waits till Election Day itself, and only a few bother showing up during the primary).

Because the Republican primary this year was by all accounts the most motivating and engaging at least since 1976, their multiplier this year will be lower than any previous value. The lowest value before was 2.8, so this time around it will probably be from, say, 2 to 2.5.

The real problem with the "monster vote" model is how it treats Democrat turnout. It's only basing its D multiplier on 2008, which was the most engaging primary in all of American history. As such, so many of the eventual D general voters had already shown up during their primary, and the result was the low multiplier of 1.9.

Naively assuming that this same multiplier applied to the 2016 D primary turnout of 30.6, we'd predict a general turnout of merely 58 million for Clinton -- down 8 million from Obama's 2012 turnout, or down 12%. The only precedent for that would be the R decline of 10 million from 1988 to '92, although about half of that is due to Perot siphoning votes. Without a massive third-party splitting Hillary's turnout, there is simply no way the Democrats will lose close to 10 million votes from 2012.

The error comes from applying a low multiplier from the most highly engaging primary ever (2008) to a primary that was somewhat engaging, but also somewhat of a coronation. Especially during the first four or five weeks, when the minority-heavy states made it a cakewalk for Hillary, and when Bernie was not really taking the fight to her. When the primary is not so engaging, it means there are likely lots of eventual Democrat voters who are just staying home during primary season, and there will be a higher multiplier.

The lowest multipliers on the D side were 1.8 to 1.9, in 2008, 1988, and 1980. These were all unusually engaging primaries -- 2008 was the chance to nominate either the first black or the first woman, 1988 was an earlier chance to nominate the first black (Jesse Jackson), and 1980 saw the incumbent President Carter be challenged by party heavyweight Ted Kennedy. These races cleared the benches of D voters, leaving far fewer left to turn out in the general.

In 2016, there was no such bench-clearing primary for the Democrats -- some novelty in nominating a woman, although sex matters less than race in identity politics, and some excitement for an anti-Establishment candidate. But it was no Carter vs. Kennedy, Dukakis vs. Jackson, or Obama vs. Clinton.

On the other hand, it was not a total coronation like sitting VP Al Gore brushing aside Bill Bradley in 2000, meaning low primary turnout and therefore a higher multiplier for the general. And it was not like 2004 where Edwards didn't distinguish himself much from Kerry other than his personal history, and where Dean flamed out early for being uber-liberal. This dynamic also made for little excitement and a high multiplier for the general when reliable D voters would eventually come out.

The best we can say is that in 2016 the D multiplier will be between 2 and 3, probably closer to 2 since it was more engaging than coronation-like.

Now notice the problem for predicting the winner in the 2016 general: the primary turnout is essentially the same on both sides, with a slight edge for Republicans (31.1 vs. 30.6). Therefore what really matters is the multiplier -- but we've seen that it will be in the same ball park for both candidates, somewhere around 2 to 2.5. With similar starting values and similar multipliers, we cannot distinguish the fine-grained difference in general turnout.

To see how murky it is, we'll make slight adjustments in the multipliers that will lead to drastically different outcomes. Suppose the D multiplier is 2.2 and the R multiplier a bit higher at 2.3 -- then the general turnout is 67 to 72 million in favor of Trump, who will win 52% of the popular vote. But suppose it's the other way around, still only a slight difference in magnitude, though -- now it's 70 to 68 million favoring Clinton, who will win 51%.

We frankly have no way to decide at a fine-grained level who will have a marginally higher multiplier, so this model makes no meaningful prediction about the difference in general turnout. The best we can conclude based on the history of 1976 to 2012 is that this year's race will be close in the popular vote (similar primary turnout, similar multipliers), and neither will win the popular vote with 55% or more.

Remember, the "monster vote" model was not just presenting a long-shot best-case scenario, it was stating the expected outcome. There is no way that the expectation is for Hillary to lose nearly 10 million voters from Obama 2012. The relatively pathetic turnout during the D primary, and the lack of enthusiasm for Hillary overall, is more likely a sign that it will be like it was for Republicans in 2008 -- lots of bored, depressed, unenthusiastic voters who would nevertheless turn out on Election Day for their party.

As it turns out, I do have a model in mind where Trump could win by quite a large margin, but it would be in the "less likely, still possible" range of likelihood, and it does not rely on assuming that the "primary-to-general multipliers" from the past couple elections apply in the present. It also does not include any way for Hillary to win by a yuge margin.

It looks at the power to draw in irregular voters, which is a longer shot the greater the size of irregular voters we're talking about, and which is effectively absent on the Crooked Hillary side. But it does not have to do with using primary turnout to predict general turnout, since the irregular and apathetic voters are mostly sitting out the primary to begin with. It is more of a "black swan" model, not based on the behavior of past well-behaved elections.

As a final warning, when I asked about the limitations of the data in the recent post at TCT, I was dismissed, and a follow-up comment that I left with data going back to 2000 instead of just 2008, was deleted. Based on that response, I'm going to strongly caution people about anything being proposed there of a quantitative nature.

They have been excellent at revealing who the key players are inside the Establishment, who their pay-masters are, what their links and relationships are, and where the balance of power is shifting among their alliances. But they're putting solid faith in a model that has little basis in reality, and are not only resistant to honest polite feedback, but censoring objections to it.

I want to keep everybody clear about what is and is not being predicted by what data we have available. Otherwise, we will harden into a deluded echo chamber, the way conservatives did back in 2012 about Romney being not only a sure win, but in a landslide -- and thinking this just days before the election!