Democratic Primary Results: Predicted vs actual (Updated with Maine)

Yesterday, the Democrats held three contests, in Louisiana, Nebraska and Kansas. I had predicted a Sanders win in Nebraska and Kansas, and a Clinton win in Louisiana, using my ever-evolving ethnicity-based projection model. Those predictions came to fruition. Like this:

Predicted on top, Actual on bottom.


Clinton did a bit better than projected in Louisiana, and Sanders did a bit better in Nebraska, but much better in Kansas than predicted.

I had projected the final delegate count to be 60:49 (Clinton:Sanders) for that day, and it turned out to be 55:49 (Clinton:Sanders). The difference is primarily in the number of actual delegates awarded to the candidate between what my model assumed and what the states (Louisiana) actually did. Overall, I'd say that the model, which currently predicts Clinton reaching lock-in on delegate count in mid or late April, is accurate, but with enough of a difference to allow for Sanders to close the gap somewhat. At this point, though, Sanders will have to start performing better in order to catch up.

Lately we've seen a discussion that runs something like this. Clinton is winning in states where a Democrat is unlikely to lose, and Sanders is doing well in states where a Democrat is likely to lose. Therefor, Clinton would lose the general election, and Sanders would win it.

This proposition fails to take into account that for the most part the two candidates are interchangeable at the level of the general election. All those people who preferred one candidate in the primary will prefer the other candidate in the general, should that other candidate win the nomination. The only way for Sanders to beat Clinton is to start winning more delegates than the model projects, and soon.

Sanders' better than predicted performance yesterday is not enough for him to overtake Clinton, but perhaps it is a sign that he is increasing his performance. Every primary or caucus is a test of the running hypothesis of status quo, and at the moment, status quo gives Clinton the nomination. Sanders will have to start falsifying that hypothesis very soon. There is no reason to say that will happen, or not happen, at this time.

By the way, a similar model (using the status quo as the determining factor in making predictions, but with no ethnic adjustment) for the Republican party predicts that Trump will lock in the nomination late enough in the process that he could actually fail to do so if his performance falters. The possibility of a brokered Republican convention is very real.

That is not the case, probably, for the Democratic convention, as the uncommitted delegates (called Super Delegates) will likely vote for the winner at the end of the process, to lock in that candidate.

UPDATE: Today, Sanders won in Maine. I had predicted a Sanders win, though Bernie got more delegates than my model had suggested.

Predicted on top, Actual on bottom.

Screen Shot 2016-03-06 at 10.05.06 PM

The Delegate total for this weekend is now 72:62 Clinton Sanders predicted, 62:64 Clinton Sanders actualized.

I will assume that the extra strong showing by Sanders in Maine is partly a result of the Favorite Son effect, and not adjust the model. Mississippi and Michigan, in just a couple of days, together with this weekend's contests, should provide excellent calibration in preparation for primaries if Florida, Illinois, Missouri, North Carolina and Ohio.

More like this

"Lately we’ve seen a discussion that runs something like this. Clinton is winning in states where a Democrat is unlikely to lose..."

Clinton's greatest strength thus far has been in southern states with large black populations that the Democrats are unlikely to win. But later, Clinton will probably win in states that Democrats are unlikely to lose.

Aside from that, an article in the latest issue of the New York Review of Books rejects the idea that polls showing Sanders beating Republican opponents are worth very much right now:

"As for Sanders, as effective as he has been, he has his own flaws, which turn chiefly on whether he is electable. His partisans point to the general election head-to-head matchups in which he fares as well as, or better than, Clinton against Trump and Cruz and Rubio. But those polls don’t mean much. The Republicans haven’t spent a dollar attacking him yet, and if he were the nominee, they and their affiliated groups would spend between $500 million and $1 billion doing so.

How would he hold up under that barrage? There is his socialist background, the honeymoon trip he and his wife took to the Soviet Union, and all that. But Republicans might not even have to go there. The tax increases that would be entailed by Sanders’s programs, including his Medicare-for-all plan, have recently been shown to be enormous, requiring, according to some estimates, a top marginal tax rate of as much as 84 percent. Sanders argues that the trade-offs he proposes for expanded Medicare—taxes but no deductibles or copayments, not even for dental care and optometry—would save people thousands of dollars a year. In some cases they surely would. But as the health care expert Harold Pollack has pointed out, Sanders’s plan would in effect require a doubling of federal income tax collections.4 The attack ads would doubtless say something like: “Bernie Sanders wants to double your taxes, limit your choice of doctor, and turn America into Cuba.” Times may have changed this country, but have they changed enough that 51 percent of voters will vote against such claims?

Finally, Sanders would present for the Democratic establishment some of the same problems that Trump and Cruz would present for Republicans. Very few elected Democrats would endorse Sanders enthusiastically, because they would calculate that being too close to Sanders would hurt their own reelection chances. So the Democrats too would be fractured, and perhaps for a very long time."

I wonder if Sanders supporters believe that Sanders, if nominated, would not be subjected to a Republican barrage (we know Clinton would), and could not be hurt by it?

(By the way, I find Sanders's politics preferable to Clinton's, but I do worry that he'd be a weaker candidate. Luckily, as I can't vote in American elections, I won't have to choose.)

By cosmicomics (not verified) on 06 Mar 2016 #permalink

they and their affiliated groups would spend between $500 million and $1 billion doing so.

Probably, but I think they could get away with much less. They've done such a good job of eliminating the distinction between Sanders' brand of Democratic Socialism and flat out communism that people would not need to be pushed much to vote against him.

cosmi and dean,

The thing is, it isn't about Republican attacks-- it just isn't the case that the majority of USA citizens support an involuntary change to a "Medicare for all" system. Right now, before the R propaganda even begins.

As I keep pointing out, President Obama had as a main selling point for ACA "you can keep your current insurance if you like it". President Obama is a very smart man, and a good politician with good advisers.

It was not a lie, as the Republicans claimed-- for all the people with employer-linked healthcare, and Medicare, very little changed. And, all those same people are quite happy with their taxpayer-subsidized insurance. They would not support Bernie on this, and so the Congress would not even come close to supporting him.

The other point I keep making is that Bernie is simply not a very appealing general election candidate on a personal level. I'm a grumpy old white guy myself, and I would much rather have a beer with Hillary than Bernie-- not that he's a bad guy, but he's the boring, grumpy old man I dread becoming some day.


A point I made earlier: The flaw in you model is that it doesn't take into account tactical shifts by the candidates on a State-by-State basis. To evaluate its skill you can only work in the aggregate; H over-performing in one State while B over-performs in another tells us nothing.

cosmicomics, not sure if Clinton has less liabilities that would be attacked in the general election. Next to the Republican Clinton evergreens, a possible indictment over her private email server, but maybe more importantly the content of those emails and all the money of foreign governments flowing into the Clinton Foundation while she was secretary of state.

Greg, I see no evidence for the claim that it does not matter whether Clinton or Sanders becomes the candidate. There are yuge differences in the head to head polls. Clearly many people do see differences. Clinton does well because of the African Americans. I see no reason why they would not vote for Sanders as well; his policies are fine, if not better. Sanders does well with independents, they may go to the Republicans. Sanders does well with people who are fed up with bribed politicians; they may well not vote if Clinton becomes the candidate.

By Victor Venema … (not verified) on 06 Mar 2016 #permalink

Zebra, that is not a flaw in my model. My model is flawless!

But seriously, this model takes into account exactly two things. Empirical data providing estimates of ethnic patterns of voting to date, and ethnic mix of voters in each state. That is why it is a good model. This model allows us to see variations from this basic, status quo prediction. So, if there are tactics being applied, if they have an effect, perhaps that will show up.

I've only evaluated the skill of the model in the aggregate (see my earlier post in which I do that). Nonetheless, there are individual primaries and caucuses, and we talk about them. Perhaps I'm missing your point here.

Victor no, there really are not huge differences in the head to head polls, and they don't tell us much anyway.

Also there is no indictment. The clinton foundation was a fundraising thing. Money flows into sucessful fundraising things. The foundation was not election money.

Please do not use my blog to spread untruthful rhetoric. You should know better than that.

One warning.

Greg - why do you keep repeating this? "...there really are not huge differences in the head to head polls, and they don’t tell us much anyway."

As I pointed out in a previous thread, the head-to-head polls even one year out predicted 12 of the last 14 eventual winners of the general election. That's telling us quite a bit. And we're no longer one year out.

Larry Hamilton makes an interesting point: "You may have seen US “horse race” polls that ask respondents who they’d vote for in hypothetical head-to-head matches. Several of these (ours included) have found Clinton leading Trump by a narrower margin than Sanders leads Trump. If you think about it, that means there exists a subgroup prepared to switch their support from Trump to Sanders, contrary to what anyone might guess from substantive issues."

By Kevin O'Neill (not verified) on 06 Mar 2016 #permalink


Mea Culpa. You have been quite restrained in talking about the individual State outcomes, and I was expressing my annoyance with all the hype coming from certain other quarters.

If one is interested in actually winning the genereral election, then one needs to look at the states that are likely to be "in play". It really doesn't matter if the Democratic candidate wins a state like Massachusetts by 20 rate rh than 25 percentage points, of looses a state like North Dakota by 15 rather than 20 points. WHat matters is the states that are close enough that a five percentage point or so shift in the vote will make a difference.

In 2012, the states that were within 5 percentage points between Obama and Romney: North Carolina, with Romney winning by 3 percentage points, and five others that Obama won, Florida, by less than 1 percentage point, Ohio, by about 2 percentage points, Virginia, by about 3, Colorado, by about 4, and Pennsylvania, by about 5 percentage points.

Virginia Is the only one of these states that we have data for yest (Florida, Ohio, and North Carolina are coming up March 15). In Virginia, Clinton won by 64%, winning with white voters with 57% and among black voters by 84%.

Sanders does extraordinary badly among black voters. And black voters are significant in Florida, North Carolina, VIrginia, Ohio and Pensylvania. Simply assuming that black voters, who thus far have shown that Sanders is not a candidate they want to support, will nevertheless turn out for him in large numbers in November seems at best naive. And in the states that actually matter, this could be the difference between winning and losing.


Good analysis. As has been pointed out by myself and many others, it is hardly a great strategy for the Bernie supporters to be so condescending to the African-American community; it simply highlights the reality of White Privilege-- and that is a good incentive to say "another old white guy president just means business as usual, why bother voting."

And then there's the flip side, which would look at Kasich voters and Republican women. The critical states are also where you would get crossover votes for Hillary, but probably not for Bernie, against some R nominees.

> " there exists a subgroup prepared to switch their
> support from Trump to Sanders, contrary to what
> anyone might guess from substantive issues.”

Er, "anyone"? People in that subgroup have been easily, too easily, swayed by Trump's fake anti-establishment rhetoric, beginning to realize that Sanders is for real on that.

How can Sanders win? I give you Adlai Stevenson's response to the woman in the crowd who called out:

""Senator, you have the vote of every thinking person!"

Stevenson called back:
"That's not enough, madam, we need a majority!".

Take yes for an answer.

By Hank Roberts (not verified) on 07 Mar 2016 #permalink

zebra, if you look at the demographics supporting the candidates, Sanders also looks more electable. Clinton does well because of the African Americans. I see no reason why they would not vote for Sanders as well; his policies and past record are fine, if not better. Sanders does well with independents, they may go to the Republicans. Sanders does well with people who are fed up with bribed politicians; they may well not vote if Clinton becomes the candidate.

Talking about "condescending":

By Victor Venema … (not verified) on 07 Mar 2016 #permalink

If you believe in head to head comparisons, your argument has some validity. But you'd have to prove that head to head comparisons during the primary season mean much. And, I'm pretty sure they don't. I expect both candidates could do well against any of the Republicans.

I gave many more arguments than just head to head polls.

By Victor Venema … (not verified) on 07 Mar 2016 #permalink

Hey Greg,

"If you believe in head to head comparisons"....

... then Bernie should drop out immediately, since he has no chance of winning according to the head-to-head polls for the primaries, which are much closer in time than the general.

Lots of money could be saved or redirected to supporting Hillary against the Republicans, which Bernie surely will want to do.

Just sayin'.

Greg writes: "’d have to prove that head to head comparisons during the primary season mean much. And, I’m pretty sure they don’t."

I'll ask again, why do you keep repeating this? It's not supported by data. The data says that head-to-head polls do convey meaning - even one year out (and we're well within that window now). I provided data sources in an earlier thread:

The Timeline of Election Campaigns:A Comparative Perspective, Jennings and Wlezien, 2013. See page 35, Figure 7. "Adjusted R-Squareds for Presidential and Parliamentary Elections Taken Separately"

There is a weak correlation one year out (r-squared = 0.6) and it increases to election day. At 250 days out the r-squared is 0.7 Combining these numbers with the fact even one year out the head-to-head polls correctly predicted the eventual winner in 12 of the last 14 Presidential elections does *not* lead me to believe head-to-head polls at this time of year are meaningless.

I can only assume that people that *do* believe this meme have never actually looked at the data.

By Kevin O'Neill (not verified) on 07 Mar 2016 #permalink

Kevin, that study looks at polls, not head to head match-ups. And, I have said many times that polls are generally good sources of information. The fact that I'm not using polls in my modeling is not because I think they are not important. In fact, in an earlier post I used a series of recent polls to verify the model at that point in time. (That model, by the way, the one verified by the polls, is the one unadjusted to suggest a Sanders win based on recent possible changes in the electorate.)

The study you refer to indicates that the percentage of variation in outcome "explained" by polling (i.e. predicted) is less for presidential races than other kinds of races. But again, that does not matter to much because they did not look at match-ups.

Victor, here's the argument.

1) The GOP has spent well over a decade vilifying Hillary Clinton. Much of this vilification has involved fabricating stories and mis-statements about her past that have been repeated so often that they are now taken as fact, even though they are not facts, by many Democrats that would otherwise vote for her. This systematic vilification gives the GOP sufficient ammunition to shoot her down if the GOP itself has a reasonable candidate.

2) Bernie Sanders is a self avowed "socialist" (even if he isn't, the GOP will convince people that this is true). Americans will never vote for a socialist. Sanders has explicitly stated that he will raise taxes on the middle class. No president (and hardly any governors or anyone else) has been elected in the US to any office for about three decades where they have declared that they will raise taxes.

Matchup polling data is not considered to be that reliable, so that does not obviate these two facts.

If you look at overall preference independent of any matchup within or between parties, both candidates do pretty well, though the GOP hate campaign has depressed Clinton's numbers somewhat. But both candidates do better than the GOP candidates. So cross checking matchup and preference data, both Democratic candidates seem to have good chances.

So, so far we have a strong argument that both are highly unelectable. And both candidates are highly electable. There is no scientific way to put these two entirely different sets of information into one comparison.

Finally, the GOP currently controls Congress. Congress is one of the most hated institutions at this time in the US. Recent events in the Senate have actually threatened the GOP with loss of the Senate, and those events are specifically anti-Obama. Whichever candidate wins the primary will ally with Obama and use this against the GOP.

If the GOP caves on SCOTUS, their own base will turn on them. If they don't cave on SCOTUS a good section of the rest of the country will turn on them. Either way they lose, and, again, the eventual Democratic nominee will be pushed towards victory.

Since you have been rather relentless in making your case, I'll say out loud what I've avoided saying to a colleague. I think you have a preference for one of these two candidates (Sanders). I applaud that, and there are good reasons to have the preference, and good for you. But, I think you are picking and choosing among the data, and biasing your interpretations, to make your case for that preference in appropriately. Not very science of you Victor!

Greg, like I wrote in the introduction of my post, I have a clear preference.

That Clinton has been vilified is what it is. That will not change between now and November. You cannot wish that away in your assessment of the situation. Just like Sanders has been ignored and his plans heavily attacked by the mass media who mostly would like the system to stay like it is.

In that situation, the numbers are what they are in that situation. People are welcome to check my facts and make their own interpretation.

You are welcome to think that a socialist cannot get elected.

Many people may also simply assume that a centrist on a left right axis has more chances of being elected. If that were the case you would expect independents to support Clinton more, but they support Sanders more. The fight this election is not the traditional left right one, but mostly the crony establishment versus the people.

Given that you only seem to have your gut feeling that a socialist cannot get elected, I feel it is inappropriate to respond to my numbers and facts with the claim that I do not behave like a scientists. The more so as this is not natural science and there is no clear cut answer to the questions we are discussing.

By Victor Venema … (not verified) on 08 Mar 2016 #permalink

Victor, you are confusing a concept (in your head) of my own thinking about this with the arguments I've stated as arguments that are on the table. Perhaps I've not made my self clear. I think both are quite electable, but also, that the arguments about their respective electability involve elements that can't be compared in any quantifiable way.

Not sure why Independents should be assumed to make what some would call a more rational choice (of supporting a centrist). Independents are not people who are positioned equidistantly between the two parties. It probably varies a great deal across space and time. For example, in Minnesota, a few years ago, "independents" went so far as to create their own party, which tended to nominate hard right candidates. Later, this was consumed by the tea party (which arguably started here, or at least, one of the birthplaces of the Tea Party was Minnesota, and a big part of that were our independents) which of course melded with (and took over to a large extent) the Republican party.

What I'm telling you is that you may be ignoring some important but unquantifiable facts, and that some of your quantifiable facts are less important than you think they are.

You say "...this is not natural science and there is no clear cut answer to the questions we are discussing." I totally agree with that!

But, you are making the argument that my model (which has so far proven very accurate) is wrong because your thinking, based on "numbers and facts" disagrees with it. But that is not true. They don't disagree. They are entirely different perspectives.

Either way, today will be a very interesting day in Michigan.

Voted this morning at my local spot in Portage Michigan, between 8 and 8:30. The folks there said turnout had been "really light - much lower than last spring."

Last spring was when local school districts had bond issues up, and there a couple county-related issues.

At least one report from Grand Rapids suggests a larger turnout there.

Good. I haven't had a chance to speak with folks who vote up here.

I don't know why turnout would be lower near my place than it has been in recent past. The weather was perfect.

Could just be time of day. Hard to estimate turnout early in the day.