21

I have seen numerous reports that describe how the Democrats are confident of overturning the House majority in the 2018 midterms. Where does this confidence spring from?

  • 5
    There isn't a definite answer to this so I expect this question to be closed, but here is an unbiased data-driven outline of why Democrats expect to retake the house - projects.fivethirtyeight.com/2018-midterm-election-forecast/… – Gramatik Nov 5 '18 at 18:51
  • There must be specific, non-partisan reasons for the consensus - for example the nature of the voting system or public faux-pas? – Ben Nov 5 '18 at 18:59
  • 3
    "There must be specific, non-partisan reasons for the consensus" That's kind of a wild notion. Remember what the Democratic party's consensus was in 2016? If their reasoning was non-partisan, how could they have gotten it so wrong? – Beanluc Nov 6 '18 at 20:44
  • 11
    Consensus was right. – Trilarion Nov 7 '18 at 8:20
  • 6
    @Beanluc Just to be clear, the polls werent as wrong in 2016 as certain people like to pretend. Clinton did win the overall count of votes by a good margin. However Trump won several states with incredibly razor thin victories. Less than a percent. The winner-take-all rules in those states granted Trump all of their electoral votes even though a nearly equal number of voters had voted against him. Given such a massive swing in the electoral college being caused by so few votes, you would have needed incredibly precise polling on the ground in those states, which they didnt have. – Tal Nov 7 '18 at 17:01
62

Polling data -- which is what is being analyzed by organizations like FiveThirtyEight -- indicates that Democrats are likely to win a majority in the House of Representatives. Polling data is the best and most objective way to predict the outcome.

However, there is no certainty. FiveThirtyEight gives Republicans a >12% chance of keeping the House. That's because polls have both random errors and potentially systematic errors. So no one is taking the outcome as a given. Polling data is imperfect, but it's the best we have.


Regarding the comparison to the 2016 presidential election: Polls generally put Clinton ahead of Trump, but only just. In the end Trump ended up winning very narrowly (indeed he lost the popular vote by a considerable margin). This is not really a "debacle" as some put it, as it was within the usual margin of error of the polls. Ultimately, it was a close race.

Before the election, FiveThirtyEight gave Trump a >25% chance of winning. Also, here is an article from 4 days before the 2016 election in which FiveThirtyEight discusses how close the polls are.

So, in 2016, we saw again that the polls are not perfect, but they did show it was close and they still are the best we have.


In terms of interpreting polls: I think a major problem is that the public misinterpret the polls and considers them to be more certain than they are. (Even experts struggle to quantify the uncertainty.)

Suppose a model gives candidate A a 70% chance of winning and candidate B a 30% chance of winning. A lot of people interpret this as saying candidate A will win, but what the model is saying is that this prediction will be "wrong" 30% of the time. You should interpret 30% uncertainty as saying the election is very close and hard to call.

Note that a 70% chance of winning does not mean the candidate will get 70% of the vote. In reality, it may be that candidate A is polling at 51% and candidate B at 49%. That 2% margin could just be a polling error and we are only 70% certain that A is ahead of B.

Perhaps a good contrasting analogy is weather forecasting. If the weather forecast is for a high of 70'F, we aren't shocked if it ends up being 75'F or 65'F. An error of 5'F is pretty good. We accept that the weather forecast isn't perfect. The problem with elections are that they are more finely balanced than weather forecasting. (The 2000 presidential election came down to <0.01% in Florida.) If you correctly predict the vote shares to within 1%, that still may not tell you who wins.

  • 29
    @Ben: What you call a debacle statisticians call a normal result. Besides, polling mechanisms are never really at fault, since it's more with how polls are interpreted and whether interpreters are able to account for a poll's unavoidable inaccuracy/bias. – Giter Nov 5 '18 at 19:28
  • 29
    @Ertai87 The gap of their chance of winning is not the same as the size of the gap in % of people voting for each, they are completely different percentages. If instead FiveThirtyEight predicted Clinton would get 70% of the total vote then indeed those models would have been very very very wrong. That wasn't the case. – Bryan Krause Nov 5 '18 at 20:07
  • 15
    In 2016, most forecasts assumed that Trump's chances of winning certain swing states were independent of each other. As a simple example, if Trump had a 30% chance of winning Ohio and a 30% chance of winning Michigan, then he had only a 9% chance of winning both. However, this ignores the fact that whatever circumstances would lead to him winning one state would equally apply to his chances of winning the other. FiveThirtyEight, I vaguely recall, made a mention of this in upgrading Trump's chances. – chepner Nov 5 '18 at 20:37
  • 9
    Joined just to post this: 538 had an article addressing exactly this – Draco18s Nov 5 '18 at 23:29
  • 11
    @Ertai87 How shocked would you be if I flipped a coin twice and it was heads both times? Would you be telling me how drastically unlikely that was? Because that's what a 25% chance of winning looks like. Not that uncommon at all... – mbrig Nov 6 '18 at 5:06
16

The position that Democrats will probably win the House is supported by data. This is what RCP says about the House predictions: 202 likely or leaning Democrat, 39 toss-ups, 194 likely or leaning Republican. This means, in all reasonable likelihood, Democrats could win up to a 47-seat majority, as bad as a 31-seat minority, and on average will probably win an 8-seat majority. Some of the leaning Democrat and leaning Republican seats could still go the opposite way, but those outcomes are simply less likely.

To oversimplify these stats, I will treat the likely and leaning races as locks, and the toss-up races as coin flips. For Democrats to control the House, they would need to win at least 16 out of 39 coin flips. Using Excel, I've calculated the probability of winning at least 16 out of 39 coin flips to be 90.02%. This is an oversimplified model, not utilizing the precise polling data of each individual race (i.e. it treats a tossup as being exactly 50/50, when more detailed data might reveal that it is 51/49). But it's reasonable to figure that the DNC's real chances of flipping the House probably fall between 80 - 95%.

fivethirtyeight.com projects an 87.5% probability of Democrats winning control of the House.

On a note related to a separate conclusion, I think it's almost a certainty that the DNC will regain some seats. RCP implies that Democrats will win, on average an 8-seat majority. Currently, Republicans hold a 42-seat majority (previously, I misstated this number as 59, because I was looking at an older number). This means, it is predicted that Democrats will win back about 25 seats, on average. It is almost a guarantee that Democrats will win back some seats, because in order for Democrats to gain no ground on the 193 seats they currently hold, Republicans would need to win all races that are likely or leaning their way, all toss-up races, and 10 out of 15 races that poll as leaning Democrat. This is essentially impossible. Strategically, Republicans would need to hope for low turnout in order to achieve this, as mid-term low turnout has historically favored Republicans. But turnout in the 2018 mid-terms is looking to be up 500% in some areas compared to 2014 turnout, which is yet more bad news for Republican hopes of pulling-off a large amount of upset victories.

On a note related to 2018 turnout and polling predictions, NBC asserts that record turnout will upend polling predictions. This article also claims "Historic levels of enthusiasm on both sides of the aisle." Some kinds of polls use survey weights to correct for their sampled ideologies by using historical weightings. E.g. if black respondents are underrepresented in the sample, then they will give higher weight to the few black respondents that they have. If registered Democrats are underrepresented in the sample, then they will give higher weight to the registered Democrats who responded. However, weighting by historical turnout models will tend to be inaccurate if 2018 turnout is higher and different than average turnout. E.g. if Democrats and young voters usually have low turnout in mid-terms, and if their turnout is high in 2018, then the historical weightings used would end-up being wrong, and the outcome would favor Democrats and young voters to a larger extent than the polls predict. The NBC article above also claims that that Democrats hold a 7-point lead, 50 vs 43, on the generic national ballot.

Some have claimed that the mid-term elections are, in part, a referendum on the president's popularity. With that in mind, Gallup's weekly presidential approval / disapproval differential has been -6 points, -14 points, and -14 points in the past three weeks. This would imply a fairly recent drop in the RNC's competitiveness within the last two weeks. It also suggests that in races where the president personally campaigned on behalf of a candidate, that the value of the endorsement has recently dropped.

Now, onto a note about whether Democrats, and only Democrats, are likely to accept the projection offered by polling data. According to Axios, 92% of Republicans think media intentionally reports fake news. Also according to Axios, 67% of Republicans fear bias in search engine, compared to 13% of Democrats. According to a September 2018 Bloomberg article, 71% of people agreed that it was extremely or somewhat likely that Democrats would win control of the House. The same Bloomberg article demonstrates that 90% of people who strongly disapprove of Trump think that Democrats will retake the House, and that only 37% of people who strongly approve of Trump think that Democrats will retake the House. Also, it shows that 70% of independents think that Democrats will retake the House. The same Bloomberg article asserts that GOP voters distrust traditional polling. However, none of the polling data in these articles are likely to be found informative by someone who doesn't trust polling data.

Is there a cause-and-effect reason why trust in polls would vary by party affiliation? Trump has been consistently telling his supporters not to trust polls, and that has likely had an effect.

  • 1
    "I think it's almost a certainty that the DNC will regain some seats." - I was gonna push back on this statement, but no you are correct. 538 has odds listed for every possible numerical outcome, and none of the ones where Republicans break even or gain House seats has greater than a 0.1% chance of occurring. Cumulatively those don't even add up to 1%. The sub-graphic shows chances of Dems gaining >59 seats at 10%, 21-59 seats at 80%, and 0-21 seats at 10%. Less than 0 doesn't make the cut. – T.E.D. Nov 6 '18 at 18:29
  • Beware: Coin flips are independent. Simultaneous elections are not. A factor which causes one race to go Republican will probably also affect other races. That means the toss-ups are much more likely to break 34-5 than you are to get only five heads in 39 coin flips. – Martin Bonner Nov 7 '18 at 10:32
5

Politicians will always try to project confidence that their side will win.

This is because voters like to vote for winners. Whiners get less votes.

In this particular case most polls happen to agree with them, but that really is irrelevant. Politicians are always confident they will win until the count is final.

@emory's excellent comment: I think you are right that "voters like to vote for winners", but I think donors and volunteers are more important. Why would I donate my $$$ or my time to someone who does not believe they will win? If one does not publicly believe they will win, it will became a self-fulfilling prophecy.

  • I think you are right that "voters like to vote for winners", but I think donors and volunteers are more important. Why would I donate my $$$ or my time to someone who does not believe they will win? If one does not publicly believe they will win, it will became a self-fulfilling prophecy. – emory Nov 6 '18 at 18:06
  • Or in the case of the 2016 election, I voted for Clinton b/c I thought she was the best candidate who had a decent shot at winning. If Clinton expressed self-doubt, then maybe I would have thought about one of the 3rd party candidates even though I would have thought they would ultimately lose. – emory Nov 6 '18 at 18:09
  • This question was worded a little awkwardly, but it seems to be asking about the general consensus (even among non-Democrats) that Democrats would probably win the House. Which, of course, they did. While what you say here certainly did play into the Democrats' PR (as well as that of the Republicans,) the overall consensus (outside of just the Democrats' PR) was that they would indeed win the House and was based more on polling than anything else. – reirab Nov 7 '18 at 21:03
-4

The same place as always: There is no better indicator of how people will vote than how people tell you they will vote.

That said, on Nov 7, 2016 (or, to be more specific, the most recent data as of Nov. 7, because I know how this site likes specific data XD) most polls showed a clear majority for Hillary Clinton, and we all know how that one turned out. Polling isn't what it once was.

  • 16
    To be fair, Clinton did win a majority of votes. But they were clustered in such a way that she didn't get enough states. The polls were as accurate as normal, but looking at the country as a whole was misleading. – Bobson Nov 5 '18 at 19:09
  • 6
    Polls directly preceding the election projected a +3.2 advantage for Clinton, who won the actual popular vote by +2.1. That is an error of 1.1 points. The stated margin of error of each individual poll was between 2.3 and 4.5. Taking the average of 10 of these polls would probably produce a lower margin of error, somewhere around 1.5 points. Most individual polls were also within their margins of error, but some were slightly outside (Reuters, NBC, Monmouth, LA Times). The average of all polls was pretty accurate, and this attack on polling is unwarranted. – John Nov 5 '18 at 19:44
  • 7
    @Bobson "To be fair, Clinton did win a majority of votes. " No, she didn't. She won more than Trump, but still only 48.2%, which is not a majority. No candidate received a majority. Approximately 6% of voters voted for someone other than Clinton or Trump. – reirab Nov 6 '18 at 8:00
  • 1
    @reirab - Good point - I also bring that up, but in the polling it also did not show Clinton with a "majority" of polling responses, so to the degree that polling did or did not track with the final results, I think the point that the results did match, with the polling showing Clinton ahead is still accurate. It's good to be precise about the language we use, though. – PoloHoleSet Nov 6 '18 at 18:16
  • 1
    @reirab You're splitting hairs on what "majority" means. Typically, when talking about voting, "majority" simply means the person who received the greater number of votes. It's perfectly accurate to say Clinton won the majority of votes. – user22277 Nov 7 '18 at 17:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.