Breaking News

TOM WOLF: Predicting the US mid-terms: Truth-time (again) for pollsters?


US President Donald Trump acknowledges supporters at a campaign rally on the eve of the US mid-term elections at the Show Me Center in Cape Girardeau, Missouri, US, November 5, 2018. /REUTERS


By Tom Wolf

After pollsters in the UK encouraged expectations of a “remain” vote on Brexit in 2016 and their colleagues in the US did the same regarding a Hillary Clinton win in the presidential election later that same year, those in the profession are still looking for a rag with which to wash some of the ‘egg’ off their faces.

(Though recall that Clinton did out-score Trump by about 2.5 per cent – very close to the 3 per cent margin most polls had predicted. It was only her failure to get an additional 80,000 votes – just 0.006 percent of all cast – in a handful of states that ensured her defeat.)

This Tuesday’s mid-term elections in the US offer a good opportunity for that. So how will they perform?

As a reminder, these US mid-term elections involve all 435 members of the House of Representatives, 35 of the 100 Senators, 36 of the 50 state’s governors, and countless state and local level officials.

As such, even if President Donald Trump has two more years of his (first?) four-year term, the outcome will have a major impact on US policy, both domestic and foreign.

And to recall, as of now, the Republicans control both Houses of Congress, with a 23 seat majority in the House and a 2 seat advantage in the Senate. That means the Democrats must pick up at least 23 Congressional seats and 2 in the Senate (just adding just one leaves the tie-breaking vote to Republican Vice President Mike Pence).

Of course, any survey used to predict election results must grapple with four constant realities. The first is the statistical margin-of-error, based on the sample size. For example, a sample size of 1,000 for a population of 500,000 generates a margin-of-error of +/-3.1%, equal to a total spread of 6.2%.

ALSO READ  Angry DP Ruto tells off Raila, says Kenya ‘has no shortage of fools’

This mean that in a winner-take-all election (where even a margin of a single vote bestows victory), the winner can be ‘predicted’ only by luck if the contest is close, that is, within the range.

The second reality is the impact of those survey respondents claiming they are still “undecided.”

Indeed, the poll results for many state-wide contests (e.g., senators, governors) show a total of less than 95 percent stated to be voting intentions for the two main (Republican and Democratic) candidates, again making close races up-for-grabs.

The third is the unpredictability of voter turnout – which is historically low in such US mid-term elections. (For example, among ‘youth’ – those aged 18 to 29 – voter turnout was only 16 percent in the last/2014 ones.)

Even if those sampled are limited (by whatever filtering mechanism) to registered voters, and even if it is possible to distinguish ‘likely voters’ among those sampled (based on such factors as previous voting, a self-declared intention to vote, etc.), a certain amount of doubt must remain. Indeed, even bad weather on voting day in certain areas can depress it.

A fourth is the impact of late-hour events that can shift especially undecided or even previously committed voters (as well as affecting turnout), given the duration-gap, however small, between the last round of polls and voting (keeping in mind the increasing proportion of the electorate taking advantage of early-voting opportunities whose votes have thus been cast (with Republicans doing better in almost all states where such figures are tracked).

But in addition to these four quite constant variables in pre-election voter-intention surveys, there are a couple of other ones that apply to these contests.

ALSO READ  Death robs Kalonzo Musyoka his father

One, again based on what was learned from the pre-Brexit and 2016 US election surveys is respondent deceit.

Based on analyses conducted after each of those events, it was clear that a significant proportion of survey respondents falsely reported their intentions – in the former case, that they would vote to “remain” when in fact they voted “leave”, and in the latter case, that they would not vote for Trump, yet they did. In both cases, “leave” and “Trump” were evidently viewed as politically ‘unacceptable’ (or at least embarrassing) to enough respondents, who were therefore unwilling to admit their true intentions.

(Indeed, one study in the US found that white women being interviewed by programmed robo-call machines were rather more likely to admit they were voting for Trump than were women of the same socio-economic identity who were interviewed by an actual person, suggesting that much less of such embarrassment, if any, is involved in the former type of data-gathering. Note here than these days it seems all such surveys are phone-based, whether land-lines or mobile phones.)

Another key one is deliberate voter-suppression, which mainly works to the advantages to the Republicans in states under their control. One such is that of Georgia, where it has been calculated that some 7 per cent of electorate will be barred from voting when (if) they show up at their polling stations, and almost all of them are likely Democratic voters, through the use of various mechanisms.

So how ‘accurate’ will these current polls be?

As of the eve of the election, the calculated average of all polls regarding all congressional races show an 8 percent advantage for the Democrats. For the gurus running the “Real Clear Politics” website, this translates into a 82 percent chance for the Democrats to capture control of the House of Representatives, but only a 15 percent chance of doing this for the Senate.

ALSO READ  Brains of overweight people look 10 years older than those of lean people - Study

Again, looking at recent data-analyses, some 110 out of the 435 House races are considered to be ‘competitive’, with the gap between the two parties’ contenders so close that they could go either way.

This is so even if spending by Democratic candidates has been greater than that of their Republican competitors in 98 of those contests.

This modest Democratic ‘advantage’ derives, in large part, because (in many cases, recently re-drawn) congressional boundaries (by Republican-controlled State legislatures) have ‘crammed’ Democratic voters into as few such constituencies as possible (mostly in/around urban areas,), allowing Republicans to win with much smaller margins in as many places as possible. (Another reason why this election is so critical: because the next national census in 2020 will provide another opportunity for reallocating these congressional seats).

Moreover, there are two other polling figures, explicitly contradictory, that suggest that this Democratic advantage may not be very firm.

That is, despite the President’s performance-approval rating being at an historic low for this point in his incumbency – only about 40 percent (in the most recent CNN poll), a near-record low, some two-thirds of respondents in several recent surveys perceive the economy as strong/good. So only post-election will it be possible to judge whether voting more strongly reflected views on Trump as opposed to the country’s perceived condition.

Undoubtedly, pollsters will be lauded or lambasted based on what the final numbers are. But the main conclusion suggested here is that there are so many factors beyond the precise estimation by such surveys, that “getting it right/wrong” is often more a question of (good/bad) luck, at least in close contests, as most elections appear to be.

Leave a Reply

By continuing to use the, you agree to the use of cookies. more information

The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you click "Accept" below then you are consenting to this.