For those who thought that based on ‘false’ predictions (or at least their failure to confirm widespread expectations) in several recent major elections ( eg Brexit and the US presidential election in 2016) the pollsters would soon be out of business, they will have to think again after the US mid-term elections.
This was so notwithstanding the fact that calculating overall outcomes (i.e., the partisan balance in the US House of Representatives (435 seats, all contested) and the Senate (35 of 100 seats contested, another one-third to be contested in 2020 and the remaining one-third in 2022) is rather more challenging than doing so for a single (usually presidential) race (though in the US ‘getting it right’ in such contests reflects the fact that whoever gets the most votes in each of the 50 states takes all of its electoral college votes, given the country’s winner-take-all system).
Based on the last round of pre-election polls, the Democrats were given a national advantage of 7.3 percent over the Republicans. This, according to the gurus running the “Real Clear Politics” website, translated into a 82 per cent chance for the Democrats to control the House of Representatives (where they needed to capture a minimum of 23 seats to achieve a 218 seat majority), but only a 15 per cent chance of doing this for the Senate.
There were two main reasons for their much lower expectations in Congress’ upper House, in which the Republicans currently hold a 51-49 advantage. First, 26/35 seats contested were held by Democrats (and in several of which Trump had defeated Clinton in 2016), making it necessary for them to retain most/all of these seats in addition to capturing any Republican-held seats. Second, each state, regardless of population, has two senators, and since the Republicans have far more support in predominantly rural areas, those states with ‘more cows than people’ are easily Republican-dominant. By contrast, in highly urbanised states such as New York and California — the two largest — the Democrats ‘waste’ their votes when their senatorial candidates win by (usually) overwhelming margins.
Taking such factors (as well as margins-of-error and unknown but estimated/expected voter turnout levels) into account, compilations from the final pre-election surveys indicated the Democrats would win close to 230 seats in the House of Representatives, but were unlikely to increase their representation in the Senate (49, as noted above). More precisely, RCP classified all races as “solid”, “likely”, or “toss-up” (meaning could go either way) for each party, which, producing the following totals: for the Republicans – 140 / 55 / 45, and for the Democrats – 182 / 10 / 3. Altogether, they estimated a total Democratic gain of 34 seats.
As of Friday, the Democrats had won 225 vs 197 for the Republicans, with 13 races undecided due to the extremely close margins between winners and losers as well as some yet-to-be-counted postal ballots. This is equal to a 51.7 to 45.3 per advantage for the Democrats, exactly what the pre-election poll average suggested.
Due to their differences of scale and level of public interest, few polls were conducted for these individual House races. This contrasts with the (more competitive) Senate contests, given their larger scale and national interest. Here, the gap between final polls and the actual results varied modestly in their accuracy.
For example, the RCP average gave the incumbent Texas Republican senator Ted Cruz a 6.8 per cent poll lead, yet he defeated Democratic Congressman Beto O’Rourke by just 1.5 per cent, evidently reflecting the excitement (if not frenzy) of his more charismatic appeal, resulting in a higher turnout among registered Democrats and younger voters in general (assisted by his significant advantage in terms of campaign contributions from all over the country as well as from Texans).
On the other hand, final polls for the Georgia gubernatorial contest gave Brian Kemp a three per cent lead over Stacey Abrams, which was largely matched (within the margin-of-error) in the actual result: Awin for Kemp by just under two per cent (though Abrams is challenging this result on several grounds).
Nationally, CNN indicated that 14 per cent more of those voters self-identifying themselves as “independents” would vote for Democratic candidates, precisely what exit polls confirmed. And all pollsters indicated a massive increase in voter turnout over past elections, which came to pass, as the 47 per cent recorded was the highest in such mid-term elections since 1966.
Additionally, many of the senatorial and congressional campaign teams conducted their own internal polls, often limited to particular sections of their constituencies/states to discover where various campaign efforts/expenditure would yield the greatest dividends in terms of enticing “undecided” voters and maximizing turnout among their already-committed supporters.
Beyond candidate vote-choice, nearly all such surveys try do ‘dig digger’ to explore the bases of such choices. These include policy preferences (the economy, healthcare, immigration, etc.), as well ‘respondent demographics’: Questions that allow for the compilation of detailed identity-portraits based on age, gender, education level, race/ethnic background, income level, employment status, and so on. Such ‘portraits’ also help reveal how political orientations/preferences of particular groups have changed, given the massive data-banks that exist going back over decades.
For example, for several complex reasons, more mainstream Protestants who used to favour the Republicans are now slightly more positively inclined to the Democrats, whereas the exact opposite (and to a much greater extent) has occurred among Evangelicals, though far more in the South than elsewhere in the country.
In other words, such surveys provide rich material not just for future campaigns, but also for university political science courses, as well as for countless academic journal articles and books.
Altogether, pollsters – and those who analyse and report their results – face challenges everywhere. But based on their performance in these US mid-term elections, one key conclusion is that even ‘getting it wrong’ on occasion can be useful in terms of refining/recalculating the tools of the trade, so as to ‘do better next time’.
Dr Wolf is a research analyst