Post image

How to deliver the most accurate election polling

How can someone with a data set and a good grasp of voter behavior get it so wrong?

Pollsters are facing unprecendented backlash this month after the shock victory of Donald Trump in the US Election. Similar to the shock when Brexit happened, it seems our best and brightest can’t call an election. All that being said, we wanted to take a closer look at the problems facing pollsters and how you can generate accurate polling data for your organization.

Firstly, it should be said, that the polls in the US election and in the Brexit referendum were not wildly inaccurate. Discounting some of the outlier polls that had Clinton 8-12 points in the lead, the majority of polls were within the margin of error, particularly when you consider she did win the popular vote very comfortably. The polls in the run-up to Brexit vote were certainly within the margin of error and, it seems, there are a couple of reasons why polling is struggling to accurately capture the way in which people will vote.

Problem Of Veracity

One of the single most important factors in the US polling failure was the fact that a sizeable minority of people were going to vote for Trump but were not willing to admit it. This ‘shame’ factor has been seen in the past, notably in John Major’s Conservative Party win in the UK in the early 1990’s. Polls were not able to pick up this section of the voting public and that 2-3 point shift was hugely important as the polls tightened closer to election day.

Problem Of Scale

Looking at the US election, really the only important polls were those in the swing states of Florida, Pennsylvania and so on. The fact that only about 120,000 votes separated Clinton and Trump in Florida, which had such a huge impact on the electoral college tally as a whole, meant that polling at scale, ie, across the US, was effectively meaningless. Polling needs to be localized as much as possible to retain its integrity.

Problem Of Interpretation

It seems unbelievable now that some media outlets and pollsters were declaring an 80-90% likelihood of a Clinton win, considering the polling was only a few points in her favor. An aggregation of polls would have been a better barometer of likelihood if the media were willing to take this less sensational line. However, the media are driven by ratings and sales and, therefore, the interpretation of the actual polls did suffer. That said, everyone did seem to be out-foxed by the polling and even conservative-biased media did not hold out much hope for Trump, even on election day.

How To Deliver Accurate Election Polling

So, how do we, as campaigners and candidates, avoid this polling fiasco happening in the future? Well, two of the problems outlined above can be simply countered by changing our attitude to polling.

In terms of the problem of veracity, if there is a sense that ‘shame’ may play a part in voter declarations to pollsters then that should be factored into the margin of error.

In terms of the problem of interpretation, there is a dark secret at the heart of polling in general and that is that the margin of error (3-4% usually) is where most election races end up on election week. On the face of it, that means that all polls are really pointless as they can predict anything other than the glaringly obvious races.

However, polling is a very useful tool in gauging mood during a campaign and understanding shifts in voter mood after particular events or debates for example. Also, if I was running for election, I would want to know if I was 3% down or 3% up in the final days of a campaign, even if they were both within the margin of error. So, polling is still hugely valuable but does need to be interpreted soberly.

Finally, and perhaps the key to delivering accurate polling is the point that polling should be conducted locally. If a broader geographic poll is required, it should be built from an aggregation of local polls, where possible. Taking a sample of 1000 voters in a constituency district is a far more valuable poll than taking a sample of 5000 at the national level.

Using voter contact technology campaigns can incorporate regular voter polling into their standard voter contact procedures like canvassing and email blasts. This allows campaign managers to see shifts in voter mood over time and to do more detailed polling with voters. It also, critically, allows candidates to gauge the mood of voters outside of their support base, thereby gaining valuable insight into what the ‘other side’ thinks. These methods of conducting and interpreting polling will allow you to avoid being blindsided on election day.

If you’d like to learn more about Ecanvasser’s polling functionality or speak with our polling expert, simply click the button below:

Back to Top