Originally Published In The Information Daily: Thursday, October 4, 2012 – 10:30 GMT
In his first article on political opinion polling for The Information Daily, Damian Lyons Lowe – Chief Executive of Survation, calls for a methodology rethink.
Polling companies are in agreement that the way we ask people who they want to vote for in an election can determine the way that the member of the public responds to the question. Some polling companies, such as YouGov, approach this by asking the question in a particular order:
Polling Prompt, YouGov
“If there were a general election tomorrow, which party would you vote for? Conservative, Labour, Liberal Democrat, Scottish Nationalist/Plaid cymru, some other party, would not vote, don’t know.
So, YouGov, in England state party selection options in their online polling prompt “CON/LAB/LD and “Some other party” in that order. In their most recent poll, (fieldwork 30th September – 1st October) this produced:
YouGov: CON 34%, LAB 43%, LD 9%, UKIP 8% GRE 3%
For a voter to indicate with YouGov that they wish to vote for “some other party” they must first click that option, they are then presented with a list of prompted alternatives, at which point they are able to select UKIP, Greens and so on.
We would have a couple of questions here.
Firstly, why display the Conservatives as an option first and the Lib Dems last? At Survation, we would be concerned that suggesting a party always as the first option may implicitly infer to a survey taker that the first option presented has more value than the others, and also that the last option (Lib Dem) has the least value. We would also have some concern that a small group of survey takers may be “speeding” through their (paid) survey and click “option one” to get through the survey quicker. YouGov regularly show the Conservatives higher than other polling companies and the Liberal Democrats lower than other companies. There are many other factors at play of course, but this is something to consider.
Certainly, Ipsos Mori may agree with some of the sense above. To avoid selection bias, Ipsos Mori, in their (random telephone number) ‘phone surveys randomise the order the survey taker is read “CON/LAB/LD options then lastly “some other party”.
Polling Prompt, Ipsos Mori
“How would you vote if there were a General Election held tomorrow? Would you vote… Conservative, Labour, Liberal Democrat [rotate order] or for some other party?”
In a recent poll, (fieldwork 15th September – 17th September) this produced:
Ipsos Mori: CON 30% LAB 41% LD 13% UKIP 4% GRE 8%
So, two different prompting approaches, two different collection methods (YouGov online self-selection, Mori telephone interview) produced two very different sets of results. Ipsos Mori randomised 3 parties. The party that YouGov displayed first (CON) was selected by more survey takers than Mori’s random order method. The party that YouGov displayed last (Lib Dem) was selected by less survey takers.
Interestingly, around the same number of responders indicated “some other party” with both firms but with YouGov, twice as many of those selected UKIP than with Ipsos Mori. Ipsos Mori have the Greens on double the vote share of UKIP, while YouGov have UKIP on more than double the vote share of the Greens.
Our second question, reflecting on both these prompting methods is “If YouGov (and others, this is the way most/all online polling companies prompt aside from Survation) are showing a similar support level for the Lib Dems and UKIP, why not ask responders in the initial prompt if they want to vote for UKIP?”
If responders need to “find” UKIP in the “Some other party” section, might YouGov be artificially depressing the UKIP vote share and potentially propping up the Conservative vote share?
We are also mindful that in a telephone interview, Mori’s callers present the interviewed member of the public with 3 options or “some other party”. We believe that many responders will select one of the three offered options (CON/LAB/LD) rather than extending their (unsolicited, unpaid) phone call. Further, they may be a social embarrassment factor with the interviewer in “getting to the unmentioned options”.
Many telephone survey respondents, on finding themselves in the “some other party” section when given options to pick from including the BNP, may find The Green Party a safe option to elect – most people do not dislike the Greens – and so don’t knows & undecideds may end up saying they would vote Green. We would be interested to hear other theories (aside from sample error / margin of error) about how a party that took 265 thousand votes in 2010 would, if you take Mori’s figures be on an implied 2.4 million today – almost an 800% increase.
What is the best way to ask the voting question?
In looking to predict General Election 2015, the party that is most important to “get right” nationally is not the Green party, or even the Lib Dems (for whom national vote share is almost meaningless to seat predictions given their localised strength) – rather it is UKIP. Why is this so?
At the 2010 General Election there were 21 constituencies where the UKIP vote was greater than the Conservative’s losing margin – and this with UKIP’s national vote share at only 3.1%
In many parts of the country, and especially in Conservative marginal constituencies, UKIP with it’s differentiated message on the UK’s place in Europe, a differentiated message on “uncontrolled” European Union migration and traditional Conservative policies on Law and Order takes vote share directly from the Conservatives.
The next years will be marked by continued financial uncertainty in the European Union with the 2014 European Elections perhaps marking the height of this. UKIP will likely be the only party with a “better off out” campaign message in the European elections and most commentators believe a repeat of 2009’s second place or better is likely for the party. If then, one year later, UKIP took, say 6% of the national vote share in the General Election it could be far more than 21 seats that would be put at risk for the Conservatives – and therefore potentially decide the next UK government.
Is UKIP currently on 4%, 8% or 12%?
The argument advanced against putting UKIP in the question prompt is that not that prompting 4 parties is not logical, rather that “there is no particular logic to it, it is just what has worked in the past” – YouGov’s Anthony Wells.
UKIP support may well diminish at the 2015 General Election if The Conservatives can convince the electorate that they have a clear pro-uk European policy, or may consider a referendum on our place in Europe or if UKIP supporters vote tactically for their Conservative MP as the “least worst” option.
The problem with this argument for predicting elections is that if one assumes (for the sake of argument) that UKIP’s support would diminish say 50% in a Westminster election – how do you quantify that without understanding the “baseline” support. Would that be 50% off Mori’s 4% (2% share) or 50% off Survation’s recent (random prompted) 12% (6% share). Without knowing the baseline, you would not have a strong view on whether UKIP will decide 10 or 50 general election seats.
Survation conduct election research via random telephone interview and online self-selection methods. Usefully, there is some evidence from this parliament that our figures randomly prompting 4 parties can produce a more accurate snapshot for accurate result prediction. We’ll leave you with two examples.
Feltham & Heston By-election
In December last year, there were two telephone polls conducted for the Feltham and Heston parliamentary by-election, one by Populus, one by Survation. Both companies used similar methodologies and the fieldwork dates overlapped. Populus (sponsor Lord Ashcroft) had a sample size however that was 3x the size of Survation’s 511 respondents (e.g. it would have a smaller margin of error). The only real difference was that Populus prompted 3 parties (CON/LAB/LD) and “other” while Survation randomly prompted (CON/LAB/LD/UKIP) and other. Despite the smaller sample size, across the parties Survation had half the error rate (5%) of Populus (10%) – where the main difference was that Populus had the Lib Dems on 2x the vote share of UKIP. The Liberal Democrats came third by around one hundred votes – a statistical tie, as the Survation method predicted.
Local Elections 2012
In the May 2012 local elections, UKIP – not traditionally a party known for local government – put up 691 candidates in around 2500 local council election contests. Their average % vote share (weighted according to total votes cast) was 13%. Of concern to Conservative marginals in 2015, UKIP’s average vote share where a Conservative candidate won the seat was 15%.
Considering the regions where there was reasonable coverage – where UKIP put up over 100 candidates – UKIP’s average vote share was 14% in Yorkshire, 10% in the North West 15% in the South East and 16% in the Eastern region.
Of course, UKIP stood in areas where they are stronger, but these figures are eyebrow raising and make a double-digit national vote share look actually, rather credible.
Without properly quantifying UKIP’s vote share by actually trying to flush it out, it is impossible to ascertain the correct support for the Conservatives and therefore impossible to predict where seats will be won and lost in General Election 2015 – surely an important consideration for the polling industry.