One of the basic questions that any campaign has is “will my candidate win?”. A poll provides a snapshot in time. It reflects voter intentions on the day the poll is conducted and does not predict necessarily what will happen on election-day. The further out from election-day the poll is conducted, the more difficult it is for a poll to predict the actual outcome. However, multiple polls establish a trend line which has more predictive potential.
In every poll there are those respondents who are undecided. Generally the further out from an election the higher the percentage of undecided voters. Typically, the percentage of undecided voters is higher in municipal elections compared to provincial and federal elections where there are a higher percentage of voters who vote along party lines. Awareness of provincial and federal campaign candidates and their associated parties are typically higher and as a result more people tend to be decided earlier on in campaigns and it stays that way right up to election-day. In provincial and federal elections it’s not uncommon to have 10% or less of voters as undecided. In contrast, the percentage of undecided voters in municipal elections can be as high as forty to fifty percent. It is therefore typical to present results as both a percentage of all respondents as well as a percent of decided voters.
To further delineate undecided voters another question could be added to the poll to ask of undecided voters. Who are you “leaning towards”? This will provide more information that assists with election strategy, especially, in the case where the percentage of undecided voters is high.
The Ballot Question
It is important to present the choice of candidate as it is presented in the real world as represented by the physical ballot on election-day. In market research there is an affect referred to as “order bias”. Order bias represents the impact of a person’s choice based on the order the choice is presented to them. If some respondents just pick the first choice on a list this could skew the result ever so slightly to the first choices they see or hear. There are two ways in which to minimize the impact of order bias. One is to rotate the order of the choices randomly. This is more difficult to accomplish with Interactive Touch Response polls, however, easier to accomplish with online polls.
With the ballot question the more appropriate solution is to present the order as it is presented to voters on election-day, in alphabetical order based on the last name of the candidate, and in the case of provincial and federal elections followed by the political party name. This is in fact how they see choices on the ballot and should order bias exist, it will be no different in the poll compared to election-day.
Sampling the Population
In order to establish results as being most representative of the population, there are two important factors that must be considered. The database in which the sample for the survey is drawn from and the weighting of the results based on certain demographics that reflects the population of “voters”.
With regards to the first factor the more “universal” the database from which the random sample is drawn from, the more representative the sample for the poll will be. This first factor is one of the major drawbacks of online polls where typically the database is derived from on-line panels that in themselves are sub-samples of the population and therefore not necessarily representative. Phone databases are more universal in nature especially when they include both residential and cell phones.
Weighting Raw Data
Weighting is the term that refers to taking raw data from polls and “weighting” them or proportioning the results to demographic statistics in the population of the municipality, riding or province as the case may be. In public opinion polling weighting is typically done by location, age and gender. Statistics are readily available for all areas of the country in order to accomplish this. Most polling firms weight their data based on population demographics. iFusion Research weights polling data by demographic but goes beyond this to include “voter turnout”.
Voter Turnout Model
In all democracies, including Canada and the USA, not all “eligible voters” end up voting.
The percentage of eligible voters that actually vote is referred to as “voter turnout”. In the last five Canadian federal elections “voter turnout” has varied from a low of 56% in 2008 to a high of 66% in 2015, the last federal election. Source: Elections Canada. Voter turnout is often lower in provincial elections and lower still in municipal elections.
This would not be a concern in polling accuracy after weighting to the population, if voter turnout was the same across demographics, but this does not actually happen. The following graph shows the average voter turnout by age and gender in Canada in the last three federal elections from post-election surveys sponsored by Elections Canada.
These statistics show, that on average, males in the 65-74 age category are almost twice as likely to vote as males in the 18-24 age category. In addition, females are slightly more likely to vote (5% points higher) in all categories up to and including the 45-54 age category and the same if not lower in higher age categories. Although there are gender differences, the age of the voter is the largest factor impacting voter turnout.
iFusion Research takes voter turnout by age and gender into account in weighting raw polling data. Our experience is that this adds a level of accuracy to our polling data, especially when you consider, that often there are large age and gender differences in support between political parties and candidates.
The added benefit of using our “Voter Turnout Model” is that we can also examine potential actual number of votes that quantifies the vote dynamic in any election that is important to campaign strategy.