Research Note: Did Undecided Voters Give the Liberals Their Majority?

For many, the Ontario majority Liberal win on June 12 was a surprise.  Our own polling showed the Liberals ahead among eligible voters for most of the campaign, but tied with the PCs among likely voters in our final release.  As Eric Grenier at threehundredeight.com has shown, the eligible voter estimates from polls were better at forecasting the result than any of the pollsters’ attempts to identify who was most likely to vote.

However, except for the automated telephone or IVR polls conducted during the election(although these polls underestimated the NDP’s vote), no other pollster predicted that the Liberals would win a majority government. So what might have happened?

I am a member of the academic CPEP team for the Ontario election and we will have some detailed analysis out soon on the election based on a post-election survey of 1,000 eligible voters.

In the meantime, I went back into our data sets to try and understand what might have happened in that final week of the campaign.

A Different Way of Measuring Vote Intention

On all of our Ontario election tracking studies, we asked respondents to select the party they were either likely to vote for or the party they were leaning towards supporting.  For the final poll, we also first asked respondents if they had voted in an advanced poll, and if yes, which party they voted for.

Did you vote in one of the advanced polling locations already?

Which party did you vote for?

If the ONTARIO election was today, which party would you vote for in your local constituency?

We noticed you said undecided, is there a party you’re currently leaning towards?

These were the questions that our ballot tracking was based on.

However, we also asked vote intention using a probabilistic question which asked respondents to rate the percent chance they would vote for each of the main parties or another party.  On a single page of the survey, respondents were shown the following question:

Untitled-1 copy

Each respondent had to input numbers ranging from 0 to 100 for each party and the total of all three boxes had to add up to 100.

This type of question was used successfully by the research team working on the Rand Corporation’s America Life Panel tracking the 2012 U.S. Presidential election.  A recent article in Public Opinion Quarterly offers more insights into the methodology.

The benefit of this question is that you get more varied data which allows analysts to get deeper insights into voting behaviour than the traditional vote intention question.  Not only do you know who the respondent is most likely to vote for, you can determine who they will not vote for and which other parties might be in their selection criteria.

The first step is turning the percent chance voting question into the traditional question.  We do this by recoding the five individual questions into a single question.  For respondents who rated their percent chance of voting for one party over an other were flagged as a likely supporter of that party.  For respondents who rated their percent chance of voting for two parties the same with no other party rated higher are flagged as an undecided voter.

Below are the results comparing the final weighted data by the percent chance question versus the traditional ballot question:

percentvotingnote

Overall the results for the parties are not that different using the two methods.  Comparing the first and third columns, the percentage of respondents who are undecided is fairly close.  The percent chance question produces a higher Liberal estimate but the other three parties are all within one percentage point.

The table below reports the results of “committed” respondents – with undecideds removed.

percentvotingnote2

Again, the results are fairly similar except for the higher “other” estimate with the percent chance question.  It is worth nothing that if we had used the percent chance question our final estimate would have been slightly more accurate – as it would have better reflected the NDP’s vote share.

Therefore, asking respondents to rate the percent chance they would vote for each party can produce accurate estimates, at least as accurate as the traditional ballot question.

The charts below report the trend over the course of the campaign for our five waves of research.  Our first poll was conducted from May 14 to 16.  As is clear, the lines don’t move much.  Among all respondents, the Liberal line continues to increase as the percentage of respondents who are undecided decreases.  The Tories drop from Wave 1 to Wave 2 and then rise a little before settling at 31% just before EDay.

Slide1

Slide2

And the chart below compares the results of the committed voter estimates of the percent chance and the traditional ballot questions.  The traditional question is the dashed line.

comparison

Generally speaking, the two lines follow the same trend except that the traditional question produces a higher estimate for the NDP by two to three points for each wave.

The benefit of the percent chance question is that we can do deeper into understand voting behaviour.

Using the percent chance question to go deeper.

One way to use the question to get further insights is to assess the intensity of support for each of the parties among those respondents who have not yet voted.  Two ways of doing this is to assess the percentage of respondents who say their chance of voting for one party is 100% and to calculate the mean rating for each party among those respondents who were flagged as one party’s supporters (those who said their chance of voting for one party was higher than any other).

The table below reports those estimates for each of the four main parties in the Ontario election.

percentvotingnote3

The estimates above tell us a few things:

  1. The size of the absolute cores for the OLP and PC Party were almost equal with the NDP in third.  Overall, 32% of respondents rated said there was a 100% chance they would vote for one of the four main parties.
  2. PC supporters were more motivated to vote for their party of choice.  Among those flagged as party supporters, the mean percent chance of voting for their party of choice among PC supporters was 85.5 compared to 79.8 for the Liberals and 78.1 for the NDP.  This confirms what we knew about PC supporters – they were more motivated to vote.  However, in the end, there were not enough of them to overcome the size of the Liberal lead.  Moreover, Liberal supporters were more motivated to vote Liberal than NDP supporters were to vote NDP indicating potential for some NDP supporters to swing to the Liberals in the final days of the campaign.
  3. The mean percent chance score among all respondents accurately predicted the order of the parties in popular vote and confirms the advantage the Liberals had with the electorate.

Looking at the 20% who were “undecided”.

The percent chance question can tell us a lot about those who have a preferred party.  But it can also tell us a lot about undecided voters by analyzing who those undecided voters are considering by looking at the percent chance they would vote for each of the parties.

Overall, undecided voters as a group were more likely to lean towards voting NDP.  Among all undecided voters (n=312) in our final poll, the overall mean percent chance of voting OLP was 28.5, 23.9 for the PCs, 29.6 for the NDP and 10.3 for the Greens.

Some other findings:

  • 16% of undecided voters were truly undecided, they rated the chance of voting for each of the parties equally.
  • 20% of undecided voters were leaning towards either the Liberals or NDP
  • 16% of undecided voters were leaning towards either the Liberals or PCs
  • 3% of undecided voters were leaning towards either the NDP or PCs

Combining these results (not including the 16% who were truly undecided), we find that the universe for each party among undecided voters was:

  • Liberal = 36%
  • NDP = 23%
  • PC = 19%

This means that if some of these undecided voters made up their minds  in the final days of the campaign and voted, the Liberals were most likely to benefit as they had the most undecided eligible voters considering Kathleen Wynne and the Liberals.

Some other findings to note from undecided voters:

  • Party holding electoral district where the undecided voter lived – OLP 34%, PC 24%, NDP 13%, Unsure 29%.
  • Self-placement ideology – Left 15%, Centre 24%, Right 7%, Unsure 54%
  • 2011 Provincial Vote – OLP 34%, PC 20%, NDP 20%, Green 6%, Did not vote 28%
  • Best Premier – Horwath 17%, Wynne 13%, Hudak 7%, Not sure 64%
  • Change vs. Re-Election – Time for change 39%, Liberals deserve re-election 12%, Not sure 49%
  • “Not really satisfied with any of the choices” – Agree 58%, Disagree 17%, Unsure 25%
  • “I would rather see another Liberal government than one led by Tim Hudak and the PCs” – Agree 40%, Disagree 29%, Unsure 31%
  • Who is going win the election? – OLP 26%, PC 15%, NDP 9%, Unsure 51%
  • What is the election about? – Don’t know 24%, Electing a government you can trust 22%, How to deal with the economic problems facing the province 15%, Which party would do best on social service 13%, Preventing Tim Hudak from winning 12%, Whether the Liberals deserve to be re-elected 9%, Whether Wynne, Hudak, or Horwath would make the best Premier 6%.
  • Reaction to Hudak government – Delighted 3%, Dismayed 41%, Wouldn’t mind 19%, Don’t know 37%
  • Reaction to Wynne government – Delighted 4%, Dismayed 28%, Wouldn’t mind 29%, Don’t know 39%.
  • Impression of Wynne – Positive 18%, Neutral 39%, Negative 33%, Don’t know 10%
  • Impression of Hudak – Positive 13%, Neutral 31%, Negative 47%, Don’t know 9%
  • Impression Horwath – Positive 25%, Neutral 47%, Negative 16%, Don’t know 11%

Conclusion

First, using the percent chance vote intention question can produce estimates that are as accurate or more accurate than the traditional ballot question.  However, the benefit of using it is even greater because of the deeper analytical power it gives us to understand the nature of the preferences and potential swings among undecided voters.

Second, those who were undecided in the final days of the campaign were quite unlikely to vote PC.  The PC vote had solidified earlier in the campaign and the question in final days of the campaign was whether to vote Liberal to stop them or to give the NDP a chance.

Moreover, while these voters were open to voting NDP, the prospect of a PC government could have scared many of them into voting for the Liberals, a choice most were not opposed to, but may not have preferred.   A plurality believed the Liberals were going to win and many had voted Liberal in 2011.  Voting for the Liberals was a safe choice for many.

It seems that in the final days of the campaign enough undecided voters may have moved over to the Liberals to ensure the Tories were defeated that the Liberals were able to win a majority.