The Final Week of the Nova Scotia Election by the Numbers

The 2013 Nova Scotia was an extraordinary result for a number of reasons.  First, no incumbent government was failed to win a second term in a 131 years.  Second, of the nine provincial elections held since 2010, the incumbent party has been defeated in only two of them (New Brunswick in 2010 and Quebec in 2012).  The defeat of the NDP government in Nova Scotia on October 8 broke that 140 year streak in Nova Scotia and went against the trend of incumbent wins in provincial elections.

During the final week of the campaign, Abacus Data and the Sun News Network teamed up to conduct nightly tracking polls in Nova Scotia on a range of issues.  All the releases from that final week are available at  Our final poll of likely committed voters released on Monday came within a percentage point of accurately forecasting the final popular vote totals for the election.

As a result of the nightly calling, we were able to build up a final data set of 1,300 eligible voters that allows us to do a more extensive analysis of public opinion, intentions and perceptions in the final week of the Nova Scotia election.  The data was collected from September 30 to October 6, 2013 using live telephone interviewers using a random sample of Nova Scotians eligible to vote.

Using this data, I answer a number of questions and conduct logistic regression to be understand why the Liberals won a fairly healthy majority and why the NDP was reduced to third party status.

What this analysis makes clear is that the Liberal Party easily won the 2013 Nova Scotia election for three primary reasons.

1. The Liberals had a partisan advantage: The partisan make up the electorate was such as that one in three Nova Scotians eligible to vote considered themselves to be Liberal or at least closely to the Liberal Party.  The NDP and Progressive Conservative Party were well back with less than 20% respectively.

2. The Liberals were seen as best to handle the issues voters cared most about.  Although their advantage shrunk in the final days of the campaign, the Liberals consistently lead both the NDP and the PC Party on the question of which party could best handle the issue voters believed were most important.

3. Liberal Leader Stephen McNeil was well liked and seen as the best to be Premier.  While none of the three main party leaders had very negative impressions, most eligible voters in Nova Scotia had a positive impression of the Liberal Leader.  He was well liked by Liberals, Progressive COnservatives, and New Democrats and was not seen as a threat in anyway.  When asked which leader would make the best premier, McNeil consistently lead the other two leaders by double digits.

This trifecta (partisan identification, issue positioning, and leader image) allowed the Liberals to cruise to victory and prevented both the NDP and the PC Party from breaking through at the end of the campaign.

The surprise of the campaign was the success of the PC Party in becoming the Official Opposition.  While our polling consistently had the PCs higher than other polls, it is likely that their support grew as the campaign entered the final week of the campaign.  In the analysis below, I argue that as it become more apparent that the Liberals would win and more importantly, the NDP would lose, it became easier for Tory supporters to return to their first preference – the PC Party.  As the logistic regression analysis will demonstrate, the PC Party support was built mainly of its core supporters and although it became official opposition, there was little chance of it becoming the alternative to the NDP since the Liberals had already position Stephen McNeil and the party as the real alternative to Darrell Dexter and the NDP.

Let’s start with describing the partisan make up of the electorate.

What was the partisan make up of the electorate?

The Liberals entered the final week of the campaign with a significant partisan advantage within the electorate.  Overall, 33% of eligible voters self-identified as Liberal (8% strong, 14% fairly strong, and 11% weak) compared with 18% for the Progressive Conservatives (4% strong, 8% fairly strong, and 6% weak) and 16% for the NDP (4% strong, 7% fairly strong, and 5% weak).  Over one in four eligible voters did not identify or feel closer to any political party while the remaining 8% identified with the Green Party or another political party.

The academic literature on partisan identification has found that party identification is stable, but not unmovable.  It moves as perceptions of political parties, party leaders, and issues shift.  That being said, the Liberals not only had a larger core of support within the electorate (strong partisans) but it also had almost twice as many partisans as either the NDP or the PC Party.


The data also demonstrates the importance of party identification in predicting vote intention.  The chart below reports the percentage of partisans who indicated they intended to vote for the party they most closely identified with.  Seventy-nine percent of Liberal partisans intended to vote Liberal while 76% of NDP partisans planned to vote NDP.  The Tories were somewhat less successful in converting partisans into voters with 69% of PC partisans saying they planned to vote PC, although this number increased somewhat in the final days of the campaign.


Which party did Liberal partisans support in 2009?

When we look more closely at the components of the Liberal Party’s partisan advantage, we find that among those who self-identify as Liberal,  41% voted for the Liberal Party in the 2009 election but another 29% voted for the NDP.  Eight percent voted PC while another 22% said they did not vote or can’t remember who they voted for in the 2009 Nova Scotia election.

Many former NDP supporters now identified more closely with the Liberal Party.  In fact, 42% of weak Liberal partisans voted NDP in 2009 compared with 11% among strong or fairly strong Liberal partisans.  This suggests that many Liberal partisans were likely NDP partisans and had become “Liberal” since the previous election.  Moreover, the Liberals were also successful in attracting some former PC supporters since 25% of those who identified fairly strongly as Liberal voted for the PC Party in the 2009 election.

The weakness of partisan attachment is not unique to Nova Scotia.  Much research has shown that most Canadians do not strongly identify with one political party and identification is likely to shift as a result of external events and changes in perceptions.  But the weakness of partisan attachment in Nova Scotia likely explains how three different political parties could form government over three consecutive elections.


Top Issues

Policy debates were central to the 2013 Nova Scotia.  While the Liberal Party’s main message was focused on positioning its leader as the most trusted, the state of the provincial economy, unemployment, health care, education policy, electricity rates, and MLA pensions all featured prominently throughout the campaign.

When respondents to our survey were asked what was the single most important issue facing Nova Scotia, jobs and the economy and health care were identified by a majority of respondents.  Education, taxes, electricity rates, and provincial debt were less likely to be identified as the single most important issue.

Throughout the final week of the campaign, the Liberal Party had an advantage on the top issues of concern for voters.  When asked which party could best handle the issue they believed was most important, 25% of eligible voters selected the Liberal Party followed by the NDP (17%) and the PC Party (14%).  Interestingly, 41% of eligible voters were not sure which party could best handle the issue they identified as most important to the province.

More specifically, the Liberals had a four-point advantage on jobs/economy, a five-point advantage on health care, and a 24-point advantage among those who identified education as the top issue.  On every prominent issue, the Liberals had an advantage over the NDP and PC Party.


It is clear not only from the results of the election, but also from the issue positioning data that the NDP and PC Party were unsuccessful in raising doubts about the Liberal Party’s plans on the economy, health care, and education.  While the Liberal advantages on health care and the economy closed in the final days of the campaign, the popularity of Stephen McNeil coupled with the fact that most voters did not believe the NDP deserved to be re-elected helped the Liberals to maintain a large lead heading into Election Day.

What explains the PC rise in the final days of the campaign?

At the end of September, CRA’s tracking poll for the Chronicle Herald had the Tories at around 17% among decided voters.  On Election Day, the PC Party received 26% of the vote.  What explains this rise in PC support?

I believe two things happened:

1. The PC Party and Jamie Baillie ran a good campaign.  The party was disciplined and focused on a core message around economic development and government accountability.  While the party’s promise to scrap MLA pensions did not appear in the unprompted top issue question, it likely appealed to the party’s traditional support base – and bringing these former supporters back under the PC tent was critical to at least matching the party’s 2009 performance.

2. As it became clear that the Liberals were going to win the election, former PC supporters who may have considered voting tactically for the Liberals to defeat the NDP, decided to come back to the PC Party.

The  chart below reports results from our polling, dividing the data into two periods from Monday to Wednesday and from Thursday to Sunday.


In the first period, the Tories were in third place among all eligible voters, three points behind the NDP and 16-points behind the Liberals.  More importantly however, only 52% of those who voted PC in 2009 said they planned to vote PC in the current election.  A full 24% of former PC supporters were planning to vote Liberal.

By the end of the final week of the campaign, former PC supporters returned to the Tory tent with 65% now saying they would vote PC, a 13-point shift in a matter of days.  With an NDP defeat all but certain, many former PC supporters could confidently vote for their first choice as opposed to voting strategically for the Liberals.

Could the NDP have won the election”?

Much has been written in the days following the election about the stunning defeat of the NDP.  Some commentators have even suggested that when the NDP government’s record is actually assessed, the party did not deserve the trouncing it received in the polls.  The results of the Nova Scotia election demonstrate the importance of understanding public perceptions.  Regardless of how the NDP government actually performed, Darrell Dexter and the NDP were defeated because most eligible voters did not believe they deserved to be re-elected and the alternative, Stephen McNeil and the Liberals, was an acceptable change.

Our polling found that by the final few days of the campaign, two thirds of eligible voters in Nova Scotia (66%) believed that it was time for a change in Nova Scotia compared with only 22% who believed that the NDP deserved to be re-elected.


Unlike previous incumbent provincial political parties, the NDP was unable to establish the perception that the Liberals and Stephen McNeil were unsuitable for government and a risky choice.  In Ontario, Alberta, and British Columbia, unpopular incumbent governments were re-elected not because voters believed they deserved to be re-elected but because too many voters felt that the alternatives were not any better than the current government.


In Nova Scotia, the Liberals effectively positioned Stephen McNeil as a safe choice, someone who voters could trust, and a vehicle to deliver the change that most voters wanted.  This was the fundamental reason why Nova Scotians went against the recent trend in provincial elections and decisively defeated an incumbent government.  The only chance the NDP had, and it’s strategy bore this out, was to make Stephen McNeil even less attractive than the NDP.

What role did Stephen McNeil’s popularity plan in the Liberal win?

In explaining the NDP defeat in Nova Scotia, I have shown that the Liberals had an advantage on key issues and few voters believed the NDP deserved to be re-elected.  The final part of the explanation lies in voter perceptions of the party leaders.

Throughout the final week of the campaign, Liberal Leader Stephen McNeil had high personal numbers.  A majority of eligible voters in Nova Scotia had a positive impression of the Liberal Leader and more importantly, few had anything really negative to say about the man.  Only 5% eligible voters had a very negative impression of him.

In contrast, impressions of Darrell Dexter were more negative with 37% having a positive impression and 47% having a negative impression.  But in a three-party system, a 37% positive rating is not that bad for an incumbent premier.  Had all those who had a positive impression of the NDP leader voted NDP, the party may have actually won the election.  The data suggests that Darrell Dexter alone was not the cause of the NDP’s defeat.  He was still liked by a large portion of the electorate.

The challenge for Darrell Dexter was not his personal image but that of his government and leadership style.  Voters wanted change, even though many still had positive impressions of the premier.


This desire for change also appears in the data on who voters believed would make the best Premier.  Throughout the final week of the campaign, Dexter trailed Stephen McNeil by double digits.  Overall, 32% of eligible voters believed that McNeil would make the best premier, with Dexter (19%) and Baillie (16%) well back.

And consider this: less than half (49%) of those who had a positive impression of Dexter believe he would make the best premier.  Another 17% picked McNeil while 10% picked Baillie.  Liking a party leader is not necessarily the same as believing he or she would make the best premier.

When we compare perceptions about best premier and 2009 voting behaviour, we find similar results as with other questions.  Those that voted NDP in 2009 were not convinced that Darrell Dexter would make the best premier since only 42% of former NDP voters picked Dexter as the best premier followed by Stephen McNeil at 28% and Jamie Baillie at 11%.


Explaining Vote Intentions in Nova Scotia

Using the valence politics model of voting behaviour developed by Clarke et al. (2009) , we are able to determine which factors were most important in determining a vote for the three main political parties in Nova Scotia.

The logistic regression model takes into account four main variables: (1) demographic and regional variables, (2) the party best to handle a respondent’s most important issue, (3) party identification, and (4) evaluations of the party leaders.  The benefit of a multivariate model is that it holds all variables constant and allows us to determine which variables were the most important predictors of voting for a party.

The analyses demonstrate that a number of predictors have statistically significant effects on voting behaviour.  As expected, all the party leader images have a positive effect on the the likelihood of voting.  As impressions of McNeil, Dexter, and Baillie increased, so too did the likelihood of a respondent voting for the party, all else being equal.

Among the demographic variables, only gender produces a statistically significant relationship with Liberal and PC voting.  All else being equal, men were more likely to vote Liberal while women were more likely to vote PC.  There was no statistically significant relationship between education level or age with voting.

Regionally, the only statistical relationship was among those living in Cape Breton.  Cape Bretoners were more likely to vote PC, all else being equal, than those living in Halifax (Halifax is the comparator group in the model).

Model of Liberal, NDP, PC Voting in the 2013 Nova Scotia Election





Predictor variables












Cape Breton




Central NS




South Shore/Annapolis Valley




Party Identification












Leader Images












Party Best Important Issue
















Cox & Snell R2




Nagelkerke R2




Per Cent Correctly Classified






The analyses in the table above tell us which factors have statistically significant effects on voting behaviour.  But they do not tell us much about the size of the impact of various factors on the likelihood of voting for each of the main political parties in Nova Scotia.  To better understand which factors were particularly important for voting intention, I take a cue from Clarke et. al, and constructed scenarios in which all of the predictors in the analysis are held at their mean values.  Then I manipulate the values of each significant predictor and note how the probability of voting for a party changes.

The analyses in the charts below demonstrate the importance of party leader image, party performance on top issues, and party identification to vote choice.

For the Liberal Party, the image of Stephen McNeil had the largest impact on the probability of voting Liberal.  As a respondent moved from a very negative to a very positive impression of the Liberal leader, the probability of voting Liberal increased by 80-points.  Issues mattered to voting Liberal as well.  If a respondent believed the Liberal Party could best handle the issue they considered most important, the probability of voting Liberal increased by 46-points, all else being equal.  Less important, but still a strong predictor of voting Liberal, was party identification.  If a respondent identified as a Liberal, then the probability of voting Liberal increased by 34-points, all else being equal.

These results suggest that attempts by the Liberal Party to position Stephen McNeil as a trusted leader were successful.  If a voter liked the leader, it had a large effect on the probability that the individual would vote Liberal.


The analysis of the factors affecting NDP voting look similar to those of the Liberal Party.  Evaluations of Darrell Dexter were the most important predictors for voting NDP.  So much so that the if a respondent moved from having a negative to positive impression of the NDP leader, the probability of voting NDP increased by 29-points from 7% to 36%, all else being equal.  If a respondent’s impression of Dexter shifted from very negative to very positive, the full effect of the factor was a 58-point shift, from a probability of 3% to a probability of 61%.

NDP issue positions was also important and demonstrates the importance of raising doubts about your opponents ability to tackle important issues and the importance of increasing trust that your party can best handle issues voters care most about.  For example, if a respondent believed the NDP could best handle the issue they cared about, the probability of voting NDP increased by 42 points from 8% to 50% all else being equal.

As with the Liberal Party analysis, party identification was also a strong predictor of voting NDP.  If a respondent self-identified as a New Democrat, the probability of voting NDP increased by 36-points from 9% to 45%.  The fact that the Liberals had such a large partisan advantage in the electorate meant that shifting vote intentions in the final week of the campaign for the NDP would be very difficult.


Finally, the model predicting vote choice for the PC Party was somewhat different than for its opponents.  Unlike evaluations of McNeil or Dexter, voters’ impressions of Jamie Baillie did not have as strong an impact on the probability of voting PC.  Instead, party identification and issue positioning were more important in predicting the probability that a voter would vote PC.  Moreover, the evaluations of other leaders, other party issue positions, and partisan identification with other political parties did not have as strong an impact on decreasing the probability of voting PC.  This suggests that for most voters, the election was really a choice between the NDP and the Liberals, even though the PC Party ultimately won more seats and received almost as much of the vote as the NDP.

Impressions of the Liberals and NDP polarized voters far more than impressions of the PC Party.  This means that those that voted PC were more likely to be Tory partisans and those who shared the party’s ideology than those looking for change.  It is clear from the analysis that voters who wanted a change from the NDP were much more likely to vote Liberal than NDP making the Liberal Party the clear vehicle for change.


If a respondent self-identified as PC, the probability of voting PC increased by 10-points from 1% to 11%. In terms of issue positioning, if a respondent believed the PC Party was best able to handle the issue they were most concerned about, the probability of voting PC increased by 10-points as well from 1% to 10%.  Put another way, if a respondent self-identified as a Progressive Conservative and believed the party was best able to handle the issue, the probability that he or she would vote PC was 47%, all else being equal.  If that same respondent had a positive impression of Jamie Baillie, the probability that he or she would vote PC increased to 62%.  The fact that the model predicts a PC vote probability of only 62% even if the respondent liked Jamie Baillie, considered themselves a Progressive Conservative, and believed the party was best able to handle the issue they were most concerned about demonstrates the fact that for many voters, the PCs were not the best vehicle to deliver change.  These dynamics likely changed as election day neared and PC support increased, but with all respondents in the survey, it shows the challenge the Tories faced in gaining traction throughout the campaign.

Sample Voting Scenarios

Since the voting models do such a good job at explaining vote choice, we can also use them to predict the likely voting behaviour of different types of voters.

The tables below report the probability of voting for each of the three parties based on different voter characteristics and responses to different attitudinal questions. These are rudimentary experiments but they show the power of a well-fit model with quality data.  Note, the probabilities don’t add up to 100 for each voter because the probabilities are based on individual binominal models for each party that predict the likelihood of a respondent voting for that party or for another party entirely.

The scenarios demonstrate the challenge both the NDP and PC Party faced in growing support in the final week of the campaign.  Voter 1 was an NDP partisan but believed the Liberals could best handle the issue she cared about.  She was slightly more likely to vote Liberal than NDP.  In contrast, voter 2 could be a likely PC voter except for the fact that he believed the Liberals would best handle the issue he cared about so the probability he votes Liberal was 89% compared to only 13% for the PC Party.



Voter 6 is also an interesting case in that she did not identify with a political party, had positive impressions of all three party leaders and was unclear which party would best handle the issue she cared about.  According to the model, if she voted, the probability should would vote Liberal was 69% compared to 46% probability of voting NDP.Slide3

David Coletto is CEO of Abacus Data and leads its Public Affairs research practice. He has a PhD from the University of Calgary and is an adjunct professor at Carleton University.  He’s an avid road cyclist.

Contact David Coletto:

T: 613-232-2806 x. 248



Sean on Twitter David on LinkedIn David on Google Plus