Canadian attitudes towards voting during the COVID-19 pandemic – Wave 3

Report prepared for Elections Canada

Peter John Loewen & Eric Merkley
PEARL (Policy, Elections, and Representation Lab)
Munk School of Global Affairs and Public Policy
University of Toronto

October 13, 2020

Introduction

In this report, we build on our two previous reports to help us better understand how the COVID-19 pandemic is affecting the administration of elections.^{Footnote 1} In previous reports, we have consistently shown how the risks posed by the pandemic are placing downward pressure on likely voter turnout. At the same time, we have demonstrated that increased safety measures can both increase perceptions of the safety of voting and turnout intentions.

This report builds on those findings. As with previous reports, we rely on a large-scale, representative survey of Canadians. We administer a variety of question batteries and survey experiments related to voter turnout. We advance five particular findings.

1. First, we have generated new findings on the perceived safety of in person voting. While in previous studies we have focused on the effects of safety measures on turnout intention – such as having a rule that social distancing must be practiced – in this report we focus on the effects of these rules on turnout – such as a voter arriving at a polling place to find that social distancing is or is not being practiced. We find that if voters see that the number of other voters is controlled, that social distancing is being practiced, and that other voters are abiding by masking requirements, this will increase perceptions of safety and the likelihood of voting in person. By contrast, the availability of hand sanitizer and unused pencils exhibit notably modest effects. Importantly, we also show that the effects of these measures are cumulative and increasing in combination. Moreover, we find that these measures close important inequities in the perceptions of election safety among various groups. By ensuring that individuals will experience safe environments, turnout can be increased.
2. Second, we examine whether learning about safety measures increases turnout intention above some baseline, or rather halts a decline from some baseline. We find that it is the latter. Essentially, safety measures have the effect of maintaining existing turnout levels, rather than increasing turnout. In short, information about safety measures has the potential to counteract the dampening effect of COVID-19 on turnout intention.
3. Third, we explore interest in working at polls during the election. We find that emphasizing payment or safety measures can have modest effects in increasing interest in working at polling stations.
4. Fourth, we explore the factors which help explain support for mandatory masking at polling places and the availability of weekend voting. We find that it is much easier to predict support for mask wearing than for weekend voting. In doing so, the most important predictors are perceived COVID risk, and previously being a voter. This suggests that it is voters who want to maintain the ability to vote safely who are driving the demand for safety measures, rather than non-voters who will be mobilized by safer polling places.
5. Fifth, and finally, we again explore determinants of voting method preferences. We find that the risk perception of COVID drives very large changes in willingness to vote in person on election day, in demand for mail voting, and in opposition to voting in advance in person or at an EC office.

In what follows, we explain our methodology, present our topline results, and then present more detailed analysis. We conclude with a short discussion of the implications of our findings.

Methodology – Survey Design

Our study relies on a survey of 2,503 adult Canadians conducted between September 21 and September 28, 2020. Our sample was provided by Dynata Inc, an international survey sample provider who controls a large number of proprietary survey panels. Our sample was constructed with national level quotas for age, gender, region, and language. We then weight the data using an iterative proportional fitting package (ipfweight) in STATA16.^{Footnote 2} Respondents completed the survey on the Qualtrics platform.

Our survey is principally focused on COVID-19. The survey also contains a battery of questions on past political participation, leader evaluations, and the like. The survey is non-partisan and no elements concerning partisan preferences have been analyzed for this report or shared with Elections Canada. In addition to COVID-19 related questions, respondents were also asked to answer a number of questions designed in consultation with Elections Canada.

In the first section of the survey, respondents were asked the following question where some respondents were primed to think about safety measures Elections Canada is likely to put in place in the next election. This question was repeated towards the end of the survey as well:

1. If a federal election was held tomorrow, how likely or unlikely would you be to vote?
   
   Or

   Elections Canada is likely to have in place multiple measures designed to keep citizens safe when casting a ballot in the next federal election such as:

   • limits on the number of people allowed in the polling station at the same time
   • a requirement to wear masks
   • physical distancing guidelines designed to keep individuals at least 2 metres apart

   If a federal election was held tomorrow, how likely or unlikely would you be to vote?

   • Certain to vote
   • Likely to vote
   • Unlikely to vote
   • Certain not to vote

In the final section of the survey, respondents were asked 6 questions. Respondents were randomized to one of the following two questions:

2. Now imagine a federal election was to be held in the coming weeks. If you were to vote, and considering the current situation with COVID-19, what voting method do you think you would use? [select only one]
   
   Or

   Now imagine a federal election was to be held in the coming weeks. If you were to vote, and considering the current situation with COVID-19, which of the following voting methods would you definitely NOT be willing to use? [select all that apply]

   • At the polling station on election day
   • At the advance polling station
   • At a local Elections Canada office before election day
   • By mail
   • Other (please specify)
   • Don't know yet

They were then asked their agreement with the following statements:

3. “All electors should be required to wear a mask to enter a polling station and vote”
4. “Polling day should be held on Saturday and Sunday (2 days), rather than on a Monday”
   • Strongly agree
   • Somewhat agree
   • Somewhat disagree
   • Strongly disagree
   • Don't know

They were also asked to rate their level of interest in working at the polls on election day. One group of respondents received information indicating they would get paid, another group was communicated the importance of the job without reference to pay, and another group received information indicating they would get paid and that Elections Canada would have safety measures in place to protect workers. A final group – the control condition – received none of the above pieces of information:

5. During an election, Elections Canada hires and pays local people to work at the polls in their area. How interested would you be in working at the polls in an election?
   
   Or

   During an election, Elections Canada needs local people to work at the polls in their area; without them it would be impossible to hold elections. How interested would you be in working at the polls in an election?

   Or

   During an election, Elections Canada hires and pays local people to work at the polls in their area. How interested would you be in working at the polls in an election, knowing that Elections Canada would put all necessary measures in place to ensure the safety of its workers?

   Or

   During an election, Elections Canada recruits local people to work at the polls in their area. How interested would you be in working at the polls in an election?

   • Very interested
   • Somewhat interested
   • Not very interested
   • Not at all interested
   • Don’t know

   Following the standard suite of questions respondents were given a conjoint experiment. First, they received a brief:

   In the following section we are going to provide you with several hypothetical scenarios related to in-person voting in the next federal election. Please read carefully. We will ask you some questions about each scenario.

   Then they were given three scenarios where the voting safety measures were randomized:

   Imagine that it is election day. You arrive at your designated polling station and observe the following:

   • There are [few/many] people inside the polling station
   • [Hand sanitizer is readily available for public use/Hand sanitizer containers are empty]
   • [Almost everyone is wearing a mask/Very few people are wearing masks]
   • Voters are [given brand new pencils to mark their ballots/reusing pencils left from previous voters]
   • People are [spaced apart in line by at least 2 metres/crowded together in line]

After each scenario they were asked the following:

6. How safe or unsafe do you think it is to vote in this polling station?
   • Very safe
   • Somewhat safe
   • Somewhat unsafe
   • Very unsafe


7. Would you proceed to vote at this polling station?
   • Definitely
   • Probably
   • Probably not
   • Definitely not

Variable descriptions can be found in Appendix A.

Results

Frequencies

We begin our analysis by considering the “topline” responses to each of our questions. Table 1 presents our respondents’ self-estimated likelihood of voting in the next federal election. The first two columns present the results of the baseline questions asked at the beginning of the survey in the previous two waves.^{Footnote 3} The next four columns present the results of the turnout likelihood question asked at the beginning and end of the survey for respondents given information about safety measures and those that were not. Turnout is relatively stable compared to previous waves, and once again we see evidence that turnout intention is reduced when respondents were primed to think of COVID-19. We return to this below with formal tests for whether information about safety measures increased turnout intention, or at least reduced the negative effect of COVID-19 priming on turnout intention.

Table 1. Self-reported likelihood of voter turnout across waves

	Wave 1	Wave 2	Wave 3
	Baseline	Baseline	Start of survey, no safety measures	End of survey, no safety measures	Start of survey, safety measures	End of survey, safety measures
Certain to vote	73.4	75.1	75.2	70.7	73.5	69.7
Likely to vote	19.1	17.2	17.7	19.6	18.2	21.1
Unlikely to vote	4.6	4.7	5.0	6.7	5.1	6.3
Certain not to vote	2.9	3.1	2.1	2.9	3.2	2.9
N	2,497	1,207	1,253	1,250	1,253	1,250

When we asked half of respondents about a preferred voting method, we find substantial preferences for non-in person voting methods. These results are shown in Table 2, with the first two pairs of columns presenting the results from the previous waves. While the majority of individuals indicate that they would vote in person, either at a polling station (30.4%), at an advance polling station (29.2%), or at an Elections Canada office before election day (3.9%), a substantial share of individuals (23.4%) indicate that they would prefer to vote by mail. Just one-in-nine respondents (11.3%) indicate that they do not know how they would vote. These results are stable compared to previous waves.

When we instead asked half of respondents about methods they definitely would not use, we find that 32% indicate they would not vote in a polling station on election day, 26% would not vote in advance at a polling station, 15% indicate they would not vote at an Elections Canada office, while 21% indicate that they would not vote by mail. For the most part, then, voters are not substantially opposed to any particular method of casting a ballot, but a concerning number indicate an unwillingness to vote on election day or using another in-person method. These results have remained stable across our three survey waves.

Table 2. Respondent preferences over voting method

	Wave 1		Wave 2		Wave 3
	Intend to use	Definitely not use	Intend to use	Definitely not use	Intend to use	Definitely not use
Election day	29.2	33.1	29.4	32.5	30.4	32.0
Advance polls	28.5	22.9	28.6	23.6	29.2	26.1
Elections Canada office	4.7	17.0	5.1	16.8	3.9	14.5
Mail	23.2	19.1	21.8	20.0	23.4	21.2
Other	2.2	1.7	2.2	2.2	1.9	2.3
Don't know	12.2	30.5	12.9	35.1	11.3	29.0
N	1,243	1,257	1,265	1,234	1,250	1,253

Table 3. Agreement with mandatory masks and weekend voting

	Wave 2		Wave 3
	Masks	Weekend voting	Masks	Weekend voting
Strongly agree	70.4	37.1	73.8	39.0
Somewhat agree	17.0	25.0	13.9	25.4
Somewhat disagree	4.8	9.5	4.7	9.6
Strongly disagree	3.6	5.0	3.5	5.1
Don't know	4.3	23.5	4.0	20.9
N	2,499	2,499	2,503	2,503

We also ask their level of agreement with statements signaling support for the mandatory usage of masks in polling stations and voting over the weekend rather than on a Monday. The results are shown in Table 3. A remarkable 74% of respondents strongly agree with the mandatory usage of marks, with only 8% reporting any disagreement. In contrast, support for weekend voting is softer and less certain. Twenty-one percent of respondents reported that they didn't know whether they agreed with the position or not. Thirty-nine percent strongly agree, while 15% report some level of disagreement. These results have remained consistent across two waves.

Finally, we ask our respondents their level of interest in working at the polls on election day. Respondents were randomly assigned into four conditions: one emphasizing pay, another emphasizing the importance of poll workers, another noting pay and safety measures, and a control condition with none of these features. This question was asked to over 25,000 Canadians in a previous survey fielded from June 12 to July 14, 2019, where the sample was split between the pay and importance conditions we use here. We combine respondents in both conditions in each survey to provide a comparison.

Table 4. Interest in working at the polls on election day

	Base (All)	Wave 3 (All)	Base (18-34)	Wave 3 (18-34)	Base (35-54)	Wave 3 (35-54)	Base (55+)	Wave3 (55+)
Very interested	23.6	20.7	20.8	21.9	21.5	21.1	27.2	19.6
Somewhat interested	27.4	25.2	31.2	30.3	26.9	26.0	25.2	20.8
Not very interested	21.2	17.1	21.2	18.2	21.7	17.0	20.8	16.4
Not at all interested	25.4	30.0	24.2	20.4	27.3	29.9	24.7	37.0
Don't know	2.4	7.0	2.6	9.2	2.6	6.0	2.2	6.3
N	25,040	1,255	6,618	350	8,181	425	10,241	480

Table 4 provides some evidence that interest in working at the polls has decreased since before the pandemic. Whereas in 2019 51% indicated some interest in working at the polls, now only 46% report the same inclination now. It is important to include a caveat to this direct comparison: the sample mode and provider are different between these two periods. We break this down by age group. The drop in interest is found primarily among older respondents. Individuals 55 and older used to exhibit the most interest in working at the polls. This is no longer the case. Indeed, this group demonstrates the largest drop in interest in working at a poll on election day. Previously, a majority were interested in working. A majority is now uninterested.

Table 5 provides the breakdown of our wave 3 results by treatment condition. 45% of respondents express some level of interest in working at the polls in the control condition, which is modestly higher in the pay (49%) and pay & safety measured conditions (51%), but not in the importance condition (43%). We return to this below with a formal test of statistical significance.

Table 5. Interest in working at the polls on election day, by treatment condition

	Control	Pay	Importance	Pay & Safety Measures
Very interested	21.7	24.4	16.8	24.0
Somewhat interested	23.6	24.4	26.1	26.7
Not very interested	18.6	15.1	19.2	18.1
Not at all interested	29.1	30.4	29.4	25.4
Don't know	7.0	5.6	8.5	5.9
N	614	645	610	634

In-depth Analyses

Improving the Perceived Safety of In-Person Voting

Previous waves of analysis have shown that priming respondents to think about COVID-19 reduces their intention to vote in the next election. Consequently, it is essential to discover ways to improve the perceived safety of in-person voting. In the last survey wave we exposed respondents to a conjoint experiment where they were given different scenarios of in-person voting on election day. Each scenario featured different possible safety measures put in place by Elections Canada, such as: 1) controls on the numbers of people allowed into a polling station; 2) social distancing measures to keep people at least 2 meters apart; 3) requiring everyone to wear a mask; 4) provision of single-use pencils; and 5) provision of hand sanitizer.

We found that controls on numbers, social distancing, and masks all substantially influence people's perception of safety and likelihood to vote in-person, while hand sanitizer and single-use pencils exhibit more modest effects. We also showed 1) that the effects of safety measures increased in combination; and 2) that safety measures closed important inequities in the perceptions of election safety among men and women, and the young and old, and between those with higher and lower levels of education.

One limitation of the previous study was its focus on explicit measures implemented by Elections Canada. It may have struck some respondents as odd to receive scenarios where Elections Canada fails to distribute hand sanitizer, like is now common for most public indoor spaces, and yet ensures people are at least 2 metres apart. Indeed, the novelty of a situation in which one obvious and common safety measure is taken but another is conspicuously and notably absent may trigger respondents to assume other precautions are not being taken. To mitigate this concern, we conducted a similar conjoint experiment where instead of being presented with different packages of safety measures, respondents imagine observing certain features of a particular polling location. Each of these features correspond to safety measures that were manipulated in the previous experiment. Ideally, we want to see our inferences from wave 2 replicated with this modified design.

We begin by instructing respondents that they will be provided with several possible scenarios related to the administration of in-person voting:

In the following section we are going to provide you with several hypothetical scenarios related to in-person voting in the next federal election. Please read carefully. We will ask you some questions about each scenario.

Then they were given three scenarios where the voting safety measures were randomized:

• Imagine that it is election day. You arrive at your designated polling station and observe the following:
  • There are [few/many] people inside the polling station
  • [Hand sanitizer is readily available for public use/Hand sanitizer containers are empty]
  • [Almost everyone is wearing a mask/Very few people are wearing masks]
  • Voters are [given brand new pencils to mark their ballots/reusing pencils left from previous voters]
  • People are [spaced apart in line by at least 2 metres/crowded together in line]

After each scenario respondents were asked: "How safe or unsafe do you think it is to vote in this polling station? (Very safe, somewhat safe, somewhat unsafe, very unsafe); and "Would you proceed to vote at this polling location?" (Definitely, probably, probably not, definitely not).

We estimate the effects of each scenario feature on perceptions of safety and vote likelihood in a pair of models. The estimates can be interpreted as the causal effect of each attribute averaging across the distributions of all other attributes in the model. This will tell us which of these features improves safety perceptions and the likelihood of in-person voting the most.

The estimates are provided in Figure 1. In much the same way as our experiment in wave 2, all features significantly improve perceptions of safety and the likelihood of voting at the p<0.001 level, but there are some differences in effect sizes. The effects of few people being in the polling station (8 and 6 points on 0-1 scales), and the availability of single use pencils (10 and 8 points on 0-1 scales) and hand sanitizer (9 and 8 points, respectively) are smaller than the effects of social distancing and mask usage, which range from 10 to 14 points. The large effects we observed in wave 2 remain the case with this modified design. This design makes it clearer that people are less concerned about numbers of people in polling locations than they are about those people being adequately spaced apart.

Figure 1. Effects of scenario features on perceived safety and the likelihood of in-person voting.

Note: 95% confidence intervals.

Text version of "Figure 1. Effects of scenario features on perceived safety and the likelihood of in-person voting."

This difference of means plot is used to identify the impact of proposed safety measures on electors’ perceived safety of voting in person during the COVID-19 pandemic (plotted as a circle) and the likelihood of electors voting in person (plotted as a square).

The Y-axis presents the variables (single-use pencils, sanitizer provided, high mask usage, spaced apart, few people).

The x-axis presents a -.05 to .2 scale.

Proposed safety measures	Number of points for proposed safety measures on electors' perceived safety of voting in-person (-.05 to .2 scale)	Number of points for the likelihood of electors voting in-person (-.05 to .2 scale)
Single-use pencils	.1	.03
Sanitizer provided	.5	.1
High mask usage	.12	.10
Spaced apart	.10	.2
Few people	.03	.04

The above analysis assumes that each feature has independent effects on safety perceptions and the likelihood of in-person voting. However, there may be interactive effects as well. It is possible that respondents see these features as fully or partially substitutable, such that there are diminishing marginal returns to additional safety features. Or, respondents may see these features as complimentary to one another, such that the effect of a feature is even higher when more additional safety features are in place. We found evidence of increasing returns in wave 2.

We do not have theoretical expectations for which features are more or less likely to exhibit interactive effects. Consequently, to test this possibility we create a continuous measure ranging from 0-5 for the number of features respondents were exposed to in a given scenario that contribute to perceptions of safety. We examine the effect of this index on safety perceptions and include a quadratic term in order to account for non-linearity. The predictions from this model are shown in Figure 2.

Figure 2. Effect of safety measures on perceived safety (left) and the likelihood of in-person voting (right).

Note: 95% confidence intervals.

Text version of "Figure 2. Effect of safety measures on perceived safety (left) and the likelihood of in-person voting (right)."

These two scatter plots present a quadratic relationship between 1) the number of safety measures used (x-axis, 1-5 scale) and the predicted perceived safety (y-axis, 0-1 scale) and 2) the number of safety measures used (x-axis, 1-5 scale) and the likelihood of electors voting in person (y-axis, 0-1 scale).

	0 safety measures	1 safety measure	2 safety measures	3 safety measures	4 safety measures	5 safety measures
Predicted perceived safety (0-1 scale)	2 points	3 points	4 points	5 points	6 points	8 points

	0 safety measures	1 safety measure	2 safety measures	3 safety measures	4 safety measures	5 safety measures
Predicted likelihood of electors voting in-person (0-1 scale)	3 points	4 points	4.2 points	5.8 points	6.2 points	8 points

As was the case for our previous experiment, the effects of safety features are increasing in the number of safety features – rather than decreasing – for both safety perceptions and the likelihood of voting. The effect of moving from no safety features to 1 feature is only 7 points, while it is 15 points when moving from 4 features to 5. Moving from scenarios with 0 safety features to all 5 increases safety perceptions from 0.20 on the 0-1 scale to 0.77. Similarly, the effect of moving from no features to 1 safety feature is 5 points in the likelihood of voting, while it is 14 points when moving from 4 to 5 safety features. Moving from scenarios with 0 safety features to all 5 increases the likelihood of voting from 0.32 on the 0-1 scale to 0.79. These results are strikingly similar to what we found in wave 2.

We can also explore whether there are important group differences in the influence of safety features on safety perceptions and vote likelihood. We test this by interacting relevant covariates: COVID-19 risk, age, education, urban location, and gender, with our safety feature index. We estimate a second set of models that includes a similar interaction with COVID-19 risk perceptions. The estimates are provided in Appendix B. Table 6 displays the predicted means at relevant values of each covariate for those exposed to scenarios with no safety features and those exposed to scenarios with all five.

We describe the effects in Table 6 as moving from scenarios with no safety features to those with all five. Women are more likely to increase their safety perceptions than men (59 vs. 54 points; model 1, column 3), which is a marginally significant difference (p<0.1). People aged 65 are expected to increase their safety perceptions by 66 points, compared to only 49 points for those aged 35. This difference is strongly significant (p<0.01). Individuals with university education are expected to increase their safety perceptions by 60 points, compared to only 52 points for those with high school education (p<0.01). COVID-19 health risk and urban/rural resident, however, are not associated with the effectiveness of these measures.

Table 6. Predicted perceptions of voting safety and proceeding to vote by values on treatment and relevant covariates

		Safety Perceptions			Proceed to vote
# of safety measures		None	All	Difference	None	All	Difference
Model 1
Female	Yes	0.11	0.70	0.59*	0.21	0.73	0.52***
	No	0.19	0.73	0.54*	0.32	0.74	0.42***
Age	35	0.23	0.72	0.49***	0.33	0.72	0.39***
	65	0.06	0.72	0.66***	0.19	0.75	0.56***
Urban	Small town	0.14	0.69	0.55	0.27	0.70	0.43
	Large city	0.16	0.74	0.58	0.26	0.75	0.49
Education	Secondary	0.19	0.71	0.52***	0.29	0.70	0.41***
	Undergrad	0.12	0.72	0.60***	0.25	0.76	0.51***
COVID risk	Yes	0.12	0.72	0.60	0.24	0.73	0.49
	No	0.17	0.72	0.55	0.28	0.73	0.45
Model 2
Risk perception	High	0.01	0.66	0.65***	0.10	0.70	0.60***
	Low	0.46	0.84	0.38***	0.64	0.80	0.16***

Note: * p<0.1; ** p<0.05; *** p<0.01 denote differences in treatment effect between groups.

We also see that people who perceive high levels of risk from COVID-19 increase their safety perceptions by 65 points compared to only 38 points for those who perceive little risk (p<0.1; model 2, column 3). We see virtually identical patterns when looking at respondents' reported likelihood of voting. These results, again, are very similar to what we saw with our conjoint in wave 2.

In short, the adoption of safety measures is more effective in improving safety perceptions and voting likelihood among women, the elderly, the highly educated, and among those who perceive high levels of risk from COVID-19. The end result is that the mean differences in safety perceptions and the likelihood of in-person voting become smaller in the presence of these safety measures.

Safety Priming Experiment

The conjoint experiments featured in this report and the last both strongly suggest that safety measures enhance people's perceptions of the safety of in-person voting and their likelihood of casting a ballot in person. These experiments, however, ask respondents to evaluate hypothetical scenarios. They do not tell us directly whether safety measures enhance prospective voter turnout in the midst of the pandemic.

There are two possibilities. First, it is possible that respondents have internalized the risk of COVID-19 as it pertains to voting, but are uncertain as to what safety measures will be in place during the next election. In this case, giving respondents information on safety measures should increase their prospective turnout in the next election. Second, it is possible that this risk has not been fully internalized. Consequently, exposure to information about COVID-19 could reduce prospective turnout. We have seen evidence of this in the previous two reports. It is possible that exposure to information on safety measures will dampen the negative effect of COVID-19 on prospective turnout.

We test these expectations with a blended between- and within-subjects experiment. Respondents were asked to report the likelihood of voting in the next federal election (Certain, likely, unlikely, certain not to) both at the beginning and the end of a 25 minute long survey featuring questions about COVID-19. The follow up question was asked before the conjoint experiment to ensure it did not contaminate the results. Recall that we have found reduced turnout intention in previous waves, particularly among respondents who perceive a high level of risk from COVID-19 infection. Our principal outcome is the change in turnout intention between time 1 and time 2 within the survey. We randomly assign our respondents into two groups. One group received the following prime before being asked to report their turnout intention at both time points:

The other group received no such information. We test two hypotheses. First, that respondents given information on safety measures report higher levels of turnout intention in the turnout intention question at the beginning of the survey. Second, that the reduction in turnout intention between time 1 and time 2 will be smaller among those given information about safety measures.

We find no evidence that information about COVID-19 safety measures increased prospective turnout. There is no significant difference in turnout intention between the control (0.89 on a 0-1 scale of turnout intention) and treatment conditions (0.88, diff=-0.01; p=0.22). However, respondents that were given information about safety measures did reduce their turnout intention at about half the rate as those in the control condition (-0.04 vs. -0.08). This difference is significant (p=0.02). In short, information about safety measures has the potential to counteract the dampening effect of COVID-19 on turnout intention.

Interest in Working at the Polls, Experimental Analysis

We asked our respondents their level of interest in working at the polls on election day. We randomly assigned our respondents into four groups: one receiving information that the position is paid, another emphasizing the importance of the role, another informing respondents both about pay and about safety measures that are likely to be in place, and a control condition with none of the above information. We rescale our outcome variable from 0-1 where 1 is for those who are very interested in working at the polls. 0.5 is for respondents who selected "Don't know We achieve similar results when instead excluding respondents who indicated "Don't know".

We find: no difference between the control condition and treatment condition focused on pay (0.02, p=0.40), no difference between the control and treatment condition focused on importance (-0.01, p=0.50), and a modest and marginally significant difference between the control and the pay and safety measures condition (0.04, p=0.06). There is, however, no significant difference between the pay condition and the pay and safety measures condition (0.02, p=0.03). We cannot be sure, from these results, whether information on safety measures, independent of knowledge of pay, increases interest in working at the polls on election day. Information on both of these elements, however, does appear to increase interest in working at the polls.

Determinants of Support for Mandatory Masks and Weekend Voting

Almost 88% of respondents somewhat or strongly agree with masks being mandatory at polling stations, while 62% agree with weekend voting at some level. Our data allows us to evaluate which groups of Canadians are more supportive or hostile to these measures. We estimate two models for each question. The first predicts support for the measure with demographic characteristics such as region, gender, age, location of residence, 2018 household income, education, and whether respondents (or someone in their household) are personally at risk of complications from COVID-19. The second model includes those same characteristics, while adding COVID-19 risk perceptions (0-1 index) and whether or not the respondent voted in the 2019 federal election (1=Yes). Each of our dependent variables are scaled from 0-1 where 1 indicates respondents who are strongly supportive of the measure. "Don't knows" are coded as a neutral point.

Our model estimates are shown in Table 7. Model 1 shows that education (p<0.01), age (p<0.01), urban location (p<0.01) and gender (p<0.01) are strongly associated with support for mandatory masks. Someone with a university level education would be expected to have support levels 5 points higher than someone who only graduated high school. An individual aged 65 would be expected to have support levels 11 points higher than someone aged 35. A big city resident would be expected to have support levels 4 points higher than those in small towns, and women score 4 points higher in their support than men. Interestingly, there is no association between COVID-19 health risk and support for mandatory masks when accounting for other factors in the model.

Model 2 shows that COVID-19 risk perceptions are strongly associated with support for masks (p<0.01), along with past voting history (p<0.01). Voters in 2019 are 5 points more supportive of mandatory masks compared to non-voters, while people with the highest perception of COVID-19 risk are 40 points more supportive than those who have the lowest perception of risk.

Table 7. OLS estimates, support for mandatory masks and weekend voting

	Mandatory masks		Weekend voting
	Model 1	Model 2	Model 3	Model 4
COVID-19 risk	0.01 (0.01)	-0.03*** (0.01)	-0.00 (0.01)	-0.03* (0.01)
Education	0.12*** (0.03)	0.13*** (0.03)	0.11*** (0.04)	0.11*** (0.04)
Income	-0.04 (0.03)	-0.03 (0.02)	-0.04 (0.03)	-0.05* (0.03)
Age	0.30*** (0.02)	0.25*** (0.02)	-0.03 (0.03)	-0.07** (0.03)
Location	0.05*** (0.02)	0.01 (0.02)	0.04** (0.02)	0.03 (0.02)
Female	0.04*** (0.01)	0.03*** (0.01)	0.01 (0.01)	0.00 (0.01)
Quebec	-0.00 (0.02)	0.02 (0.02)	0.03 (0.03)	0.04 (0.03)
Ontario	-0.01 (0.02)	-0.01 (0.02)	0.01 (0.03)	0.01 (0.03)
West	-0.01 (0.02)	-0.00 (0.02)	0.03 (0.03)	0.03 (0.03)
COVID-19 risk perception		0.40*** (0.03)		0.20*** (0.03)
Voted in 2019		0.05*** (0.02)		0.06*** (0.02)
Constant	0.66***	0.40***	0.62***	0.47***
R2	0.07	0.21	0.01	0.04
N	2236	2236	2236	2236

Note: Standard errors in parentheses; * p<0.1, ** p<0.05, *** p<0.01. All covariates re-scaled from 0-1 for their minimum and their maximum

Support for weekend voting is not as well predicted by our model. Model 3 shows that education and urban location are the only demographic determinant of support for this measure (p<0.01 & p<0.05). Someone with a university level education would be expected to have support levels 5 points higher than someone who only graduated high school, while a big city resident would be expected to have support levels 5 points higher than those from small towns. Model 4 highlights significant associations between COVID-19 risk perceptions (p<0.01) and voting history (p<0.01), on the one hand, and support for this measure, on the other. Electors who voted in 2019 are 6 points more supportive of weekend voting, while people who perceive the most risk from COVID-19 are 20 points more supportive of this measure than those who perceive the least amount of risk.

Determinants of Voting Method Preferences

COVID-19 concern and perceptions of election safety are also closely tied to voting methods people are willing to use or avoid. We estimate two models predicting respondents' preference for each of the following voting methods: on election day, at an advance poll, at an Elections Canada office, and by mail, as well as their opposition to each of those methods. The first model includes demographic characteristics such as region, gender, age, location of residence, 2018 household income, education, and whether they (or someone in their household) are personally at risk of complications from COVID-19. The second model includes those same characteristics, while adding level of COVID-19 concern and whether or not the respondent voted in the 2019 federal election.

Table 8. Predicted probability of supporting or opposing voting method by values on relevant covariates

		Support				Oppose
		Election day 1	Advance poll 2	EC office 3	Mail 4	Election day 5	Advance poll 6	EC office 7	Mail 8
Model 1
Female	Yes	0.27***	0.29	0.03	0.26**	0.32	0.25	0.15	0.21
	No	0.36***	0.28	0.05	0.21**	0.33	0.27	0.14	0.22
Age	35	0.32	0.25***	0.06***	0.23	0.35*	0.27	0.15	0.20
	65	0.31	0.34***	0.01***	0.24	0.30*	0.25	0.13	0.23
Urban	Small town	0.35**	0.25**	0.04	0.24	0.34	0.25	0.14	0.22
	Large city	0.29**	0.32**	0.04	0.23	0.32	0.27	0.15	0.21
Income	Low	0.29	0.27	0.02***	0.27*	0.36*	0.22**	0.15	0.22
	High	0.34	0.31	0.06***	0.20*	0.30*	0.30**	0.14	0.21
Education	Secondary	0.32	0.27	0.06**	0.20*	0.33	0.31***	0.17**	0.21
	Undergrad	0.31	0.30	0.03**	0.26*	0.32	0.23***	0.12**	0.22
COVID risk	Yes	0.27**	0.32**	0.04	0.26*	0.35	0.27	0.18**	0.20
	No	0.34**	0.27**	0.04	0.22*	0.31	0.25	0.13**	0.22
Model 2
Risk perception	High	0.25***	0.30	0.04	0.33***	0.35	0.32***	0.20***	0.21
	Low	0.49***	0.26	0.04	0.08***	0.28	0.16***	0.06***	0.23
Voted in 2019	Yes	0.35***	0.32***	0.04	0.21***	0.33	0.27	0.14	0.21
	No	0.10***	0.15***	0.05	0.41***	0.30	0.21	0.19	0.22

Note: * p<0.1; ** p<0.05; *** p<0.01.

The estimates of our models are provided in Appendix C, while predicted probabilities are shown in Table 8 to illustrate the substantive importance of the effects since logistic regression coefficients are not directly interpretable. We find that women are 9 points less likely to choose election day than men (Table 8, model 1, column 1; p<0.01). People in large cities are 6 points less likely to select election day compared to those from small towns (p<0.05). Those at risk of COVID-19 complications are also 7 points less likely to choose election day (p<0.05). COVID-19 risk perception is also negatively related with this choice (model 2, column 1). People who have the highest risk perceptions are 24 points less likely to vote on election day than people who have the lowest risk perception (0.25 vs. 0.49; p<0.01). People who voted in 2019 are 15 points more likely to choose election day (p<0.01).

When asking respondents what voting methods they would definitely not use, younger people and low-income respondents appear more inclined to oppose election day voting (model 1, column 5). Individuals aged 35 are expected to be 5 points more likely to oppose this voting method than someone aged 65 (0.35 vs. 0.30; p<0.1), while low income respondents are 6 points more likely to oppose election day voting (0.36 vs. 0.30; p<0.1).

Older respondents are modestly more inclined to select advance polls (model 1, column 2). Those aged 65, for example, are 9 points more likely to vote at advance polls than those aged 35 (0.34 vs. 0.25, p<0.01). They are also 5 points less likely to vote at an Elections Canada office than those aged 35 (p<0.01; model 1, column 3). People living in large cities are 7 points more likely to use an advance poll (p<0.05; model 1, column 2). High income respondents are 4 points more likely to vote at an Elections Canada office (p<0.01; model 1, column 3) and they are 8 points more likely to oppose voting at an advance poll (p<0.05; model 1, column 6). University educated respondents are 3 points less likely to vote at an Elections Canada office (p<0.05; model 1, column 3), but they are also 8 and 5 points less likely to oppose voting at an advance poll (p<0.01; model 1, column 6) and an Elections Canada office, respectively (p<0.05; model 1, column 7). People at risk of COVID-19 complications are 5 points more likely to choose voting at an advance poll (p<0.05; model 1, column 2), and 5 points more likely to avoid voting at an Elections Canada office (p<0.05; model 1, column 7). COVID-19 risk perceptions are also associated with 16 points more opposition to voting at an advance poll (p<0.01; model 2, column 6) and 14 points more opposition to voting at an Elections Canada office (p<0.01; model 2, column 7).

Finally, there are important demographic differences in willingness to vote by mail as shown in Table 6. Women are 5 points more likely than men to say they would use mail-in ballots (0.26 vs. 0.21; p<0.05; model 1, column 4). Higher income respondents are 7 points less likely to vote by mail than those with lower income (0.20 vs. 0.27, p<0.1; model 1, column 4), while university educated respondents are 6 points more likely to vote by mail than those with only secondary school education (0.26 vs. 0.20; p<0.1; model 1, column 4). People at risk of COVID-19 complications are also 4 points more likely to use mail (0.26 vs. 0.22, p<0.1; model 1, column 4). COVID-19 risk perceptions are associated with voting by mail in the expected direction, with people with the highest level of risk perceptions being 25 points more likely to vote by mail than those with the lowest level (0.33 vs. 0.08; p<0.01; model 2, column 4).

In short, respondents' concern about COVID-19 and perceptions of election safety are powerful determinants as to whether or not respondents will choose to vote in person – especially on election day – or whether to vote by mail.

Discussion

We have advanced five findings in this report:

1. The perceptions of safety measures and the experience of safe voting places are as likely to matter as the measures themselves. Voters need to see that voting is safe, rather than simply know that safety measures are in place.
2. Safety measures will only maintain turnout intentions – they are unlikely to raise them above previous levels.
3. Emphasizing safety measures can increase interest in individuals working at polling places.
4. COVID risk continues to be a driver of voters supporting safety measures, such as mandatory masking at polling places.
5. COVID risk also drives support for alternative voting measures.

Taken together, these findings have three important implications. Safety measures which can practically ensure safe voting places are important. They are also possible. Canadian polling places are traditionally not crowded, as the number of voters per polling place is comparatively low (notwithstanding bunching at certain hours of the day). Moreover, there are increasing returns from more safety measures. Nonetheless, our evidence suggests that these mostly serve to maintain levels of turnout, rather than increase them. Turnout increased measurably in 2015 and in 2019 for reasons quite apart from polling place safety. Our current results suggests that maintaining those hard won gains in increased turnout over the last two elections during a pandemic will rely on clear communication and experiences of safe voting in this election.

Second, given that in-person voting will not realize an increase in frequency, there is likely a need for more alternative, non-in-person voting. The only readily available and secure method on offer within the current legislative framework for federal elections is mail-in voting. Our intuition would be that perceptions of the safety and integrity of this will require education and reinforcement, especially after the Presidential election in the United States.

Finally, as a second wave of the COVID-19 pandemic onsets and more voters are exposed to the virus and consequent disease, it will be important to monitor increases in risk perceptions. These will serve to drive turnout down, and so it is important to closely monitor and test the effects of other potential safety measures as public health directives change; and to understand the effect of these on perceptions of safety.

Appendix A – Variable Descriptions

Table A1. Variable descriptions

Variable	Description
COVID-19 risk	"Is anyone in your household in a high-risk group for which the annual seasonal influenza vaccine would usually be recommended by the Public Health Agency of Canada? (These conditions include, individuals who are pregnant or those with chronic respiratory disease, chronic heart disease, chronic kidney disease, chronic liver disease, chronic neurological disease, diabetes (all types), cancer, immunosuppression, dysfunction of the spleen, and/or BMI > 40)"; 1=Yes
Education	No schooling, some elementary, elementary, some secondary, secondary, some technical, technical, some university, bachelor's degree, master's degree, doctorate or professional, rescaled 0-1, don't knows coded as missing
Income	No income; 1-30,000; 30,001-60,000; 60,001-90,000; 90,001-110,000; 110,001-150,000; 150,001-200,000; 200,000+, rescaled 0-1, don't know/prefer not to answer coded as missing
Age	In years, rescaled 0-1 from minimum (18) to maximum (99)
Location	Rural area, small town, large town, mid-sized city, large city, rescaled 0-1
Female	1=Yes
Region	1=Atlantic; 2=Quebec; 3=Ontario; 4=West
COVID-19 risk perception	"How serious of a threat do you think the coronavirus (COVID-19) is to yourself?" "How serious of a threat do you think the coronavirus (COVID-19) is to Canadians?" Not serious at all, not very serious, somewhat serious, very serious, summed and rescaled 0-1
Voted in 2019	1=Yes
In-person voting safety perceptions	Very unsafe, somewhat unsafe, somewhat safe, very safe rescaled 0-1
Voting turnout	Certain to not vote; unlikely, likely, certain to vote, rescaled 0-1

Appendix B – Estimates for Conjoint Heterogeneous Effects Analysis

Appendix B – Estimates for Conjoint Heterogeneous Effects Analysis

	Safety		Voting
	1	2	3	4
Continuous treatment	0.04** (0.01)	0.00 (0.02)	0.00 (0.02)	-0.05*** (0.02)
COVID-19 Risk	-0.05** (0.02)	0.01 (0.02)	-0.05** (0.02)	0.02 (0.02)
Risk * Treatment	0.01 (0.01)	0.00 (0.01)	0.01 (0.01)	-0.00 (0.01)
Age	-0.46*** (0.05)	-0.41*** (0.05)	-0.39*** (0.06)	-0.33*** (0.05)
Age * Treatment	0.09*** (0.02)	0.09*** (0.02)	0.10*** (0.02)	0.09*** (0.02)
Education	-0.18*** (0.05)	-0.21*** (0.05)	-0.11* (0.06)	-0.14** (0.05)
Education * Treatment	0.04** (0.02)	0.05*** (0.02)	0.05*** (0.02)	0.06*** (0.02)
Location	0.02 (0.03)	0.06* (0.03)	-0.00 (0.04)	0.03 (0.03)
Location * Treatment	0.01 (0.01)	0.01 (0.01)	0.01 (0.01)	0.01 (0.01)
Female	-0.08*** (0.02)	-0.08*** (0.02)	-0.12*** (0.02)	-0.11*** (0.02)
Female * Treatment	0.01* (0.01)	0.01* (0.01)	0.02*** (0.01)	0.02*** (0.01)
COVID-19 risk perception		-0.46*** (0.04)		-0.54*** (0.05)
Risk perception * Treatment		0.05*** (0.01)		0.09*** (0.02)
Constant	0.48***	0.75***	0.56***	0.87***
R2	0.18	0.23	0.11	0.17
N	7476	7476	7476	7476

Note: Clustered standard errors in parentheses; * p<0.1, ** p<0.05, *** p<0.01

Appendix C – Logistic Regression Estimates

Table C1A. Logit estimates, preferred method

	Election day		Advance polls
	1	2	1	2
COVID-19 risk	-0.34** (0.14)	-0.24 (0.15)	0.27* (0.14)	0.23 (0.15)
Education	-0.07 (0.36)	-0.36 (0.37)	0.44 (0.37)	0.36 (0.38)
Income	0.39 (0.31)	0.17 (0.32)	0.35 (0.31)	0.22 (0.32)
Age	-0.14 (0.34)	-0.52 (0.36)	1.22*** (0.35)	0.95*** (0.36)
Location	-0.42* (0.21)	-0.43* (0.23)	0.47** (0.23)	0.40* (0.23)
Female	-0.40*** (0.14)	-0.42*** (0.14)	0.05 (0.14)	0.04 (0.14)
Quebec	-0.27 (0.27)	-0.51* (0.29)	0.16 (0.28)	0.12 (0.29)
Ontario	-0.33 (0.26)	-0.43 (0.27)	-0.07 (0.27)	-0.09 (0.28)
West	-0.57** (0.27)	-0.68** (0.28)	-0.22 (0.28)	-0.22 (0.28)
COVID-19 risk perception		-1.15*** (0.30)		0.22 (0.29)
Voted in 2019		1.65*** (0.29)		0.91*** (0.25)
Constant	0.11	-0.04	-2.21***	-2.88***
N	1,117	1,117	1,117	1,117

Note: Standard errors in parentheses; * p<0.1, ** p<0.05, *** p<0.01

Table C1B. Logit estimates, preferred method

	Elections Canada		Mail
	1	2	1	2
COVID-19 risk	-0.06 (0.34)	-0.00 (0.35)	0.28* (0.15)	0.10 (0.16)
Education	-1.64** (0.75)	-1.59** (0.75)	0.76* (0.40)	1.11*** (0.41)
Income	2.08*** (0.67)	2.14*** (0.67)	-0.68* (0.36)	-0.44 (0.37)
Age	-4.94*** (1.07)	-4.84*** (1.10)	0.27 (0.37)	0.58 (0.38)
Location	0.01 (0.52)	0.07 (0.50)	-0.07 (0.24)	-0.15 (0.25)
Female	-0.40 (0.33)	-0.38 (0.34)	0.32** (0.15)	0.33** (0.15)
Quebec	-0.80 (0.71)	-0.82 (0.71)	0.39 (0.37)	0.66* (0.37)
Ontario	-0.50 (0.60)	-0.50 (0.60)	0.56 (0.35)	0.63* (0.36)
West	0.26 (0.61)	0.27 (0.61)	0.80** (0.36)	0.89** (0.36)
COVID-19 risk perception		-0.20 (0.60)		1.88*** (0.36)
Voted in 2019		-0.41 (0.40)		-1.03*** (0.20)
Constant	-1.33*	-1.03	-2.27***	-3.21***
N	1117	1117	1117	1117

Note: Standard errors in parentheses; * p<0.1, ** p<0.05, *** p<0.01

Table C2A. Logit estimates, opposed method

	Elections Canada		Advance polls
	1	2	1	2
COVID-19 risk	0.18 (0.14)	0.15 (0.14)	0.11 (0.15)	0.01 (0.15)
Education	-0.11 (0.35)	-0.15 (0.35)	-1.03*** (0.37)	-1.10*** (0.37)
Income	-0.50* (0.30)	-0.51* (0.30)	0.69** (0.31)	0.70** (0.31)
Age	-0.70** (0.33)	-0.80** (0.34)	-0.34 (0.36)	-0.57 (0.37)
Location	-0.11 (0.22)	-0.13 (0.22)	0.13 (0.24)	0.07 (0.24)
Female	-0.05 (0.13)	-0.06 (0.13)	-0.13 (0.14)	-0.17 (0.14)
Quebec	0.32 (0.31)	0.32 (0.31)	0.26 (0.29)	0.30 (0.30)
Ontario	0.29 (0.31)	0.28 (0.30)	-0.24 (0.29)	-0.26 (0.29)
West	0.29 (0.31)	0.28 (0.31)	-0.22 (0.29)	-0.24 (0.30)
COVID-19 risk perception		0.31 (0.28)		0.96*** (0.30)
Voted in 2019		0.18 (0.20)		0.31 (0.22)
Constant	-0.42	-0.67	-0.55	-1.25***
N	1119	1119	1119	1119

Note: Standard errors in parentheses; * p<0.1, ** p<0.05, *** p<0.01

Table C2B. Logit estimates, opposed method

	Election day		Mail
	1	2	1	2
COVID-19 risk	0.37** (0.18)	0.26 (0.18)	-0.14 (0.16)	-0.13 (0.16)
Education	-0.99** (0.47)	-0.89* (0.47)	0.20 (0.39)	0.22 (0.39)
Income	-0.26 (0.39)	-0.16 (0.39)	-0.09 (0.36)	-0.08 (0.36)
Age	-0.32 (0.47)	-0.37 (0.48)	0.44 (0.38)	0.47 (0.39)
Location	0.05 (0.30)	-0.04 (0.30)	-0.08 (0.24)	-0.07 (0.24)
Female	0.14 (0.18)	0.12 (0.18)	-0.10 (0.15)	-0.10 (0.15)
Quebec	0.51 (0.43)	0.62 (0.44)	0.11 (0.35)	0.11 (0.35)
Ontario	0.41 (0.42)	0.42 (0.43)	0.09 (0.33)	0.10 (0.34)
West	0.51 (0.42)	0.53 (0.43)	0.19 (0.34)	0.19 (0.34)
COVID-19 risk perception		1.29*** (0.35)		-0.10 (0.32)
Voted in 2019		-0.36 (0.24)		-0.07 (0.22)
Constant	-1.63***	-2.23***	-1.51***	-1.42***
N	1119	1119	1119	1119

Note: Standard errors in parentheses; * p<0.1, ** p<0.05, *** p<0.01

Footnotes

Return to source of ^{Footnote 1} Report 1 was released in June, 2020, while report 2 was released in August, 2020.

Return to source of ^{Footnote 2} Minimum and maximum untrimmed weights were 0.73 and 1.49. The average weight is 1 and the standard deviation is 0.21. All the results presented below replicate substantively in the absence of weights.

Return to source of ^{Footnote 3} Wave 1 was conducted between June 15 and June 18, 2020. Wave 2 was conducted between August 10 and August 12, 2020.