ORIGINAL PAPER
Fair weather voters: do Canadians stay at home
when the weather is bad?
Daniel Stockemer1
& Michael Wigginton1
Received: 1 August 2017 /Revised: 29 November 2017 /Accepted: 23 January 2018 /Published online: 1 February 2018
# ISB 2018
Abstract
What is the relationship between precipitation and the temperature on turnout? Using data on the 2004, 2006, 2008, 2011, and
2015 Canadian federal elections, we try to answer this question. Through bivariate and multi-variate statistics, we find that each
millimeter of precipitation decreases turnout by more than 0.1 percentage points. When it comes to the temperature, our results
indicate that higher temperatures trigger higher turnout. However, we also find that these relationships are influenced by season
and only apply to spring, summer, and fall elections. In the winter 2006 elections, the association was inversed; warmer
temperatures in this election triggered lower turnout, in particular when it was combined with precipitation.
Keywords Precipitation . Temperature . Weather . Turnout . Canada
Introduction
Does bad weather in the form of rain, snow, and low temperatures
prevent citizens from voting on Election Day?
Utilizing a rational choice perspective, two scenarios are
possible: on the one hand, unpleasant weather may dissuade
citizens from voting by increasing the costs associated
with going to the polls. In rainy, snowy, or cold weather,
it is unpleasant to leave one’s home and citizens might try to
avoid getting wet or cold while standing in line at or traveling
to the polls. On the other hand, particularly pleasant
weather could also deter voters by increasing opportunity
costs, as voting may interfere with the enjoyment of a nice
day either at the beach, in the mountains, or just in one’s
garden. If either, which of the two perspectives holds? In
this article, we address this question by focusing on five
Canadian legislative elections (i.e., the general elections
of 2004, 2006, 2008, 2011, and 2015).1
Specifically, we
test the influence of two weather-related factors (the
amount of precipitation and the temperature) on voter turnout.
We expect that more citizens will participate in elections
in warm and dry weather.
Our dependent variable is the percentage of registered
voters in Canada that cast their ballot, measured at the level
of the electoral district. Our independent variables are the
amount of precipitation on Election Day measured in millimeters
and the temperature measured in Celsius. Controlling for
each district’s median age and income, number of polling
stations, population density, the percentage of residents who
are immigrants, the electoral competitiveness, as well as for
northern districts, we find that both weather-related proxies
generally influence turnout. The former of the two measures,
precipitation, decreases turnout, while the latter, higher temperatures,
increases turnout.
This article proceeds as follows: first, we will succinctly
summarize the previous literature on weather-related
factors and turnout, as well as formulate our hypotheses.
Second, we will present the data, case, and control variables.
In the third part, we will quickly describe the sta-
1
Although desirable, including additional elections prior to 2004 was not
feasible as Canada underwent major redistricting prior to the 2004 elections.
Electronic supplementary material The online version of this article
(https://doi.org/10.1007/s00484-018-1506-6) contains supplementary
material, which is available to authorized users.
* Daniel Stockemer
dstockem@uottawa.ca
1
University of Ottawa, Ottawa, Canada
International Journal of Biometeorology (2018) 62:1027–1037
https://doi.org/10.1007/s00484-018-1506-6
tistical procedures employed in our analysis. Then, we
will present and discuss the results of our quantitative
study. Finally, we will situate our study within the current
literature and provide some avenues for future research.
Weather and the turnout literature
The turnout literature is one of the largest and most developed
literatures in the field of political behavior (Blais 2006). There
are two strands of turnout studies: first, individual-level studies
which try to decipher which personal characteristics render one
citizen more likely to vote than another. In general, the one’s
likelihood of voting increases with age, education, income, and
political interest and knowledge (Smets and van Ham 2013).
Second, macro-level analysis tries to determine how the context
in which the election takes place influences turnout. These
studies concur that turnout generally increases under compulsory
voting, under proportional representation in large districts,
and when the elections are important and decisive or close
(Geys 2006; Stockemer 2017). Aside from these institutional
factors, there is some, albeit inconclusive, support that the composition
of the district (e.g., urban versus rural, ethnic composition,
and level of affluence) as well as corruption and income
inequality hamper turnout (Stockemer et al. 2013).
In this study, we examine another macro-level factor, the
weather, and its influence on turnout. There are two types of
studies examining how weather- or climate-related factors can
influence turnout: one array of studies looks at seasonal
effects, and the second at the precise weather on Election
Day. When it comes to studies looking at seasonal effects,
there is some evidence that elections conducted in the winter
months trigger lower turnout than elections conducted in the
spring, and moderately lower turnout than elections held in the
summer and fall. For example, LeDuc and Pammett (2006, p.
307) highlight that Canadian elections held in the winter had,
on average, 6 percentage points lower turnout than elections
conducted in the spring (67.7% versus 74.4%). Studlar (2001)
confirms this finding; according to his analysis, spring elections
consistently had the highest turnout since 1940 in
Canadian federal elections (for a similar result for local
elections in the United Kingdom, see Rallings et al. 2003).
When it comes to the causal mechanisms behind the finding
that spring is the season that triggers the highest turnout,
there is some evidence that the length of daylight plays a role
in voters’ decision to turn out. For example, Rallings et al.
(2003) posit that the longer the day-length the higher the turnout,
as voters may be more reluctant to vote on darker days.
Eisinga et al. (2012b) contextualize this finding and report that
each additional hour of daylight corresponds to approximately
a one-half percentage point increase in turnout. For a different
seasonal factor, research has shown that seasonal holidays
may put a drain on turnout, as well, because voters might have
planned free-time activities, such as vacationing or visiting
friends out of town during the holiday. To support this conjecture,
Dubois and Lakhdar (2007) report that turnout in French
Presidential Elections is approximately 1.7 percentage points
lower in departments that are experiencing school holidays
compared to departments with no school holidays.2
When it comes to studies looking at the influence of the
weather on Election Day on political behavior, earlier studies
(e.g., Ludlum 1989) have speculated for quite some time that
poor weather such as low temperatures and precipitation, deter
citizens from going out to vote. This speculation has been
echoed by electoral bodies. For example, the Nova Scotia
House of Assembly asserted in a 2009 report that poor
weather and icy roads deter citizens from voting, though
they cited no evidence to support this claim. The academic
literature empirically supports these claims, but only finds a
fairly minor drop in turnout triggered by poor weather. For
example, Artés (2014) finds in the case of Spanish elections
that a one millimeter increase of rainfall results in a turnout
decrease of 0.053 percentage points. For the United States
1982 general elections, Merrifield (1993) reports that per each
millimeter of rainfall in a region, turnout decreases by 0.015
percentage points. Fifteen years later, Gomez et al. (2007)
report a slightly higher decrease; in their study on the US,
the reported turnout decrease is 0.035 percentage points per
millimeter increase in rainfall. Very similarly, and using the
Netherlands, as a case, Eisinga et al. (2012b) find that per each
millimeter of rainfall turnout decreases by 0.041 percentage
points in the Dutch parliamentary elections. A contrary finding
applies for the Norwegian municipal elections, where increased
rainfall is actually associated with increases in turnout
(Lind 2014).3
Arnold (forthcoming, p. 22) reports some more nuanced
findings. Focusing on the Bavarian mayoral elections, his first
finding is that the absence of rain is associated with a slight
decrease in turnout, though substantial rainfalls also have a
slight turnout-depressing effect. Arnold also reports that the
depressing effect of rain on turnout is lessened in more competitive
Bavarian mayoral elections. He argues that voters’
perception that their vote can potentially have a great impact
on the election outcome lessens the importance of costs such
2
In the Canadian context, a similar negative effect is often found during winter
elections for so-called Bsnowbirds^ who spend the winter outside of Canada in
warmer climates. For them, postal voting might be too difficult and costly
(Elections Canada 2006, p. 8; LeDuc and Pammett 2006).
3
A slightly different type of analysis distinguishes between two forms of
precipitation, rainfall and snowfall, and their influence on turnout. For
example, two US studies, by Gomez et al. (2007) and Fraga and Hersh
(2010) find that rain has a stronger negative effect on voting than snowfall,
despite its overall moderate influence.
1028 Int J Biometeorol (2018) 62:1027–1037
as rain. Fraga and Hersh (2010) observe similar results in US
presidential elections, where precipitation only has an impact
on turnout in uncompetitive elections. One of the few studies
which finds no significant effect is that of Persson et al. (2014)
who evaluate the effect of rainfall on turnout using both
individual- and aggregate-level data in Swedish parliamentary
elections; regardless of the data they use, the authors find no
significant and robust effect between the two variables.4
There are also some, albeit few, studies, which discuss the
type of individuals that might be enticed to stay at home due to
rain. For example, Knack (1994) highlights that rain only
depresses turnout for citizens with a low sense of civic duty,
leading him to conclude that only Bperipheral^ voters are deterred.
Bassi’s (2013) more psychological study suggests that
poor weather influences the mood of some voters, and can
impact their vote choice. While this laboratory study did not
address abstention directly, we deem it likely that some citizens
may conceivably experience a similar psychological effect;
that is, the mood change induced by the weather may
entice them to stay home rather than turn out on Election
Day.5
As second measure of the Election Day weather, the role
of temperature in determining electoral turnout has been far
less studied than that of precipitation. We only know of four
studies: (1) Artés’ (2014) study of Spanish elections includes
temperature data, but finds no significant effect on
turnout. (2) Eisinga et al. (2012b) do find significant effects
from temperature on Dutch election turnout, estimating a
modest increase of 0.119 percentage points per degree
Celsius of temperature increase. (3) Focusing on French
parliamentary elections, Ben Lakhdar and Dubois (2006)
examine the effect of temperature on turnout in terms of
the difference from the average temperature for the area,
finding a one-point increase in turnout for a 3 °C increase
above the average normal temperature. Finally (4),
Matsusaka and Palda (1999) incorporate temperature data
in their study on turnout in the 1980 US federal elections;
they conclude that temperature does not affect citizens’
likelihood to cast their ballot.6
To sum up, the majority of studies looking at the influence
of rainfall on turnout find a small negative influence
of precipitation on turnout. However, this finding is not
conclusive and possibly conditional to certain conditions
on the ground (see Fraga and Hersh 2010; Persson et al.
2014). Many existing studies also have some inherent
methodological weaknesses. They are conducted in relatively
small geographic areas and thus capture a limited
range of climatic variation. Furthermore, some research
examines only a single election, and most studies are conducted
in systems with unvarying election dates (e.g., US
presidential elections are always in November, Norwegian
municipal elections in September, etc.). This focus on one
electoral cycle or date might mask the possibility that the
relationship between precipitation and turnout might be
different based on the season. For example, warmer temperatures
might be beneficial for increased participation in
winter, while in the summer the reverse might be true.
When it is 30 °C, voters might either deem it too hot to
leave their home or prefer other uses of their free time,
such as swimming. This possible variation in the relationship
between weather and turnout becomes more important
given that the few existing studies—particularly in the
temperature turnout realm—provide inconclusive results.
In addition, no study has examined if the effect of precipitation
and the temperature on turnout is similar in the
winter, spring, summer and fall periods. Therefore, it is
important to engage in a renewed analysis of the relationship
between precipitation and temperature as the righthand
side variables in regression equations and turnout as
a left-hand side variable. We do so using district level data
for five Canadian elections (i.e., the 2004, 2006, 2008,
2011, and 2015 general elections). Three considerations
inform this choice: (1) the size of the country, (2) the fact
that elections are held at varying times of the year, and (3)
that there are a lot of public discussions of the possible
relationship between the weather and turnout in Canada.
First, thanks to its size and the various weather patterns in
various parts of the country Canada is geographically wellsuited
for this analysis. It provides enough variation on the
independent variables, the temperature and the amount of precipitation.
For any election between 2004 and 2015, the
amount of precipitation covered fluctuated between 0 and
15 mm of rainfall and for some elections between 0 and
4
In addition, several studies have focused on the partisan electoral implications
of weather-related turnout decreases. In the majority, these studies concur
that conservative parties benefit from this turnout decline. For example,
Gomez et al. (2007) report that in US presidential elections every inch of rain
above the election day normal results in a 2.5% increase in the Republican vote
share (0.098% per 1 mm increase). Gatrell and Bierly confirm these results for
the 2012 general election in the State of Kentucky. Similarly, Eisinga, et al.
(2012a) find that right-leaning parties benefit from inclement weather; if the
weather is bad, each right leaning party gains one seat in the Dutch parliament,
whereas, left-leaning parties lose approximately one seat. Arnold and Freier
(2016) highlight that in German local and state elections, rain-related turnout
decreases also benefit the success of the conservative CDU, at the expense of
the more left-wing SPD. Finally, Artés (2014) confirms that turnout increases
due to dry and pleasant weather hurt the vote share of the Spanish conservatives,
primarily to the benefit of small left-wing parties.
5
The studies, we cite here, measuring the influence of precipitation on turnout,
normally do not control for temperature.
6
Matsusaka and Palda (1999) also report that rainfall does not influence electoral
participation in their study.
Int J Biometeorol (2018) 62:1027–1037 1029
30 mm. The temperature difference between electoral districts
was over 15° for the 2004 elections and over 20° for other
elections. We can find the same variation in the dependent
variable turnout, which fluctuated by more than 30 percentage
points between electoral districts in any of the five election
years.
Second, Canada is one of the few countries which
holds elections in irregular intervals. Following the
British Westminster model, the Canadian Prime Minister
can ask the Governor-General to call an election any day
within a five-year framework from the previous elec-
tion.7
Contrary to most countries like the US or France,
this means that elections can and do happen in different
seasons. For example, the five elections we cover in this
study took place in all four seasons. In more detail, our
data cover the winter elections in 2006 (i.e., January 23,
2006), the spring elections in 2011 (i.e., May 2, 2011),
the summer elections in 2004 (i.e., June 23, 2004) and
the two fall elections in 2008 and 2015 (i.e., October 14,
2008 and October 19, 2015).
Third, the role of the weather in influencing elections has
been a matter of public discussion in Canada over the past
decades. In particular, the expected cold temperatures in the
January 2006 federal elections, the first federal elections to be
held during winter in over 25 years, became a topic of public
discussion. Commentators on the radio and on TV affirmed
loud and strong that this election timing would drive turnout
even further below the historic low of 60.5% achieved in
2004. However, in reality turnout rose to 64.9%, possibly as
a result of increased voter awareness campaigns and the unseasonably
pleasant temperatures on Election Day (LeDuc
and Pammett, 2006; Elections Canada, 2006, p. 8). Even beyond
the 2006 elections, the weather-turnout nexus has
remained important in the Canadian context. Statistics
Canada even included the weather as a possible reason for
non-voting in the 2011 Labour Force Survey. The survey data
reveal that voters do not seem to make a conscious decision to
stay home due to inclement weather—when asked about the
reasons for their abstention, only 0.1% of non-voters cited
poor weather as their main reason for not casting a ballot, with
the majority citing a lack of either interest (27.7%) or time
(22.9%) (Statistics Canada 2012).
Are voters’ self-assessments confirmed in district level
macro-level data? Does bad weather depress turnout?
Building on most previous research we think it does and expect
the following: (1) the higher the amount of precipitation
the lower the district level turnout, and (2) the higher the
temperature the higher the district level turnout. We are also
interested if these relationships hold across spring, summer,
fall and winter elections, or if different dynamics are at play in
different seasons. We test our expectations based on data for
almost all electoral districts in Canada for the 2004 to 2015
elections.8,9
In the next section, we present the variables and
data.
Variables and data
The dependent variable is the official turnout statistic for
each electoral district; it measures the percentage of registered
voters that cast their ballot during the respective federal
election.10
The data come from Elections Canada’s
official turnout results, which are published following each
federal general election (Elections Canada 2017). Our two
weather-related proxies or independent variables, measure
the amount of precipitation in millimeters and the average
daily temperature in degrees centigrade in each electoral
district. The data for our weather proxies come from
Environment Canada’s historical data service, which provides
historical meteorological readings from weather stations
across the country (Government of Canada 2017). In
larger electoral districts with multiple available weather
stations, we used the weather station for the riding’s largest
population centre (e.g., for Central Nova which includes
multiple municipalities, we took weather data for the town
of New Glasgow). In more concentrated urban areas with
multiple electoral districts in the same municipality, we
sometimes had to use the same weather station for multiple
ridings (e.g., all downtown Toronto districts shared the
same weather data). In total, we used 135 weather stations
with slight year-to-year variations as stations became disused
or new stations appeared.
Since the weather is likely not the only variable which
influences district level turnout, we control for seven likely
predictors of turnout, all at the district level; median age and
income, population density, the percentage of immigrants,
the number of polling stations, and the northern location of
7
Officially, Canada has had fixed election dates since 2007, with elections
occurring on the third Monday in October. In practice, this rule was not
followed either in 2008 or 2011. Although this law increases pressure to hold
elections at regular intervals, to date only the 2015 election was held on the
prescribed date, and the instability of minority governments makes early elections
possible and even likely in the federal elections to come.
8
We excluded Canada’s three artic territories from the analysis. Additionally,
we excluded the electoral district of Durham for the year 2004 due to the
unavailability of census data.
9
Before the 2015 election, federal electoral districts underwent regular decennial
redistricting that moved some electoral boundaries and created 30 new
ridings. To account for this, our pooled models include only those 2015 districts
which lost no more than 40% of their territory in the redistricting, and
gained no more that 40% new territory, as calculated by the Pundit’s Guide to
Canadian Elections (Funke 2017). In total, we matched 93% of the 2004–2011
districts to a 2015 district.
10
We used electoral districts instead of polling sub-divisions as census data
were unavailable for these smaller areas. The same applies to more precise
weather data.
1030 Int J Biometeorol (2018) 62:1027–1037
the districts. We sourced the first five of these predictors
from Statistics Canada’s 2006 and 2011 Censuses of
Population,11
as well as the National Household Survey of
2011.12
We collected the sixth control variable, the number
of polling stations per district from the Elections Canada
data (Elections Canada, 2017).
When it comes to our control variables, we first hypothesize
that districts with older electorates should have higher
turnout. Research (e.g., Franklin 2004; Melo and Stockemer
2014) has established that voting is more of a habit for older
individuals than for younger people. Normally, the older citizens
are, the more civic minded they are, the more integrated
they are in society, and the more stable political ideologies
they have, all of which should contribute to a higher likelihood
to vote. In this study, we further assume that what applies to
the micro-level should also apply to the macro-level; that is
older electorates should trigger higher turnout. We
operationalize age by the median age in the electoral district.
We use a similar logic for the second control variable, the
average income per district; that is on the micro-level it is an
established finding that individuals’ likelihood to vote increases
with their socio-economic status (SES) (Gallego
2010). In other words, individuals in the higher echelons of
society have the personal and financial resources to get politically
engaged, as well as the network connections (Verba
et al. 1995). As it is standard in the literature, we measure
somebody’s SES by their income. Because we have an aggregated
analysis, we use the median individual income per
district.
The third control variable gauges the degree of urbanization
in a district. Because of higher levels of religiosity and
closer links between politicians and parties, among other
things, turnout has traditionally been higher in the countryside
than in cities and urban hubs in Canada and other Western
countries (Corvalan and Cox 2013; Smets and van Ham
2013). We have no reason to assume that this relationship
should not hold for our study. Hence, we hypothesize that
more rural districts should have higher turnout than more urban
districts. We measure the degree of urbanization by the
population density per square kilometer, which we derived
using the census values for total population and area of the
electoral district.
Forth, we control for the percentage of citizens who are not
born in the country. We expect those individuals who are born
in the country of current residence (i.e., those born in Canada)
to have a stronger attachment to Btheir^ country than
Canadians who came to Canada as immigrants. Native-born
Canadians are more familiar with the political institutions and
parties in their country; they have become politically socialized
in their home country, and they have developed partisan
attachments, all of which should contribute to higher turnout
(Xu 2005). However, due to data unavailability on the percentage
of naturalized citizens, we have to employ a suboptimal
proxy variable for the percentage of naturalized citizens,
namely the percentage of the population within any district
that have not been Canadian citizens from birth. For sure, this
includes naturalized citizens in the majority, but also permanent
residents and individuals residing in Canada on a visa.
Ideally, we would have liked to exclude the latter two categories,
as they are ineligible to vote, but this is not possible with
the data we have. We nevertheless assume that our proxy
provides a somewhat valid operationalization measuring the
influence of the proportion of non-Canadians by birth in a
district on turnout.
Fifth, we control for the average number of polling stations
per square kilometer in each district. The idea is that the closer
the polling station is to somebody’s home the quicker and less
costly voting should be. In contrast, if voters have to drive
several kilometers, or even tens of kilometers to get to their
station, voting becomes more costly in terms of both time and
money, and thus turnout should be lower (Aldrich 1993;
Feddersen 2004). We calculated the average number of
polling stations by dividing the absolute number of polling
stations per district (provided in the Elections Canada data)
by the size of the district measured in terms of square kilometers
(as reported in the census data).
The sixth control variable is the competitiveness of the
race. Following rational choice theory, we assume that the
more competitive the electoral race is, the higher the turnout
should be (Grofman et al. 1998). A competitive race increases
every voter’s chance that her vote will be decisive.
Competitive races also trigger more campaign activity and
increased media attention for the local race; both of which
should also increase turnout (Cancela and Geys 2016). We
measure competitive races as it is standard in the literature
by the gap in votes between the winner and the runner-up of
in the district.
Finally, we control for those districts that are located in
Canada’s north. In more detail, we identify, via a dummy
variable, all 14 ridings whose territory fall within the
BNorthern^ or BNorthern Transition Zones,^ as defined
by Statistics Canada (2006).13
These districts are generally
large, sparsely populated and can experience extreme
11
Ideally, we would have liked to use the 2016 census for the 2015 election;
however, not all the necessary data had been released at the time of data
collection.
12
Stephen Harper’s Conservative government abolished the mandatory longform
census, replacing it with the optional National Household Survey, which
collected the same information. The mandatory short-form census did continue
to exist for more basic demographic data, and was used for population and age
data in our sample.
13
We exclude the three districts of Canada’s artic territories due to their size,
extreme weather, low population and incredibly different social and political
realities from the rest of Canada.
Int J Biometeorol (2018) 62:1027–1037 1031
weather conditions, all of which can have an influence on
turnout.
Statistical procedures
Our data capture the population of electoral districts for
our five elections. We use these data for the following
types of analyses: first, we use the raw data and present
some bivariate analyses (i.e. scatterplots) between either
of our two weather proxies and district level turnout, respectively.
Second, we present three pooled models. On
the left-hand side is the dependent variable, district level
turnout. On the right-hand side are the two independent
variables; the precipitation per day measured in millimeters,
and the daily average temperature, as well as the
control variables. Our main model is a unit fixed effects
model. This model is the most conservative model we
could choose as it controls for other non-observed district-level
characteristics (Allison 2009). It gauges if
year-to-year weather changes within a district change
electoral turnout. As robustness checks, we run the same
model as a panel random effects model and a generalized
least squares (GLS) model. To account for the fact that
overall turnout and district level turnout differed in the 5
elections we cover, we include election dummies for the
2006, 2008, 2011, and 2015 general elections. The 2004
elections serve as the reference category in our three main
models.
Results
The bivariate scatterplots mainly confirm our initial hypotheses;
that is Figs. 1 and 2 generally confirm (1) that
more precipitation leads to lower turnout and (2) higher
temperatures lead to higher turnout.14
In Fig. 1, which
gauges the influence between precipitation and turnout,
all scatterplots show a flat or robust negative fitted line,
confirming the notion that if it rains, some voters might
prefer to stay at home rather than casting their ballot.
Substantively, the scatterplots display a 0 to 5 percentage
points’ difference in turnout between district where it does
not rain and districts where it rains a lot during Election
Day (i.e., 15 mm of rainfall and more).
When it comes to temperature, our assumption that
higher temperatures trigger higher turnout is confirmed in
three out of the five scatterplots (Fig. 2). Except for the
2006 winter election, where higher temperatures led to
fewer individuals casting their ballot and the 2015 election,
where higher temperatures seem to have no impact on electoral
participation, the remaining three elections show a
solid relationship between warmer temperatures and higher
turnout. This relationship seems particularly strong for the
2008 fall election. As such, the third scatterplot highlights
that turnout was approximately 5 percentage points higher
in districts with an average temperature between 15 and
20 °C, as compared to districts with an average temperature
between 0 and 5°. In 2004, and 2011, the relationship
was a bit less strong, but still in the vicinity of 2 to 3
percentage points between cold districts and warm
districts.
The main exception of our general pattern is the 2006
winter general election. These elections reveal an opposite
than expected relationship; in this January election, turnout
was significantly higher in Bcold^ districts, where the average
daily temperature was in the vicinity of −10 or −15
C, than it was in less cold districts, with a temperature
between 0 and 5 °C. How can this rather unexpected finding
be explained? Would not it be logical that voters turn
out more in relative warm temperatures (for winter) of
around 0 or 5 °C than in relatively cold weather with average
temperatures of −10 or −15 °C? Possibly, this relationship
could be explained by the distribution of precipitation
on Election Day. In other words, the warmer districts
might have been the ones where it has rained or snowed a
lot. To examine this possibility, we first run a correlation
analysis; the Pearson correlation coefficient is 0.41, indicating
some medium-strong association between warmer
weather and more precipitation. We examine this interaction
effect below in more detail, by displaying a separate
OLS model for 2006 with the addition of an interaction
variable between the two weather proxies (see model 4).
Because the winter elections of 2006 do not fit the general
pattern, we also run models 1 to 3 excluding these elections
(see models 5 to 7).
Our main regression models confirm the nuance from
our bivariate analysis with the raw data. Models 1 to 3
indicate that precipitation moderately decreases turnout.
In more detail, the three equations predict that per every
millimeter of rain- or snowfall decreases turnout by approximately
0.1 points. In other words, 10 mm of precipitation
decreases electoral participation by 1 percentage
point. When it comes to temperature, Models 1 to 3 illustrate
that for every degree the temperature gets warmer,
turnout increases by 0.05 to 0.07 percentage points. In
other words, a ten-degree temperature increase triggers approximately
a half percentage point increase in participation
rates. However, if we exclude the 2006 winter elections,
the substantive influence of the variable temperature
is more than twice as high as in our main equations. In
other words, our additional regression models (see models
14
Because of the different seasons during which our five elections took place,
it is impossible to create one graph displaying the influence of temperature or
precipitation on turnout.
1032 Int J Biometeorol (2018) 62:1027–1037
5 to 7), which exclude the winter 2006 elections predict
that turnout increases at least 0.15 points for every degree
Celsius the temperature is warmer. In other words, the
models predict a 1.5 percentage point increase in turnout
for a 10° increase in temperature (Tables 1 and 2).
Model 4 confirms that the 2006 winter election reveals
a specific pattern confirming the bivariate finding that
higher temperatures decrease turnout for this particular
election. The same applies to high precipitation. In particular,
in areas with relatively warm winter temperatures
and a lot of precipitation turnout is higher. We can only
speculate about the rather counterintuitive findings for the
winter 2006 elections. Though the higher turnout could be
an anomaly, it may still be explained in terms of opportunity
costs. In those areas with no precipitation, the temperature
was colder than in areas with precipitation, but
still unseasonably warm for January; 5 °C is essentially
spring weather for most of Canada, and −10 °C a mild
winter day. These rather pleasant temperatures in combination
with sunshine might have made enjoying outdoor
activities more enticing than voting for some citizens.
Conversely, the presence of precipitation may negate the
Fig. 2 Scatterplot: Mean
temperature on turnout for the
June 28, 2004, January 23, 2006,
October 14, 2008, May 2, 2011,
and October 19, 2015 elections,
respectively
Fig. 1 Scatterplot: Precipitation
and turnout for the June 28, 2004,
January 23, 2006, October 14,
2008, May 2, 2011, and October
19, 2015 elections, respectively)
Int J Biometeorol (2018) 62:1027–1037 1033
desire to enjoy the warm weather; hence, the combination
between warm weather and precipitation might have
Bdecreased^ the opportunity costs, leading more citizens
to turn out. A similar logic is used by Lind (2014) to
explain the positive relationship between rain and turnout
in Norwegian municipal elections, although with no analysis
of the role of temperature.
Conclusion
This article pushes our understanding of the weather–turnout
nexus forward. As the first study that looks at the association
between the weather and turnout in all four seasons we make
important observations. First, we confirm most prior research
that highlights that precipitation decreases turnout, whereas
Table 1 Multiple regression
models measuring the influence
of precipitation/temperature on
turnout
Model 1:
Fixed
effects model
Model 2:
Random
effects model
Model 3: GLS
model
Model 4: The 2006
interactive model
(OLS with Huber
White standard
errors)
Average temperature 0.074***
(0.015)
0.070***
(0.015)
0.052*
(0.028)
−0.104**
(0.043)
Total precipitation −0.113***
(0.020)
−0.108***
(0.020)
−0.082**
(0.037)
−0.125**
(0.079)
Temperature
∗ Precipitation
0.036**1
(0.019)
Median district level age −0.0000005
(0.00001)
−0.0000006
(0.00001)
−0.00003
(0.00003)
0.541***
(0.082)
Median district level
income
0.0004***
(0.00005)
0.0004***
(0.00003)
0.0005***
(0.00002)
0.0007***
(0.00006)
Population density 0.0003
(0.0005)
0.00005
(0.0004)
−0.0007
(0.0005)
0.001
(0.001)
Average number of voting
stations per square
kilometer
0.023
(0.185)
0.118
(0.017)
0.455*
(0.238)
−0.614
(0.526)
Percent immigrants per
district
−0.106***
(0.039)
−0.125***
(0.014)
−0.120***
(0.010)
−0.067**
(0.017)
Margin of victory −0.015***
(0.005)
−0.024***
(0.005)
−0.071***
(0.007)
−0.048***
(0.013)
North −1.61
(2.27)
−0.199
(1.59)
0.318
(1.23)
−1.38
(1.27)
2006 5.38***
(0.370)
5.29***
(0.370)
4.90***
(0.691)
2008 −1.45***
(0.197)
−1.47***
(0.198)
−1.47
(0.387)
2011 −0.705**
(0.286)
−0.812***
(0.246)
−1.13***
(0.415)
2015 6.34***
(0.316)
6.16
(0.281)
5.65
(0.469)
Constant 50.88***
(1.46)
50.91
(0.949)
51.32***
(0.822)
25.95***
(4.50)
R2
0.51 0.53 0.51
Log likelihood −4172.33
Number of observations 1463 1463 1463 289
Number of groups 339 339 339 289
1
The estimate of 0.036 in model 4 is the effect of the product of temperature and precipitation (i.e., interaction
effect). To interpret the interaction effect, we have to interpret the effect of temperature, precipitation, and the
interaction
Notes: Standard errors in parentheses. Significance: * p < 0.1; ** p < 0.05; *** p < 0.01 (two-tailed)
1034 Int J Biometeorol (2018) 62:1027–1037
higher temperatures increase turnout. This association seems
to hold reasonably well for spring, summer and fall elections.
Substantively, we find that for every millimeter of precipitation
turnout decreases by 0.1 percentage points. The relationship
for temperature is of a somewhat smaller magnitude; that
is for every degree the temperature increases turnout increases
by approximately 0.7°. These impacts are slightly larger in
comparison to most other studies, hinting that rain/snowfall
and the daily temperature on Election Day could explain fluctuations
in turnout of several percentage points.15
Second, our results also indicate that for the winter
elections, this general pattern might not hold. In particular,
we find that colder temperatures trigger higher turnout;
this finding applies, in particular, to regions with no
precipitation. In contrast, in districts with high precipitation,
the combination between mild winter temperatures
and high amounts of precipitation triggers higher turnout.
Hence, the winter elections do not confirm the general
relationship. The costs and opportunity costs associated
with winter voting might be more complex and require
further inquiry not only in Canada, but in other countries
and regions, as well.
For policy, this study allows for the tentative conclusion
that scheduling elections during more pleasant weather
could moderately increase turnout. If it is warm and does
not rain on Election Day, more individuals will turn out as
compared to a relatively cold and rainy day. Nevertheless,
addressing the effects of the weather alone will not dramatically
change turnout. Rather, to reach turnout figures
achieved in the 1950s and 1960s in Canada and elsewhere
in Western countries, it is necessary that citizens achieve
equally high amounts of political interest and knowledge
than citizens had decades ago. On the institutional side, the
only measure that would significantly increase turnout
15
Unfortunately, we could not consider early voting, which ranged from 5.5 in
2004 to 20.5% in 2015.
Table 2 Multiple regression
models measuring the influence
of precipitation and the
temperature on turnout excluding
the year 2006
Model 5: Fixed
effects model
Model 6: Random
effects model
Model 7: GLS
model
Average temperature 0.169***
(0.020)
0.164***
(0.020)
0.150***
(0.037)
Total precipitation −0.084***
(0.023)
−0.076***
(0.023)
−0.42*
(0.043)
Median district level age −0.0000007
(0.00001)
−0.0000008
(0.00002)
−0.00003
(0.00003)
Median district level income 0.0004***
(0.000005)
0.0004***
(0.00003)
0.0005***
(0.00002)
Population density 0.0004
(0.0005)
0.0001
(0.0005)
−0.0008
(0.0006)
Average number of voting
stations per square kilometer
−0.021
(0.207)
0.111
(0.192)
0.515*
(0.266)
Percent immigrants per district −0.146***
(0.041)
−0.142***
(0.015)
−0.134***
(0.011)
North −0.727
(2.31)
0.081
(1.62)
0.418
(1.37)
2008 −0.937***
(0.209)
−0.965***
(0.210)
−0.950**
(0.416)
2011 −0.036
(0.314)
−0.171
(0.266)
−0.445
(0.451)
2015 7.46***
(0.367)
7.27
(0.324)
6.85
(0.551)
Constant 50.39***
(1.54)
50.02***
(0.987)
50.57***
(0.940)
R2
0.53 0.54
Log likelihood −3369.13
Number of observations 1174 1774 1774
Number of groups 338 338 338
Notes: Standard errors in parentheses. Significance: * p < 0.1; ** p < 0.05; *** p < 0.01 (two-tailed)
Int J Biometeorol (2018) 62:1027–1037 1035
would be the implementation of mandatory compulsory
voting, which, in turn, could reveal other ethical
considerations.
A possible avenue for future research is to further investigate
which Canadian voters are deterred by inclement
weather. Other studies have found patterns in partisan affiliation,
such as US Republicans being less affected, as well
as that those with low senses of civic duty (Knack 1994;
Gomez et al. 2007). Theoretically, we could also expect that
elderly voters with decreased mobility may worry about
slipping in ice or rain.
References
Aldrich JH (1993) Rational choice and turnout. Am J Polit Sci 37(1):246–
278. https://doi.org/10.2307/2111531
Allison PD (2009) Fixed effects regression models. SAGE publications.
https://doi.org/10.4135/9781412993869
Arnold F (Forthcoming) Turnout and closeness: evidence from 60 years
of Bavarian mayoral elections. Scand J of Economics. https://doi.
org/10.1111/sjoe.12241
Arnold F, Freier R (2016) Only conservatives are voting in the rain:
evidence from German local and state elections. Elect Stud 41:
216–221. https://doi.org/10.1016/j.electstud.2015.11.005
Artés J (2014) The rain in Spain: turnout and partisan voting in Spanish
elections. Eur J Polit Econ 34:126–141. https://doi.org/10.1016/j.
ejpoleco.2014.01.005
Bassi A (2013) Weather, mood, and voting: an experimental analysis of
the effect of weather beyond turnout. Social Science Research
Network, Rochester, NY
Ben Lakhdar C, Dubois E (2006) Climate and electoral turnout in France.
French Politics 4(2):137–157. https://doi.org/10.1057/palgrave.fp.
8200100
Blais A (2006) What affects voter turnout? Annu Rev Polit Sci 9(1):111–
125. https://doi.org/10.1146/annurev.polisci.9.070204.105121
Cancela J, Geys B (2016) Explaining voter turnout: a meta-analysis of
national and subnational elections. Elect Stud 42:264–275. https://
doi.org/10.1016/j.electstud.2016.03.005
Corvalan A, Cox P (2013) Class-biased electoral participation: the youth
vote in Chile. Latin American Politics and Society 55(3):47–68.
https://doi.org/10.1111/j.1548-2456.2013.00202.x
Dubois E, Lakhdar CB (2007) More on the seasonal determinants of
turnout: holidays and French presidential elections. French
Politics; Basingstoke 5(2):144–159. https://doi.org/10.1057/
palgrave.fp.8200123
Eisinga R, Grotenhuis M, Pelzer B (2012a) Weather conditions and political
party vote share in Dutch national parliament elections, 1971–
2010. Int J Biometeorol 56(6):1161–1165. https://doi.org/10.1007/
s00484-011-0504-8
Eisinga R, Grotenhuis MT, Pelzer B (2012b) Weather conditions and
voter turnout in Dutch national parliament elections, 1971–2010.
Int J Biometeorol 56(4):783–786. https://doi.org/10.1007/s00484-
011-0477-7
Elections Canada (2006) Report of the Chief Electoral Officer of Canada
on the 39th General Election of January 23, 2006
Elections Canada (2017) Elections Canada’s official reports—official
voting results. Elections Canada, In http://www.elections.ca/
content.aspx?section=res&dir=rep/off&document=index&lang=e#
officialvoting. Accessed 16 May 2017
Feddersen TJ (2004) Rational choice theory and the paradox of not voting.
J Econ Perspect 18(1):99–112. https://doi.org/10.1257/
089533004773563458
Fraga BL, Hersh E (2010) Voting costs and voter turnout in competitive
elections
Franklin MN (2004) Voter turnout and the dynamics of electoral competition
in established democracies since 1945. Cambridge University
Press. https://doi.org/10.1017/CBO9780511616884
Funke A (2017) Pundits’ guide to Canadian elections. https://www.
punditsguide.ca/
Gallego A (2010) Understanding unequal turnout: education and voting
in comparative perspective. Elect Stud 29(2):239–248. https://doi.
org/10.1016/j.electstud.2009.11.002
Geys B (2006) Explaining voter turnout: a review of aggregate-level
research. Elect Stud 25(4):637–663. https://doi.org/10.1016/j.
electstud.2005.09.002
Gomez BT, Hansford TG, Krause GA (2007) The republicans should
pray for rain: weather, turnout, and voting in U.S. presidential elections.
The Journal of Politics 69(3):649–663. https://doi.org/10.
1111/j.1468-2508.2007.00565.x
Government of Canada (2017) Historical Climate Data. Retrieved on July
14, 2017 from www.climate.weather.gc.ca
Grofman B, Collet C, Griffin R (1998) Analyzing the turnout–competition
link with aggregate cross-sectional data. Public Choice 95:233–
246
Knack S (1994) Does rain help the Republicans? Theory and evidence on
turnout and the vote. Public Choice 79(1-2):187–209. https://doi.
org/10.1007/BF01047926
LeDuc L, Pammett JH (2006) Voter turnout in 2006: more than just the
weather. In: Pammett JH, Dornan C (eds) The Canadian federal
election of 2006. Dundurn Press, Toronto ON, pp 305–326
Lind JT (2014) Rainy day politics—an instrumental variables approach to
the effect of parties on political outcomes. Social Science Research
Network, Rochester, NY
Ludlum D (1989) The weather factor. American Meteorological Society,
Boston
Matsusaka JG, Palda F (1999) Voter turnout: how much can we explain?
Public Choice 98(3/4):431–446. https://doi.org/10.1023/A:
1018328621580
Melo DF, Stockemer D (2014) Age and political participation in
Germany, France and the UK: a comparative analysis.
Comparative European Politics 12(1):33–53. https://doi.org/10.
1057/cep.2012.31
Merrifield J (1993) The institutional and political factors that influence
voter turnout. Public Choice 77(3):657–667. https://doi.org/10.
1007/BF01047865
Persson M, Sundell A, Öhrvall R (2014) Does election day weather affect
voter turnout? Evidence from Swedish elections. Elect Stud 33:335–
342. https://doi.org/10.1016/j.electstud.2013.07.021
Rallings C, Thrasher M, Borisyuk G (2003) Seasonal factors, voter fatigue
and the costs of voting. Elect Stud 22(1):65–79. https://doi.org/
10.1016/S0261-3794(01)00047-6
Smets K, Van Ham C (2013) The embarrassment of riches? A metaanalysis
of individuallevel research on voter turnout. Elect Stud
32(2):344–359
Statistics Canada (2006) What is north? In: Canada year book 2006.
Statistics Canada, Ottawa
Statistics Canada (2012) Reasons for not voting. http://www.statcan.gc.
ca/pub/11-402-x/2012000/chap/sc/sc01-eng.htm. Accessed 15
Apr 2017
Stockemer D (2017) What affects voter turnout? A review article/metaanalysis
of aggregate research. Gov Oppos 52(04):698–722. https://
doi.org/10.1017/gov.2016.30
Stockemer D, LaMontagne B, Scruggs L (2013) Bribes and ballots: the
impact of corruption on voter turnout in democracies. International
Political Science Review 34:74–90
1036 Int J Biometeorol (2018) 62:1027–1037
Studlar DT (2001) Canadian exceptionalism: explaining differences over
time in provincial and federal voter turnout. Canadian Journal of
Political Science 34:299–319
Verba S, Schlozman KL, Brady HE (1995) Voice and equality: civic
voluntarism in American politics. Harvard University Press
Xu J (2005) Why do minorities participate less? The effects of immigration,
education, and electoral process on Asian American voter registration
and turnout. Soc Sci Res 34(4):682–702. https://doi.org/10.
1016/j.ssresearch.2004.11.002
Int J Biometeorol (2018) 62:1027–1037 1037