The data used in this study originate from a survey conducted in 2019 among Finnish citizens residing abroad. The goal was to study the implications of external voting in the most recent parliamentary elections held on 14 April 2019. The survey approached a disproportionate stratified random sampleFootnote 1 of 10,000 Finnish emigrants entitled to vote. The sample was drawn from the Population Register Center of Finland. Selected individuals were invited to participate via a letter sent by post to their registered physical address. The invitation included a unique code and instructions on how to access the online survey via their own devices. The sample included residents from the 17 largest diasporas (countries with more than 1000 citizens with the right to vote in Finnish elections). In each country, the survey approached 500 persons; the exception was Sweden (i.e., an exceptionally large Finnish diaspora), where the sample consisted of 1500 persons. An additional 500 Finnish non-resident citizens were invited from the rest of the world (i.e., residing in any other country than one of the 17 largest diasporas).Footnote 2
The responses were collected with an online survey questionnaire which covered a wide range of questions related to political attitudes and electoral participation. The questionnaire was available in three languages: in the two official languages of Finland—Finnish and Swedish—as well as in English. The data collection took place between 23 May 2019 and 30 September 2019, starting more than a month after the parliamentary election. Only ten days after we dispatched the invitation letters, the new Finnish government presented its program (see Arter, 2020). Therefore, we do not expect that the development in Finnish politics could have any strong influence over respondents’ answers in the survey during the duration of data collection.
Four months may sound like a relatively long data collection period, especially when compared to the surveys collecting data within a single country. However, it is due to logistical reasons: We needed to allow a sufficiently long time for the invitation letter to reach selected participants (which might take up to eight weeks). Then, we needed to wait until all respondents who wanted to participate could find a convenient moment to fill in the questionnaire. Therefore, we monitored the response rate and closed the survey at the end of the month when no additional responses arrived for two consecutive weeks.Footnote 3
The effective response rate was slightly over 20%. Although the response rate may seem rather low in comparison to similar surveys collected among resident citizens, it is largely in line with other surveys collected among citizens abroad. In two previous larger data sets collected from non-resident citizens, the response rate was 20–30% (Peltoniemi, 2018; University of Gothenburg, 2016). It should be mentioned that any kind of representative random sample of non-resident citizens is rare (Ahmadov & Sasse, 2016). The fact that our survey was able to reach Finns living abroad depending on their current country of residence was possible only thanks to the Finnish population register system, which also includes non-resident citizens.
Subset and Variables
Altogether 2101 respondents participated in our survey, of which 983 indicated a voting method employed in the 2019 Finnish parliamentary elections. Of the respondents, 928 declared that they did not vote. Another 188 respondents skipped the questions, and two participants answered that they voted but did not specify their method of electoral participation. However, due to the missing values of some control variables which enter analysis as potential theoretically relevant confounders, the final number of suitable cases for the analysis decreased from 983 (all respondents who specified their method of electoral participation) to 664 individuals (see Fig. 2).Footnote 4
Beginning with the 2019 national elections, the Finnish electoral system offered voters residing in a foreign country four options to cast their electoral ballot. Two methods allow voting while staying abroad: (1) voting via post and (2) advanced voting abroad at one of the Finnish representative offices (e.g., embassy, consulate). The two remaining methods require voters to travel to Finland: (3) advance voting in Finland and (4) voting in Finland on election day. In the conducted survey, respondents who stated that they turned out in the 2019 elections were asked, “How did you vote / which method did you use to cast your vote?” and offered the above-mentioned set of four options. Based on their responses, we were able to construct a binary variable postal voter, which was used as a dependent variable in the binomial logistic regression models estimated for this research.
Overall, the survey includes 11% of respondents who reported using postal voting as their method of electoral participation (see Fig. 2). If we focus only on the subsample of voters (n = 985), early adopters of postal voting (given its recent implementation in 2019) already constitute a substantive group of voters among Finnish citizens residing abroad. More than every fifth respondent (21%) declared having delivered their ballot via post in the 2019 elections in Finland.
Our theoretical propositions focus on two independent variables: (1) trust in postal voting as a voting method and (2) distance from the polling station (for an overview of variable constructions used in the analysis, see Appendix A).
Trust in postal voting is measured by two survey items introduced with the question, “Do you agree or disagree with the following statements regarding postal voting in Finnish parliamentary elections?” (1) “Postal voting jeopardizes the secrecy of the ballot” and (2) “Postal voting enables electoral fraud.” Respondents were offered five options ranging from 1 = Agree strongly to 5 = Disagree strongly (see Appendix A for details). These two responses were combined into a single interval ranging from 0 = Complete distrust to 1 = Complete trust. Internal consistency of the measure is Cronbach's alpha = 0.85.Footnote 5
Distance from the polling station is measured in two ways: subjective and objective. Subjective (perception of the) distance to the nearest polling stations is based on a numerical response to a write-in question: “What is the distance from where you live to the nearest polling station where you can vote in Finnish elections (approximately, kilometers)?” Objective distance to the polling station represents the length of the route between the respondent’s address where the invitation to participate in the survey was delivered and the nearest polling station organized by Finnish administration abroad. These two measures do not have to be identical because individuals might respond with (a) an approximate distance, (b) distance from their work or other place of interest, or (c) distance from the temporary address if they visit Finland during the elections. In addition, (d) Google maps can sometimes approximate the respondent’s address (if, e.g., no street number or only zip code is known to its database). Despite these potential shortcomings, the Spearman's rank correlation between the two distance measures is high—ρ = 0.84 (for a visual depiction, see Figure A1 in the online appendix), mitigating reliability concerns related to either of the measures. Since the distribution of both distance measures is right-skewed towards shorter distances, the analysis uses their decimal logarithmic transformation.
Lastly, the analysis controls for a set of basic sociodemographic factors that could potentially confound the results. Control variables include gender and age (in years; calculated from birth year), the latter because older voters may be increasingly hesitant to adopt a newly implemented voting technique. In addition, the models control for education (1 = Still in school [comprehensive school, high school, vocational school, etc.], 2 = Elementary school [folke schoole, kansakoulu], 3 = Comprehensive school, 4 = Vocational school, 5 = Gymnasium or abitur, 6 = Polytechnic school, 7 = University, 8 = Licentiate or doctoral degree), because the more educated respondents may better understand the risks related to casting a ballot via post.Footnote 6 Also, Blais and Daoust (2020) demonstrate that explaining one’s motivation to turn out needs to take civic duty and political interest into account. Therefore, we included the civic duty as measured by the response to the statement “Finnish citizens residing abroad have a duty to vote in Finland’s elections,” on a five-point scale ranging from “Agree strongly” to “Disagree strongly.” Political interest was measured as “How interested would you say you personally are in [Politics in Finland]?” offering four responses from “Very interested” to “Not at all interested.” (For a detailed overview of all variables included in the analysis, see Appendix A.)
One can argue that we should also control for numerous factors acknowledged to influence turnout among individuals. First, as suggested by Dyck and Gimpel (2005), considerations about whether to turn out and which voting method to use constitute two relatively separate processes, even though they may be two subsequent decisions. Therefore, it could be expected that these two choices are driven by two separate sets of factors that may overlap, but only to a certain degree. Of course, the absence of a sufficiently convenient voting method may result in abstention. However, that should exceptionally apply to the 2019 Finnish elections because voters residing abroad had two options: postal voting and voting in person at a diplomatic mission. Second, the turnout literature has found that all the above-mentioned control variables (in some form) play a role in individuals’ decisions to turn out (Blais, 2006; Smets & van Ham, 2013). Therefore, even though our set of controls is specifically deduced to control potential confounders in the adoption of postal voting, it also controls for some relevant correlates of turnout.
Estimation Strategy
The point of departure for our analysis is the two models, 1 and 2, which solely examine the effects of the main variables of interest: distance and trust (see Table 1). The reason for two models is that model 1 includes objective and model 2 includes subjective distance measures. An analogous reason applies to all subsequent pairs of models (i.e., models 3 and 4, models 5 and 6, and models 7 and 8). Then, the specification includes an interaction term between trust and distance (models 3 and 4), which allows inspection of the effect of distance across voters at various levels of the moderator trust.
Table 1 Probability to cast a ballot via post—examination of distance and trust: Results of binomial logistic regression The same logic is followed in the second set of estimated models (models 5 to 8), which adds an above-discussed set of theoretically relevant control variables. The results are similar, which indicates that the results are robust towards the inclusion of potential confounders.
Due to the challenges related to the interpretation of coefficients in binomial logistic models and interaction terms (included in some specifications), the results are visualized and interpreted with respect to the figures. These figures are based on full model specifications controlling for potential confounders—models 5 to 8—depending on the hypothesis under consideration.