Intro

On Sunday, 25 October 2020, one year since massive social protests took the streets of Chile, a referendum to vote for the option to write a new constitution took place. This day was likely to be historical. 30 years had passed since the end of Pinochet’s dictatorship, but these 30 years of democracy had still been ruled by Pinochet’s constitution.

At the same time, the election marked the second occasion in which Chileans abroad were able to vote in their countries of residence - the first time was in 2017 for Presidential Elections. Nevertheless, this option to vote is limited to an in-person fashion, through the embassies or consulates, forcing citizens to travel to the nearest city with an electoral circumscription to exercise their right to vote.

In the context of the pandemic, there were several issues with guaranteeing that all people who wanted to vote were able to. All over the world, restrictions of movement were put back in place around that time, due to the rise of a second infection wave. In most countries only the capital city had an available polling place. Voting abroad requires a huge amount of effort and commitment for some citizens, and during the ongoing COVID-19 crisis, these costs are probably even higher. What are the main factors that influence the decision of voting, when the costs of doing so are high? How did the pandemic affect turnout in the election?

For this study, I will use publicly available data on the election results and turnout in each city outside of Chile, for both the 2020 plebiscite, and the 2017 Presidential elections. I will combine these with city level data on the COVID-19 incidence during the week of the plebiscite, and in addition, results of an individual survey on the motivations and costs of voting. This survey was spread with help of the consulates, social media groups, and social networks, reaching 648 completed surveys within seven days.


Summary stats

It is important to acknowledge the particularities of this sample. It does not represent the whole population of Chileans abroad, given the ways the survey was spread. Even though I used several mediums, I barely got responses from Chileans within South America, thus this sample is biased toward Chileans in North America and Europe. This is a highly educated sample, mostly female, and a fair mix of students and employees.


Summary statistics for the individual survey of Chileans abroad

This table presents means and standard deviations of demographic characteristics for the share of sample that voted (1), did not vote (2), was registered to vote with their current address (4), and was not registered (5). Columns (3) and (6) show the differences between those means; (1)-(2) and (4)-(5) respectively. *** 1%, ** 5%, * 10% significance.

Within the sample, respondents who vote do not significantly differ from the ones who did not, and the same happens for registered and non-registered voters. The only dimension that makes these groups significantly different is their political inclination. Of those who voted, less people declared identifying themselves with a right-wing political party, when compared to those who did not vote.


Number of surveys by country (limited to those with the highest share of responses)

Country Number of surveys Percentage of surveys Percentage of registered voters in 2017
United States 147 22.69 13.69
Spain 118 18.21 7.94
Germany 59 9.10 4.63
UK 49 7.56 2.88
France 38 5.86 3.53
Australia 36 5.56 5.14
Argentina 29 4.48 19.18
New Zealand 24 3.70 1.33
Canada 19 2.93 6.59
Italy 13 2.01 0.95

Individual survey

I start analyzing the data by running simple regressions of several variables on a dummy indicating whether the person voted. I include fixed effects by electoral circumscription for the within specifications, and cluster standard errors at the same level for all specifications. The share of sample that voted is 0.71. The estimates plotted below show the change in percentage points of the likelihood to vote.

Importantly, these estimates should be interpreted as mere correlations. For example, respondents are more likely to know other Chileans in their city if they live in their electoral circumscription (more populated city), and thus confront lower costs of voting.


Political engagement on the probability of voting

Across circumscriptions
protested <- lm_robust(voted ~ protested , data = survey)
politics_sm <- lm_robust(voted ~ politics_sm, data = survey)
politicallyinvolved<- lm_robust(voted ~ politicallyinvolved, data = survey)
politics <- plot_summs(protested, politicallyinvolved, politics_sm,
           coefs = c("Defines themself as politically engaged"="politicallyinvolved", "Uses social media to portray political views" = "politics_sm", "Has participated in a protest" = "protested"),
         point.shape=FALSE, robust = "HC1", inner_ci_level = .9, exp = FALSE, cluster=cl)
politics + theme(legend.position = "none") + labs(x = "\n Effect on the probability of voting \n ", y = NULL) 

Within circumscriptions
protested <- lm_robust(voted ~ protested , data = survey, fixed_effects = ~ fe)
politics_sm <- lm_robust(voted ~ politics_sm, data = survey, fixed_effects = ~ fe)
politicallyinvolved<- lm_robust(voted ~ politicallyinvolved, data = survey, fixed_effects = ~ fe)
politics <- plot_summs(protested, politicallyinvolved, politics_sm,
           coefs = c("Defines themself as politically engaged"="politicallyinvolved", "Uses social media to portray political views" = "politics_sm", "Has participated in a protest" = "protested"),
         point.shape=FALSE, robust = "HC1", inner_ci_level = .9, exp = FALSE, cluster=cl)
politics + theme(legend.position = "none") + labs(x = "\n Effect on the probability of voting \n ", y = NULL) 


Networks on the probability of voting

Across circumscriptions
livewithchileans<- lm_robust(voted ~ livewithchileans, data = survey)
knowchileans<- lm_robust(voted ~ knowchileans, data = survey)
mostfriendschilean<- lm_robust(voted ~ mostfriendschilean, data = survey)
speakspanish<- lm_robust(voted ~ speakspanish, data = survey)

networks <- plot_summs( livewithchileans, knowchileans, mostfriendschilean, speakspanish,
           coefs = c( "Lives with Chileans in the same household"="livewithchileans", "Knows other Chileans in the city"="knowchileans", "Most of their friends are Chilean"="mostfriendschilean", "Mainly speaks Spanish in the country of residence" ="speakspanish"),
         point.shape=FALSE, robust = "HC1", inner_ci_level = .9, exp = FALSE, cluster=cl)
networks + theme(legend.position = "none") + labs(x = "\n Effect on the probability of voting \n ", y = NULL) 

Within circumscriptions
livewithchileans<- lm_robust(voted ~ livewithchileans, data = survey, fixed_effects = ~ fe)
knowchileans<- lm_robust(voted ~ knowchileans, data = survey, fixed_effects = ~ fe)
mostfriendschilean<- lm_robust(voted ~ mostfriendschilean, data = survey, fixed_effects = ~ fe)
speakspanish<- lm_robust(voted ~ speakspanish, data = survey, fixed_effects = ~ fe)
networks <- plot_summs( livewithchileans, knowchileans, mostfriendschilean, speakspanish,
           coefs = c( "Lives with Chileans in the same household"="livewithchileans", "Knows other Chileans in the city"="knowchileans", "Most of their friends are Chilean"="mostfriendschilean", "Mainly speaks Spanish in the country of residence" ="speakspanish"),
         point.shape=FALSE, robust = "HC1", inner_ci_level = .9, exp = FALSE, cluster=cl)
networks + theme(legend.position = "none") + labs(x = "\n Effect on the probability of voting \n ", y = NULL) 

Across circumscriptions
noplanstoreturn<- lm_robust(voted ~ noplanstoreturn, data = survey)
morethan10yabroad<- lm_robust(voted ~ morethan10yabroad, data = survey)
followslocalnews<- lm_robust(voted ~ followslocalnews, data = survey)

networks <- plot_summs( noplanstoreturn, morethan10yabroad, followslocalnews,
           coefs = c( "Does not plan to return to Chile"="noplanstoreturn", "Has lived abroad for more than 10 years"="morethan10yabroad", "Follows local news more than Chilean news"="followslocalnews"),
         point.shape=FALSE, robust = "HC1", inner_ci_level = .9, exp = FALSE, cluster=cl)
networks + theme(legend.position = "none") + labs(x = "\n Effect on the probability of voting \n ", y = NULL) 

Within circumscriptions
noplanstoreturn<- lm_robust(voted ~ noplanstoreturn, data = survey, fixed_effects = ~ fe)
morethan10yabroad<- lm_robust(voted ~ morethan10yabroad, data = survey, fixed_effects = ~ fe)
followslocalnews<- lm_robust(voted ~ followslocalnews, data = survey, fixed_effects = ~ fe)

networks <- plot_summs( noplanstoreturn, morethan10yabroad, followslocalnews,
           coefs = c( "Does not plan to return to Chile"="noplanstoreturn", "Has lived abroad for more than 10 years"="morethan10yabroad", "Follows local news more than Chilean news"="followslocalnews"),
         point.shape=FALSE, robust = "HC1", inner_ci_level = .9, exp = FALSE, cluster=cl)
networks + theme(legend.position = "none") + labs(x = "\n Effect on the probability of voting \n ", y = NULL) 


Chile’s future and turnout

In a scale of 1 to 5, how much do you agree with the following statements?

Across circumscriptions
s5_define_chilefuture <- lm_robust(voted ~ s5_define_chilefuture_std , data = survey)
s5_importance_chilefuture <- lm_robust(voted ~ s5_importance_chilefuture_std, data = survey)

future <- plot_summs(s5_define_chilefuture, s5_importance_chilefuture, coefs = c("\"This plebiscite will define Chile's future\" (SD)" = "s5_define_chilefuture_std", "\"Chile's future is very important for me\" (SD)" = "s5_importance_chilefuture_std"),
         point.shape=FALSE, cluster=cl,
           robust = "HC1", inner_ci_level = .9)
future + theme(legend.position = "none")  + labs(x = "\n Effect on the probability of voting \n ", y = NULL)

Within circumscriptions
s5_define_chilefuture <- lm_robust(voted ~ s5_define_chilefuture_std , data = survey, fixed_effects = ~ fe)
s5_importance_chilefuture <- lm_robust(voted ~ s5_importance_chilefuture_std, data = survey, fixed_effects = ~ fe)

future <- plot_summs(s5_define_chilefuture, s5_importance_chilefuture, coefs = c("\"This plebiscite will define Chile's future\" (SD)" = "s5_define_chilefuture_std", "\"Chile's future is very important for me\" (SD)" = "s5_importance_chilefuture_std"),
         point.shape=FALSE, cluster=cl,
           robust = "HC1", inner_ci_level = .9)
future + theme(legend.position = "none")  + labs(x = "\n Effect on the probability of voting \n ", y = NULL)


Interest and information on turnout

Across circumscriptions
interest_std <- lm_robust(voted ~ interest_std , data = survey)
s5_plebiscite_informed_std <- lm_robust(voted ~ s5_plebiscite_informed_std, data = survey)

future <- plot_summs(interest_std, s5_plebiscite_informed_std, coefs = c("Interest on voting in the plebiscite (SD)" = "interest_std", "How informed they were about the plebiscite (SD)" = "s5_plebiscite_informed_std"),
         point.shape=FALSE, cluster=cl,
           robust = "HC1", inner_ci_level = .9)
future + theme(legend.position = "none")  + labs(x = "\n Effect on the probability of voting \n ", y = NULL)

Within circumscriptions
interest_std <- lm_robust(voted ~ interest_std , data = survey, fixed_effects = ~ fe)
s5_plebiscite_informed_std <- lm_robust(voted ~ s5_plebiscite_informed_std, data = survey, fixed_effects = ~ fe)

future <- plot_summs(interest_std, s5_plebiscite_informed_std, coefs = c("Interest on voting in the plebiscite (SD)" = "interest_std", "How informed they were about the plebiscite (SD)" = "s5_plebiscite_informed_std"),
         point.shape=FALSE, cluster=cl,
           robust = "HC1", inner_ci_level = .9)
future + theme(legend.position = "none")  + labs(x = "\n Effect on the probability of voting \n ", y = NULL)


Costs of voting and turnout
In a scale of 1 to 5, how relevant were these factors in your decision about whether to vote?

Across circumscriptions
s7_cost_COVID_movrestr <- lm_robust(voted ~ s7_cost_COVID_movrestr_std , data = survey)
s7_cost_distance <- lm_robust(voted ~ s7_cost_distance_std , data = survey)
s7_cost_money <- lm_robust(voted ~ s7_cost_money_std, data = survey)
s7_cost_COVID_fear <- lm_robust(voted ~ s7_cost_COVID_fear_std, data = survey)
s7_cost_lackinfo <- lm_robust(voted ~ s7_cost_lackinfo_std, data = survey)
s7_cost_electiondefined <- lm_robust(voted ~ s7_cost_electiondefined_std, data = survey)

costs <- plot_summs(s7_cost_COVID_movrestr, s7_cost_distance, s7_cost_money, s7_cost_COVID_fear, s7_cost_lackinfo, s7_cost_electiondefined,
                    coefs = c("Movement restrictions due to COVID-19 (SD)" = "s7_cost_COVID_movrestr_std", "Distance to the polling place (SD)" = "s7_cost_distance_std", "Monetary cost of getting to the polling place (SD)" = "s7_cost_money_std","Fear of getting infected with COVID-19 (SD)" = "s7_cost_COVID_fear_std", "Lack of information from the consulate (SD)" = "s7_cost_lackinfo_std","Idea that the election results were already defined (SD)" = "s7_cost_electiondefined_std"),
         point.shape=FALSE, cluster=cl,
           robust = "HC1", inner_ci_level = .9)
costs + theme(legend.position = "none")  + labs(x = "\n Effect on the probability of voting \n ", y = NULL)

Within circumscriptions
s7_cost_COVID_movrestr <- lm_robust(voted ~ s7_cost_COVID_movrestr_std , data = survey, fixed_effects = ~ fe)
s7_cost_distance <- lm_robust(voted ~ s7_cost_distance_std , data = survey, fixed_effects = ~ fe)
s7_cost_money <- lm_robust(voted ~ s7_cost_money_std, data = survey, fixed_effects = ~ fe)
s7_cost_COVID_fear <- lm_robust(voted ~ s7_cost_COVID_fear_std, data = survey, fixed_effects = ~ fe)
s7_cost_lackinfo <- lm_robust(voted ~ s7_cost_lackinfo_std, data = survey, fixed_effects = ~ fe)
s7_cost_electiondefined <- lm_robust(voted ~ s7_cost_electiondefined_std, data = survey, fixed_effects = ~ fe)

costs <- plot_summs(s7_cost_COVID_movrestr, s7_cost_distance, s7_cost_money, s7_cost_COVID_fear, s7_cost_lackinfo, s7_cost_electiondefined,
                    coefs = c("Movement restrictions due to COVID-19 (SD)" = "s7_cost_COVID_movrestr_std", "Distance to the polling place (SD)" = "s7_cost_distance_std", "Monetary cost of getting to the polling place (SD)" = "s7_cost_money_std","Fear of getting infected with COVID-19 (SD)" = "s7_cost_COVID_fear_std", "Lack of information from the consulate (SD)" = "s7_cost_lackinfo_std","Idea that the election results were already defined (SD)" = "s7_cost_electiondefined_std"),
         point.shape=FALSE, cluster=cl,
           robust = "HC1", inner_ci_level = .9)
costs + theme(legend.position = "none")  + labs(x = "\n Effect on the probability of voting \n ", y = NULL)


Pros of voting and turnout

In a scale of 1 to 5, how much did these factors influence your decision about whether to vote?

Across circumscriptions
s7_pro_changechile <- lm_robust(voted ~ s7_pro_changechile_std , data = survey)
s7_pro_endviolence <- lm_robust(voted ~ s7_pro_endviolence_std, data = survey)
s7_pro_civicduty <- lm_robust(voted ~ s7_pro_civicduty_std, data = survey)
s7_pro_historicday <- lm_robust(voted ~ s7_pro_historicday_std, data = survey)
s7_pro_networkschile <- lm_robust(voted ~ s7_pro_networkschile_std, data = survey)
s7_pro_networkscountry <- lm_robust(voted ~ s7_pro_networkscountry_std, data = survey)

pros <- plot_summs(s7_pro_changechile, s7_pro_endviolence, s7_pro_civicduty, s7_pro_historicday, s7_pro_networkschile, s7_pro_networkscountry, 
                    coefs = c("Desire of changing Chile (SD)" = "s7_pro_changechile_std", "End violence in Chile (SD)" = "s7_pro_endviolence_std", "Sense of civic duty (SD)" = "s7_pro_civicduty_std","Participate in a historical day (SD)" = "s7_pro_historicday_std", "Networks in Chile also voting (SD)" = "s7_pro_networkschile_std","Networks in country of residence also voting (SD)" = "s7_pro_networkscountry_std") , cluster=cl ,
         point.shape=FALSE, 
           robust = "HC1", inner_ci_level = .9)
pros + theme(legend.position = "none")  + labs(x = "\n Effect on the probability of voting \n ", y = NULL)

Within circumscriptions
s7_pro_changechile <- lm_robust(voted ~ s7_pro_changechile_std , data = survey, fixed_effects = ~ fe)
s7_pro_endviolence <- lm_robust(voted ~ s7_pro_endviolence_std, data = survey,  fixed_effects = ~ fe)
s7_pro_civicduty <- lm_robust(voted ~ s7_pro_civicduty_std, data = survey, fixed_effects = ~ fe)
s7_pro_historicday <- lm_robust(voted ~ s7_pro_historicday_std, data = survey, fixed_effects = ~ fe)
s7_pro_networkschile <- lm_robust(voted ~ s7_pro_networkschile_std, data = survey, fixed_effects = ~ fe)
s7_pro_networkscountry <- lm_robust(voted ~ s7_pro_networkscountry_std, data = survey, fixed_effects = ~ fe)

pros <- plot_summs(s7_pro_changechile, s7_pro_endviolence, s7_pro_civicduty, s7_pro_historicday, s7_pro_networkschile, s7_pro_networkscountry, 
                    coefs = c("Desire of changing Chile (SD)" = "s7_pro_changechile_std", "End violence in Chile (SD)" = "s7_pro_endviolence_std", "Sense of civic duty (SD)" = "s7_pro_civicduty_std","Participate in a historical day (SD)" = "s7_pro_historicday_std", "Networks in Chile also voting (SD)" = "s7_pro_networkschile_std","Networks in country of residence also voting (SD)" = "s7_pro_networkscountry_std") , cluster=cl ,
         point.shape=FALSE, 
           robust = "HC1", inner_ci_level = .9)
pros + theme(legend.position = "none")  + labs(x = "\n Effect on the probability of voting \n ", y = NULL)


Residing in a city different from the electoral circumscription on turnout

othercity <- lm_robust(voted ~ s3_othercity , data = survey , fixed_effects = ~ fe)
othercity1 <- lm_robust(voted ~ s3_othercity, data = survey)

livesothercity <- plot_summs(othercity, othercity1, coefs = c("Lives in another city" = "s3_othercity"), model.names = c("Within electoral circumscription", "Across electoral circumscription"), robust = "HC1", legend.title="", cluster=cl, inner_ci_level = .9)

livesothercity + labs(x = "\n Effect on the probability of voting \n ", y = NULL)


COVID-19 perceptions on turnout

Standardized values of scales from 1 to 5. A higher value implies a better situation.

Across circumscriptions
pandemic_relchile_std <- lm_robust(voted ~ pandemic_relchile_std , data = survey)
s10_pandemic_habits_std <- lm_robust(voted ~ s10_pandemic_habits_std, data = survey)
pandemic_country_std <- lm_robust(voted ~ pandemic_country_std, data = survey)

future <- plot_summs(pandemic_country_std, pandemic_relchile_std, s10_pandemic_habits_std, coefs = c("Current situation in your country with respect to the pandemic (SD)" = "pandemic_country_std", "Pandemic situation in your country relative to Chile (SD)"= "pandemic_relchile_std", "My daily routine has gone back to pre-pandemic normalcy (SD)" = "s10_pandemic_habits_std"),
         point.shape=FALSE, cluster=cl,
           robust = "HC1", inner_ci_level = .9)
future + theme(legend.position = "none")  + labs(x = "\n Effect on the probability of voting \n ", y = NULL)

Within circumscriptions
pandemic_relchile_std <- lm_robust(voted ~ pandemic_relchile_std , data = survey, fixed_effects = ~ fe)
s10_pandemic_habits_std <- lm_robust(voted ~ s10_pandemic_habits_std, data = survey, fixed_effects = ~ fe)
pandemic_country_std <- lm_robust(voted ~ pandemic_country_std, data = survey, fixed_effects = ~ fe)

future <- plot_summs(pandemic_country_std, pandemic_relchile_std, s10_pandemic_habits_std, coefs = c("Current situation in your country with respect to the pandemic (SD)" = "pandemic_country_std", "Pandemic situation in your country relative to Chile (SD)"= "pandemic_relchile_std", "My daily routine has gone back to pre-pandemic normalcy (SD)" = "s10_pandemic_habits_std"),
         point.shape=FALSE, cluster=cl,
           robust = "HC1", inner_ci_level = .9)
future + theme(legend.position = "none")  + labs(x = "\n Effect on the probability of voting \n ", y = NULL)