Getting out the vote in different electoral contexts: the effect of impersonal voter mobilization techniques in middle and high salience Norwegian elections

ABSTRACT

This article reports on two sets of field experiments testing text messaging (SMS) among native Norwegian voters and live phone calls among young voters eligible to vote in their first election. The two sets of experiments are done in both local and national elections. Most previous studies of election salience and voter mobilization use data from countries with generally lower levels of turnout than in Northern Europe (including Norway). This article addresses how the effectiveness of voter mobilization varies by election salience, in a generally high-salience political context. We find the two experiments to be more effective in low/middle salience elections, with turnout around 60 percent, than in high salience elections with turnout higher than 75 percent.

Introduction

Field experimental scholarship has tested if impersonal voter mobilization contacts such as SMS text messaging and live calls from volunteers increase turnout (Green, McGrath, and Aronow 2013; Green and Gerber 2019; Dale and Strauss 2009; Michelson, Garcia Bedolla, and McConnel 2009; Nickerson 2007; Bergh, Christensen, and Matland 2019). Treatment effects are usually measured in low salience elections or so-called second-order elections. These elections, such as local elections, are characterized as second-order simply because they matter less than national (first-order) elections (Reif and Schmitt 1980). Turnout is lower in second-order elections, when compared to national elections in the same country, but turnout also varies substantially between countries. The level of voter engagement in a low-salience election in one country might be as high as in a high-salience election in another. We argue that GOTV efforts will be most effective in “middle salience” elections – elections that garner substantial interest in the electorate, but where turnout is low enough that there is an ample supply of voters to mobilize. This would fit the profile of local elections in Northern Europe. High salience national elections in this part of the world reach levels of turnout that substantially limit the number of potential voters that can be mobilized. Most voters that are contacted will have the intention to vote anyway, which would limit the effectiveness of GOTV campaigns.

We have limited insight into how impersonal GOTV contacts increase turnout in high salience elections. Fieldhouse et al. (2013) conduct a mail and a telephone experiment in a European Parliament election and a general election in the UK. They find larger effect sizes in the high salience national election. However, turnout in the two elections in their study was 34 and 65 percent, respectively. The latter corresponds approximately to turnout in Norwegian local elections. In a sophisticated re-analysis of 11 face-to-face mobilization experiments, Arceneaux and Nickerson (2009) analyze the complex relationship between a voter’s propensity to vote, the saliency of the election and GOTV treatment effects. An important takeaway from that study is that we should not look for single-factor explanations for the success or failure of mobilization efforts. Arceneaux and Nickerson (2009) find that voters at the “cusp” of voting are most likely to be mobilized. If your goal is to get low propensity voters to the polls, you are most likely to succeed in high turnout elections. These 11 experiments were also conducted in contexts with generally lower levels of turnout than in Norway.

We have previously found (Bergh, Christensen, and Matland 2019; 2020) that voter mobilization in Norwegian local elections – relatively speaking, a low-salience election – is most effective among low-propensity voters. This suggests that the dynamics of election salience may work differently in a Northern European context than in the contexts of these previous studies.

It may be that this type of “middle salience” election is a good context for mobilizing voters. The higher salience national elections in Norway may reach a ceiling effect where it becomes more difficult to mobilize additional voters.

This article reports the results of impersonal GOTV contacts in Norway. We present the results of two field experiments conducted in the 2017 parliamentary (Storting) election. The first replicates an SMS text messages voters’ study from the 2015 municipal elections among native Norwegian voters finding strong mobilization effects especially among young voters (Bergh, Christensen, and Matland 2019). The experiment found text messages to produce strong effects. The “intent to treat” (ITT) effect among native Norwegians below the age of 30 was 4.6 percentage points (with a baseline turnout of 45.3 percent). The text messages had an ITT effect of 1.0 percentage points among native Norwegian voters over the age of 30.¹ These effects are substantially stronger than the effects found in the United States, and stronger than the effects found in the one previous European study of text messages in the 2013 Danish local elections (Bhatti et al. 2017). Thus the 2015 Norwegian study found that younger voters less plugged into the political process and vote less frequently, responded mostly to the text messages. These results indicate text messages can be an effective (and economical) tool for mobilizing voters, especially those who tend to participate less than the average Norwegian voter does. Even if turnout is relatively high in Norwegian local elections (60.2 percent in 2015), there were fewer voters to mobilize in the 2017 parliamentary election. Thus we anticipate a more modest effect in 2017 compared to 2015.

The second experiment use live phone calls to young voters eligible to vote in their first election. This was not done in the 2015 election, so to have comparative data on a low/middle salience election, we replicated the 2017-experiment in the 2019 local elections. To our knowledge, this study is the first to test the effect of phone calls outside of the USA and the UK (c.f. Fieldhouse et al. 2013; John and Brannan 2008). The experiment was both designed and conducted by the Norwegian Children and Youth Council (LNU). The LNU had young volunteers calling randomly assigned young voters encouraging them to take part in the election. Thus the phone calls involved a personal contact not provided by the SMS messages.

Turnout in the 2017 parliamentary election was 78.2 percent, 18 percentage points higher than the 2015 municipal elections, and 14 percentage points higher than the municipal election in 2019. Thus we test the effect of impersonal voter mobilization efforts in an electoral context with a high number of habitual voters. Since almost 8 out of 10 voters participated in the 2017-election, additional voters should be harder to mobilize.

The study finds that the SMS experiment modestly increased turnout among native Norwegian voters by 0.42 percentage points in 2017. The more “personal” volunteer live phone call experiment barely and insignificantly increased turnout among young voters eligible to vote in their first election in 2017: the complier average causal effect (CACE) was 1.1 percentage points. The phone experiment was a bit more effective in 2019: CACE of 3.1. Thus our findings suggest that the efficiency of these voter mobilization contacts depends on the type and salience of the election. These tools of voter mobilization are more promising in what we have dubbed “middle salience elections”.

The next section covers previous research and our expectations. The second section describes the data and design of the two field experiments. In the third section, we present the results. The article concludes by discussing our findings and possible directions for future research.

Previous research and expectations

Given the large number of previous studies of GOTV experiments, we now have a fairly sophisticated understanding of the effectiveness of various modes of mobilizing voters. Though some studies have also looked at the meaning and importance of context, we know less about what works when, and in different types of elections. In this section, we briefly review previous research on the relevant modes of voter mobilization: SMS and telephone calls. We then look at what previous research tells us about variation in GOTV effectiveness between various types of elections and political contexts.

Not all GOTV methods are equally effective. Some existing research find that face-to-face mobilization techniques get more voters to the polls than impersonal contact tactics such as emails, direct mail, telephone calls and SMS messages (Arceneaux and Nickerson 2009; Michelson and Nickerson 2011; Matland and Murray 2012; Garcia Bedolla and Michelson 2012; Green, McGrath, and Aronow 2013; Green and Gerber 2019). On the other hand, there is evidence that impersonal tools (mailings), matched with social pressure can be effective (Gerber, Green, and Larimer 2008; Bergh, Christensen, and Matland 2020). While text messages may not match canvassing in terms of increasing turnout, if one calculates effectiveness as costs per extra vote, text messages and volunteer phone calls can be highly effective (Green and Gerber 2019).

The Noticeable Reminder Theory holds that a simple nudge in the form of an SMS text message is enough to mobilize voters. In proposing the theory, Dale and Strauss (2009) emphasize that voters in their U.S. study have already shown themselves interested in voting by registering to vote and agreeing to receive an SMS. Hence, they do not need to be convinced to vote, they simply need a reminder (see also Thaler and Sunstein 2009). The Norwegian context differs. In Norway, all voters are automatically registered to vote, there is no need to physically register, and as such the Norwegian case represents a tougher test for the theory. Dale and Strauss (2009) collected information from a sample of young people and people who recently had moved. These new registrants provided their mobile phone numbers and agreed to get a text message reminding them to vote. Dale and Strauss found that SMS reminders produced a statistically significant positive “intent-to-treat” effect of 3.0 percentage points. In the second U.S. test, Malhotra et al. (2011) ran two tests, during different elections, contacting a sample of voters in a single California county with cold calls (no previous contact). They find modest, but statistically significant effects, a 0.7 percentage point increase in the local election and a 0.9 percentage point increase for the state-wide election.

In addition to the Norwegian 2015 local election SMS experiment, a team of Danish researchers has done the only other study to test text messages outside the United States (Bhatti et al. 2014: 2017). They ran three distinct text messaging campaigns during the 2013 local elections and a fourth experiment during the 2014 EU parliamentary elections. Three of the text messaging campaigns targeted young ethnic Danish voters below 30 years of age. A fourth study was conducted on a broad sample of Danish residents of all ages with an oversampling of immigrants. The Danish team tested a series of distinct messages and found no difference in effectiveness across message content. The three campaigns aimed at young voters produced a statistically insignificant 0.72 percentage point increase in turnout, a statistically significant effect of 0.63 percentage points with a larger sample and finally a quite strong turnout impact of 1.8 percentage point increase in turnout. The fourth SMS experiment resulted in an insignificant 0.33 percentage point increase in turnout, although turnout among first-generation immigrants increased by a significant 1.0 percentage points.

The 2015 Norwegian study found text messages to be more effective than in neighboring Denmark (see above and Bergh, Christensen, and Matland 2019). Turnout in the Danish local election (used for three of the experiments²) was 12 percentage points higher than in the 2015 Norwegian local elections. There were in other words more potential voters to mobilize in Norway. The design of the experiments also differed when it comes to message content and the population pulled for the experiments (see Bergh, Christensen, and Matland 2019).

Turning to telephone as a mode of mobilization, such GOTV contacts fall into three categories (Green, McGrath, and Aronow 2013): Pre-recorded phone calls, live calls from commercial phone banks, and live calls from volunteer phone banks. Phone experiments are generally subject to noncompliance since citizens often fail to pick up the phone (Nickerson 2006; Blackwell 2017). Therefore, phone experiment effects are usually measured among those actually contacted, the so-called Complier Average Causal Effect (CACE). The design of our experiment falls in the category of live calls from volunteer organizations, and according to previous field experiments, these calls should be the most effective. As pointed out by Green, McGrath, and Aronow (2013, 33) “live interactions with human beings seem to vary in effectiveness depending on whether GOTV messages are delivered in a routinized way by a commercial phone bank or in a more authentic manner by a volunteer phone bank”. The calls are personal, and subjects are talking to a live caller with an overreaching goal to make personal contact with the voter (Nickerson 2006). A meta-analysis of 32 volunteer live calls shows an average CACE of 2.8 percentage points (Green and Gerber 2019: Appendix C).

These types of averages hide a lot of variation between individual experiments that have been fielded in different elections. Some studies have looked at that variation, in relation to the context of each election. Malhotra et al. (2011, 667) argue that the effect of text messaging depends on a combination of the salience of the election and individuals’ voting history. They find text messaging increases turnout only among habitual voters in the lowest salience elections, and only among casual voters in a more salient election. Voters with low levels of participation were not mobilized. Fieldhouse et. al. (2013) test two types of impersonal mobilization techniques, mail and telephone, in two types of elections: a low salience European Parliament Election and a high salience UK general election. They find that both techniques, but especially telephone, are more effective in the high salience election. Finally, in a much-cited re-analysis of face-to-face field experiments, Arceneaux and Nickerson (2009), analyze how election salience interacts with voters’ propensity to vote. In high salience elections, most voters will not need to be mobilized by a canvasser. However, low-propensity voters are at the cusp of voting in such an election and may therefore benefit from a GOTV contact. In low salience elections with generally low levels of turnout, most voters will not vote irrespective of their contact with a canvasser. High propensity voters are most likely to be affected by a GOTV contact in those elections.

The 11 experiments included in Arceneaux and Nickerson’s (2009) study are all from the USA, in elections that have generally lower levels of turnout than Norway or other Northern European countries have. Some of our previous studies from Norwegian elections (Bergh, Christensen, and Matland 2019, 2020) suggest that a different dynamic is at work than in Arceneaux and Nickerson’s (2009) results. Our GOTV experiments were most effective among low propensity voters in low salience elections. In the Norwegian context, “low salience” means local elections with turnout levels above 60 percent. Another potential reason for the difference in findings is the mode of mobilization: face-to-face in the case of Arceneaux and Nickerson (2009), and text messages and letters in the mail in Bergh, Christensen, and Matland (2019, 2020). Another general take away from Arceneaux and Nickerson (2009), which is nevertheless relevant, is that what works on who in what election is complex, and varies with the saliency of the election and a voter’s propensity to vote. Additionally, it may depend on interactions between the sender of the message, message content, delivery method, type of election and the population targeted.

In the present study, we test if the effect of text messaging on similar groups (native Norwegian voters below and above the age of 30) varies between elections. Even if Norwegian local elections are often referred to as “second-order elections” (Mjelde et al. 2016) and receive less attention and interest compared to parliamentary elections, turnout in these elections is still high³, and vastly higher than the elections where U.S. studies of mobilization techniques are fielded.⁴ Our assumption is that the Norwegian local election, with turnout approximating that of high salience elections in the UK for instance (c.f. Filedhouse et al 2013), could be seen as a middle salience election, ideal for GOTV mobilization. There is widespread interest in and knowledge about the election in the general public, but there are also a substantial number of prospective non-voters that could be mobilized. In the high salience Norwegian parliamentary election, turnout is so high (78 percent) that there may not be much additional mobilization potential in the electorate. We expect smaller effects in the 2017 parliamentary election.

Experimental setting, data, design, and method

To design the experiments in 2017, we got access to the electoral rolls for approximately 3 million Norwegian voters living in 249 municipalities that have adopted electronic registration of turnout. Because practically all of Norway’s larger towns and cities have electronic registration of turnout, our dataset includes a majority of Norway’s eligible voters: 75 percent. The National Population Register [Folkeregisteret] provided information with respect to birthdate, sex, country of origin, and citizenship. We contracted with an IT company to provide cell phone numbers tied to the individuals in the voting registry. Of the approximately 3 million voters in the registry, the IT firm came up with just over 2.5 million cell phone number matches (82 percent).⁵

We began with just over 2.3 million native Norwegians who were eligible to vote, and where we had cell phone numbers. The samples for both experiments were pulled from these data. For each group of voters, we randomly assigned individuals to the control or treatment groups. Table 1 displays the composition of the control and treatment groups. 23,305 native Norwegians under the age of 30 received a text message with 344,440 in the control group. 116,214 native Norwegians over the age of 30 received a text message with 1,770,871 in the control group. The design of the phone call experiment in 2017 targeted exclusively native Norwegian young voters eligible to vote in their first election, and this group of voters was therefore excluded from the SMS experiment. 32,311 youngsters were pulled to receive a call from the Norwegian Children and Youth Council (LNU) while the remaining 54,780 first time young voters were assigned to the control group.

Table 1. SMS text campaign/telephone campaign: control and treatment groups.

SMS experiment in 2017	Treatment	Control
Below the age of 30	23,305	344,440
Over the age of 30	116,214	1,770,870
Total	139,519	2,115,310
Telephone experiment
Young voters eligible to vote for the first time in 2017	32,311	54,780
Young voters eligible to vote for the first time in 2019	26,081	88,273

For the telephone experiment in 2019, we received a data file of so-called first-time voters, meaning those aged from 18 through 21. The file included 233,940 voters, of whom 114,354 had a registered phone number. We randomly pulled 26,081 individuals to receive treatment; the remaining 88,273 constitute the control group.

Because we wanted to compare the results from the SMS experiment with that of the 2015 municipal elections we used similarly phrased text messages (Bergh, Christensen, and Matland 2019). The text messages included a reminder of the upcoming election, with a short civic duty appeal (see below). Message content was developed in agreement with the Ministry of Local Government and Modernization (KMD). KMD has overall responsibility for all elections in Norway and we used their election webpage address valg.no (election.no) as the sender of messages. To test if the sender of the messages had any impact on the results, we randomly assigned 9398 voters in Oslo (the capital) to receive identical messages, but with the municipality of Oslo [Oslo Kommune] as the sender. 10,147 voters in Oslo received identical messages with valg.no as the sender of the message.

The live volunteer phone call experiment was conducted (and designed) by the Norwegian Children and Youth Council (LNU). LNU is an umbrella organization of approximately 100 Norwegian children and youth organizations, and primarily concerned with questions regarding youth participation. LNU ran a campaign called “Young Voices” during the election campaign trying to mobilize young voters to take part in the election. To conduct the experiment LNU recruited young voters to deliver the treatment to voters of the same age and marked them as compliers or not. They ran the experiment in the same way in 2019 as in 2017.

The design of the two experiments captures contact in a real-world setting. The receivers of the SMS and the phone calls had not given prior consent to receiving them. This strengthens the external validity of our experiments since our results cannot be explained as unique to recipients who previously agreed to receive the treatments.

One advantage of controlled experiments is that random assignment facilitates the simplicity of analysis (Angrist and Pischke 2008). The intent-to-treat effect (ITT) is a conservative standard for analyzing results from clinical trials; it evaluates subjects as they were randomized regardless of whether they received/completed the treatment or not (Gerber and Green 2012). ITT analysis maintains the benefits of randomization and provides an accurate measure of how effective a treatment is given realistic conditions where coverage is less than complete. The ITT is simply obtained by subtracting the turnout rate of the treatment group from that of the control group. Due to noncompliance additional analysis is, however, usually necessary. The Complier Average Causal Effect (CACE) is relatively liberal and based only on those who actually received the treatment. The contact rate in the text messaging experiment was as high as 88.7 percent, and in the phone call experiment it was 23.6 percent. We report ITT effects based on linear probability models (OLS) while the CACE effects is calculated by performing a two-stage least squares regression of vote on actual contact using selection into the experimental group as an instrumental variable (Gerber and Green 2012). The latter is equivalent to dividing the ITT effect by the contact rate (see online appendix for the full regression results).

Experimental results in middle salience elections

The first round of Norwegian text message experiments was done in the 2015 local elections. We report the results in Bergh, Christensen, and Matland (2019), using two samples of native Norwegian voters, divided by age. Native Norwegians below the age of 30 had a control group turnout of 45.3 percent (see Table 2). Text messaging has a remarkably strong effect in this group: 4.6 percentage points which decreases to 3.3 percentage points when controls are added. The CACE effect is estimated as 6.2 percentage points. Young Norwegians are mobilized by text messages. On the other hand, native Norwegians 30 and above show a positive but only modest effect. The already high turnout in this group rises with a statistically significant .96 percentage points in the treatment group. Adjusting for the contact rate, we find a modest CACE effect of about 1.4 percentage points.

Table 2. Experimental results in 2015: base turnout and change in turnout.

	Control group turnout	Intention to Treat (ITT)		ITT with control variables		Complier Average Causal Effect (CACE)		N
		Effect	s.e.	Effect	s.e.	Effect	s.e.
Below 30	45.33	4.58	.48	3.31	.47	6.16	.64	66,086
30 or more	72.66	.96	.20	.38	.19	1.36	.27	389,107

*The following social background control variables are included: sex, age, and married.

There was no phone call experiment in 2015; the first GOTV effort by phone was done in the 2017 election. To be able to compare those results with those in a middle saliency election we replicated the study in the local election in 2019. We start by presenting those results here, before getting to the comparison: GOTV via SMS and telephone in the 2017 (high saliency) parliamentary election.

The Norwegian Children and Youth Council (LNU) conducted the experiment in the last 2 weeks before the election. The volunteer callers where themselves young voters and they attempted to contact 26,081 first-time voters, in the age range of 18 through 21. They completed 5398 phone conversations. Of all the attempted calls, 79 percent was not successful, either because of no reply or because of rejection.

The callers had a script but were encouraged to try to make a personal connection with the voters. The callers also had a list of answers to possible questions from the subjects. The treatment (script) was patterned after “social pressure” messages (2 out of 3 first-time voters use the right to vote) (Gerber, Green, and Larimer 2008), “voting plan” messages (When and where do you intend to vote?) (Nickerson and Rogers 2010), and “gratitude” (thank you for voting) messages (Panagopoulos 2011) in the voter mobilization literature. The script had the following content:

Hi! Do I speak to [name]?

 

My name is [full name] and Ím calling from the campaign Young Voices. Im calling because this year is the first time you and I are eligible to vote in the municipal elections. 2 out of 3 first-time voters use the right to vote, so I wonder:

 

Question 1:

Do you intend to vote?

If yes:

Great! When and where do you intend to vote? (can I help you find your nearest polling station)

If no:

Can I ask why not? (Reasons not to vote)

Question 2:

Are you wondering about something concerning the election I can help with?

If yes:

Practical help concerning the election/reasons to vote

If no:

Great! If you wonder about something, you can check out the webpage valg.no

Thank you for the talk and thank you very much for using your voting rights.

 

The results of the 2019 phone experiment are in Table 3. It is fair to say that the effects sizes are small. The low ITT effect to some extent reflects the low contact rate, which is common in these types of experiments. The effect on those who were actually contacted (CACE) is somewhat more substantial. The effects are statistically significant in a one-tailed test. In sum, the phone experiment had an effect on turnout among young voters in 2019, but it is certainly a weaker effect than that of text messages. Given the vast differences in costs between setting up a phone bank and sending out text messages, the latter seems like the preferable option in these middle saliency elections. The cost of the text messages used in this study was 0.70 kroner (7.0 US cents) per message, compared to 1.05 kroner (10.5 US cents) per call.⁶

Table 3. Experimental outcomes for the 2019 phone call study (N = 114,354).

	Turnout (%)	ITT (s.e)	ITT with controls* (s.e.)	CACE (s.e)
Voted Control group	54.2
Voted Treatment group	54.8	.63 (.35)	.66 (35)	3.14 (1.68)

*The following social background control variables are included: gender and age.

We then move on to the experiments conducted in a different setting, albeit in the same country: the 2017 national parliamentary election.

Experiment 1: SMS text messages to native Norwegian voters

The text messages experiment was conducted in the last week up to and including Election Day (From Monday September 7 to Monday September 14). The number of text messages sent ranged from a minimum of 11,318 messages to a maximum of 24,980. These went out at 7 pm each night except Saturday leading up to the election, summing to 136,041 messages (this includes messages to a sample of immigrants). The second round of messages was sent on Election Day, starting at 8 am and then every hour until 6 pm: 51,245 messages were sent on Election Day (also including immigrants). The text messages were sent by a polling agency with the technical equipment to send large numbers of text messages simultaneously. Furthermore, the firm kept track as to which messages were delivered. That enables us to calculate precise treatment effects.

The text messages are listed below. SMS#1 was used for messages sent the week before the election. SMS#2 was used for messages sent the day of the election. The SMS conveyed the following content:

SMS#1: Hi! This is a friendly reminder concerning the parliamentary election on September 14. Democracy needs your voice/vote⁷, so remember to participate in the election! Regards valg.no. [Regards Oslo municipality/valg.no]

SMS#2: Hi! Have you voted? If not, you can still make it. Polling stations are open from XX to YY today. Participate in the election! Regards valg.no [Regards Oslo municipality/valg.no]

Our main interest is to test if SMS text message increase turnout in both middle- and high salience elections. If the effect is ineffective in high salience elections this has implications for how turnout can be maximized in different types of elections. To do this we compare the effect on similar groups of voters eligible to vote in both elections: native Norwegian voters under the age of 30 and native Norwegian voters above the age of 30.

The results of the SMS experiment are given in Table 4. The first column shows the control group turnout, the second the turnout in the treatment group, and the third shows the ITT effect (see online Appendix A for bivariate regressions). We also ran regression models with control variables shown in column 4 (see online Appendix B). The fifth column shows the Complier Causal Effect (CACE) measuring the effect only on those who received the treatment.

Table 4. Experimental outcomes for the text messaging study.

	Voted Control group (%)	Voted Treatment group (%)	ITT (s.e)	ITT with control variables	CACE (s.e)	N
Whole sample	82.56	82.92	0.36(.10)	0.04(0.11)	0.41(.12)	2,254,829
Age 30 or more	84.88	85.32	0.44(.11)	−0.14(0.11)	0.49(.12)	1,887,084
Below age 30	70.67	71.02	0.35(.31)	0.43(0.31)	0.39(.35)	367,745

The following social background control variables are included: male, age, age-squared, and household size.

We begin with the whole sample. The already high turnout among native Norwegians (82.6 percent in the control group) rises with a statistically significant 0.36 percentage points in the treatment group with a standard error of 0.1. Among Norwegians aged 30 and above turnout increased by 0.44 percentage points (baseline turnout was 84.9 percent). Turning to native Norwegians below the age of 30, this group had a control group turnout of only 45.3 percent in the 2015 municipal election (Bergh, Christensen, and Matland 2019), but in the 2017 parliamentary election turnout in the control group was as high as 70.7 percent. Thus turnout among young voters below the age 30 was 25.4 percentage points higher in the 2017 election compared to the second-order local elections in 2015. In 2017 text messaging insignificantly increased turnout by 0.35 percentage points among young voters (standard error 0.31). Hence, it seems like text messaging mobilize young Norwegians in middle salience elections, but not in high-salience national elections. However, using standard errors (or P values) to distinguish between “an effect” and “no effect” among these groups of voters is too simplistic. The sample of young voters is much smaller compared to older native Norwegian voters, and consequently the standard errors are much larger. The confidence interval surrounding the effects size for the whole sample [95% CI: 0.16–0.57 percentage points] also include the effect size among voters under the age of 30 [0.35] indicating that the effect size for the two groups of native Norwegians may be indistinguishable. Adding our control variables returns insignificant ITT effect in all three groups. Adjusting for the contact rate, we find modest CACE effects ranging from 0.39 percentage points (under 30 years of age) to 0.49 (aged 30 or more).

Overall, the core finding of our 2017 text messaging experiment is that the effect of text message is modest in a high salience election such as the 2017 Norwegian parliamentary election. With a baseline turnout as a high as around 80 percent SMS messages produce a slight increase in turnout.

Experiment II: phone calls to voters eligible to vote in their first election

The phone call experiment was conducted by the LNU in the last 2 weeks up to the election (From Monday August 28 to Sunday September 11). The 35 callers matched the population targeted in the sense that all were young voters eligible to vote in their first election. The callers were recruited from one of the different organizations under the LNU umbrella, and many of them were active in key positions in their respective organizations. The callers managed to randomly contact 32,469 young voters of the 40,000 pulled to receive a call. After deleting duplicate cell phone numbers from the file from the LNU we are left with 32,311 unique young voters in the treatment group. The remaining sample of young first-time voters with a registered cell phone number was allocated to the control group (N = 54,780), and no calls were directed to this group.

The callers used the same script and approach as in the 2019 study (see above), and registered whether the call was received and how long the listener spent on the line. The conversations with those successfully contacted lasted roughly 1.5 min on average (SD = 1.1 min). 73 percent did not answer the phone, 3 percent of the conversation was not completed while a successful completion of the script was achieved with 23.6 percent of the youngsters.

Table 5 reports voter turnout rates for both the treatment and the control group (see online Appendix C for bivariate and multivariate regressions). The table reports effects both for subjects randomly assigned to receive the call (the ITT), the ITT with control variables and the effect for those successfully contacted (the CACE). The turnout rates are 69.5 percent in the control group and 69.7 percent in the treatment group. Turnout in the treatment group is estimated to be 0.25 percentage points higher than the control group (s.e = 0.32). Thus the difference is indistinguishable from zero. The experiment produced an insignificant CACE of 1 percentage point (standard error 1.37). Hence, the volunteer phone calls did not increase turnout among voters eligible to vote in their first election in 2017.

Table 5. Experimental outcomes for the phone call study (N = 87,091).

	Turnout (%)	ITT (s.e)	ITT with controls	CACE (s.e)
Voted Control group	69.48	−		−
Voted Treatment group	69.73	0.25(.32)	0.20(0.31)	1.05(1.37)

The following social background control variables are included: male, age, and household size.

Conclusion

The message from the existing literature is that impersonal mobilization tactics can produce positive results (Green and Gerber 2019). The data and analyses in Bergh, Christensen, and Matland (2019) based on an SMS text message experiment conducted in the 2015 Norwegian local elections confirm that suggestion. The data and analyses presented in this paper based on a replication of the 2015 study in the 2017 Norwegian parliamentary election show that the mobilization effects uncovered in 2015 was for the most part dependent on the salience of the election. The effect of text messaging in the 2017 parliamentary election among similar groups as in the 2015 local election is modest. Thus it is harder to mobilize voters in an election with many habitual voters and few casual voters. Still, even with a baseline turnout of around 85 percent text messaging increased turnout among voter aged 30 or more by 0.44 percentage points.

Compared to the simple nudge delivered by the SMS text campaign the design of the volunteer live phone call experiment targeting young first voters involved a personal contact. Not only did voters get to talk to a live person, but the callers were also recruited to perfectly match the targeted population. Young voters called other young voters trying to make a personal connection with the person on the other end of the line and encouraging them to take part in the election. Still, the calls increased turnout insignificantly by a modest 0.25 percentage points (CACE 1 percentage point) from a baseline turnout rate at 69.5 percent.

The contact rate for phone calls is much lower than for SMS text messages, especially among the young (see also Nickerson 2006). It is also worth pondering what people consider to be more “personal”. Perhaps a text message, which is a very common form of interaction between friends and family, feels as much as a personal direct interaction as a phone call.

The core finding in this article is that the effects of impersonal GOTV contacts such as the simple nudge involved in text messaging and live phone calls depend on the context of the election. Every election in every country has a unique story; the parties, candidates, political issues and saliency vary from one election to the next. It is difficult, if not impossible, to test which of these varying factors explain variation in GOTV effectiveness. Previous GOTV studies have focused on election saliency, with varying results (Arceneaux and Nickerson 2009; Fieldhouse et. al. 2013; Malhotra et al. 2011). Our findings suggest that a middle salience election, one with turnout around 60 percent, may be the best context in which to mobilize voters. Really high salience, high turnout elections, such as general elections in Northern Europe that reach turnout levels close to 80 percent, do not seem to be conducive to voter mobilization. In those cases, one probably gets close to a ceiling of possible mobilizable voters, making it increasingly difficult to get that extra person out to vote.

While our study proposes that the saliency of an election is important for the effectiveness of GOTV-efforts, there are of course limitations to the analysis. First, the context of an election is more than just saliency. The issues and level of mobilization may also play a role. For example, the 2019 local elections in Norway saw a surprising rise in turnout among young voters, for whom climate change and the environment were a primary concern. This political mobilization among the young is a plausible, but not definitive, explanation for the relatively weak effects of the phone bank experiment in that election. Second, we analyze a very limited set of experiments. Who is mobilized with what and in which election is presumably more complex than simply comparing the effect size with the turnout rate. The effect may depend on who is sending the message, the content of the message, the method of delivery and the population targeted. Based on this study there is work to be done especially on understanding what mobilizes young voters (Nickerson 2006b). Young voters seem to be less responsive to social pressure messages compared to older voters (Panagopoulos and Abrajano 2014), while experiments using online ads and social media show promising results (Jones et al. 2017; Haenschen and Jennings 2019).

One of the most striking findings of our research is how effective SMS text messages can be in getting the vote out in certain elections and certain contexts. This is by far the most cost-effective means of mobilizing voters. More research is needed to specify even more precisely the situations under which text messages can be used to mobilize low-turnout groups. If the use of text messaging become more widespread it is also possible that their impact may drop. Therefore, studying variation in the efficacy of this method between different types of elections, different countries and different groups of voters is a promising avenue for future research.

Acknowledgements

We dedicate this article to the late Richard Matland who initiated a research collaboration that led to a number of publications, including this article. We are also grateful to the three anonymous reviewers of this journal for comments and suggestions to this article.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the Norwegian Research Council: [Grant Numbers 249687 and 249687].

Notes

1 The 2015 SMS experiment increased turnout with 2.9 percentage points (from a baseline turnout of only 21.7 percent) among foreign nationals who had become newly eligible to vote in local elections. Among voters with immigrant backgrounds but who had been eligible to vote for at least one election previously, the ITT effect was 2.7 percentage points. After 3 years of legal residence, immigrants to Norway automatically receive voting rights in local elections. In national parliamentary elections, on the other hand, Norwegian citizenship is required to vote. Because we want to compare the effects on similar groups, we focus on native Norwegians in this paper.

2 The fourth experiment was conducted in a European Parliament election, with comparable levels of turnout to Norwegian local elections.

3 The turnout in the 2015 local election was the second-lowest level of turnout in Norwegian local elections since World War II. The record-low was set in the 2003 elections (59.0 percent).

4 Turnout in the 2006 US midterm elections – the setting for the Dale and Strauss (2009) study – was 41 percent (of the voting eligible population). Malhotra et al. (2011) did their study in San Mateo County in California during two separate elections, a local election and a statewide primary election. The voter turnout in the county was 4 percent in the local election and 9 percent in the statewide election.

5 See our pre-analysis plan of the 2015 experiments for a discussion of design and possible heterogenous effects (https://osf.io/du7c2). The argument that mobilizing efforts are most effective in mid-salience elections is not a preregistered hypothesis. The phone call experiment was designed by the LNU and not preregistered.

6 Unfortunately, we do not have data on staff costs, office space, telephone services and training to calculate precise cost-per-vote estimates.

7 We used the Norwegian word “stemme”, which means both “voice” and “vote”.