Introduction
Do the benefits often enjoyed by campaign contact – in the form of higher rates of political participation – continue at subsequent elections? While Get Out the Vote (GOTV) research is well developed (Green and Gerber Reference Green and Gerber2015), “one of the most important questions regarding voter mobilization is whether its effects endure” (Bedolla and Michelson Reference Bedolla and Michelson2012, 173). If a campaign boosts participation not only at the original election but also at subsequent elections, then this represents positive civic externalities of campaigning. But analysing downstream effects can also shed light on other aspects of voter behaviour such as habit-formation and socialisation (Green and Gerber Reference Green and Gerber2002).
A line of research has focussed on this question (Cutts, Fieldhouse, and John Reference Cutts, Fieldhouse and John2009; Davenport et al. Reference Davenport, Gerber, Green, Larimer, Mann and Panagopoulos2010; Gerber, Green, and Shachar Reference Gerber, Green and Shachar2003). Studies test whether the difference in turnout between control and treatment groups after an initial GOTV experiment persists at the next election(s), with no additional treatment being administered. Existing evidence on downstream effects is mixed. Some experiments have had enduring effects on participation (e.g. Cutts, Fieldhouse, and John Reference Cutts, Fieldhouse and John2009; Gerber, Green, and Larimer Reference Gerber, Green and Larimer2008; Mann Reference Mann2010; Panagopoulos Reference Panagopoulos2010), while the effects of others decay, and at varying rates (e.g. Davenport Reference Davenport2010; Reference Gerber, Green and Larimer2010). Scholars have argued that a ‘habit’ effect could explain why sudden, artificial shocks to voter turnout have enduring effects on participation (Gerber, Green, and Shachar Reference Gerber, Green and Shachar2003). But there is no consensus on why the effects of some experiments endure, suggesting habit-formation among subjects, while others dissipate.
While this remains something of an open question (Davenport et al. Reference Davenport, Gerber, Green, Larimer, Mann and Panagopoulos2010), there is evidence that underlying turnout and the saliency of the original and subsequent elections could play a role in determining downstream effects (Arceneaux and Nickerson Reference Arceneaux and Nickerson2009; Davenport et al. Reference Davenport, Gerber, Green, Larimer, Mann and Panagopoulos2010). For instance, effects detected in higher saliency elections (e.g. John and Brannan Reference John and Brannan2008) have downstream effects at subsequent lower saliency elections (Cutts, Fieldhouse, and John Reference Cutts, Fieldhouse and John2009). Meanwhile, the effects of experiments conducted at lower saliency elections tend to persevere at subsequent lower saliency contests but not at higher saliency elections such as presidential elections (Davenport Reference Davenport2010; Gerber, Green, and Larimer Reference Gerber, Green and Larimer2008; Gerber, Green, and Larimer Reference Gerber, Green and Larimer2010). There is reason to expect, therefore, that if a campaign can boost turnout at a higher saliency election, then effects will persist – but if turnout is mobilised at a lower saliency election, effects will be more likely to decay. But, due to the amount of time between elections, the effect of context and saliency on downstream effects can be muddied with the decay that might solely be down to time passing.
This paper contributes to this discussion by reporting the results of a pre-registered analysis of the downstream effects from a partisan campaign experiment conducted in the 2017 local elections in the UK to the general election a month later. The paper provides a unique experimental setting in which to test downstream campaign effects. While existing literature suggests there is a relationship between electoral context and the strength (or decay) of downstream effects, it is often difficult to unpick this relationship from simple temporal decay. This study is unique due to the close proximity, but vastly differing saliencies of the two elections – a result of UK prime minister Theresa May’s decision to call a ‘snap’ general election. The study also represents one of the few downstream analyses to be conducted outside of the US, and the first to examine the downstream effects of partisan campaign treatments in Britain.
While the original experiment – conducted in a local election in which underlying turnout was just 35.4% – found that leaflets and canvass visits increased turnout by 3.6 percentage points, the effects virtually disappeared at the higher turnout general election a month later. The results are in line with research showing that campaign effects can be determined by underlying turnout and saliency (Arceneaux and Nickerson Reference Arceneaux and Nickerson2009).
Downstream Effects Of Gotv Campaigns
Since Gerber, Green, and Shachar (Reference Gerber, Green and Shachar2003) first detected downstream effects from their 1998 campaign experiment, a series of downstream studies have been conducted. These have tended to conclude that while effects decay over time, both negative (Atkinson and Fowler Reference Atkinson and Fowler2011) and positive (Bedolla and Michelson Reference Bedolla and Michelson2012; Cutts, Fieldhouse, and John Reference Cutts, Fieldhouse and John2009) shocks to turnout can persist at subsequent elections.
But downstream effects are mixed, and there is no consensus regarding our understanding of the basis for persistence in mobilisation effects over time. Some have suggested that the memorability of the initial treatments and the timings between elections could be relevant factors (Davenport et al. Reference Davenport, Gerber, Green, Larimer, Mann and Panagopoulos2010). Further, saliency and underlying turnout could be key. Davenport et al. (Reference Davenport, Gerber, Green, Larimer, Mann and Panagopoulos2010) track the mobilisation effects of six social pressure experiments over time. They find that initial turnout effects persist in some cases, but in others initial effects decay almost entirely at subsequent elections. A pattern emerges that when turnout is mobilised at a lower saliency contest, effects are more likely to persist at other subsequent lower saliency contests than at higher saliency elections.
For instance, Davenport et al. (Reference Davenport, Gerber, Green, Larimer, Mann and Panagopoulos2010) find that the effects of Gerber, Green, and Larimer’s (Reference Gerber, Green and Larimer2008) experiment during the August 2006 primaries in Michigan persisted at the November 2006 midterms, the January 2008 primaries, and the August 2008 primaries, but not at the 2008 presidential election in which underlying turnout was around 50 percentage points higher. Similarly, the initial effects ranging from 1.4 to 4.9 points detected by Gerber, Green, and Larimer’s (Reference Gerber, Green and Larimer2010) experiments during the November 2007 midterms persisted at the January 2008 primaries but disappeared at the 2008 presidential election. The effects of Davenport’s (Reference Davenport, Gerber, Green, Larimer, Mann and Panagopoulos2010) study at the 2007 municipal elections likewise continued several months later at the 2008 primaries, but not at the 2008 presidential election. Finally, in a supplementary analysis Hill and Kousser (Reference Hill and Kousser2015) also find that letters that mobilised turnout in the 2014 primary elections had no downstream effects at the November presidential.
This pattern is not clear-cut, however. Davenport et al. (Reference Davenport, Gerber, Green, Larimer, Mann and Panagopoulos2010) note that some studies conducted in lower saliency contests had strong downstream effects across higher and lower saliency elections (e.g. Mann Reference Mann2010; Panagopoulos Reference Panagopoulos2010). However, even in the case of Panagopoulos’ (Reference Panagopoulos2010) series of experiments at municipal elections in 2007, strong downstream effects at the 2008 presidential election were only detected in the area that had the highest underlying level of turnout in the original election – 31% in Monticello, Iowa.Footnote 1
The significance of electoral context and underlying turnout is analogous to research by Arceneaux and Nickerson (Reference Arceneaux and Nickerson2009) that shows that mobilisation effects differ according to the saliency of an election. Examining the effects of 11 GOTV field experiments, Arceneaux and Nickerson (Reference Arceneaux and Nickerson2009) found that the effects of campaign contact are a function of how close an individual is to his or her ‘threshold of indifference’ between voting and abstaining. They find that those closest to this threshold are more responsive to campaign effects, as they are more likely to be tipped over the threshold and into voting by campaign contact. Meanwhile, individuals who are either habitual voters or habitual abstainers are further away from their thresholds. As such, campaign contact is less effective among such voters. Crucially, they find that the type of voter that is closest to this threshold can change from election to election.
Habit-formation and socialisation-based research gives reason to anticipate that downstream effects will be detected in this analysis (Gerber, Green, and Shachar Reference Gerber, Green and Shachar2003). However, the competing hypothesis would suggest that experiments conducted in lower saliency contests detect effects due to the saliency of the election and because most voters were close to their indifference threshold between voting and abstaining. At higher saliency general and presidential elections meanwhile, most voters will have shifted further away from this threshold and be much more inclined towards voting due to the high saliency of the election. As such, we would expect the effects of the original campaign contact to be much reduced.
Besides saliency, another relevant factor to this discussion is time, namely, that we would expect effects to decay naturally over time regardless of saliency or context. But it is difficult to separate the natural decaying effect of time from the role played by the saliency patterns of the elections. For instance, while effects might decay from a lower turnout election to a higher turnout election a year later, the notion that this is related to the differing saliency of the elections is undermined by the fact that a year passes between the contests. In other words, would the downstream effects persist if an election of the same saliency was held much sooner after the original election? This paper tests for downstream effects by exploiting an unusual situation in which the time between elections is very small, while the saliency differs significantly. To test for potential downstream effects from a low saliency election to a high saliency election, I examine the downstream effects of a partisan experiment conducted in the UK local elections in 2017. I exploit the timing of the ‘snap’ general election called by Prime Minister Theresa May for June 2017 and analyse the turnout records of the same voters involved in the original election experiment, to test whether the initial effects persisted a month later.Footnote 2
May 2017 Experiment
The original experiment (Townsley Reference Townsley2018) was conducted during the May 2017 round of local elections in cooperation with the local Liberal Democrats (Lib Dems). Nationally, turnout at the local elections was 35.1% – typical for local elections in the UK (Electoral Commission 2017). The study population comprised all registered voters in the Thedwastre North division within Suffolk County Council, after several villages were excluded having been subject to a pilot experiment beforehand. The experiment population totalled 6,525 registered voters in 3,371 households. Turnout in May among all subjects in the study was 35.4%.
Subjects were split between postal voter households and non-postal voter households, then cluster randomised (at the household level) into either a control group or one of two treatment groups (Figure 1). As such, all subjects within a given household were assigned to the same treatment (or control) condition (Townsley Reference Townsley2018). As with most experiments involving canvassing, noncompliance was an issue in the original study, with 28% of subjects in the canvass + leaflet group being successfully treated.Footnote 3 Households that were not successfully canvassed are still included in the ITT estimates in this paper.
Field experiment research design process (Townsley Reference Townsley2018).
The control group was not exposed to any direct Liberal Democrat campaigning, while one treatment group received a Lib Dem leaflet, and the other treatment group received the same leaflet followed by a canvass visit.Footnote 4 The leaflet was partisan in nature and advocated local issues, as well as profiling the candidate briefly. The canvass visits were carried out by a Liberal Democrat volunteer and were likewise partisan in nature mentioning local issues and the party’s candidate.Footnote 5
After treatment was conducted, subjects’ turnout data at the 2017 local election were obtained from the local authority’s official records, which were released after the election and matched to the experimental records. Intent-to-treat (ITT) results were estimated using linear regression, with standard errors clustered at household level. The covariate-adjusted effects adjust for pre-treatment covariates in a multivariate linear regression framework (Gerber and Green Reference Gerber and Green2012: 109). Covariate data were obtained from the local party’s voter records, which were made available prior to random assignment. The covariate data include sex, party support (whether a Lib Dem supporter or not), ward (approximately village-sized boundaries within the wider division of Thedwastre North), postal voter registration, previous turnout (at the 2009 local elections) and age group (under 35, 35–59, and 60+), and was matched to the experimental records.
The original experiment found that subjects that were exposed to campaign contact – in any form – were 3.6 percentage points more likely to vote, significant at the 95% level. Turning to the specific treatments, both (a) leaflets and (b) leaflets followed by canvass visits increased turnout in the non-postal voter experiment (by +4.3 and +4.9 percentage points, respectively). In the postal voter experiment, meanwhile, neither treatment combination had any significant effects on turnout.
Because the analysis is essentially a panel design, I examine the rate of attrition in the sample of voters between the May and June elections. This shows that the rate of attrition (3.9%) did not vary significantly between assignment groups.Footnote 6 A balance test shows that after attrition is taken into account, the covariates still do not vary appreciably between assignment groups. Following Gerber and Green’s protocol (Reference Gerber and Green2012), I run a multinomial logistic regression model to show that all covariates taken together do not significantly predict assignment to treatment conditions, increasing confidence that attrition did not affect the balance obtained from the original randomisation process.
June 2017 General Election
On 18th April 2017, just over a fortnight before the May local elections, UK prime minister Theresa May announced her intention to hold a ‘snap’ general election, 3 years earlier than the next regularly scheduled election in 2020. The calling of an early general election had long been the prerogative of the prime minister but was limited by the Fixed Term Parliaments Act 2011, which required a two-thirds majority of MPs to vote to dissolve Parliament and thus trigger an election. On 19th April Parliament supported the move, and the election was subsequently held on 8th June 2017. Theresa May’s decision came largely as a surprise but was taken in light of strong opinion polls for her Conservative Party (Cowley and Kavanagh Reference Cowley and Kavanagh2018). Despite the ‘snap’ nature of the election, the period between prime minister declaring her intention to hold the contest and the actual election was relatively long (ibid.).
As we might expect in a ‘snap’ general election, across Britain the intensity of local campaigning was reduced (39% reported being contacted by a party, compared to 52% in 2015 (Cowley and Kavanagh Reference Cowley and Kavanagh2018, 301). However, campaigning at the 2017 election was typical of recent general elections in terms of style – relying heavily on leaflets and letters. In terms of saliency too, there is evidence that the election was broadly similar to typical general elections. For instance, turnout was similar to that of previous contests – 68.7%, compared to 66.4% in 2015 and 65.1% in 2010 – and in line with increases in turnout at recent general elections since 2001. Locally, the Thedwastre North area in which the original experiment was conducted falls within the Bury St Edmunds parliamentary constituency, which tends to experience less-intense campaigning due to its status as a ‘safe’ Conservative seat (BBC 2017). Turnout in Bury St Edmunds was 70.8%, slightly above that of the national level. Among subjects in the original May experiment specifically, turnout was 77.7%.
That the overall turnout among experimental subjects most than doubled from 35.4% in May to 77.7% in June is indicative of the higher saliency status of the general election. The increased turnout and the close temporal proximity of the May and June elections – separated by just 35 days – provides an interesting test for downstream effects. From one perspective, the proximity of the second election to the first election suggests effects should persist. But if downstream effects are influenced by underlying turnout – specifically, by much-increased underlying turnout, then we would expect the original effects to decay. Would we expect downstream effects to differ between postal voters and non-postal voters? We might anticipate that the latter would be more likely to exhibit downstream effects as they were mobilised by the original experiment. Treatment effects among postal voters, meanwhile, suffered from an apparent “ceiling effect” wherein their high underlying rate of turnout meant that further increases were unlikely (Arceneaux and Nickerson Reference Arceneaux and Nickerson2009).
Downstream Results
Table 1 presents descriptive information about the turnout rates of subjects in both experiments, namely, how the rate of turnout in June differs between those who voted in May and those who did not. The June turnout data were gathered in the same way as the May turnout data. The local party obtained the official turnout records from the local authority, which were then made available and matched to the experimental records. Unexpectedly, voting in May is strongly associated with voting in June. While 77.7% of subjects voted in the June general election, turnout was higher among those who had previously voted in May (97.3%) compared to those who had not (66.6%). This is in line with existing research that shows previous turnout is a strong predictor of turnout (Green and Shachar Reference Green and Shachar2000).
Turnout Rates in June, by Turnout in May
Did the effects of Lib Dem campaign contact in May endure at the general election in June? Table 2 presents the ITTFootnote 7 effects on turnout in May (left half) and June (right half) to allow comparison. Results were estimated using a linear regression framework, the same as was used in May, with standard errors clustered at household level. All effect sizes that are statistically significant at conventional levels of at least 95% confidence are highlighted in bold. Assignment to partisan campaigning had statistically significant effects on voter turnout in the original May local election. In May, the effect of Lib Dem campaigning was 3.6 percentage points across the full sample and 4.6 percentage points among non-postal voters. Meanwhile, postal voters were not significantly mobilised by exposure to the leaflet or leaflet and canvassing treatments. The leaflet and canvass treatments likewise had stronger effects among non-postal voters.
Intent-to-Treat (ITT) Effects of Treatment in May and in June (Covariate-Adjusted)
Robust standard errors (in parentheses) clustered on households. All tests two-tailed. **p < .01, *p < .05.
The right half of Table 2 shows the effect of the treatments on turnout in June. Across the full sample, those assigned to receive Lib Dem campaign contact in May were slightly less likely to vote in June than the original control group (ITT = −0.2, SE = 1.4). This is despite this group being 3.6 percentage points more likely to vote in May. Turning to the separate treatment groups, the leaflet group was also less likely to turnout than the control group (ITT = −0.9, SE = 1.7). Meanwhile, the canvass/leaflet combination groups were 0.5 percentage points more likely to vote than the control groups (SE = 1.6). These effects, however, fall short of statistical significance. The pattern is similar in the non-postal voter and postal voter experiments, separately. In both cases, the leaflet group was less likely to vote in June, while the canvass and leaflet combination had a small positive downstream effect, but both fall short of statistical significance.
There is, therefore, weak evidence of any downstream effects of the original treatments. Across the full sample the ITT effects fall from 2.8 and 4.5 for leaflets and the canvass/leaflet combination, respectively, to −0.9 and 0.5 in June. The 0.5 percentage point downstream effect of the canvass and leaflet combination is positive, suggesting that there may be some enduring impact of the canvass and leaflet treatment combination. Indeed, we might expect this treatment combination to be the most likely to have an enduring effect, as it was the most powerful treatment in the original experiment. However, the drop-off in their effect between May and June is substantial, and with the current sample size it is not possible to reject the null hypothesis of the downstream effect of the canvass and leaflet combination being zero.
Discussion And Conclusion
The aim of this paper was to examine whether exposure to partisan campaign activities in a low saliency, low turnout election has downstream effects on turnout at a subsequent higher turnout general election. This paper exploits the timing of the ‘snap’ UK general election called for June 2017 to examine whether effects persist from a low saliency to a high saliency contest taking place only a month later. The study represents one of the few downstream analyses to be carried out in a non-US context, and the first such partisan study to be conducted in Britain. The results show that while campaign contact had strong effects at boosting turnout at the May 2017 local elections, their effects quickly dissipated and were ultimately only temporary as the difference in turnout between those in the control and treatment groups was negligible only a month later.
That almost all of the initial mobilisation effects detected in May had decayed by June may surprise some campaign scholars from one perspective. Not only because many existing studies have found that over half of the initial mobilisation effect often carries on to the next election (Cutts, Fieldhouse, and John Reference Cutts, Fieldhouse and John2009; Gerber, Green, and Shachar Reference Gerber, Green and Shachar2003; Panogopoulos 2010), but because in this study the original experiment was conducted only a month earlier than the subsequent election. With the May local election and the June general only 35 days apart, the rate of decay in the effect of the original campaign contact is one of the fastest detected (e.g. Davenport Reference Davenport2010; Gerber, Green, and Larimer Reference Gerber, Green and Larimer2010).
However, the findings are in line with evidence that downstream effects tend to be more likely between lower saliency elections, rather than between low and high saliency contests (Davenport et al., Reference Davenport, Gerber, Green, Larimer, Mann and Panagopoulos2010). The results are also in line with Arceneaux and Nickerson’s (Reference Arceneaux and Nickerson2009) research into how effects differ according to the underlying propensity of voters to participate at a given election. Indeed, many experiments show that campaign effects are strongest when the underlying turnout at the election is lower – and vice versa (Green and Gerber Reference Green and Gerber2015). Arceneaux and Nickerson (Reference Arceneaux and Nickerson2009) show that the effect of campaign contact is a function of how close an individual is to his or her ‘threshold of indifference’ between voting and abstaining. While turnout in the original May experiment was 35.4%, at the June general election it was 77.7%. This paper suggests that turnout effects were detected in the former as many voters were close to this threshold. At the latter, however, those who were formerly close to this threshold were much further away due to the higher saliency of the June election, hastening the decay of the original effects.
Why downstream effects struggle from low to high saliency contests is not directly tested in this study, however. Alongside Arceneaux and Nickerson’s framework, it could be that the campaign noise generated by the higher saliency general election in June led subjects to resort more rapidly to their pre-existing behavioural patterns when it came to voting. Other factors could also have contributed to the effect decay that are connected to the partisan nature of the original treatments. The partisan treatments in the May 2017 experiment meant that the stimuli were very specific to that election, for instance, by referring to the local election candidate and issues at stake at that contest. As well as the candidates, the issues at stake generally in the June election were different to those in May. As such, voters may have been less likely to associate the original treatments with the turnout decision at hand at the very different general election. It could be that partisan-based prompts to vote are ultimately less likely to endure than appeals that trigger more deeply held and generally applied feelings of having a sense of civic duty to vote.
The downstream effects resulting from campaign experiments are worthy of further research. This paper suggests that the rate of decay in the original effects could be a function of the salience of the subsequent election. However, future research might consider the different contexts in which effects are most likely to persist in order to find the optimal elections in which GOTV campaign can reap the most civic rewards.
Supplementary Material
To view supplementary material for this article, please visit https://doi.org/10.1017/XPS.2019.20.
