Negative Descriptive Social Norms and Political Action: People Aren’t Acting, So You Should

While others classify descriptive norms as applying to both the individual and group level (Gerber et al. 2008, 2016; Mann 2010), in this paper we focus exclusively on group-level descriptive norms or descriptions of what others in the subject’s community are doing. We believe this fits more consistently with the literature on descriptive and injunctive norms in psychology (Cialdini 2003).

Hassell and Oeltjenbruns (2016) collected a sample of 1400 emails sent out by congressional campaigns in 2012 from 100 randomly selected congressional districts to analyze campaign negativity. Analysis of a randomly selected subset of 100 of those emails shows that 19% of emails sent by congressional campaigns contained norm inducing language. Thirteen of the 19 emails contained negative descriptive norms while 11 contained positive descriptive norms.

Perhaps even more notable than the two quotes that lead this paper was a prominent fundraising email of the 2012 re-election campaign of President Barack Obama which used the subject line “I will be outspent.” This statement not only implied that donors were not giving enough to his campaign, but that many people were giving more to his political opponent, Mitt Romney. In the tests that the Obama campaign ran before sending the email, this appeal outperformed other subject line options by over a half million dollars (Green 2012). While this email could be construed more as an indication of what Obama’s opponents were doing, it is only one example of campaigns’ continued use of negative descriptive norms (the House Majority PAC and NRCC emails described above being others) that are used in political calls to action.

While not explicitly testing social normative messaging, research suggests that the social setting an individual is in can also affect partisanship (Klar 2014; Sinclair 2012) and political activism generally (Han 2008, 2016; Teske 1997). Yet these effects are more likely attributed to personal relationships rather than social norms, although the latter have not been explicitly tested.

Campaigns explicitly play on these purposive motives to arouse emotions. As one ideological donor explained, “I like to read junk mail. It really pumps you up. It’s like, our side is on the march and the other side is scum. I know this is silly, but it makes you want to contribute to the cause” (Francia et al. 2003, p. 95).

Gerber and Rogers (2009) actually note that for those who regularly vote, negative descriptive norms are marginally more effective at increasing vote intentions (but not significantly significant using a very small sample size (n < 80). This may be because descriptions of low turnout raises anger levels for individuals who have a high sense of civic duty. It could be that, for these individuals, civic duty is not a social motivation but rather a purposive motivation. If this is the case, these individuals are more likely to become incensed when they hear that others are not turning out to vote, thus increasing their likelihood of voting. The same would not be the case for those with low levels of civic duty.

As part of the pre-treatment battery of questions, respondents were asked whether they would describe themselves as an activist and were given the options of “No,” “Yes, somewhat,” and “Yes, definitely.” Because there was only a very small percentage that identified “definitely” as an activist (less than 4%), for the purposes of this analysis we have combined those who indicated “Yes, somewhat” and “Yes, definitely” into a single variable indicating that the respondent identified as an activist. We felt that individuals who even somewhat identified as being an activist were likely to have a higher propensity for taking political action than those who did not identify as an activist, while the difference between “somewhat” activists and “definitely” activists was smaller. This measure also shows a high level of consistency (Chronbach’s alpha = 0.96) that other combinations do not.

Full details about the balance of the two groups on a number of different pertinent co-variates is included in the online appendix.

There was no significant difference in the amount of time that individuals in the two groups took to complete the letter and return to the online survey. The group that received the positive social norm took 87.1 s (112.5 s excluding those who spent less than 20 s overall) on average while the group that received the negative social norm treatment took 91.7 s (108.8 s), p < 0.60.

Although organizations regularly use click-through rates to analyze the effectiveness of their messages sent to organizational members (Congressional Management Foundation 2008), because we did not have the ability to assess whether individuals actually signed the petition, we were forced to rely on an assumption that individuals did fill out the form and then returned to the survey. Respondents were asked to fill out their name, address, and to enter a short text to send to their local official. Respondents who spent more than 20 s spent a minute and a half on average before returning to the survey, with the longest spending 6 min filling out the form. Of the 384 individuals who indicated that they were willing to write a letter to their local official, 308 (80%) of them spent twenty-seconds or more. Raising the minimum amount of time necessary to be considered as having completed the form to 30 or 60 seconds or eliminating the minimum time necessary to be counted as having written the local public official (essentially counting click through rates) has no significant effect on the results. More details are available in Table A.3 in the Online Appendix.

Some concerns may arise that efforts to improve respondent attention may increase social desirability bias (Clifford and Jerit 2015) although merely presenting the manipulation check does not (Berinsky et al. 2014). Increasing social desirability bias, however, would merely increase response rates across the board rather than bias response rates for one or the other as the injunctive norms for both treatments are the same. Again, however, if we remove those individuals who failed the manipulation check the results are the same, as seen in Table A.3 in the Online Appendix.

Because our hypotheses are directional, we use one-tail tests as our baseline assumption. However we present data with two-tailed tests to emphasize more confident results. We admit that some will quibble with results that are p < 0.05 on a one tailed test, but we believe the evidence from the two experiments we present here provides a consistent reaction of the substance of the effect.

Another explanation for this positive (but not quite significant) effect is that positive descriptive social norms are most likely to have an effect on those individuals who identify with the group being described (Lapinski et al. 2007). While this might be the case that activists are more likely to act when they hear of other people doing activist type things, the scripts used do not describe activists but rather describe normal people acting or not acting in their community on behalf of an issue (see the print appendix for the full text of the positive and negative descriptive norms).

It is possible the effect of norms could be conditional on age as those who are older might be less susceptible to negative descriptive framings because they are more aware of socially undesirable behavior. While we find no interaction effect of the norm treatment and age on political behavior in the Mechanical Turk sample, we recognize that a sample from Mechanical Turk is younger than the general population and could mask the effects. While we find no effect, we think this would be important to investigate in future studies.

Education and efficacy are strongly correlated with being an activist which may explain their insignificance in the model. Removing the activist variable from the model makes efficiency significant. Efficiency and education are also significant predictors of other past behaviors in our data such as attending a rally, writing a public official, and volunteering for a campaign. Moreover, as demonstrated in model 1 of Table 3, removing these variables from the model has no significant effect on our key variables of interest.

Respondents were asked “How does this make you feel?” and then were told to “Please indicate the extent to which the above statement makes you angry” and to “Please indicate the extent to which the above statement makes you enthusiastic.” For each question respondents were asked to indicate their feelings on a five-point Likert scale ranging from “not at all” to “extremely”

These results are almost identical if we only compare individuals who saw the same type of appeal in the recycling call to action and in the littering descriptive text. Likewise, there is no difference in the relative increase of anger of self-identified activists and non-activists.

We recognize that others may be concerned about the MTurk’s sample and our ability to generalize using that particular population. Our student sample does not explicitly solve that problem. As we explained previously, however, although neither sample may not be entirely representative of the general population, which Mechanical Turk is not, its usefulness depends on the variation along key moderating characteristics (Druckman and Kam 2011). In this case, as shown in Table 1, there is significant similarity between the general population and the MTurk sample as well as substantial variation in the key variables in which we are interested.

While it would have been nice to have measures of activism or emotion in the survey, this was a field experiment with an outside organization which ultimately requires collaborative efforts. The outside organization wanted to keep the survey at no more than eight questions and was only willing to allow us to manipulate the call-to-action portion of the survey.