Call Your Legislator: A Field Experimental Study of the Impact of a Constituency Mobilization Campaign on Legislative Voting

This research was approved by an Institutional Review Board.

Note that we are not suggesting that the data provided by constituent contacts present an unbiased representation of the preferences of the population of all constituents. We discuss bias in contacting behavior below.

To assess the balance of the treatment assignment, treatment category was regressed on control variables using logit and multinomial regression (Table 4). Control variables included W-NOMINATE score, party, indicators for term, an indicator for gender, percentage voting for the Republican candidate in the previous election, an indicator for membership on the education committee in either chamber and an indicator for state senators. There is no statistically significant relationship between the control variables and the four category treatment variable (p = .92) or a two-category treatment variable that collapses the three treatment categories into one category (p = .70).

The variables used to stratify legislators include indicators for members of the state senate, party identification, sponsorship of the anti-bullying law under consideration in the state house (HB 4163) and the state senate (SB 45) as of late summer 2011, membership in each chamber’s education committee, and a variable divided into four roughly equal categories based on proportion of two-party support for the Republican candidate in the previous election, and a variable based on voting on an anti-bullying bill in the previous legislative session (voted on in the house and not the senate) with categories for support for the bill, opposition to the bill, or senate members/not voting/not in the legislature yet at the time of the vote.

If a child was detected, callers asked to speak with an adult. If the caller was told that they were speaking to someone who was not registered to vote, callers asked to speak with a registered voter in the household, and if none was available, the call was terminated. Otherwise, the call was completed with the person who answered the phone.

Calls were evenly spaced over the 2 weeks for legislators in each of the treatment groups.

See http://​www2.​ed.​gov/​rschstat/​eval/​bullying/​state-bullying-laws/​state-bullying-laws.​pdf.

There were few other campaign activities on the issue. The School-Community Health Alliance of Michigan reports that they carried out few other activities on the antibullying legislation (personal communication with author). The group invited its members to encourage young people to email and contact elected officials on social media, although there is no documentation of whether members responded to these requests. Some other coalition members were active in direct lobbying, and at least one conservative Christian group opposed the measure.

Some Republican legislators opposed the measure because they saw it as a means to promote gay rights (http://​swampland.​time.​com/​2011/​11/​04/​why-does-michigans-anti-bullying-bill-protect-religious-tormenters/​).

We thank an anonymous reviewer for these observations.

W-NOMINATE scores were created separately for each chamber. Although there is considerable overlap in the legislation voted on in both chambers, the roll calls used in either chamber are not identical. However, regressing the W-NOMINATE score on an indicator for chamber, party, and an interaction term for these two variables found no statistically significant main effect of state senate membership or interaction effect of state senate X Democrat. The W-NOMINATE score correlated highly with an interest-group-created vote score (Americans for Prosperity-Michigan) created from bills voted on in 2011 (r = .95). We use the vote score as a control instead of other measures of prior legislative behavior (including sponsorship of and house voting on prior bullying bills). Including the latter as controls does not substantively change the results.

Using the actual number of calls as an independent variable or instrumenting the actual number of calls on treatment group (0, 22, 33 or 65 intended calls) while controlling for the treatment indicator for receiving any calls does not change the substantive results.

The coefficients for the regression with controls are presented in an Appendix (Table 5). Of the control variables, only the coefficient for state senators is statistically significant. The state senate had earlier in the year passed an anti-bullying bill that was criticized for including language that exempted certain behaviors from what the bill classified as bullying. Media attention to this bill may have increased pressure on the state senate to support the more recent version of the bill.

As an exploratory analysis, we also explored the interaction of the treatment indicator with a variety of controls, including closeness of the prior election (|proportion of two party vote-.5|, Republican percentage of the two party vote in the prior election, first term legislator, final term legislator (legislators at the time of the vote who were serving their final term due to term limits), state senate membership, gender, party, education committee membership, liberalism (W-NOMINATE score), and extremism (|W-NOMINATE score|). For each of these ten variables, we ran a separate regression including a treatment, each of these controls, and a treatment X control interaction term. The only statistically significant interaction was for treatment X Republican (p < .05, two tailed), although after accounting for multiple comparisons, this coefficient is not statistically significant. See Appendix Table 6.