Pre-Analysis plan: What messages encourage political participation in young South Africans?

Once recruited and consented into the study, participants completed a brief electronic pre-survey with an enumerator to gather demographics and a baseline for their intention to register and vote in the forthcoming 2016 election.

Subjects were then randomly assigned to one of eight conditions (the seven messages or a control, which received no message). Respondents received the message read out face-to-face by enumerators. All messages were delivered in English and the consent and survey process were conducted in English. Given that our sample is from urban Gauteng, levels of English proficiency are high.

These messages were crafted from theories of political participation. In addition, we conducted eight focus groups in May 2015 in the study area with both registered and unregistered voters aged 18-30. We discussed whether participants think voting is important, their reasons for (not) registering and voting, where they get information from, and who or what influences their decisions. Many of the phrases in the messages, such as politics being ‘an old person’s game’, every vote ‘being counted’, the youth ‘jobs crisis’, or non-voters ‘letting their community down’, were drawn from these focus groups. We also asked focus group participants to recommend adaptations to messages to make them more appealing to young people.

The different messages focused on one's obligation to vote (as a citizen, as a young person, or stemming from South Africa's history), on the idea that every vote counts, on the fact that young people are underrepresented in parliament, on the idea that voting may force politicians to pay more attention to issues which affect young people, and on the idea that one's community will know if one does not vote.

2015-07-20

2015-09-04

Intention to register to vote, intention to vote, intention to obtain information about registering and voting, partisanship, intention to register and vote among others like you, whether the respondent would vote for the ANC, compared to voting for an opposition party or not voting, whether the respondent believes ‘Opposition parties should regularly examine and criticize government policies and actions’, preference for democracy over other forms of government, overall satisfaction with democracy, index of satisfaction with municipal performance, political efficacy index, importance of accountability index, belief that voting matters index, agreement that voting is a duty, strength of identification with being a young person

See pre-analysis plan

Once recruited and consented into the study, participants completed a brief electronic pre-survey with an enumerator to gather demographics and a baseline for their intention to register and vote in the forthcoming 2016 election. Subjects were then randomly assigned to one of eight conditions (the seven messages or a control, which received no message).

See pre-analysis plan

Randomisation of individuals to eight treatment arms using surveyCTO software on tablets at the lab-in-the-field site

Individual

No

The treatment is randomised at individual level, not at cluster level. We treated individuals at 23 sites.

3,200

400

We calculate the minimum detectable effect size (MDES) in standard deviations for an individual-level trial(voting district) with power = 0.8, alpha = 0.05, covariate R2 = 0.5. We calculate MDES for 3,200 subjects, divided into eight arms. For marginal effects, such as message 1 vs the control, the MDES is 0.14. For our main outcome, registration rates, we assume a baseline registration rate in the control of 50% (before registration drives for the 2014 election in Gauteng, 48% of young voters were registered), so the standard deviation is ~0.5. For marginal effects on registration, we can thus detect changes of ~7.00 percentage points. For a comparison of all treatments to the control, the MDES is 0.1058, so we can detect changes of ~5.29 percentage points. For interactions with marginal effects (e.g. message 1 vs the control by gender), the MDE is 9.9 percentage points. For interactions with marginal effects (e.g. all messages vs the control by gender), the MDE is 7.5 percentage points.

PAP