Primary Outcomes (explanation)
After the experiment is concluded and we will have collected all the data, we will test the impact of the informational treatment on the vote choice reported in the second questionnaire. Besides, we will also test coherence between attitude reported in the first questionnaire and actual vote choice reported in the second questionnaire.
We will estimate the models using the Logistic model when the response variable is a binary indicator.
A – Impact on vote behavior
The impact on vote behavior will comprise a series of statistical tests where we will compare treated and untreated units' vote behavior. Vote behavior will be evaluated among several dimensions, including:
- A1: Turnout (vote extensity)
we regress an equation:
V_i= a_1*T_i+u_i
, for each individual i, where T_i is treatment and V_i(D2_1) is a dummy equal to 1 if individual i voted and 0 otherwise.
Are more informed individuals more/less likely to turn out?
H0: No, a_1=0
H1: Yes, a_1≠0
- A2: Expression of a preference for councilor candidate and mayor candidate (vote intensity)
we regress an equation:
P_i= a_2*T_i+u_i,
for each individual i, where T_i is treatment and P_i (D2_3) is a dummy equal to 1 if individual i expressed a preference and 0 otherwise.
Are more informed individuals more/less likely to cast a preference vote?
H0: No, a_2=0
H1: Yes, a_2≠0
- A3: Characteristic of the candidates that received the preference (vote choice): (D2_2-D2_4)
We will compare voter preferences of treated and untreated groups across several dimensions of the candidates, including:
-Gender, education, experience and, ideology.
we regress a set of equations V_i= a_3*T_i+u_i, for each individual i, where T is treatment and V can be Gender, Education, Experience and Ideology (party membership) of the candidates voted by the individual (both for mayors and councillors).
Do more informed individuals more/less likely to cast a preference vote for a Women/Educated/Experienced candidate?
H0: No, a_3=0
H1: Yes, a_3≠0
Mayors and councilors characteristics will be measured as follow:
-Gender (Female=0; Male=1)-Education (University Degree=1; Else =0)
-Experience (At least one previous political experience =1; Else=0)
-Ideology (Left (PAP, PRC and others at the left of PD); Centre-Left (PD and civic lists supporting Gualtieri ); 5 Star Movements; Centre-Right and Right)
We will explore the heterogeneity of the treatment effect along several dimension including:
-Political participation (party-member), civic participation (a member of the activist group), education level. (D1_6, D1_7, D1_3)
we regress a set of equations:
V_i= b_1*T_i+b_2*P_i+a_4*P_i*T_i+u_i,
for each individual i, where T is treatment and V can be Gender, Education, Experience and Ideology (party membership) of the candidates voted by the individual. P can be the level of political participation, civic participation, or education level of the individual.
Does a prior level of political participation influence the informational effect of casting a preference vote for a Women/Educated/Experienced candidate?
H0: No, a_4=0
H1: Yes, a_4≠0
B – Attitudes and vote behavior
The second part of the results will analyze voters’ perception about transparency and demand for candidates’ quality and their actual voting behavior. We will thus carry a series of statistical tests comparing answers reported in the first survey compared to answers reported in the second survey.
B1 - Ideal Candidate and vote behavior
We will compare the matching between:
-the profile of the ideal candidate reported by the interviewer and the actual preference expressed.
we regress an equation:
〖AC〗_i= b_1*T_i*〖AI〗_i+b_3*〖AI〗_i+u_i,
where AI (D1_11) is the attribute of the ideal candidate as reported by the individual i. And AC is the same attribute of the candidate voted by the individual. The attribute can be the level of education or the level of experience. For example, we compare how the individual responds to the following question “Do you think the ideal candidate should have at least a university degree?”, and compare their answer with the level of education of the candidate for which they voted. Or, for example, we ask “Do you think women should be more represented in Politics?” and compare if the voter actually expressed a preference for a women candidate. T_i is a dummy if the individual belongs to the treated units.
Does the informational effect influence the matching between preference expressed and actual vote behavior?
H0: No, b_1=0
H1: Yes, b_1≠0
Does preference expressed predict actual vote behavior?
H0: No, b_3=0
H1: Yes, b_3≠0
B2 – Demand for transparency and use of transparency.
We will compare the matching between
-the demand for transparency expressed by voters and the actual use of the information.
we regress an equation:
R_i= b_2*D_i+u_i,
where D is the answer of the individual to the question “are you interested in political transparency”, “do you think political transparency is important” (D1_16), and R is a dummy on whether the respondent opened the link with the information set.
Does the revealed demand for transparency predicts the use of transparency?
H0: No, b_2=0
H1: Yes, b_2≠0
Representativity of the sample
We will analyze and discuss the representativity of the final sample (respondent of the first and second questionnaire) with respect to:
Sample of all the respondents of the first questionnaire
Total population of Rome (using the survey “Aspetti della Vita Quotidiana” from the National Institute of Statistics (ISTAT) that contains the same questions)
The analysis will comprise balance tests and computation of selection propensity score (probability of being in the final sample) among sex, age, Education, Area of residence (Municipio), Political Participation, Participation in the electoral campaign (D1_1 to_D1_9 of the baseline questionnaire).
After having computed the propensity scores we will repeat and provide results of testing the hypothesis of the experiment using the propensity scores as weight in the regression models (as suggested by Stuart et al. 2011).