Political influence on homicide reports under civil conflict 

Get Complete Project Material File(s) Now! »

Direct effects of text message intervention

The main results on the effect of the text message intervention on early childhood development are reported in Table 1.3. Panel A reports ITT estimates for the aggregate cognitive development index for all children 1-7 year old (column 1), as well as separately for children 1-3 year old (column 2), and 3-7 years old (column 3). The estimated impact on early childhood development is very small and not significantly different from zero.17 This result holds for the different age groups. Table 1.17 in appendix shows that this also holds for the individual tests when they are analyzed separately. Table 1.18 shows consistent results across the treatment variations focusing on nutrition, stimulation or both, as well as depending on whether the intervention targets the mother, the father, or both. Tables 1.19 and 1.20 further show that there is no effect on the overall behavioral (socio-emotional) index in column 1, which holds for most of its individual components as well (in additional columns).
Panel B of Table 1.3 reports ITT estimates by level of education of the main caregiver. This shows a small negative coefficient of -0.12 standard deviations among children of the least educated caregivers (those with 3 or less years of education), significant at the 5% level (column 1). In contrast, interaction terms are positive for children whose caregiver completed primary school, though the treatment effect is only significant at the 10% level for the youngest children (with an effect size of 0.16 standard deviations). Finding more variations among children aged between 1 and 3 years old is in line with the literature suggesting higher malleability at a young age. It is also consistent with the hypothesis that interventions in early childhood can be particularly important for cognitive development, a key rationale to target this age group. Differences by education levels suggest that caregivers’ ability to read and understand the messages could (intuitively) be important for the messages to be effective. There is no such heterogeneity, however, for older children for whom there is little impact overall.
Table 1.4 shows results for intermediate outcomes for all children between 12 and 83 months old. Positive ITT estimates are found for all intermediate outcomes, with magnitudes between 0.07 to 0.16 standard deviations. Results for individual questions used to construct the indices are reported in the Appendix Tables 1.21, 1.22, 1.23. Panel B of Table 1.4 reports estimated effects on the 7 intermediate outcome, separately for each (randomized) variation in the content of messages. The impacts on the indices for nutrition, micro-nutrients intake, proteins and hygiene are larger for households assigned to the nutrition and preventive health messages (though not significantly so). Impacts on stimulation and caregivers’ attitudes toward ECD are significantly larger for those assigned to the stimulation and home environment messages. Overall these results show that changes in parental investments and practices broadly reflect the content of the text messages that caregivers received. That said, none of the thematic variations lead to significant changes in early childhood development outcomes (as mentioned above in relation to Table 1.18).

Impact of opinion leaders’ exposure to text message intervention

Table 1.5 reports the estimated b1 coefficients from equation 1.3 on the main outcome, as in Table 1.3. Panel A shows the estimated effect of living in a village where opinion leaders were sent the text messages. It shows that opinion leaders receiving messages has a negative spillover effects on cognitive outcomes of children from other (nonleader) households in the same village. Children living in villages where the leader was assigned to treatment have a score on average 0.11 standard deviations lower than children from households in villages where leaders were not treated (column 1). While the spillover coefficient is very small and insignificant for the younger children, it is negative and highly significant for children between 3 and 7 years old (-0.14 standard deviations). When considering impacts on individual tests, Table 1.24 shows that the negative results are strongest for the two language scores (Denver language and the receptive vocabulary test), which are often considered the best proxies for cognition. On the other hand, Tables 1.19 and 1.20 show no significant impact of the leader treatment on socio-emotional outcomes.
Panel B reports the leader spillover effects on cognitive outcomes distinguishing by caregivers’ level of education. Results show a negative spillover effect from leaders’ exposure to text messages for all age groups among children whose caregivers have less then 4 years of schooling. Children with the least educated parents hence appear the most affected by the negative leader effect. That said, for the older children, and for all children together, the effect is also negative for those whose parents completed primary education (p-value for the joint significance test for the high educated group is 0.05). Similarly consistent with this last result, when distinguishing by education levels of the leaders, Table 1.25 shows that the negative effects are found for leaders with the lowest and the highest education levels.
Exposure of opinion leaders to parenting information by text messages leads to a deterioration of early childhood outcomes among children in their village. As such, the effects of opinion leaders’ exposure to text messages go in the opposite direction than anticipated, and suggest a possible negative influence of these opinion leaders. Since this is arguably a surprising finding, we use an alternative specification to test its robustness. We define a variable measuring the number of dwellings between each household and the closest opinion leader. The variable captures physical distance to opinion leaders, which can be used to test whether the negative leader effects are driven by households that are closer to the leaders. Of course, the variable could also capture remoteness more generally (if opinion leaders live in more central locations), or social distance (if, for instance, members of the same extended families live closer to each other), among other factors. Even so, the interaction effect between the distance variable and the leader treatment provides a useful check about the plausibility that the negative experimental leader effects comes from exposure to those opinion leaders. The interaction effect in Table 1.6 (panel A) shows that the negative leader effect is indeed stronger for households living close to the leaders, and weakens as distance increases (column 1). This effect is particularly strong for younger children (column 2). For them, the significant interaction term indicates that the negative leader spillover effect disappears for households living 6 or more dwellings away from the leader’s house.

Sample, balance across treatments and descriptive statistics

READ  Comparison of a contract of sale and contracts for the execution of works

The sample was drawn from five French regions, presented in Figure 2.6 in the Appendix. These five regions were those with the highest score for the FN in the regional elections of 2015 (as presented on the left of Figure 2.7 in the Appendix) and were chosen to guarantee a sufficient proportion of MLP supporters among respondents. The regions are Hauts de France, Provence-Alpes-Côte d’Azur, Occitanie, Grand Est et Centre Val de Loire.15 Most of our sample comes from the region Hauts-de-France (35,8%), followed by Provence-Alpes-Côte d’Azur (26,1%) and Grand Est (19%).16 MLP indeed did relatively well in these regions in the 2017 election: they ranked 1st, 2nd, 3rd, 6th, and 7th out of 13 regions of mainland France in terms of MLP’s vote share in the first round of the presidential election (see the map on the right of Figure 2.7 in the Appendix).
We stratified our sample on education, age and gender by treatment. The sampling quotas were designed to make the sample as representative of the French adult popula-tion eligible to vote as possible.
For a broad range of variables, Table 2.1 presents the means by treatment group (Columns 1 to 4 show the means in Alt-Facts, Fact-Check, Facts, and Control groups, respectively) and the p-values for the test of the equality of these means across different treatment groups (columns 5 to 10). In column 11, we correct for multiple hypotheses testing. The table suggests that the four randomized groups are largely balanced in observable characteristics. The largest imbalance that we observe is in the proportion of wage earners vs. pensioners: wage earners are 7 and 5 percentage points more frequent in the Fact-Check group and in the Facts group, respectively, compared to Control and the Alt-Facts groups; and there are no significant differences between Control and Alt-Facts groups and between Facts and Fact-Check groups. In all regressions that we present below, we control for a dummy indicating whether respondent is a wage earner as well as other socio-economic characteristics.
In line with the results of the European elections of 2014, regional elections of 2015, and the presidential elections of 2017 in the regions from which the sample was drawn, 22% of the sample voted for Marine Le Pen in the previous presidential election. Television is the main source of information for the majority of respondents, that is 61% of the sample, whereas about 22% of the sample prefer to get information from the Internet and only 10% of the respondents use radio as their main source of information. In addition, we observe that our sample has a strong representation of Catholics (57%) and of those who reported no religion (37%). Table 2.6 in the Appendix provides summary statistics for the main variables of interest in the full sample.

Past election outcomes

As it is often harder to influence voting intentions of those voters who once already voted for the candidate (Mullainathan and Washington, 2009), we asked respondents whom they voted for in the 2012 presidential elections. In order not to contaminate the experiment by framing effect or other aspects of cognitive dissonance, we asked this question after the experiment (in the third part of the survey). This, however, means that the answers to could potentially be affected by the treatment. We check this and find that the past vote for each candidate, including MLP, is balanced across treatment and control groups as reported in Table 2.1. 21.6% of respondents reported having voted for MLP in 2012, which is consistent with the aggregate election results for the regions in our sample.

Prior knowledge

In order to test how the effects of alternative facts and fact checking depends on the knowledge of voters about the subject matter, we need a measure of prior beliefs. In the first part of the survey, before the experiment, all participants were asked about their beliefs on the rate of unemployment among the immigrant population in 2015. In particular, they were asked to chose their response from ten 10-percentage-point intervals. Unemployment rate among working-age foreign-born residents of France in 2015 was 18%, thus falling into the second category. Overall, 27.1% have a correct prior, 9.6% of respondents (238 people) underestimate the unemployment rate among immigrants, and 63.3% of respondents overestimate the unemployment rate among immigrants to a varying degree. 39% of respondents overestimate the unemployment among immigrants grossly, i.e., by at least two categories (believing that unemployment among immigrants is 31% or above).22 In the analysis below, we differentiate between respondents with “correct priors,” “overestimated priors” and “underestimated priors.” The priors are balanced across the four treatments as can be seen from the last four rows of Table 2.1.

Table of contents :

1 Texting Parents about Early Child Development: Behavioral Changes and Unintended Social Effects 
1.1 Introduction
1.2 Intervention, Study Design and Data
1.2.1 Experimental Design
1.2.2 Incentives and Compliance
1.2.3 Data
1.2.4 Balance and Attrition
1.3 Empirical Specification
1.4 Results
1.4.1 Direct effects of text message intervention
1.4.2 Impact of opinion leaders’ exposure to text message intervention
1.5 Mechanisms
1.5.1 Confusion
1.5.2 Boycott
1.5.3 Crowding out
1.6 Concluding remarks
1.7 Appendix
1.7.1 Supplementary tables
1.7.2 Lasso prediction of quiz participation
2 Facts, Alternative Facts, and Fact Checking in Times of Post-Truth Politics 
2.1 Introduction
2.2 Related literature
2.3 Experimental design
2.3.1 Context
2.3.2 Facts and alternative facts
2.3.3 Setup of the experiment
2.3.4 Sample, balance across treatments and descriptive statistics
2.3.5 Variables
2.3.5.1 Voting intentions
2.3.5.2 Past election outcomes
2.3.5.3 Prior knowledge
2.4 Results
2.4.1 The average treatment effect
2.4.2 Heterogeneity with respect to the prior knowledge
2.4.3 Interpretation
2.4.3.1 Nonlinearities in mapping facts to votes
2.4.3.2 Experimenter demand effects
2.4.3.3 Signal about the candidate
2.4.3.4 Salience
2.5 Additional results
2.5.1 Credibility of self-reported voting intentions
2.5.1.1 Evidence from the dictator games
2.5.1.2 Evidence from the list experiments
2.5.2 Heterogeneity with respect to other observables
2.6 Concluding remarks
3 Political influence on homicide reports under civil conflict 
3.1 Introduction
3.2 Elections and conflict persistence in the literature
3.3 Context: The Colombian conflict and elections
3.4 Effect of close electoral outcomes on murders
3.4.1 Econometric framework
3.4.2 Data
3.4.3 Testing Assumptions
3.4.4 Graphical analysis
3.4.5 Results
3.4.6 Placebo tests
3.4.7 Robustness checks
3.5 Mechanisms
3.5.1 Reaction
3.5.2 Reporting biases
3.5.2.1 From unknown perpetrator to guerrillas
3.5.2.2 A benchmark data source
3.5.2.3 Subgroup analysis of municipalities with false positive cases
3.5.2.4 Subgroup analysis of official figures for false positive municipalities
3.5.2.5 Overcount of coalition vs undercount of opposition
3.5.3 Anecdotal evidence
3.6 Concluding remarks
3.7 Appendix

GET THE COMPLETE PROJECT

Related Posts