Economic Anxiety, Security Concerns, and Discontent: Tracing Public Attitudes Toward Democracy in Latin America

We aim to study the extent to which economic anxieties and insecurity drive citizens' democratic attitudes and shape expressions of discontent. For this purpose, we designed three versions of the questionnaire (A, B, C), which contain the same questions but differ in the order of their sections.
- In Survey A, we begin by asking about insecurity experiences, followed by sections on discontent and concerns; perceptions of inequality, injustice, and discrimination; political preferences and democracy; preferences for redistribution; expressions of discontent; and self-exclusion.
- In Survey B, we start with questions on economic anxiety and expectations, followed by the same sections as in Survey A.
- In Survey C, questions on insecurity experiences and economic anxiety and expectations are asked at the end.

In our intervention, Survey C serves as the control group. Surveys A and B aim to prime respondents’ concerns about (i) insecurity experiences and (ii) economic anxiety and expectations, respectively, to measure how these affect levels of discontent and democratic attitudes.

The randomization of survey versions ensures a source of exogenous variation, as the assignment is not influenced by any characteristics of the respondent, interviewer, region, or time of day. By comparing responses across these three versions, we aim to answer our research question on the role of insecurity and economic anxiety in shaping political attitudes and expressions of discontent.

Each version of the survey will be randomized. This randomization ensures a source of exogenous variation as it is not based on any characteristics of the respondent, the interviewer, the region, or the time of day. Our goal is to compare these three versions to answer our research questions regarding priming.

What are the main drivers of citizen discontent in Latin America? How do economic distress and insecurity shape citizens' perceptions and legitimation of the social contract? To answer these research questions, we partnered with the National Statistics Offices (NSOs) of various countries in Latin America and the Caribbean to conduct an experimental study on subsamples of the national household surveys. This study received Institutional Review Board (IRB) or ethics committee approval from both The World Bank and Harvard University. An advance agreement with the implementation group ensures that all results can be published.

Each NSO will conduct the surveys using one of three modes: phone, hybrid, or in-person interviews. At the time of writing, significant progress has been made in Colombia, Peru, Guatemala, and Ecuador. While surveys in additional countries are under consideration, their feasibility remains uncertain. Collaboration with the NSOs is critical, as it allows us to leverage the detailed information collected in each country’s household survey.
For these subsamples, we developed a survey instrument called VOCES, designed to measure multiple aspects of respondents' lives through approximately 50 questions. The survey includes sections assessing trust, opinion contagion channels, perceptions of corruption, insecurity experiences, discontent and concerns, perceptions of inequality, injustice, and discrimination, political preferences and democracy, preferences for redistribution, expressions of discontent, self-exclusion from public services in favor of private services, and economic anxiety and expectations.

As part of the experimental design, we created three versions of the questionnaire (A, B, and C), each containing the same set of questions but differing in the order of sections:
• Survey A begins with questions on insecurity experiences, followed by sections on discontent and concerns, perceptions of inequality, injustice, and discrimination, political preferences and democracy, preferences for redistribution, expressions of discontent, and self-exclusion.
• Survey B starts with questions on economic anxiety and expectations, followed by the same sections as in Survey A.
• Survey C asks questions on insecurity experiences and economic anxiety and expectations last.
In this intervention, Survey C serves as the control group. Surveys A and B aim to prime respondents’ concerns about insecurity experiences and economic anxiety and expectations, respectively, to measure how these concerns influence levels of discontent and democratic attitudes.
Each version of the survey was randomized independently of any characteristics of the respondent, interviewer, region, or time of day. By comparing responses across the three versions, we aim to address our research questions regarding the drivers of discontent in Latin America.

We follow a complete randomization process. While we did not stratify based on any respondent covariate, the risk of imbalance in subgroup proportions is minimal given our sample size of over 2,000 observations per country (Kernan et al., 1999; Bruhn and McKenzie, 2009).
Each NSO uses different software platforms to collect and program the surveys, including tools such as MySurveySolutions and CSPro. Some platforms allow built-in randomization, ensuring that interviewers automatically receive one of the three survey versions when starting a new survey. For NSOs using less advanced software, we developed a manual randomization algorithm. This method involves storing participants’ contact information in an Excel sheet, re-ordering the table randomly, and assigning one of the three versions sequentially (A, B, C) in a repeating pattern.

Beyond the experiment, we will leverage the rich data from the NSO household surveys in each country to further characterize individuals’ socio-economic status and explore how these factors influence discontent. For instance, we are particularly interested in examining how both sociotropic and egotropic economic factors shape citizen discontent and its expressions. In Peru, we will take advantage of the availability of panel data from the household surveys to capture intertemporal variance, offering deeper insights into changes over time. Additionally, we will use this fine-grained socio-economic characterization to explore the extent to which the effects of our primes interact with socio-economic factors.

For Peru, we reordered six variables, enabling heterogeneous effects analysis for that country only. Additionally, each country has a unique question (Q38). In Peru, this question is included in our strong leader preference outcome index.

2024-11-05

2025-09-30

These outcome variables allow us to gain a deeper understanding of broader societal attitudes and individual behaviors across these themes. We will analyze 13 outcome variables:
1) Anti-elite discontent
2) Broken system – general discontent
3) Perceptions of inequality/meritocracy
4) Strong leader sentiment
5) Belief in liberal democracy
6) Support for democracy
7) Satisfaction with democracy
8) Majoritarian preferences
9) Service dissatisfaction
10) Perception of primary national issues
11) Exit - Disengagement tendencies
12) Voice - Expression of discontent
13) Preferences for redistribution

In our questionnaire, we have included several questions that aim to measure different dimensions of our outcomes of interest. As such, we will aggregate some of these questions into indices by standardizing the variables (using control group statistics) and averaging them. Below, we provide a brief overview of each variable. A comprehensive Excel sheet in the appendix summarizes the outcome variables (indices), their components, the moderators, and our measures of treatment intensity.

1. Anti-elite discontent:
a. Q36.2 Agreement with: "The main division in society is between citizens and economic/political elites"
b. Q36.3 Agreement with: "The country's economy is manipulated to favor the rich and powerful"

2. Systemic disillusionment:
a. Q17.2 Level of anger or frustration regarding Colombia's current situation
b. Q17.3 Level of fear or anxiety regarding Colombia's current situation
c. Q24 Perceived fairness of access to justice in Colombia
d. Q25 Acceptability of inequality levels in Colombia
e. Q27 Perceived determinants of wealth in Colombia
f. Q34 Satisfaction with democratic function in Colombia
g. Q36.3 Agreement with: "The country's economy is manipulated to favor the rich and powerful"

3. Perceptions of inequality/meritocracy:
a. Q25 Acceptability of inequality levels in Colombia
b. Q27 Perceived determinants of wealth in Colombia

4. Strong leader sentiment:
a. Q36.4 Agreement with: "To fix the country, we need a strong leader willing to break the rules"
b. Q45 Justification frequency for law enforcement actions to control protests or demonstrations
c. (Peru only) Q38_new Preference for leaders who prioritize agreements vs. those who defend their positions

5. Belief in liberal democracy:
a. Q33 Perceived essentiality of democratic characteristics:
i. Freedom to criticize the government
ii. Existence of political parties
iii. Congressional checks on presidential power
iv. Regular, fair, and free elections

6. Support for democracy:
a. Q36.1 Agreement with: "Democracy may have problems, but it's better than any other form of government"

7. Satisfaction with democracy:
a. Q34 Satisfaction with democratic function in Colombia

8. Majoritarian preferences:
a. Q33 Perceived essentiality of "always fulfilling the will of the majority" for democracy

9. Service dissatisfaction:
a. Q24 Perceived fairness of access to justice in Colombia
b. Q47 Quality rating of state-provided services (education, healthcare, security, public transport)

10. Perception of primary national issues:
a. Q18 Perceived most serious problem facing the country
b. Q19 Most concerning problem for the individual and their family

11. Disengagement tendencies:
a. Q35 Voting intentions in hypothetical upcoming presidential elections (focus on non-voting)
b. Q49 Primary reason for considering living abroad (excluding those who haven't considered it)

12. Expression of discontent:
a. Q35 Voting intentions in hypothetical upcoming presidential elections (focus on opposition, outsiders, blank votes, and vote annulment)
b. Q44 Justification frequency for various protest actions

13. Preferences for redistribution:
a. Q42 Perceived appropriate recipients of government aid on a 10-step economic ladder
b. Q43 Perceived appropriate starting point for taxation on a 10-step economic ladder

We have embedded a survey experiment in a broader survey to subsamples of national household surveys conducted by the National Statistics Offices (NSOs) in participating countries. The experiment involves three versions of the survey questionnaire (A, B, and C), which contain identical questions but differ in the order in which key sections are presented.
• Survey A begins with questions about insecurity experiences, followed by sections on discontent and concerns; perceptions of inequality, injustice, and discrimination; political preferences and democracy; preferences for redistribution; expressions of discontent; and self-exclusion.
• Survey B starts with questions on economic anxiety and expectations, followed by the same sections as in Survey A.
• Survey C serves as the control group, presenting the sections on insecurity experiences and economic anxiety and expectations at the end of the survey.
We randomize which version individuals receive. This design allows us to measure how priming respondents to think about insecurity or economic anxiety shape their levels of discontent and democratic attitudes.

Detailed Account of Sampling, Survey Procedure, and Fieldwork

1. The Sample

1.1. Requirements for National Statistics Offices (NSOs):
- The survey must use a subsample drawn from the sample of the country’s National Household Survey, including sociodemographic characteristics and total household income. The subsample must be selected from the 2022 or 2023 wave of the National Household Survey and may include unipersonal households if they are randomly selected.
- The subsample must be representative of per capita income quintiles for the urban population, urban centers, or the country’s capital city. Survey respondents must be adults between the ages of 18 and 50, randomly selected from the household members in this age range. The random selection criteria must be verifiable (e.g., selecting the adult in the age range who most recently had a birthday).
- Participation is restricted to adults who perceive their household's socioeconomic conditions (household composition, occupations, and income) as being similar to the last time the household was contacted. This will be verified through a filter question at the beginning of the survey.
- The survey is designed to take approximately 20 minutes to complete.
- Each survey will yield around 2,500 effective observations.

1.2. Measures to Increase Response Rates:
- If the initially selected individual is unavailable, another individual from the same household in the same age range may be selected using the same randomization criteria. A maximum of two replacements is allowed. If the third randomly selected individual does not respond, the household will no longer be part of the sample.
- An individual is considered unavailable under the following circumstances: refusal to answer, absence (for in-person surveys), failure to answer the phone (in phone surveys), or health or disability conditions that prevent participation. Only one individual per household will be surveyed.
- For phone surveys, enumerators are required to make at least six contact attempts before considering an observation "lost." Contact attempts should be made at varying times of the day and on different days of the week, including weekends. An observation will only be classified as "lost" after exhaustive efforts to reach the respondent.
- A protocol will be implemented to minimize non-response rates and ensure quality data collection.

2. Fieldwork
2.1. The NSO receives the initial version of the questionnaire and adjusts it for local language usage. This is followed by a cognitive test and a pilot survey administered to a small number of households.
- The cognitive test ensures that the survey concepts are well understood across different contexts and socioeconomic backgrounds, independent of responses to the pilot survey.
- The pilot survey helps identify issues with survey flow and respondent comprehension.
2.2. Based on the findings of the cognitive test and pilot survey, the questionnaire is refined, and the official survey is launched.
2.3. Fieldwork is expected to last no longer than four weeks per country, though schedules vary based on contract start dates.

3. Current Progress
Colombia and Guatemala are currently in the field implementing the survey. Peru is set to begin in about a week. The exact start date for Ecuador remains unclear. Guatemala and Colombia are expected to complete data collection around mid-December.

This study aims to thoroughly analyze and characterize discontent within the region. We will carefully identify socioeconomic groups exhibiting the highest levels of discontent and those most sensitive to treatment effects. Our analysis will consider 17 sources of heterogeneity:
1. Institutional Trust index 1: (Q30) Respondents rated their trust (1-5 scale) in the following institutions:
* The National Government
* Local Government (municipality/city)
* Congress
* The Judicial Branch (judges and magistrates)
* Political Parties
* Labor Unions
* Media (print, radio, TV, etc.)
* Electoral Process
* National Police
* Armed Forces (Army, Navy, Air Force)
2. Institutional Trust index 2: (Q30) Respondents rated their trust (1-5 scale) in the following institutions:
* The National Government
* Local Government (municipality/city)
* Congress
* The Judicial Branch (judges and magistrates)
* Political Parties
* Electoral Process
* National Police
* Armed Forces (Army, Navy, Air Force)
3. Corruption:
* (Q39) Respondents reported the frequency with which their acquaintances pay bribes or gifts to public officials to obtain services.
* (Q40) Respondents estimated the percentage of individuals involved in corruption within various groups.
4. Ideology (Peru Only): (Q37) Respondents placed themselves on a 0-10 scale (0 = Left, 10 = Right) to indicate their political ideology.
5. Policy Stances (Peru Only): (Q36.5-Q36.8) Respondents indicated their level of agreement (e.g., 1-5 scale) with the following statements:
* Abortion should be legal.
* Same-sex marriage should be legal.
* The healthcare system should be 100% public.
* The state, rather than the private sector, should own the country's most important companies.
6. Social Class:
* (Q6) Respondents answered yes/no to statements indicating (public school attendance, English conversational ability, international travel, public transport usage, time/money spent on entertainment).
* (Q7) Respondents self-identified their social class.
* (National Household Survey): Monetary social class.
7. Education: (National Household Survey): Highest level of schooling attained.
8. Gender
9. Race/Ethnicity
10. Occupation
11. Income: (National Household Survey): Income quintile, poverty line, distance to median income.
12. Formal Employment Status
13. Private vs. Public Service Use: (National Household Survey):
* Private vs. public school attendance.
* Private vs. public healthcare usage.
14. Mismatch between True and Perceived Decile
15. Mismatch: Monetary vs. Non-Monetary Social Class
16. Social Mobility: Parental education vs. respondent's education.
17. Information Exposure:
* (Q31) Frequency of news consumption (news media, social media news, social media political discussion).
* (Q32) Individuals with whom respondents discuss politics.
The treatment primes respondents’ concerns regarding (i) insecurity experiences and (ii) economic anxiety and expectations. These primes will measure the effects on discontent and democratic attitudes. Respondents will answer questions about their levels of concern/perception related to these topics. We will then analyze whether individuals with higher levels of concern exhibit different changes in our outcome variables. We have three options for measuring these treatment variables:
1. Insecurity Index:
* (Q20) Victim of a crime in the past year?
* (Q21) Presence of armed groups in the respondent's neighborhood?
* (Q23) Agreement (1-5 scale) with statements regarding:
* Increased crime/violence in the past year.
* Authorities' inability to protect citizens.
* Authorities' complicity with criminals.
* Migrants as the main cause of insecurity.
* Prevalence of armed individuals.
2. Economic Distress/Pessimism:
* (Q9) Comparison of the current national economic situation to that of a year ago.
* (Q10) Five years ago, what were the respondent's aspirations regarding work opportunities and professional development?
3. Unfulfilled Expectations:
* (Q12) Comparison of current income to income expectations five years ago.
* (Q13) Attainment of various life goals (homeownership, neighborhood/city change, vehicle purchase, home improvements, international travel, university education, debt reduction, savings).
* (Q14) Comparison of accomplishments to expectations.
* (Q15) Comparison of accomplishments to those of most people.
* (Q16) Assessment of the return on investment in education (for respondent and/or children).

Our main specification will be Y_{i,c} = \beta_0 + \beta_1 \text{Version A}_{i,c} + \beta_1 \text{Version B}_{i,c} + \varepsilon_{i,c}, where, $Y_{i,c}$ represents one of the 13 outcome variables. $\text{Version A}_{i,c}$ is a binary variable indicating if the participant was assigned to the first treatment group, with zero representing assignment to the control or second treatment group. Similarly, $\text{Version B}_{i,c}$ indicates assignment to the second treatment group. $\varepsilon_{i,c}$ is the error term. We will estimate the coefficients using OLS.

Additionally, because treatment is assigned at the individual level and we are not using multiple time periods, clustering is unnecessary. Standard robust Eicker-Huber-White errors will be appropriate for our analysis, as discussed in \cite{abadie2023clustering}.

Since we have 13 outcome variables, we will adjust for multiple hypothesis testing, following \cite{shaffer1995multiple}. We do not want to aggregate multiple outcomes into groupings as interpreting these average effects in our case can be problematic. We are more interested in individual outcomes. We will also analyze individual variables separately.

We will leverage the extensive set of individual characteristics, as specified above, in our dataset to analyze heterogeneous effects across different population subgroups and countries. Using interactions, we will create a difference-in-differences (DiD) design to examine whether priming with economic anxiety and security concerns generates different effects across these subgroups, allowing us to assess how discontent and political attitudes vary not only across countries but also within specific demographic markers. We are also interested in assessing the intensity of the treatment through interactions. While some of the data are from our survey, a significant portion of the information comes from household surveys and is observational in nature.

Complete randomization done in office by a computer

Individual (survey respondent)

No

10000 individuals, 2500 in each country is the goal.

10000 individuals

We expect approximately 3333.3 in each of the 3 treatments. But as it is randomized by a computer and it will also depend on phone response rates, this could vary in some degree.

We calculate the Minimum Detectable Effect (MDE) for each type of outcome variable. The MDE represents the smallest true effect size that can be detected with a specified level of power, given the study's sample size and design. We opted for this approach because we have already established a fixed sample size of N=2,500 with each National Statistical Office (NSO), rather than determining the sample size that maximizes statistical power. Our analysis encompasses 25 outcome variables, distributed as follows: - 14 binary variables - 5 ordinal variables (scale: 1 to 5) - 5 ordinal variables (scale: 0 to 10) - 1 count variable To accommodate this diverse set of outcomes, we calculated the MDE using each of the 4 types of outcomes. Furthermore, we simulated various potential distributions for these four types of variables, exploring different combinations of means and standard deviations. This comprehensive approach allows us to assess the study's sensitivity to detect effects across a range of plausible scenarios. Notably, our calculations indicate that our study design has sufficient statistical power to detect even relatively small effect sizes. Our MDEs resulting from the most conservative assumptions are: for binary outcomes is 0.04, for ordinal variables (1 to 5) is 0.12, for ordinal variables (1 to 10) is 0.28, and for count variables is 0.18.