Historical segregation, neighborhood social capital, and human capital accumulation

Participants are randomly assigned to one of two arms:

TREATMENT (Map-First): An interactive map-drawing exercise is administered BEFORE the main outcome elicitation. Participants view a satellite map centered on their household and draw the boundary of their neighborhood (colonia/barrio), then briefly explain their boundary choice.

CONTROL (Map-After): The identical map-drawing exercise is administered AFTER the main outcome elicitation.

Both arms complete the same modules; only the timing of the map-drawing exercise varies. This design tests whether making neighborhood boundaries cognitively salient affects educational investment decisions.

Randomization occurs at the household level with 1:1 allocation, stratified by geographic location (Analco vs Poniente neighborhoods).

2026-05-09

2026-08-15

The primary outcome is willingness to pay for educational investment, measured through a Becker-DeGroot-Marschak (BDM) incentive-compatible mechanism using a Multiple Price List (MPL) design.

Primary outcome variables:
1. Count of rows (out of 11) where respondent chose one hour of online math tutoring over an Amazon gift card (range: 0-11)

2. Willingness to pay estimate (in MXN pesos) constructed following the method in Jack, McDermott, & Sautmann (2022), which accounts for randomization of MPL rows and columns

Jack, B. K., McDermott, K., & Sautmann, A. (2022). Multiple price lists for willingness to pay elicitation. Journal of Development Economics, 159, 102977. https://doi.org/10.1016/J.JDEVECO.2022.102977

Secondary outcomes are organized into three categories:

1. NARRATIVE MECHANISMS (open-ended responses):
- Coded reasons for why young people in the neighborhood stop studying (open-ended question)
- Categories coded from responses: economic necessity, family obligations, perceived low returns, peer influences, neighborhood-specific factors, institutional barriers

2. MODEL PRIMITIVES:

Neighborhood orientation:
- Standardized trust in neighbors minus average trust in family, coworkers, and strangers (continuous, standardized)
- Agreement with "I feel part of this colonia/barrio" (1-7 Likert scale)
- Amount allocated to in-group in disinterested dictator game (0-500 MXN)

Network benefit:
- Which family receives more help from neighbors in emergency - Martínez (stays local) vs Ramírez (leaves for university)? (binary: Martínez=1, Ramírez=0)
- Number of times in past year received help from neighbors in emergencies/loans/job-finding (count, 0-3)

Time/security substitution:
- Typical hours per week spent in neighborhood activities (continuous)

Conformity costs:
- trajectory_social_cost_vignette: Which trajectory creates more difficulty maintaining good relations with neighbors - Martínez (stays) vs Ramírez (leaves)? (binary: Ramírez=1, Martínez=0)
- perceived_peer_disapproval: Expected negative reaction from neighbors if child pursues university away from neighborhood (Likert scale)

3. ALTERNATIVE MECHANISMS:

Intergenerational Educational Mobility:
- Mother and father educational levels

Returns to education (perceived):
- Difference in expected age-30 income for university graduate vs high school graduate from respondent's neighborhood (continuous, MXN)
- Minimum income that would justify cost of university (continuous, MXN)

Beliefs and norms:
- "Of every 10 families in your neighborhood, how many would choose tutoring over 100 MXN gift card?" (0-10)
- Personal agreement with "Young people should work after high school rather than continue studying" (1-7 Likert)
- Belief about share of neighbors who would agree with above statement (proportion)

Aspirations:
- Highest educational level parent wants focal child to complete (categorical)
- Highest level parent realistically expects focal child to complete (categorical)

NARRATIVE MECHANISMS:
The open-ended question "Why do you think young people in this neighborhood sometimes don't continue studying after high school?" will be coded by two independent coders following a pre-specified coding scheme. Categories include economic necessity, family obligations, low perceived returns, peer influences, neighborhood-specific factors (safety, norms, opportunities), and institutional barriers. Inter-rater reliability will be assessed via Cohen's kappa; disagreements resolved through discussion.

MODEL PRIMITIVES:
These measures operationalize the parameters in the theoretical model. θ (neighborhood orientation) captures the share/intensity of bonding social capital through trust premiums, identification, and parochial preferences. δ (substitution) captures opportunity cost of education through time spent in neighborhood networks and benefits received from those networks. κ (conformity) captures social penalties for deviating from local educational norms, elicited through vignettes and direct questions about peer reactions.

Vignette structure: Respondents hear about two families (Martínez and Ramírez). The Martínez son completes high school, works near the neighborhood, remains active in local life. The Ramírez son completes high school, attends university across town, gradually disengages from the neighborhood. Respondents indicate which family would receive more help from neighbors in an emergency, and which trajectory would create more difficulty maintaining good neighborhood relations.

ALTERNATIVE MECHANISMS:
Returns to education: Subjective wage premium elicited by asking expected monthly income at age 30 for two hypothetical young people from respondent's neighborhood—one who completed only high school, one who completed university. Minimum income threshold asks: "What is the minimum monthly income at age 30 that would make university worth the cost?"

Beliefs: First-order beliefs are respondent's own choices and views. Second-order beliefs are perceptions of what neighbors think/do. The peer tutoring belief is elicited after the BDM ("You chose tutoring in X rows. Of every 10 families in your neighborhood, how many do you think would choose tutoring when the gift card is 100 pesos?")

Aspirations vs expectations: Following the literature on aspiration gaps, we distinguish what parents want (aspiration) from what they think will happen (expectation). The gap identifies perceived constraints on educational mobility.

This is a two-arm randomized controlled trial testing whether making neighborhood boundaries cognitively salient affects parents' willingness to invest in their children's education.

Participants complete a survey including: (1) an interactive map-drawing exercise in which they trace their neighborhood boundary, and (2) a Becker-DeGroot-Marschak mechanism eliciting willingness to pay for educational tutoring.

Treatment arm: Map-drawing occurs BEFORE the educational investment elicitation
Control arm: Map-drawing occurs AFTER the educational investment elicitation

The design tests whether activating awareness of neighborhood boundaries reduces revealed willingness to pay for tutoring, and whether this effect differs by geographic location and by individual neighborhood orientation.

Target sample: ~250 parents/caregivers with children aged 6-20, recruited from neighborhoods near a historical boundary in Guadalajara, Mexico.

Not available

Computer-based randomization within survey software (SurveyToGo) at survey start, before any modules are administered.

Fair binary draw assigns each household to Map-First (treatment) or Map-After (control) with equal probability using SurveyToGo's built-in random number generator.

Assignment revealed to enumerator only when software prompts the corresponding module sequence.

Stratification implicit through geographic quotas: 180 households in Analco, 70 in Poniente. Within each area, randomization is unconstrained.

Individual household. Each household receives an independent random assignment to treatment or control with no further clustering.

No

Not applicable. Randomization occurs at the household level with no clustering. The design uses geographic stratification (Analco vs Poniente) but treatment is assigned individually to each household, not at the cluster level.

~ 250 households total - ~200 households in Analco (eastern side of historical boundary) - ~50 households in Poniente (western side of historical boundary)

Total n=250 households with 1:1 randomization to Map-First vs Map-After:

ANALCO (n=200):
- Treatment (Map-First): 100 households
- Control (Map-After): 100 households

PONIENTE (n=50):
- Treatment (Map-First): 25 households
- Control (Map-After): 25 households

OVERALL (n=250):
- Treatment (Map-First): 125 households
- Control (Map-After): 125 households

With n=250 (125 per arm), α=0.05 two-tailed, and 80% power: MINIMUM DETECTABLE EFFECTS (80% power): - Overall sample (n=250, 125 per arm): MDE = Cohen's d = 0.36 - Analco subsample (n=200, 100 per arm): MDE = Cohen's d = 0.40 - Poniente subsample (n=50, 25 per arm): MDE = Cohen's d = 0.57 POWER AT SPECIFIC EFFECT SIZES: The study is designed to test H1 (main treatment effect) in the Analco subsample where theory predicts the effect is concentrated. Based on pilot data (overall d=0.68, Analco d=1.27), we calculate power under two scenarios: Conservative scenario (d=0.50, representing 39% of pooled pilot estimate): - Analco subsample (n=200): 79% power - Overall sample (n=250): 97% power Pilot-based scenario (d=0.68, overall pilot estimate): - Analco subsample (n=200): 90% power - Overall sample (n=250): 99% power The unbalanced geographic allocation (80% Analco, 20% Poniente) maximizes power for H1 while retaining sufficient power to test H2 (geographic heterogeneity via Treat × Analco interaction) and H3 (individual heterogeneity via Treat × θ interaction in Analco subsample). Minimum Detectable Effect (MDE) = Cohen's d = 0.36 for overall sample For primary hypothesis H1 tested in Analco subsample (n=180, 90 per arm): MDE = Cohen's d = 0.42 The study is powered at 96% to detect d=0.5 for H1, which represents a conservative scenario (39% of pooled pilot estimate d=1.27, p=0.003).