Gender and Development

Carolina Torreblanca

University of Pennsylvania

Global Development: Intermediate Topics in Politics, Policy, and Data

PSCI 3200 - Spring 2026

Agenda

  • Development and Gender

  • Chattopadhyay and Duflo, 2004

  • Interaction Models

Upcoming Deadlines

  • Expanded Research Design – April 15, 2026
    • Refined research plan: RQ, hypothesis, design, limitations
    • Submit via Slack by 11:59pm EST
  • Final Project – April 29, 2026
    • Full analysis + professional webpage
    • Submit via Slack by 11:59pm EST

Gender and Representation

Gender and Representation

  • Women hold only 27.0% of legislative seats (IPU, 2023)
  • 23% of cabinet positions (UNWOMEN)
  • Disputed gender gap in turnout, large variance

Gender and Development

Gender and Development

  • Female labor force participation rate: ~47% vs. ~73% male (ILO, 2023)
  • Gender wage gap (Pew)
  • 1/3 of women have experienced GBV
  • Literacy gap: 90% male vs. 83% female (World Bank, 2022)

Gender, Development, and Representation

  • The two are undoubtedly related

  • But could improving gender representation produce development?

From Representation to development

Why might having more women in office improve development?

  • Women may want different things

  • Women may behave differently

  • Women may be better at their job

  • Women representatives may teach about equality

Our Expectations

  • Different outcomes?
    • Do women prioritize different policy domains?
  • Better outcomes?
    • Potential improvements in service delivery and reduced corruption
  • Conditionally different outcomes?
    • Effectiveness depends on institutional context
    • Critical mass may be necessary
    • Interactions with other social identities

Empirical Test?

  • Seems like a good theory with important implications that we want to validate empirically

  • Imagine we have data on share of female legislators and many development outcomes

  • We compare places with female politicians to places with male politicians and see development differences

  • What is the problem with this?

A (more realistic) DAG

Gender Quotas: What are they?

  • Rules requiring minimum women’s representation in politics

  • Goal: Improve gender parity in government

  • Vary by country, region, and level of government

  • Typically set minimum thresholds (20-50%)

Gender Quotas: Types

  • Voluntary quotas:
    • Set by political parties themselves
    • No legal enforcement
  • Mandatory quotas:
    • Required by law
      • Reserved seats:
        • Specific seats only women can hold
      • Candidate quotas:
        • Parties must nominate certain % of women

Gender Quota Adoption

  • Started in 1990s, expanded rapidly

  • Now used in about 130 countries

  • Most common in newer democracies

  • Less common in older democracies (US, Japan)

  • Highest representation achieved in Rwanda (61%)

Chattopadhyay and Duflo, 2004

  • India reserved 1/3 of village council (panchayat) head positions for women (1993)

  • Reserved positions assigned by random lottery

  • Randomization creates natural experiment

  • Solves confounding problem \(\rightarrow\) can identify causal effects!

The Research Question

Does mandated representation of women change policy outcomes?

  • Reserved positions assigned by random lottery
  • Do villages with female heads invest differently than those with male heads?

Theory

Hypothesis: Reserved villages invest more in goods women prioritize

  • Mechanism: Heterogeneous priorities
    • Women and men have different policy preferences
    • Female leaders translate those preferences into policy
  • If true: representation has substantive, not just symbolic, value

Research Design

Treatment: Village council randomly assigned a female head (reserved seat lottery)

  • Identification: Randomization via lottery
    • Reserved vs. non-reserved villages are comparable on all other dimensions
  • Outcome: Public investment decisions (type and quantity)
  • Data: Village councils in West Bengal and Rajasthan, India

Measuring \(D_i\): Gender Differences in Requests

  • Authors surveyed villagers: “Have you made a request to the GP in the past 6 months?”
  • Requests = complaints or petitions for specific public goods (water, roads, irrigation…)
  • For each good \(i\), \(D_i\) = share of women requesting it minus share of men requesting it
  • Positive \(D_i\): women want it more. Negative: men want it more.

A bit more detail


\(Y_{ij} = \beta_1 + \beta_2 * R_j + \beta_3 D_i * R_j + \sum_{l=1}^{N} \beta_l d_{li} + \epsilon_{ij}\)

Where:

  • \(Y_{ij}\) is an outcome of interest

  • \(R_j\) takes the value of 1 if GP is reserved

  • \(D_i\) is gender difference in requests for that good (women - men)

  • \(d_{li}\) are good Fixed Effects!

Main Findings

Interaction Terms

What are Interaction Terms?

  • An interaction term is the product of two variables used as a covariate in a regression model

  • Captures how the effect of one variable depends on the value of another variable

  • Powerful for testing conditional hypotheses  
    Key Modeling Assumption

An increase in \(X\) is associated with an increase in \(Y\) only when a specific condition is met

Recap: OLS Interpretation Basics


\(Y_{i} = \alpha + \beta_1 * X_i + \beta_2 * Z_i + \epsilon_{i}\)

  • \(\hat{\beta_1}\) represents the change in \(Y\) when \(X\) increases by one unit

  • \(\hat{\beta_2}\) represents the change in \(Y\) when \(Z\) increases by one unit

  • Why?

Coefficients as Marginal Changes

A partial derivative of the equation shows how \(Y\) changes when \(X\) changes by 1:


\(Y_{i} = \alpha + \beta_1 * X_i + \beta_2 * Z_i + \epsilon_{i}\)

  • \(\frac{\partial Y}{\partial X} = \beta_1\)

  • \(\frac{\partial Y}{\partial Z} = \beta_2\)

Interaction Effects

  • Idea, the effect of X on Y depends on Z

\(Y = \alpha + \beta_1X + \beta_2Z + \beta_3(X*Z) + \epsilon\)

Think of it as another covariate!

X Z X.Z
0 0 0
1 0 0
0 1 0
1 1 1
  • What needs to happen for the new covariate to be 1?

New model


\(Y_{i} = \alpha + \beta_1 * X_i + \beta_2 * Z_i + \beta_3 * Z_i * X_i + \epsilon_{i}\)

  • How to interpret \(\beta_3\): Expected change in Y when the product of \(X \times Z\) increases by 1 unit.

  • How to interpret \(\beta_1\): Change in Y when X increases by 1 unit AND \(Z =0\)

  • How to interpret \(\beta_2\): Change in Y when X increases by 1 unit AND \(X =0\)

Back to Math!

  Why does the interpretation of \(\beta_1\) and \(\beta_2\) change?

\(Y_{i} = \alpha + \beta_1 * X_i + \beta_2 * Z_i + \beta_3 * Z_i * X_i \epsilon_{i}\)

  • \(\frac{\partial Y}{\partial X} = \beta_1 + \beta_3*Z_i\)

  • \(\frac{\partial Y}{\partial Z} = \beta_2 + \beta_3*X_i\)

Three Simple Rules for Interaction

  1. Use for conditional hypotheses

  2. Always include all constituent terms

  3. Interpret \(\beta_1\) and \(\beta_2\) correctly

Marginal Effects: The Idea

Recall:

\[\frac{\partial Y}{\partial X} = \beta_1 + \beta_3 \cdot Z\]

  • The effect of \(X\) on \(Y\) is not a single number – it depends on \(Z\)
  • marginaleffects evaluates this at specified values of \(Z\) and computes uncertainty

Two cases: \(Z\) is binary vs. \(Z\) is continuous

Binary Moderator: Two Estimates

# Effect of women's representation on policy outcome, by quota status
slopes(model_binary,
       variables = "women_rep",
       by = "quota")

      Term quota Estimate Std. Error    z Pr(>|z|)     S  2.5 % 97.5 %
 women_rep     0   -0.547      0.177 -3.1  0.00194   9.0 -0.893 -0.201
 women_rep     1    2.351      0.185 12.7  < 0.001 120.6  1.989  2.714

Type:  response 
Columns: term, quota, estimate, std.error, statistic, p.value, s.value, conf.low, conf.high

Binary Moderator: Plot

Continuous Moderator: A Line, Not Two Points

When \(Z\) is continuous, the marginal effect varies across the full range of \(Z\)

Back to C&D logic: as \(D_i\) increases (women want a good more relative to men), does the effect of reservation on investment grow?

Continuous Moderator: Plot

Show code 
plot_slopes(model_cont,
            variables = "reservation",
            condition = "pref_gap") +
  geom_hline(yintercept = 0, linetype = "dashed", color = "grey40") +
  labs(x = expression(D[i] ~ "(gender gap in requests: women - men)"),
       y = "Effect of GP reservation\non public investment",
       caption = "Simulated data") +
  theme_minimal(base_size = 16)

What to Look For

Binary moderator - Two point estimates with CIs – does the effect change sign or magnitude?

Continuous moderator - A line with CI band – where does the effect become significant? Where does it flip?

Both are \(\beta_1 + \beta_3 \cdot Z\) evaluated at different values of \(Z\) – marginaleffects handles the standard errors