# Conflict, Information Access, and Women's Digital Exclusion

**Airah Balogun** · MS International and Development Economics · University of San Francisco · May 2026

---

## Abstract

Armed conflict has well-documented effects on income, health, education, and labor markets, but its consequences for women's access to information have not been quantified at the individual level across multiple country contexts. This thesis estimates the relationship between conflict exposure and women's access to traditional media and digital information channels using individual-level data from the Demographic and Health Surveys (DHS) linked to geocoded conflict event data from ACLED, covering five countries (Nigeria, the Democratic Republic of Congo, Ethiopia, Jordan, and Tajikistan) across 18 survey waves between 2007 and 2024. A two-way fixed effects difference-in-differences specification exploits within-region variation in five-year log conflict intensity. A one-unit increase in log conflict reduces the probability that a woman has weekly exposure to any of television, radio, or newspaper by **5.8 percentage points** (p < 0.05) and reduces internet use by **16.2 percentage points** (p < 0.01), while mobile-phone ownership is essentially unchanged. The use-versus-ownership asymmetry in digital outcomes, the sharpening of the media effect under a violent-events-only restriction, and the channel-wide deterioration across television, radio, and newspaper jointly point to **infrastructure disruption** as the dominant mechanism. Social connectedness, measured by Meta's Social Connectedness Index, does not moderate the conflict effect. A complementary stacked-sample specification finds that the marginal conflict effect on information access does not differ systematically by gender. The findings extend the empirical conflict literature to the information environment and argue for treating telecommunications and broadcast continuity as a humanitarian priority alongside food, shelter, and medical care.

---

## Key Results

| Outcome | Conflict effect | p-value |
|---------|----------------|---------|
| Weekly media access (TV, radio, or newspaper) | −5.8 pp | < 0.05 |
| Internet use (past 12 months) | −16.2 pp | < 0.01 |
| Mobile phone ownership | ~0 | n.s. |

*A one-unit increase in log five-year cumulative conflict intensity per 100,000 population. TWFE specification with country + wave fixed effects, standard errors clustered by region.*

---

## Data and Methods

- **Survey data**: DHS Individual Recode (IR) files — 18 waves, 5 countries, ~331,000 women (2007–2024)
- **Conflict data**: ACLED geocoded conflict events aggregated to DHS admin-1 regions
- **Population**: WorldPop 100m rasters for per-capita conflict scaling
- **Social connectedness**: Meta Social Connectedness Index (admin-1 level)
- **Identification**: Two-way fixed effects (country + wave) exploiting within-region variation in 5-year log conflict intensity
- **Robustness**: Urban subsample, 1-year conflict window, wild cluster bootstrap, husband-characteristics controls

---

## Countries and Waves

| Country | Waves | Years |
|---------|-------|-------|
| Nigeria | 4 | 2008, 2013, 2018, 2024 |
| DR Congo | 3 | 2007, 2013, 2023 |
| Ethiopia | 2 | 2011, 2016 |
| Jordan | 4 | 2007, 2012/2014, 2017, 2023 |
| Tajikistan | 3 | 2012, 2017, 2023 |
| Yemen | 1 | 2013 (media proxy only) |

---

## Policy Implications

1. **Telecommunications continuity is a humanitarian priority.** Conflict knocks women offline through infrastructure destruction — broadcast and cellular systems should be protected alongside food, shelter, and medical care.
2. **Protect overlapping channels.** The effect runs uniformly across television, radio, and newspaper. Investing in a single resilient channel is insufficient; redundancy across systems is necessary.
3. **Focus on keeping connected women online.** The conflict penalty falls most heavily on internet *use* rather than device ownership, suggesting the policy leverage point is connectivity infrastructure, not device distribution.

---

## Citation

```
Balogun, A. (2026). Conflict, Information Access, and Women's Digital Exclusion.
MS Thesis, Department of Economics, University of San Francisco.
```

---

## Replication Package

All R scripts for data construction, cleaning, and analysis are in this repository. Raw microdata are not included — see data requirements below.

### Data Requirements

| Source | Files needed | Access |
|--------|-------------|--------|
| DHS Program (IR files) | Individual Recode (IR) .DTA files for NG, CD, ET, JO, TJ, YE | Free account at [dhsprogram.com](https://dhsprogram.com/data) |
| DHS Program (MR files) | Men's Recode (MR) .DTA files for NG, CD, ET, JO | Same DHS account |
| ACLED | Country CSVs for NGA, COD, ETH, JOR, TJK | Free account at [acleddata.com](https://acleddata.com) |
| WorldPop | 100m population rasters by country-year | Free at [hub.worldpop.org](https://hub.worldpop.org) |
| Meta SCI | Social Connectedness Index at admin-1 level | [data.humdata.org](https://data.humdata.org) (HDX) |
| NASA Black Marble | VIIRS VNP46A4 annual composites | Free account at [urs.earthdata.nasa.gov](https://urs.earthdata.nasa.gov) (script 33 only) |
| GADM | Admin-1 shapefiles (level 1) | Via `geodata::gadm()` in R (auto-downloaded) |

### Setup

Set the `THESIS_BASE` environment variable to the root of your local data tree before running any script:

```bash
export THESIS_BASE="/path/to/your/project"
```

Or edit the default directly in `config.R`. The expected directory structure is:

```
$THESIS_BASE/
├── data/
│   ├── raw/
│   │   ├── dhs/microdata/         # DHS IR and MR .DTA files
│   │   │   ├── nigeria/
│   │   │   ├── DRC/
│   │   │   ├── ethiopia/
│   │   │   ├── jordan/
│   │   │   ├── tajikistan/
│   │   │   ├── yemen/
│   │   │   └── men/               # MR files for gender analysis
│   │   ├── acled/
│   │   ├── worldpop/
│   │   ├── gadm/
│   │   └── sci/
│   └── clean/                     # auto-created
└── output/
    ├── tables/                    # auto-created
    └── figures/                   # auto-created
```

Script 33 additionally requires a NASA Earthdata bearer token:

```r
Sys.setenv(EARTHDATA_TOKEN = "your_token_here")
```

### Script Pipeline

All scripts source `config.R` at startup. Run in order.

**Data construction**

| Script | Description |
|--------|-------------|
| `02_worldpop_download_extract.R` | Download WorldPop rasters; extract population by admin-1 |
| `05_dhs_process_DRC.R` | Process DRC DHS IR files → standardised format |
| `06_dhs_process_JOR.R` | Process Jordan DHS IR files |
| `07_dhs_process_TJK.R` | Process Tajikistan DHS IR files |
| `08_dhs_process_YEM.R` | Process Yemen DHS IR files |
| `09_dhs_process_NGA.R` | Process Nigeria DHS IR files |
| `10_dhs_process_ETH.R` | Process Ethiopia DHS IR files |
| `11_stack_master.R` | Stack all six country files into one DHS microdata panel |
| `12_acled_aggregate.R` | Aggregate ACLED event-level data to admin-1 × year |
| `13_region_crosswalk.R` | Build ACLED ↔ DHS region name crosswalk |
| `14_conflict_per_capita.R` | Merge ACLED + WorldPop → conflict per capita |
| `15_master_panel.R` | Build admin-1 × wave panel with all covariates |
| `25_conflict_5yr_window.R` | Compute 5-year cumulative conflict exposure (main measure) |
| `26_individual_analysis_data.R` | Build individual-level analysis dataset |
| `27_extract_individual_controls.R` | Extract age, parity, religion, marital status from raw DHS |
| `29_fix_nigeria_sci.R` | Assign Meta SCI values at correct GADM admin-1 level for Nigeria |
| `30_conflict_1yr_window.R` | Compute 1-year conflict window (robustness check) |
| `31b_extract_digital_vars.R` | Extract internet frequency (v171b) and mobile (v169a) variables |
| `34_men_recode_extraction.R` | Extract men's survey data (MR files) for gender comparisons |

**Analysis and output**

| Script | Description |
|--------|-------------|
| `31_information_access_regressions.R` | **Main results**: TWFE regressions — no_media, internet, mobile |
| `32_internet_freq_regressions.R` | Internet frequency analysis |
| `33_viirs_ntl.R` | Nighttime lights extraction and NTL robustness check |
| `35_gender_descriptives_table.R` | Descriptive statistics by gender |
| `36_gender_interaction_internet.R` | Gender × conflict interaction models |
| `37_robustness_urban_wildboot.R` | Urban subsample and wild cluster bootstrap robustness |
| `38_sci_moderation_rerun.R` | SCI moderation of conflict–access relationship |
| `39_thesis_defense_map.R` | Map: conflict exposure across study regions |
| `40_husband_controls_robustness.R` | Robustness check: husband characteristics as controls |

### R Packages

```r
install.packages(c(
  "tidyverse", "haven", "dplyr", "readr", "purrr", "tidyr", "stringr",
  "fixest", "modelsummary",
  "sf", "terra", "exactextractr", "geodata",
  "ggplot2", "ggrepel", "patchwork", "scales",
  "stringdist", "httr", "jsonlite",
  "blackmarbler"   # script 33 only
))
```

### Notes

- Yemen (2013) contributes only media-access observations; the 2013 DHS lacks internet and mobile variables.
- Script 33 (nighttime lights) is a supplementary analysis not required to reproduce the main results.