๐ฆ Deconstructing the MeToo Movement and the Blue Wave in the 2018 House Elections
๐ Summary
At the peak of the #MeToo movement in 2018 is the historical performance of women candidates in the 2018 midterm elections.
I examine three ways in which the movement could have an electoral impact using geocoded #MeToo tweets that text analyses show engage core movement discourse and figures, and that geographically predict individual attitudes.
I find no credible evidence that the movement increased Democratic vote shares or turnout.
Instead, I find that in districts with a stronger movement, Democratic women are more likely to challenge their co-partisan male incumbents, a pattern absent in the Republican Party and in 2016.
Hence, the more likely channel is the rise of new Democratic women through a weakening of party norms against same-party challenges.
Shen, Lucas. 2026. "Deconstructing the MeToo Movement and the Blue Wave in the 2018 House Elections." Computational Communication Research 8 (3): 1โ33. https://doi.org/10.5117/CCR2026.3.4.SHEN
๐ Manuscript
๐ Repo Structure and Scripts
.
โโโ build/ # Python data pipeline
โ โโโ 1-downloadTweets/ # raw tweet collection (GetOldTweets-python)
โ โโโ 2-downloadGeolocation/ # Twitter user profile locations (Tweepy)
โ โโโ 3-parseGeolocation/ # parse user locations to US counties
โ โโโ 4-classifyAccounts/ # Botometer bot/human classification
โ โโโ 5-election2018Data/ # 2018 House results + candidate ethnicity
โ โโโ 6-ACS_2015_19/ # ACS 5-year demographics
โ โโโ 7-merge/ # โ panel.dta (master analysis input)
โ โโโ 8-VOTER/ # VOTER survey processing
โ โโโ 9-election2016Data/ # 2016 House placebo data
โ โโโ Makefile # `make data`
โ
โโโ text/ # Python computational text analysis
โ โโโ 0_tweet_content.ipynb # entry point โ gc_human_tweets.parquet
โ โโโ bertopic.ipynb # topic modeling
โ โโโ hashtags.ipynb # hashtag tabulation + ECDF
โ โโโ hashtag-network.ipynb # hashtag co-occurrence network
โ โโโ mentions.ipynb # mention tabulation
โ โโโ mentions-network.ipynb # mention network
โ โโโ urls.ipynb # domain tabulation
โ โโโ ner.ipynb # named-entity counts
โ โโโ zeroshot-classify-corroborate.ipynb # GPT-4o validation vs manual labels
โ โโโ fewshot-unlableled.ipynb # GPT-4o few-shot classification (5000 tweets)
โ โโโ helpers.py
โ
โโโ analysis/ # Stata regression analysis
โ โโโ ado/ # custom Stata commands
โ โโโ data/ # static input CSVs (cawp, tweets-ts)
โ โโโ set_environment.do # env + load panel.dta
โ โโโ preamble.do # data prep + variable definitions
โ โโโ deconstruct.do # orchestrator (all tables/figures)
โ โโโ ... # do files
โ
โโโ ms/ # LaTeX manuscript
โ
โโโ assets/
โ
โโโ Makefile # `make setup` creates venv
โโโ readme.md

Software
- Python 3.10 with
pandas==2.2.3, pyjanitor==0.27.0 โ full list in assets/requirements.txt. Bootstrap with make setup.
- Stata 13+ (15+ for CI-band transparency..)
- Manuscript inspection targets in
ms/Makefile use LSYS/texCheckmate, my shell/Make utilities for LaTeX manuscript hygiene (word counts, acronym tally, repeated-word checks, hardcoded-number finder, textidote, etc.).
๐ Stage 1: Data Pipeline (Python)
| Script |
Output |
build/1-downloadTweets/get-old-tweets.ipynb |
tweets-data/*.csv (raw monthly #metoo tweets, one-time scrape) |
build/2-downloadGeolocation/get-twitteruser-loc.ipynb |
out-userGeolocations/twitter-users-*.csv (Tweepy, one-time) |
build/3-parseGeolocation/ParseGeolocation.ipynb |
parsed-geoloc.csv (county-level user location) |
build/4-classifyAccounts/botometer.ipynb |
bot-classification.csv (Botometer scores) |
build/4-classifyAccounts/botometer-description.ipynb |
botometer_score_distribution.pdf (cited figure) |
build/5-election2018Data/parse-election-data.ipynb |
counties-election-results.csv, politicians.csv |
build/5-election2018Data/get-ethnic.ipynb |
politicians-with-raceindicators.csv |
build/6-ACS_2015_19/prep-census.ipynb |
acs_2015_2019.csv |
build/7-merge/merge.ipynb |
panel.dta (master analysis input) |
build/8-VOTER/get-voter-data.ipynb + processVOTER.ipynb |
voter-individual.csv |
build/9-election2016Data/get-house2016-data.ipynb |
house-2016-stata.csv (2016 placebo) |
Running the pipeline
make data runs the 3 terminal notebooks (7-merge/merge, 8-VOTER/processVOTER, 9-election2016Data/get-house2016-data). Steps 1โ6 are one-time fetches whose outputs are committed โ Twitter, Botometer, and Census API access are not needed to rebuild panel.dta.
๐ Stage 2: Computational Text Analysis (Python)
text/0_tweet_content.ipynb reads raw tweets from build/1-downloadTweets/ and bot scores from build/4-classifyAccounts/, filters to geocoded human tweets, and writes text/gc_human_tweets.parquet. The other 9 notebooks consume that parquet and produce the appendix figures (topic map, hashtag/mention/URL tabulations, networks, NER counts) and the GPT-4o classification artifacts.
Run notebooks individually in Jupyter as needed.
๐ Interactive topic map of #MeToo tweets โ hover for tweet text, zoom into clusters.
๐ Stage 3: Statistical Analysis (Stata)
cd analysis
do deconstruct.do
deconstruct.do calls each producer script in paper-exhibit order, wrapped in preserve/restore. Outputs go directly to ms/tables/ and ms/figures/.
Citations
If this work is useful to you, please consider citing:
Shen, Lucas. 2026. "Deconstructing the MeToo Movement and the Blue Wave in the 2018 House Elections." Computational Communication Research 8 (3): 1โ33. https://doi.org/10.5117/CCR2026.3.4.SHEN
BibTeX:
@article{deconstruct,
author = {Shen, Lucas},
title = {Deconstructing the MeToo Movement and the Blue Wave in the 2018 House Elections},
journal = {Computational Communication Research},
year = {2026},
volume = {8},
number = {3},
pages = {1--33},
doi = {10.5117/CCR2026.3.4.SHEN},
url = {https://www.aup-online.com/content/journals/10.5117/CCR2026.3.4.SHEN},
issn = {2665-9085}
}