# SCCC codebook

Scaled and Classified Congressional Communication (SCCC), public release.
Source article: Kistner, Heseltine, Alvarez, Fitch, Lothamer and Simas (2026), *American Political Science Review*, doi:10.1017/S0003055426101841.

The public files contain **no message text**. Each row carries a platform identifier so that researchers with their own API access can retrieve the underlying message.

## Files

| File | Unit of observation | Notes |
|---|---|---|
| `twitter_<congress>.csv.gz` / `.parquet` | one tweet | 111th–117th Congress |
| `facebook_<congress>.csv.gz` / `.parquet` | one Facebook post | 111th–117th Congress |
| `newsletters_<congress>.csv.gz` / `.parquet` | one email newsletter | 111th–117th Congress; categories and scores rolled up from the sentence level (see below) |
| `newsletter_sentences_<congress>.csv.gz` / `.parquet` | one sentence bigram from an email newsletter | the unit the classifiers and scaling model operated on; several rows per newsletter |
| `members.csv` | one member × Congress | member attributes plus message counts, category proportions and mean partisanship (pooled and per platform) — the simplest file for member-level analysis |
| `summary.csv` | one member × Congress × platform | the same measures in long format, plus mean engagement; the table behind the website explorer |

Congress numbers map to years as follows: 111 = 2009–10, 112 = 2011–12, 113 = 2013–14, 114 = 2015–16, 115 = 2017–18, 116 = 2019–20, 117 = 2021–22. Data for the 111th Congress do not cover the full session.

## Message-level variables

| Variable | Type | Platforms | Description |
|---|---|---|---|
| `TweetID` | integer | Twitter | Tweet status ID. |
| `Handle` | string | Twitter | Account handle at time of collection (official, campaign or personal account of the member). |
| `PostURL` | string | Facebook | URL of the post on facebook.com. |
| `NewsletterID` | integer | Newsletters | Newsletter identifier (DCinbox). |
| `Sentences` | integer | Newsletters (newsletter level) | Number of sentence-bigram units in the newsletter. |
| `BigramNumber` | integer | Newsletters (sentence level) | Position of the sentence bigram within the newsletter, starting at 1. |
| `Date` | YYYY-MM-DD | all | Date the message was posted/sent. |
| `Congress` | integer | all | Congressional session (111–117). |
| `MemberICPSR` | integer | all | ICPSR legislator ID (as used by Voteview). Join to `members.csv`. |
| `Likes`, `Retweets`, `Replies`, `Quotes` | integer | Twitter | Engagement counts at time of collection. |
| `Likes`, `Shares`, `Comments` | integer | Facebook | Engagement counts at time of collection (CrowdTangle). Early-period posts may show zeros where counts were unavailable. |
| `Advertising` | 0/1 | all | Effort to disseminate the member's name in a favourable way (Mayhew 1974). |
| `CreditClaiming` | 0/1 | all | Claims personal responsibility for a government action or outcome (Mayhew 1974). |
| `PositionTaking` | 0/1 | all | Judgmental statement on a matter of interest to political actors (Mayhew 1974). |
| `ConstituentService` | 0/1 | all | Presentation of self in the district or allocation of resources to constituents (Fenno 1978). |
| `NegativePartisan` | 0/1 | all | Attack on the policies or politicians of the opposing party (Russell 2018). |
| `Bipartisan` | 0/1 | all | Advocates the value of bipartisan collaboration (Russell 2018). |
| `CreditConstituent` | 0/1 | all | Subtype of credit claiming. **[Authors: add definition.]** |
| `CreditPolicy` | 0/1 | all | Subtype of credit claiming. **[Authors: add definition.]** |
| `PartisanScore` | float, −1 to 1 | all | Text Partisanship Score: 2πᵣ − 1 from a class affinity model (Perry and Benoit 2017) fit separately by Congress. −1 = most Democratic language, 0 = nonpartisan, +1 = most Republican. Missing where the model was not fit (e.g. messages dated outside the session's scaling window). |
| `PartisanExtremity` | float, 0 to 1 | all | Absolute value of `PartisanScore`. |

Categories are neither mutually exclusive nor exhaustive; roughly 28% of tweets, posts and newsletter sentences fall in none of the six main categories.

**Newsletter roll-up.** Classification and scaling were performed on sentence bigrams. In the newsletter-level files, a category is 1 if *any* sentence in the newsletter was labelled 1, and `PartisanScore` / `PartisanExtremity` are the means across the newsletter's sentences. Because newsletters are long, newsletter-level category shares are much higher than tweet or post shares and are not directly comparable to them. The member-level tables, the explorer and the article all use the sentence level for newsletters, so their newsletter figures are shares of sentences and means over sentences.

Classifier labels come from BERTweet models trained on approximately 43,000 hand-coded messages. Out-of-sample macro F1 by category: 0.83–0.95 (Twitter), 0.81–0.92 (Facebook), 0.77–0.93 (newsletters). See Section G–H of the article's supplementary material.

## `members.csv`

| Variable | Description |
|---|---|
| `MemberICPSR` | ICPSR legislator ID. |
| `Congress` | Congressional session. |
| `MemberName` | Name as recorded in the source data. |
| `MemberParty` | Democrat, Republican or Other. |
| `MemberChamber` | House or Senate. |
| `MemberState` | Two-letter state abbreviation. |
| `MemberDistrict` | House district number, or `S` for senators. |
| `Source` | `voteview` if attributes were taken from Voteview for that member-session; `messages` if inferred from the internal files. |

The remaining columns summarise that member-session's messages. Each appears four times: without a suffix (pooled across all three platforms) and with the suffixes `Twt` (tweets), `FB` (Facebook posts) and `NL` (newsletter sentences). Suffixed columns are blank where the member has no messages on that platform.

| Variable | Description |
|---|---|
| `NumMessages` | Number of tweets, posts and/or newsletter sentences. |
| `NumNewslettersNL` | Number of distinct newsletters (NL block only). |
| `PartisanScore` | Mean Text Partisanship Score over scored messages. |
| `PartisanExtremity` | Mean Text Partisan Extremity Score over scored messages. |
| `NumAdvertising`, `NumCreditClaiming`, `NumPositionTaking`, `NumConstituentService`, `NumNegPartisan`, `NumBipartisan`, `NumCreditConstituent`, `NumCreditPolicy` | Number of messages labelled 1 for the category. |
| `PropAdvertising`, …, `PropCreditPolicy` | The corresponding share of messages (`Num… / NumMessages`). |

For example, `PropNegPartisanTwt` is the share of a member's tweets in that Congress containing a negative partisan attack, and `PartisanScore` (no suffix) is the mean partisan score across all of their tweets, posts and newsletter sentences.

## `summary.csv`

One row per `Platform` × `Congress` × `MemberICPSR`, with the member attributes above plus:

| Variable | Description |
|---|---|
| `n_messages` | Number of tweets, posts or newsletter sentences. |
| `n_newsletters` | Number of distinct newsletters (Newsletters rows only). |
| `n_scored` | Number of rows with a non-missing `PartisanScore`. |
| `mean_partisan_score`, `mean_partisan_extremity` | Means over scored rows. |
| `p_<Category>` | Share of rows labelled 1 for that category. |
| `mean_likes`, `mean_shares`, `mean_replies`, `mean_quotes` | Mean engagement per message (social media only; `Retweets`→`shares`, `Comments`→`replies`). |

To aggregate across rows, weight proportions by `n_messages` and scores by `n_scored`.

## Known limitations

- Twitter and Facebook content from accounts deleted before collection is absent; the 2009–2017 counts are lower bounds.
- Facebook engagement counts for the earliest years are frequently zero because CrowdTangle did not return them.
- Some member-sessions lack a name or party in the source data; these show as blank or `Other`. Where Voteview attributes are available (`Source = voteview`) they should be preferred.
- Newsletter-level category flags use an any-sentence rule, so they rise with newsletter length; sentence-level shares are in `newsletter_sentences_*`.

## Licence

Data: CC BY 4.0. Please cite both the article and the Dataverse record (doi:10.7910/DVN/19MIBB).