The data

People

The people are faculty at the top 150 US research universities, collected for Angeli and Lowe, “Virtue Signals”. Their names, departments and Twitter handles were collected by hand, department by department, by research assistants and student volunteers, with quality checks on each department. Adjunct, clinical and teaching faculty, postdocs, emeriti, visitors, administrators, people on leave, people who had died or left the US, and lab accounts were removed.

A joiner is a faculty member with a Twitter handle. Handles were matched to account profiles collected on 10 May 2022; the entry year is the calendar year the account was created. Joiners who opened their accounts between 2007 and 2020 are kept. 2006 is dropped because the platform launched in the middle of that year. A never adopter is a faculty member in the same departments with no handle.

Publication records

Each person was linked to an OpenAlex author profile by searching their name within their university and scoring candidates on name similarity, search relevance, and how closely the profile's research topics match the person's department. Each person keeps their best-scoring profile, each profile is used once, and the lowest-scoring 4 percent of links are dropped. Works published before 1966, and works before a gap of more than five years in a record before 2005, are dropped as likely belonging to someone else with the same name.

The panel runs from 2001 to 2020, one row per person and year. It starts in 2001 because citations are counted from that year.

Outcomes

• Citations: citations received in the year by all of the person's works, whenever those works were published.
• Publications: works published in the year.
• Weighted publications: works published in the year, each weighted by its venue's two-year mean citedness.
• Distinct coauthors: different people the person published with in the year. New coauthors counts those not seen in any earlier year; distinct coauthor slots sums, over the year's works, the number of other authors on each.
• Disciplinary breadth: entropy of the fields of the year's works.
• New topics and new affiliations: research topics and institutions appearing on the person's works for the first time.
• Citations to pre- and post-adoption papers: the year's citations split by whether the cited work was published before the entry year or from it onward.

PhD years

PhD years were collected separately from OpenAlex, from CVs, faculty pages, dissertation repositories and similar public sources, with the source archived for each person and a confidence score for whether the source is the right person. The pages use a PhD year when that confidence is at least 0.7. If no PhD is recorded, a terminal doctorate counts (DPhil, ScD, EdD, MD, JD, DVM, PharmD, DBA, MFA and combined degrees). People without a usable year do not appear on the raw-means page.

Samples

All joiners

Every joiner who entered between 2007 and 2020: 22,791 people, 5,242 of them with a usable PhD year.

Never adopters

Every never adopter: 71,363 people, 25,422 of them with a usable PhD year. On the raw-means page, never adopters are compared with joiners in the same PhD decade and calendar year, and averaged so that each entry year and PhD decade gets the weight of its joiners.

Not-yet joiners

Joiners from later entry years, observed only in years before they joined. They are drawn from the same 22,791 joiners.

PhD-matched sample

2,975 joiners and 4,318 never adopters (2,940 and 4,285 with a PhD year at the 0.7 confidence floor). Each joiner is paired with up to three never adopters who share their OpenAlex field, whose total citations in the four years before the joiner's entry are within 5 percent of the joiner's, and whose PhD year is within one year of the joiner's. Controls inherit their joiner's entry year. A never adopter can serve as a control for several joiners. On the raw-means page each control is weighted by one over the number of controls in its joiner's cell, so each matched cell counts once on each side.

Headline matched sample (event studies)

5,884 joiners and 9,629 never adopters. Joiners are grouped into cells by entry year and first-publication year, keeping those whose first publication came at least three years before entry. Each joiner is matched to up to three never adopters in the same cell and OpenAlex field whose citations in each of the four years before entry are within 25 percent of the joiner's (or within two citations) and whose citation slope over those years is similar. One of the main-figure series drops each person's years before their first publication.

Joiners with no comparable never adopter (event studies)

The 16,907 joiners outside the headline matched sample, compared with joiners who had not yet entered.

Splits

• Gender and ethnicity are recorded in the faculty data for joiners. Under-represented minority means Black or Hispanic, compared with Caucasian, East Asian or South Asian; other and uncertain categories are left out.
• Field is the OpenAlex field most common among the person's works before entry, grouped into life, health, physical and social sciences.
• Pre-entry citations are quartiles of mean yearly citations in the four years before entry, among joiners with a PhD year.
• Twitter account type classifies each joiner's original tweets from January 2020 to March 2022 (up to 100 per account) into research, academic life (teaching, students, conferences, jobs), politics and policy, and personal (everything else, including replies and thanks). The classifier was trained on 2,000 hand-labelled tweets and gets 78 percent of 500 held-out tweets right. An account takes a topic's name when more than 40 percent of its tweets fall in it, and is mixed otherwise. Joiners with no original tweets from those years in the data form their own group. Accounts typed research or academic tweet far less than those typed personal or politics (a median of 27–41 original tweets over the period against 155–180), so the types partly reflect how much people tweet. The tweets come years after most people joined.
• Department groups (event studies) group departments into STEM, life and environmental sciences, health sciences, social sciences, business and professional, and humanities and arts.
• Platform use (event studies) comes from the accounts' tweets and profiles: followers (on a log scale), the shares of tweets that promote the person's own work, that are political, and that are replies, and overall tweeting volume. These are measured around 2020–22, after most people joined.

Matched controls take their joiner's split value. Never adopters have no recorded gender or ethnicity and no entry year, so they appear only in the field split.

Values

The raw-means page shows plain means. The event-study estimates winsorize outcomes at the 1st and 99th percentiles (2nd and 98th for coauthor counts).

Who is in each sample

People with a usable PhD year, as on the raw-means page. Means are per person-year; pre-entry means cover the four years before entry.