Starbird, K., Prochaska, S., & Yamron, B. (2025). What is going on? An evidence-frame framework for analyzing online rumors about election integrity. Proceedings of the ACM on Human-Computer Interaction, 9, 1–37. https://doi.org/10.1145/3757522
Summary
This paper reconceptualizes online election misinformation as a problem of framing rather than merely false facts. Adapting Klein et al.’s data-frame theory of sensemaking — and integrating it with framing theory from Entman and Benford & Snow — the authors develop an “evidence-frame” framework to analyze how misleading rumors about election integrity are produced. Through a grounded, mixed-method study of 479 tweets sampled from a 1.8M-tweet corpus gathered during the 2022 Arizona midterm, they show that misleading rumors frequently emerge from factually accurate evidence combined with politically charged frames. The quote-tweet/comment structure of Twitter is treated as an observable site of collective sensemaking, where framing actions transform true evidence into false “rigged election” narratives.
Key Contributions
- Theoretical: Adapts data-frame sensemaking theory to online rumoring, bridging individual sensemaking with collective rumor dynamics via framing theory.
- Methodological: A replicable coding scheme and stratified sampling design for analyzing evidence-frame interactions across quoted tweets and their comments.
- Empirical: A case study documenting how accurate reports of Arizona voting-machine problems were reframed into false fraud narratives.
- Conceptual reframing: Shifts the locus of the misinformation problem from the veracity of isolated claims to the interaction between evidence and political frames.
Methods
The authors used a five-stage grounded, interpretive mixed-method process: exploratory analysis, iterative coding-scheme development, stratified sampling, closed coding by three trained coders, and integrated qualitative/quantitative analysis. Data were collected in real time via the Twitter Streaming API (Nov 8–9, 2022). The coding scheme operationalized rumors along four dimensions — relevance, evidence (asserted vs. referenced), interpretations (mapped to frames), and framing (support/counter, explicit/implicit) — applied separately to quoted tweets and comments. Ten sampling strata (top-20, random-top-100, random-top-500 quoted tweets with top and random comments, plus a 60-tweet random sample) yielded n=479. Reliability was assessed with Krippendorff’s Alpha; comparisons used chi-square tests and one-way ANOVA.
Findings
- ~70% of sampled voting-related tweets concerned election administration, showing integrity dominated discourse over candidates or outcomes.
- Highly quoted tweets disproportionately contained asserted evidence (82% vs. 45% in the random sample), indicating evidence-bearing posts attract amplification and reframing.
- 94% of election-administration tweets carried a frame, but only ~39% framed explicitly; framing was less explicit among highly quoted tweets, consistent with influencers leaving frames implicit.
- The “poor election integrity” meta-frame dominated (78%), with election fraud the most common specific frame; “robust election integrity” framing was rare (12%).
- Comments added, echoed, escalated (17%), or countered (27%) the quoted frames; unframed quoted tweets often attracted comments uniformly supplying a poor-integrity frame — a “call and response” rumoring pattern.
- Framing actions correlated with engagement, with distinct retweet-to-comment ratios depending on whether comments added, escalated, countered, or aligned with frames.
- A “poll worker” video shared by conservative influencers became central evidence repeatedly reframed into fraud narratives despite the originating tweets being largely accurate.
Connections
This work extends Starbird’s broader program on collective sensemaking and rumoring; it pairs closely with Prochaska2025-ef, co-authored by a shared author and grounded in the same election-integrity discourse. Its argument that misinformation resides in framing rather than false facts speaks to work distinguishing misinformation types and audience dynamics such as Marwick2025-ov and Marwick2026-qd, and complements platform-scale studies of election-related content like Pierri2025-hm. It contrasts with veracity- and source-quality-based operationalizations of misinformation, offering a discourse-analytic alternative.