PaperlayerView exact revision on arXiv

Source-linked AI summary

Double-Edged Sword of Mediated Visibility: How Visual Framing Undermines Congresswomen's Perceived Competence

Bryce J. Dietrich, Hyein Ko, Myriam Shiran

arXiv:2609.31970v1cs.CV

TL;DR

Research on congresswomen’s television visibility has largely measured who is mentioned rather than how they appear on screen. Using facial recognition and two pre-registered experiments, the paper finds that visual framing and verbal register have distinct gendered consequences for perceived competence and warmth.

  • Problem

    Prior research largely measures whether congresswomen are named or mentioned, leaving limited evidence about who appears on screen and how they are visually framed.

    Source: §1 · p. 2

  • Method

    The paper uses facial recognition to analyze CNN and Fox News broadcasts and two pre-registered experiments to test responses to visual framing and verbal register.

    Source: §1 · p. 2, §4.2 · p. 13

  • Results

    Congresswomen appear disproportionately in split-screens; split-screen lowers their perceived competence in static frames but not congressmen’s, while outraged language reduces their warmth roughly three times as much as men’s.

    Source: §1 · p. 3, §6 · p. 31

  • Takeaways & Limitations

    The findings suggest that how congresswomen are shown, not only how often they appear, can work against the symbolic gains their visibility is meant to produce.

    Source: §1 · p. 4

    The framing account distinguishes static excerpts from dynamic appearances and treats visual composition and emotional register as separate factors in its expectations.

    Source: §3 · pp. 9, 11

Abstract

from arXiv · show

Women's presence in Congress has stalled below 30%. That institutional underrepresentation is mirrored by their limited visibility on televised news, where visual presentation can shape perceived competence and authority. Yet research tells us more about whether congresswomen appear than how they are visually framed. Facial recognition analysis of 695,464 image-text segments from CNN and Fox News (2011-2021) reveals that congresswomen appear disproportionately in split-screen rather than solo shots. Two pre-registered experiments with 6,220 participants show that, in static frames, split-screen framing reduces congresswomen's perceived political competence, but not congressmen's. In dynamic videos, the competence penalty disappears; instead, outraged language reduces a congresswoman's perceived warmth roughly three times as much as a congressman's. Suggestive evidence (p = .057) indicates that women viewers report greater external political efficacy after watching a congresswoman appear alone. We conclude that visual framing shapes both congresswomen's mediated visibility and their perceived capabilities.

1 Introduction

The paper asks not only whether congresswomen appear on television, but how they are shown and what those frames mean for viewers. It finds that split-screen visibility is common and can carry different costs depending on whether viewers see a still frame or a dynamic exchange.

1 Introduction

  • Congresswomen appear on cable news more often than their share of seats predicts, but disproportionately in split-screens, typically beside male colleagues.The authors argue that the disadvantage lies less in how often congresswomen appear than in how they are shown.Source: §1 · pp. 3–4
  • In static frames, split-screen presentation lowers perceived competence for congresswomen, with no similar effect for congressmen.In video, the competence penalty is no longer detected, suggesting that what a legislator says can override how she is framed.Source: §1 · p. 3
  • In video, outraged rather than measured remarks reduce a congresswoman’s perceived warmth roughly three times as much as a congressman’s.The authors also report that the screen arrangement neither produced nor detectably altered this cost.Source: §1 · p. 3
  • Women viewers who watched a congresswoman appear alone reported greater external political efficacy, but the evidence is suggestive (p = .057).An exploratory analysis suggests the movement comes from solo appearances delivered in a measured register.Source: §1 · p. 3
  • The authors conclude that routinely pairing congresswomen with men, and the competence cost of still-frame appearances, can work against the symbolic gains of visibility.The solo frame is also the condition in which congresswomen remain furthest from parity with male colleagues.Source: §1 · p. 4

2 Literature Review

The literature review distinguishes being visible from being recognized in a particular visual form, and connects both to symbolic representation. It reviews how split-screen and solo compositions shape viewers’ interpretations, while identifying distinct gendered penalties tied to visual form and verbal register.

2.1 Mediated Visibility, Television, & Symbolic Representation

  • Symbolic representation concerns how women’s presence in power shapes attitudes toward competence and women’s sense of political efficacy, but evidence of these effects in the United States is mixed.The literature suggests these effects depend on citizens seeing and registering women’s representation, often through mediated exposure.Source: §2.1 · p. 4
  • Symbolic effects involve two partly independent dimensions: judgments of a politician’s capability and women viewers’ beliefs that government can respond to people like them.Competence judgments may rely on compositional cues when performance evidence is limited; efficacy is tied to recognition of the congresswoman as a person speaking to viewers.Source: §2.1 · p. 5
  • Most television studies establish whether congresswomen appeared, not what viewers saw; much U.S. evidence centers on Sunday programming, while still photographs show visual gender stereotyping.Text-based visibility measures cannot establish on-screen appearance or framing.Source: §2.1 · pp. 5–6

2.2 Split-Screen as a Style of Visibility & Its Consequences

  • Split-screens can heighten perceived conflict, while solo shots can foreground a politician; prior debate research found Clinton rated higher on competence-related traits when shown alone.Cable-news exchanges may be hostile or cordial, so dynamic viewing can provide evidence against what a divided frame implies, unlike a static frame.Source: §2.2 · pp. 7–8
  • A divided frame may lead viewers to read a congresswoman’s participation as contested, particularly when brief or detached viewing leaves the frame as their main account of the exchange.Separately, agentic penalties attach to assertive language and can reduce women’s perceived warmth; the review distinguishes a visual-form double standard from a verbal-register double bind.Source: §2.2 · pp. 8–9

3 Theoretical Expectations

The theoretical expectations distinguish penalties arising from visual composition from those tied to a speaker’s emotional register. They also ask whether these effects depend on static or dynamic viewing and whether solo appearances affect women viewers’ external efficacy.

3 Theoretical Expectations

  • The account expects solo shots to foreground congresswomen and project authority, while split-screens may make their authority seem contested.The first hypothesis predicts that congresswomen in solo shots will be perceived as more competent than those in split-screens.Source: §3 · pp. 9–10
  • The predicted competence effect of composition should be strongest in static excerpts and smallest when viewers can observe a politician’s performance over time.Whether the split-screen penalty carries into dynamic viewing is presented as an open question.Source: §3 · p. 10
  • The authors expect women viewers to report higher external political efficacy after seeing a congresswoman alone rather than in a split-screen.This predicted benefit should be clearest in dynamic appearances, where viewers can recognize a person speaking rather than a figure arranged within a shot.Source: §3 · p. 10
  • The contested-advocacy account predicts that outraged words will compound split-screen competence costs by making a congresswoman’s claims appear more conflictual.Under this account, forceful words that count as advocacy in a solo shot may count as evidence of conflict in a split-screen.Source: §3 · p. 11
  • The rival agentic-backlash account predicts that identical outrage will reduce a congresswoman’s warmth more than a congressman’s, regardless of screen arrangement.The experiments cross member gender with visual format and emotional register, and measure perceived contestation and warmth to distinguish the proposed processes.Source: §3 · p. 11

4 Data and Methods

The study uses archived CNN and Fox News broadcasts to measure congressional members’ on-screen visibility, then applies a facial-recognition approach designed for limited training images. The archive window spans six Congresses and multiple political contexts.

4.1 TV News Archive Data

  • The archive analysis covers CNN and Fox News broadcasts from May 2011 to March 2021, spanning six Congresses and 953 members.The period includes 374,643 CNN and 320,821 Fox News image-text segments, each corresponding to one minute of broadcast.Source: §4.1 · p. 12

4.2 Small-Sample Facial Recognition Using Siamese Neural Networks

  • Facial recognition identifies which members of Congress appear on screen, addressing limitations of mention-based measures that cannot distinguish physical appearance from reference.The custom Siamese Neural Network learns from image pairs, reducing the number of training examples needed when images are scarce across many legislator categories.Source: §4.2 · p. 13

Appendix presents the architecture and a worked example of the output.

The paper identifies legislators and classifies split-screen frames using calibrated facial similarity and a CLIP-based vision-language pipeline. It validates the face matching and checks the resulting patterns against transcript-based evidence, while noting that transcripts cannot establish on-screen appearance or composition.

Appendix presents the architecture and a worked example of the output.

  • The 500-epoch training example gives Pelosi’s self-comparison a contrastive loss of 0.00032 and her comparison with Emma Watson a loss of 1.71736.The example illustrates the model’s separation of faces sharing head position and facial structure.Source: Appendix presents the architecture and… · p. 13
  • Each extracted face is compared with legislators and 20 gender- and position-matched celebrity faces; a 95-percent confidence interval calibrates classification across 953 legislators.A face exceeding the calibration interval is classified as a member of Congress.Source: Appendix presents the architecture and… · p. 14
  • The model distinguishes legislators from matched non-legislators: celebrity comparisons average 2.12483 versus 0.0003162278 for self-comparisons (p < 0.0001).The reported test is t = 245.48, df = 19074.Source: Appendix presents the architecture and… · p. 14
  • Transcript-based measures yield the same substantive results, but cannot show whether a congresswoman appeared on screen or how the shot was composed.The comparison uses more than 120,000 transcripts from 2011–2021, matched to one-minute caption blocks; it checks rather than substitutes for the visual approach.Source: Appendix presents the architecture and… · p. 14
  • CLIP similarities to descriptions of split-screen and single-camera shots classify two-legislator frames; a 0.70 threshold identifies 27,658 CNN and 25,519 Fox News split-screen frames.The analysis starts with frames containing one or two legislators; the resulting split-screen frames form the basis for subsequent analyses.Source: Appendix presents the architecture and… · p. 15

5 Results

Congresswomen are disproportionately shown in split-screens, often beside a congressman, despite appearing more often than their share of seats would predict. In the static-frame experiment, split-screen presentation lowered perceived competence for congresswomen but not congressmen; evidence for an external-efficacy effect was limited.

5.1 How are Congresswomen Visually Framed?

  • 51.6% of 53,177 split-screen frames showed women, compared with 32.6% of 226,067 solo-legislator frames (χ2 = 6682.6, p < 0.0001).The pattern was significant across CNN and Fox News; women most often appeared beside a congressman, in roughly two-fifths of split-screen frames.Source: §5.1 · p. 15, Figure 1 · p. 16
  • The male-to-female ratio was 3.9 to 1 in Congress, but 2.1 in solo frames and 1.7 in split-screens; the disparity was no stronger on Fox News than CNN.Solo appearances by congressmen remained the most common frame overall, accounting for 69.9% of CNN solo frames and 65.4% of Fox News solo frames.Source: §5.1 · p. 16

ceived as Being More or Less Competent?

  • β = 0.122 (95% CI [0.044, 0.199], p = .002): congresswomen were rated as more competent in solo than split-screen frames, while congressmen showed essentially no penalty.The congresswoman’s effect remained significant with covariate adjustment (β = 0.130, 95% CI [0.053, 0.207], p = .001); the difference between penalties was β = 0.096 (90% CI [0.005, 0.187], one-sided p = .041; two-sided p = .082).Source: Figure 2 · p. 19, ceived as Being More or Less Competent? · pp. 19–20
  • The treatment did not significantly shift external political efficacy at the 0.05 level across model specifications, and registered equivalence bounds excluded all but small effects.The outcome averaged agreement with three items, each scaled from −3 to 3, with higher scores indicating greater efficacy.Source: ceived as Being More or Less Competent? · pp. 20–21

Dynamic Video?

In dynamic video, split-screen framing produced no detectable competence penalty for the congresswoman, unlike the static-frame result. Instead, outraged language reduced her perceived warmth more than the congressman’s, while evidence of higher external efficacy among women viewers who saw her alone was suggestive.

Dynamic Video?

  • The experiment crossed MC gender, solo versus split-screen format, and outraged versus measured register in eight roughly 40-second cable-news-style videos.Within each gender-and-register combination, audio was identical across visual conditions; the scripts differed in emotional intensity and closing message.Source: Dynamic Video? · pp. 21–22
  • β = 0.009 (95% CI [−0.062, 0.079], p = .813): the congresswoman’s dynamic split-screen versus solo expertise difference was essentially zero.The null was precise enough to exclude a penalty larger than half the static experiment’s estimate (β = 0.122).Source: Dynamic Video? · p. 24, Figure 3 · p. 25
  • β = −0.268 (95% CI [−0.448, −0.087], p = .004): identical outraged words cost the congresswoman more perceived warmth than the congressman.The estimated warmth reduction was β = −0.415 for the congresswoman and β = −0.147 for the congressman—roughly three times as much for her.Source: Dynamic Video? · pp. 25–26
  • The outrage diagnostic was inconclusive, and the four-item perceived-contestation measure had low reliability (α = 0.44), warranting caution about those treatment effects.The diagnostic could not rule out a penalty of the prespecified smallest effect size, though it ruled out a benefit of that size.Source: Dynamic Video? · p. 26
  • β = −0.085 (p = .033): viewers perceived slightly less contestation in split-screen than solo, while outraged language increased perceived contestation by β = 0.434 (p < .001).The frame did not supply the contested-advocacy account’s proposed first link; viewers heard contest in the language instead.Source: Dynamic Video? · pp. 24–26
  • β = 0.158 (95% CI [−0.005, 0.321], p = .057): women viewers who saw the congresswoman alone reported higher external efficacy, but the estimate fell just short of significance.This pattern was exploratory; among roughly thirty covariate-adjusted subgroup contrasts, only the measured-register estimate crossed the conventional threshold, and its unadjusted counterpart did not.Source: Dynamic Video? · pp. 27–28

6 Discussion and Conclusion

The study finds that split-screen framing carries a competence cost for congresswomen in static encounters, while outraged language carries a warmth cost even in dynamic video. The discussion bounds these findings and identifies implications for how appearances circulate and are produced.

6 Discussion and Conclusion

  • In 695,464 CNN and Fox News segments, congresswomen appeared in 51.6% of split-screen frames versus 32.6% of solo frames; static split-screens lowered their competence ratings, while dynamic split-screens did not measurably affect competence.Outraged language reduced congresswomen’s warmth more steeply than congressmen’s, regardless of screen arrangement.Source: §6 · p. 29
  • The competence penalty appeared in static frames but not dynamic anchor interviews; because medium and split-screen partner changed together, the informational explanation remains an interpretation.The authors argue that viewers fill in the ambiguous contest implied by a static frame, while a live exchange shows whether the congresswoman is being challenged.Source: §6 · p. 29
  • The experiments used disclosed fictional members of Congress and, in the dynamic study, synthesized voices, trading realism for experimental control and omitting real politicians’ partisanship, incumbency, and records.Those real-world attributes could invite motivated evaluations that the nonpartisan stimuli do not engage.Source: §6 · p. 30
  • Static frames approximate information-poor encounters such as muted clips and screenshots, but the study did not track appearances onto platforms; claims about their downstream effects remain interpretive.The authors identify roughly 27,000 corpus split-screen frames containing a congresswoman as candidate thumbnails.Source: §6 · pp. 30–31
  • The registered gender-by-register interaction was β = −0.268; outraged language reduced warmth by −0.415 for congresswomen and −0.147 for congressmen.The design bundles speaker gender with in-group advocacy, since the congresswoman discusses women while the congressman does not.Source: §6 · p. 31
  • Static exposure did not appreciably change women viewers’ external efficacy; a dynamic-experiment estimate for women watching a solo congresswoman approached significance (p = .057) but was exploratory.It was one of roughly thirty exploratory contrasts.Source: §6 · pp. 31–32
  • Future work should test the editorial and production logics behind split-screen assignment, examine which appearances are clipped and recirculated, and isolate the contribution of on-screen text.The authors note that all experimental conditions included text, so the design cannot identify its contribution.Source: §6 · pp. 32–33
  • The authors propose that symbolic effects may depend on visibility’s form and register, not volume alone, and locate potential remedies in editorial choices and platform defaults rather than asking women to temper their register.They present the proposed explanation for mixed findings as a testable hypothesis.Source: §6 · pp. 32–33

Online Appendix

The supplied passages give the paper’s title, which identifies visual framing and congresswomen’s perceived competence as its subject.

Online Appendix

  • The paper examines how visual framing relates to congresswomen’s perceived competence.Source: Online Appendix · p. 37

1 The Siamese Neural Network

A Siamese Neural Network compares paired inputs through two identical subnetworks that produce embeddings. The study uses Euclidean distance and contrastive loss to distinguish similar from dissimilar faces.

1 The Siamese Neural Network

  • A Siamese Neural Network takes pairs of inputs and predicts whether they are meaningfully related, rather than classifying a single input.Source: §1 · p. 37
  • Its identical subnetworks transform each input into an embedding, which a distance function compares; training pulls similar pairs together and pushes dissimilar pairs apart.The study used Euclidean distance and contrastive loss.Source: §1 · p. 37
  • The worked example’s contrastive loss is near zero for Pelosi compared with herself (0.00032) and higher for Pelosi compared with Emma Watson (1.71736), indicating less similarity.Training ran for 500 epochs on one GPU node for approximately eight hours; validation loss fluctuated around 0.05–0.07 after early training.Source: §1 · p. 37

2 The MC-versus-Celebrity Calculation

The study matched congressional and celebrity faces by orientation to construct comparison pairs, then used a Siamese network to decide whether an extracted frame face was a member of Congress.

2 The MC-versus-Celebrity Calculation

  • The image set comprised official photos of 953 legislators who served in the 112th–117th Congresses and 1,100 celebrity images.Source: §2 · p. 38
  • The researchers used pan, tilt, and roll to characterize head orientation and prioritize face pairs with similar poses, reducing noise from pose differences.They calculated three-dimensional Euclidean distance between each legislator and celebrity pair.Source: §2 · p. 38
  • For each legislator, the researchers selected 20 celebrities with the closest facial orientations; the closest celebrity to Nancy Pelosi had a distance of 1.8271.Source: §2 · p. 38
  • Each frame face was compared with the member’s official photo and 20 matched celebrity faces; it counted as the member if its similarity score exceeded the celebrity-score mean plus 1.96 standard deviations.The procedure used the Siamese Neural Network to produce the member and celebrity similarity scores.Source: §2 · pp. 38–39

3 Validation of the Facial Recognition Pipeline

The section checks the facial-recognition pipeline against self-matches and a transcript-based measure. The transcript approach reproduces the main visibility patterns but cannot establish on-screen presence or shot composition.

3.1 Emma Watson Calculation

  • The validation replicated the Pelosi–Watson calculation by comparing each MC’s Congress.gov photo with itself and with 20 randomly selected same-gender, same-orientation celebrities.The procedure calculated a mean for each comparison type across MCs.Source: §3.1 · p. 39
  • The self-match mean was 0.0003162278, while the mean similarity to randomly selected same-gender, same-orientation celebrities was 2.12483.The authors interpret this contrast as evidence that the SNN reasonably differentiates an MC’s face from the celebrity faces.Source: §3.1 · p. 39

3.2 Transcript-Based Calculation

  • The transcript measure matched more than 120,000 CNN and Fox News transcripts to one-minute closed-caption blocks, using semantic similarity and speaker identifiers to identify MC appearances.MC names were matched to the congress-legislators repository and cross-checked against on-air introductions; a frame was treated as containing the speaking MC when a human face was detected.Source: §3.2 · p. 39
  • Transcript-based identification reproduced the findings that congresswomen appeared more often than their congressional representation predicted and disproportionately in split-screen shots.This convergence supports the facial-recognition findings as not being artifacts of that pipeline.Source: §3.2 · p. 39
  • Transcript data cannot establish whether an MC was on screen or how the shot was composed, so the two validation approaches are not interchangeable.Source: §3.2 · p. 40

4 Corpus Retrieval and Shot Classification

The corpus uses a published dictionary to retrieve CNN and Fox News image-text segments, then applies CLIP to classify two-legislator frames as split-screen or not.

4.1 Per-Term Search Frequencies

  • The corpus search applied the 16 Pearson and Dancey (2011) woman-related terms to CNN and Fox News broadcasts; a term counted when its exact form appeared in a segment’s paired minute of captions.Per-term counts can sum to more than the total segments because terms often co-occur within a minute.Source: §4.1 · p. 40, Table 1 · p. 40

4.2 Shot Classification with CLIP

  • CLIP classified frames containing exactly two legislators by comparing each image with text prompts for prototypical shots and converting similarities into split-screen probabilities.A frame was classified as split-screen above the prespecified 0.70 threshold; this yielded 27,658 CNN and 25,519 Fox News frames.Source: §4.2 · pp. 40–41

5 Experiment 1: The Fully Crossed Static Design

Experiment 1 used a fully crossed set of static frames and fielded outcome batteries, then reported registered estimates and equivalence tests. The women-viewer efficacy estimate is small, with a confidence interval that permits a broad range of effects.

5.1 Samples of the Two 2026 Experiments

  • Table 2 presents demographic breakdowns for the two 2026 experiments and four pilot samples; analyzed samples were 2,619 for Experiment 1 and 3,601 for Experiment 2.Source: §5.1 · p. 41, Table 2 · p. 41

5.2 Stimuli and Vignette

  • Participants saw a healthcare-policy vignette alongside an assigned broadcast frame, with the text describing proposals to support frontline healthcare providers.The bracketed wording varied to match whether one or multiple members of Congress appeared.Source: §5.2 · p. 41

5.3 The Complete Set of Treatment Frames

  • The six conditions used the same four individuals, graphics, and accompanying text; only the MC’s gender and the shot format varied.The evaluation stage re-displayed the assigned frame with a green box identifying the MC being evaluated.Source: §5.3 · p. 42, Figure 2 · p. 42, §5.2 · p. 42

5.4 Outcome Measures as Fielded

  • Political capability was rated on four policy areas using a five-point scale, with healthcare benchmarked against the other three; external efficacy averaged three seven-point items.Source: §5.4 · pp. 42–43

5.5 Full Results

  • For women respondents’ external efficacy, β = 0.031 with a 90% CI of [−0.141, 0.203], ruling out effects outside |d| = 0.153 at 90% confidence.No equivalence threshold was registered for the man MC’s penalty; the largest |d| consistent with its 90% CI is 0.121.Source: §5.5 · p. 43

6 Experiment 2: The Dynamic Design

This section describes the development and evaluation of Experiment 2’s dynamic video design, including corpus analyses, manipulation checks, and pilot studies of competence ratings across solo and split-screen formats with measured and outraged scripts.

6.1 Stimulus Frames

  • The solo condition cut to whoever was speaking, whereas the split-screen condition kept the MC and anchor together in a two-shot.The video stimuli displayed rolling transcripts in both formats.Source: §6.1 · p. 44, Figure 3 · p. 46

6.2 Full Scripts for the Segments

  • The measured and outraged statements used the same position and matched structure, differing in emotional intensity and closing stance; scripts were identical across genders and audio was identical across visual conditions.The outraged statement had 36 tokens and the measured statement 39; lexical scoring found anger- and fear-laden vocabulary only in the outraged script.Source: §6.2 · pp. 45–46, §7 · p. 52

6.3 Sample Waterfall

  • 3,838 of 3,845 randomized respondents passed the audio check on the first attempt, and 93.7% passed the attention check.The three registered exclusion regimes retained 93.7%, 100.0%, and 93.5% of randomized completes, respectively.Source: §6.3 · pp. 46–47, Table 8 · p. 47

6.4 Full Results

  • β = 1.850 (95% CI [1.765, 1.935], p < .001) for the outraged-register emotional-intensity check; 96.5% and 93.3% correctly identified solo and split-screen formats, respectively.Tables 9 and 10 report registered estimates for perceived health expertise and external efficacy, but the supplied passages do not provide their values.Source: §6.4 · p. 47, Table 9 · p. 48, Table 10 · p. 49

6.5 Exploratory Causal-Mediation Estimates

  • Causal-mediation analyses were exploratory because they require sequential ignorability, which treatment randomization does not guarantee; the contestation measure was least reliable (α = 0.44).The supplied passages do not provide mediation effect estimates.Source: §6.5 · p. 49

7 The Emotional Register of the Corpus

  • 93.1% of corpus segments contained at least one anger- or fear-connoting token, but the average rate was 29.0 tokens per 1,000 words, and measured presentation was most common.The passage cautions that segment captions mix speakers and the lexical instrument has limits in handling negation and plural forms.Source: §7 · pp. 53–54, Table 17 · p. 54

7.1 The Register of the Coverage by Shot Type

  • 28.9 anger- and fear-laden words per 1,000 appeared in congresswomen’s solo minutes versus 27.3 in minutes with a congressman; solo-minute rates showed no meaningful gender difference.The outraged script’s concentration exceeded that of 98.2% of tokened corpus segments, and the measured-side classification covered 77.6%. These counts describe surrounding captions, not legislators’ own speech.Source: §7.1 · pp. 55–56

8.1 The 2024 Pilot Waves

  • The two 2024 pilot waves showed only a woman MC, so they could not establish whether the observed penalty was gendered.The fully crossed Experiment 1 was designed to address that question.Source: §8.1 · p. 56
  • β = 0.119 (ATE, p < 0.006) and β = 0.115 (CTE, p < 0.009) in the pooled Qualtrics and student samples support the split-screen competence penalty.The treatment effect was significant among women participants but not men in the individual samples.Source: §8.1 · p. 57

8.2 The December 2024 Pilot and Its MTurk Replication

  • The Prolific experiment held emotionally charged caption text constant across solo and split-screen conditions, and the competence penalty persisted.Because outrage was not randomized in this study, it could not establish whether outrage itself affected competence judgments.Source: §8.2 · pp. 58–60
  • β = 0.186 (ATE, p < 0.006) and β = 0.182 (CTE, p < 0.006) in the pooled Prolific and MTurk samples support a split-screen competence penalty.The pooled effect did not differ appreciably by respondent gender.Source: §8.2 · pp. 59–60
  • Across model specifications in both pilot waves, the external-efficacy treatment coefficient was never statistically significant at the 0.05 level.Its direction was generally consistent with the expectation that women viewers exposed to solo shots would report greater external efficacy.Source: §8.1 · p. 57, §8.2 · p. 60

8.3 Word Frequencies When Congresswomen Appear Alone

  • Negative-emotion terms such as “terrified” and “forgotten” topped the BERT completion lists for “Women are [MASK]” on both networks and informed the December pilot caption.The analysis used closed-caption segments where the pipeline identified a single congresswoman on screen.Source: §8.3 · p. 60

Related papers

Loading 2609.31970v1…