PaperlayerView exact revision on arXiv

Source-linked AI summary

On-Screen Inertia: Persistent Racial and Gender Disparities in Hollywood Film (1900-2024)

Hazem Ibrahim, Talal Rahwan, Yasir Zaki, Minsu Park

arXiv:2608.28613v1cs.CY

TL;DR

Existing research has documented cast underrepresentation and modest visibility gains, but has not fully examined whether increased inclusion changes actors’ structural narrative positioning. This paper analyzes 76,815 U.S. English-language films and over 3.1 million cast and crew entries across visibility, narrative structure, institutional pathways, and market outcomes. It finds on-screen inertia: visibility improves modestly while structural hierarchies persist, without a consistent market penalty for diversity.

  • Problem

    Research has documented underrepresentation and modest visibility gains, but structural positioning within cinematic narratives remains less examined.

  • Method

    The paper analyzes 76,815 U.S. English-language films and over 3.1 million cast and crew entries across visibility, structural, institutional, and market dimensions.

  • Results

    The analyses reveal on-screen inertia: modest visibility gains coexist with persistent structural hierarchies, while crew diversity predicts cast inclusion only along matching demographic lines and not narrative centrality.

  • Takeaways & Limitations

    The findings indicate that Hollywood’s representational inequalities are institutionally sustained rather than a consistent market response.

  • Takeaways & Limitations

    The analyses are restricted to U.S. English-language films, and network centrality, occupational coding, and poster-based visual analysis are proxies with narrower coverage than the full dataset.

Abstract

from arXiv · show

Hollywood has diversified its casts. Whether this has translated into structural change in how those actors are positioned within narratives remains largely unexamined. Drawing on 76,815 U.S. English-language films (1900-2024) and over 3.1 million cast and crew entries, we move beyond headcounts to examine long-term inclusion trends through network centrality, occupational stereotypes, crew-to-cast diversity pathways, and financial outcomes. We find evidence of what we term on-screen inertia. While the raw inclusion of women and racial minorities has increased modestly, White actors have become more overrepresented relative to the U.S. Census in recent decades, not less. Within the visibility layer, women face a consistent longevity penalty with significantly shorter careers than men, and visual depictions framing men as dominant and women as sensual have remained stable since the 1950s. Structurally, White actors retain disproportionate network centrality; women achieve parity in centrality and lead billing yet cluster in secondary co-lead roles; and occupational stereotypes anchoring racial and gender groups to specific labor categories persist largely unchanged across the pre- and post-2000 periods. Crew diversity associates with cast inclusion only along matching demographic lines (i.e., racial with racial, gender with gender) and does not extend to narrative centrality, revealing a structural ceiling on hiring-based interventions. Critically, we find no consistent market penalty for diversity across decades of box office returns and audience ratings, eliminating the primary rationalization for these practices. Together, these findings demonstrate that Hollywood's representational inequalities are not a rational market response, but are an institutionally sustained choice.

Introduction

Hollywood representation has become more visible, but whether that progress changes narrative positioning and institutional hierarchies remains unresolved. This paper evaluates representation across visibility, structural, and institutional layers to identify where inequalities persist.

  • Persistent racial and gender inequalities reflect institutional structures shaping access, hiring, casting, and creative positions, not isolated anomalies.
  • Computational work found modest increases in marginalized groups’ facetime, but visibility metrics do not capture agency or structural positioning.
  • The paper evaluates representation through visibility, structural positioning, and institutional ecosystem layers.These layers cover demographic inclusion and portrayal, narrative centrality and occupational stereotypes, and crew composition, director identity, and market incentives.
  • The framework distinguishes surface-level progress from structural change and identifies where Hollywood’s representational hierarchies have yielded or held firm.
  • 76,815 films and over 3.1 million cast and crew entries reveal on-screen inertia: modest visibility gains coexist with resistant structural hierarchies.The dataset also includes posters from 11,951 films and scene-level co-appearance data from 3,265 films.

Results

Representation of racial minorities and women in cast compositions has increased modestly, and the paper reports the same pattern extending to crews. However, White individuals remain consistently overrepresented.

  • Modest increases in racial-minority and women’s cast representation also extend to crews.

A B C

Across visibility, narrative structure, occupational roles, institutional pathways, and market outcomes, Hollywood shows modest inclusion gains alongside persistent racial and gender inequalities. These patterns indicate that surface-level diversification has not produced comparable structural change.

  • Visibility: White individuals occupied 88% of cast and 89% of crew roles in the 2020s, while remaining increasingly overrepresented relative to U.S. Census shares.Black actors remained persistently underrepresented, Hispanic representation declined relative to population share, and Asian/Pacific Islander representation showed no systematic deviation.
  • Visibility: White actors had the longest observed careers at 13.27 years, while Asian/Pacific Islander actors had the shortest at 7.31 years.White actors exited at significantly older ages than all non-White groups (∆exit = 4.49, p < 0.001).
  • Visibility: Dominant portrayals remained overwhelmingly male at 92.1%, while sensual portrayals remained almost exclusively female at 99.0% across eight decades.Neither pattern changed significantly, indicating stable gendered visual framing since the 1950s.
  • Structural Inertia: White actors occupied more central narrative-network positions than expected, while all racial minority groups showed large negative normalized centrality values before and after 2000.Male and female actors had broadly comparable centrality overall, with women moving toward greater centrality after 2000.
  • Structural Inertia: Women matched men in top billing but were overrepresented in positions 2 through 11, consistent with prominent secondary roles rather than equivalent star positioning.In male-led films, female co-stars also concentrated interactions around the male star, an asymmetry absent in female-led films.
  • Institutional Gatekeeping and Market Outcomes: Occupational stereotypes remained stable across pre- and post-2000 periods, while crew diversity associated with cast diversity only along matching racial or gender dimensions.Racial occupational rankings were preserved across periods (ρ = 0.45, p = .004), and racially diverse crews correlated with racially diverse casts (r = 0.631, p < 0.001).

Discussion

Across more than a century, Hollywood increased marginalized groups’ visibility while leaving structural hierarchies in narrative prominence, occupational portrayal, and visual framing largely intact. These patterns, together with the absence of a consistent financial penalty for diversity, point to institutionally sustained inertia rather than economic inevitability.

  • Surface-level visibility improved modestly, but structural hierarchies governing narrative prominence, occupational portrayal, and visual framing remained remarkably resistant to change.
  • Organizations can update demographic composition while leaving their cultural logic intact, making increased numerical presence conditional on persistent portrayal and career structures.
  • Female actors remain relationally oriented toward male stars, while occupational stereotypes linking identity to labor persist with minimal compression across pre- and post-2000 periods.
  • Crew diversity predicts cast inclusion only along matching demographic lines and does not shift narrative centrality, creating a ceiling for hiring-based interventions.
  • Neither racial diversity nor female cast composition consistently penalizes profit, weakening the market-based rationale for Hollywood’s demographic disparities.
  • Name-based demographic classification may disproportionately exclude minority individuals, so minority underrepresentation estimates should be interpreted as conservative lower bounds.
  • The findings are restricted to U.S. English-language films, use proxies for centrality and occupation, draw network analyses from 3,265 films, and remain correlational rather than causal.
  • Hiring-based interventions alone have not dismantled structural hierarchies, while the absence of a market rationale indicates that on-screen inertia is an institutionally sustained choice.

Data and Methods

The study builds a century-scale Hollywood dataset from TMDB credits and supplements it with poster imagery to analyze demographic representation and visual portrayal over time. It applies filtering, name-based demographic inference, automated poster classification, and human validation.

  • 332,356 U.S. English-language films released from 1900 to 2024 were initially queried from TMDB, with metadata and full cast and crew credits retrieved.
  • 76,815 films and 3,118,706 cast and crew entries remained after excluding sparse credits and low-confidence demographic classifications.
  • NamePrism inferred four broad racial groups, while Genderize assigned binary Male/Female categories from names and confidence scores.
  • 11,951 posters spanning eight decades were collected to characterize visual framing across body composition, posture, and portrayal type.
  • GPT-4 with Vision assigned poster images to dominant, sensual, or submissive categories using keyword sets derived through FastText nearest-neighbor lookup.
  • Human validation yielded 88.1% accuracy for portrayal classification and 99.5% accuracy for gender classification.
  • Temporal portrayal gaps were assessed by regressing decade-level male dominant shares and female sensual shares on a linear decade index, with face-only gaps as a comparison.
  • Relative representation was calculated as film role shares compared with corresponding U.S. Census population shares for each group and decade.

Age and career-trajectory analysis

The paper measures actor ages and career spans, then models narrative prominence through scene co-appearance networks and standardized centrality measures. It also tests star-oriented interaction patterns, billing-position representation, and occupational stability across time.

  • Age and career-trajectory analysis: Per-role ages were calculated as release year minus birth year, excluding missing values, ongoing careers, and implausible ages outside 5-100.
  • Age and career-trajectory analysis: Career entry and exit were defined as the earliest and latest release years of an actor’s appearances, with career length equal to exit year minus entry year plus one.
  • Movie co-appearance networks and narrative centrality: 3,265 U.S. movies supplied scene boundaries and visible-character data for weighted actor co-appearance networks.
  • Movie co-appearance networks and narrative centrality: The disparity filter retained unexpectedly strong edges under a uniform-strength null model, with centrality results robust across threshold choices.
  • Movie co-appearance networks and narrative centrality: Degree, closeness, and betweenness centrality were mean-centered within groups defined by measure, cast-size octile, and release year.
  • Movie co-appearance networks and narrative centrality: Star ratio measures the share of an actor’s total interaction weight connected to the top-billed actor.
  • Movie co-appearance networks and narrative centrality: Billing-position representation was compared with a 1,000-permutation null model that preserved each film’s actors and demographic labels.
  • Occupational representation: Character descriptors were matched to O*NET occupations and broader fields, whose normalized representation was compared before versus after 2000 using rank correlations and paired tests.

Crew-cast diversity relationships

The study operationalizes cast and crew diversity, tests director-actor demographic pairings and crew-cast associations against permutation-based expectations, and models financial and audience outcomes with controlled regressions and genre robustness checks.

  • Diversity measurement: Film-level racial diversity equals 1 − P(race = White), while gender diversity equals 1 − P(gender = Male).
  • Director-actor pairings: Director-actor pairing scores compare observed top-billed demographic pairings with a null model that randomizes billing order while preserving actors and demographics.
  • Crew-cast relationships: Crew-cast relationships were assessed with Pearson correlations, while narrative effects used betweenness-centrality gaps between marginalized and majority groups.
  • Market outcomes: Profit was defined as revenue minus budget, and TMDB vote average measured audience evaluation among films with at least 50 ratings votes.
  • Market outcomes: Four pooled OLS specifications varied decade versus year controls and included models adding crew composition controls to isolate cast-diversity effects.
  • Robustness checks: Genre-stratified analyses and multi-genre indicators tested whether results were robust across seven frequent genres and alternative genre specifications.
  • Statistical testing: Two-sided t-tests and χ2 tests used p < 0.05 as the significance threshold, with exact p-values reported when available.

Data and code availability

The paper provides reproducibility materials and identifies the third-party sources underlying its analyses.

  • All code for the figures and statistical analyses is publicly available in a GitHub repository.The repository contains self-contained Python scripts and aggregated role-level data tables.
  • The analyses use derived tables covering Census shares, O*NET occupational mappings, film metadata, financial records, and scene-level co-appearance data.
  • Film metadata, credits, budgets, revenues, posters, and audience ratings come from the TMDB API, while scene-level co-appearance data come from the augmented Amazon X-Ray dataset.

Competing Interests

The supplementary document identifies the paper, its authors and affiliation, its correspondence contacts, and its absence of competing interests.

  • The authors declare no competing interests.
  • The paper is titled On-Screen Inertia: Persistent Racial and Gender Disparities in Hollywood Film (1900-2024).
  • The authors are affiliated with New York University Abu Dhabi, UAE, and correspondence is directed to Yasir Zaki and Minsu Park.
  • The supplementary document is organized into supplementary notes, tables, and figures.

Supplementary Note 1: Age analysis of racial and gender

The supplementary age analysis finds persistent racial and gender differences in appearance age, career timing, and observed career length.

  • Racial age patterns: White actors appear oldest on average at 42.56 years, followed by Black actors at 41.78, Hispanic actors at 40.86, and API actors at 39.00.All pairwise comparisons are statistically significant.
  • Racial age patterns: White actors enter Hollywood at 34.45 years and exit at 45.95 years, while Hispanic actors enter at 34.51 years but exit earlier.
  • Gender age patterns: Male actors are older than female actors at appearance by Δ= 7.13 years, with both groups’ mean appearance ages increasing slightly across the dataset.
  • Gender age patterns: Female actors enter and leave at younger ages than male actors, with career-start and career-end gaps of 5.48 and 6.97 years, respectively.
  • Gender age patterns: Women’s observed careers are shorter than men’s by 1.77 years, consistent with a documented double standard of aging in entertainment.

Supplementary Tables

The supplementary tables document dataset composition, classification and missingness, occupational-role coverage, demographic network measures, pairing analyses, and regression specifications.

  • Dataset coverage: Supplementary Table 1 reports the number of films and credited cast and crew by decade.
  • Data quality: Supplementary Tables 2 and 3 report demographic classification shares and missing birth-year information by decade.Both tables express shares between 0 and 1.
  • Occupational coverage: Supplementary Table 4 reports the number of roles by occupational field and decade.
  • Network measures: Supplementary Tables 5–7 report normalized betweenness, closeness, and degree centrality differences across demographic groups for varying α values.
  • Financial outcomes: Supplementary Table 17 reports pooled OLS models linking cast and crew diversity with profit and audience ratings using controls, fixed effects, and robust standard errors.

Supplementary Figures

The supplementary analyses extend representation, centrality, billing, star-ratio, gatekeeping, and financial-outcome tests across demographic groups, genres, decades, and model specifications. Together, they document how these dimensions are measured and tested for robustness.

  • Visibility: Supplementary Figure 1 compares appearance ages and career lengths across racial and gender groups, with pairwise differences tested using independent two-sample t-tests.Panels show distributions, decade trends, group means, and annotated test statistics.
  • Representation and centrality: Supplementary Figures 2 and 6 examine cast and crew representation relative to U.S. Census shares, director demographics, and associations between crew diversity and cast centrality gaps.Figure 6 separates racial and gender centrality analyses and organizes crew diversity into quartiles.
  • Billing position: Supplementary Figures 4 and 13 measure billing-order representation as deviations from overall or permuted cast shares across billing positions.The genre-specific analysis uses 1,000 permutations of the billing-order column.
  • Portrayal and narrative relationships: Supplementary Figures 5 and 14 assess visual portrayal and star ratio, including male–female differences in face-only poster depictions and non-lead co-appearance weight tied to leads.Figure 14 reports female-versus-male significance using two-sided t-tests.
  • Robustness analyses: Supplementary Figures 7–11 test financial, rating, representation, and gatekeeping patterns using pooled, genre-stratified, multi-genre, decade-specific, and three-genre analyses.Figure 7 reports pooled OLS coefficients with 95% confidence intervals across four temporal and crew-control specifications; Figures 8–11 provide genre and institutional checks.
Loading 2608.28613v1…