Social change, cultural resistance: a meta-analysis of the influence of television viewing on gender role attitudes

ABSTRACT

In the last decades, there have been substantial changes in public attitudes toward gender roles and in television’s landscape and messages. Our meta-analysis of nearly 50 years of studies of television’s contribution to gender role attitudes is based on 485 effect sizes from 69 independent samples (N = 57,542) and reveals an overall effect size of .102. While we found no evidence of any decline of this association over time, it is significantly weaker for gender role attitudes related to the public sphere. Our findings imply that television viewing may strengthen an “egalitarian essentialism ideology,” that is, a discrepancy between the endorsement of gender equity in the public sphere/workplace and the persistence of traditional views regarding the private/domestic sphere.

KEYWORDS:

• Television
• gender role attitudes
• meta-analysis
• cultivation

Data availability statement

The raw data, codebook, and full list of studies included in this meta-analysis can be accessed via the following link: https://osf.io/7mqsr/

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 We recognize that “sex” and “gender” are very different concepts (and both are increasingly acknowledged to be non-binary), but the literature uses both “gender”/”gender role” and “sex role” in combination with “stereotypes,” “attitudes,” and similar terms interchangeably, and we do so here as well.

2 Other media besides television (e.g., video games, social media) may make similar contributions to sex role attitudes, but in the tradition of cultivation theory, this study focused exclusively on television.

3 There were overlaps in their samples – five studies were in both meta-analyses – but Morgan and Shanahan (1997) looked exclusively at studies based explicitly on cultivation theory.

4 Only 5 of these were also included in the studies of Herrett-Skjellum and Allen (1996) and Oppliger (2007); 23 were published after the 1997 study of Morgan and Shanahan.

5 We excluded conference papers and dissertations because conference papers before 2004 are not indexed, and many dissertations are available only on microfilm. To avoid a biased sub-sample of unpublished work, we thus focused on published papers.

6 The authors coded the entire sample themselves. Any questions or ambiguities were resolved by discussion among the authors. We did not conduct a systematic reliability test for the coding of this particular sample. However, in the larger project, the intercoder reliabilities measured by Gwet’s AC₁ ranged from .790 to 1.000.

7 Two authors independently determined the coding of the dependent variable. Some measures touched on both the private and the public spheres, and these were resolved by discussion. For instance, we decided that the weight of the items in the Attitudes Toward Women Scale (Galambos et al., 1985; Spence & Helmreich, 1972) leaned more heavily toward the public sphere (i.e., support for egalitarianism in terms of rights and opportunities in society). Dependent variables that equally captured both spheres or that could not be unambiguously assigned to one category were coded as “Other/Mixed.”

8 We used Peterson and Brown’s (2005) formula to transform standardized regression betas (b*) to r. To convert gamma to r, we applied the formula provided by Morgan and Shanahan (1997). If only an F test or unstandardized b* and standard error values were provided, we calculated r from N and p values or by computing t values, respectively (Rosenthal & Rubin, 2003).

9 For example, some studies presented all data separately for males and females, without providing data for the sample as a whole.

10 Of the 485 distinct findings we coded, 37.1% are based on Pearson r, 20.4% are partial correlations, 23.5% are gammas, and 17.7% are betas. These four represent 98.7% of the coefficients we coded. Further breakdown revealed lower ESs based on converted betas (r = .074, k = 15), partial correlations (r = .073, k = 6), and gammas (r = .080, k = 8).

11 Following Borenstein et al. (2021), we only analyzed subgroups with at least five samples.

12 Following the same procedure using stereotypes about innate qualities as a dichotomized predictor did not reveal significant evidence of moderation (b = .042, p = .090).

13 The funnel plots for each dependent variable can be found in the Online Supplementary Materials.