Cumulative exposure to state-level structural sexism and risk of disordered eating: Results from a 20-year prospective cohort study

Abstract

Background:

Gendered inequities in disordered eating are well-documented, yet few studies have examined their structural drivers. To help fill this gap, we investigated whether cumulative exposure to state-level structural sexism from childhood through young adulthood potentiates differences in disordered eating risk between cisgender girls/women and boys/men.

Methods:

Participants came from the Growing Up Today Study (N = 16,875), a cohort of children aged 9–14 years in 1996 who we followed through 2016. Using a composite index of relevant state policies and social inequalities from the Institute for Women’s Policy Research, we categorized states as having high or low levels of structural sexism and summed the number of years participants had lived in a high structural sexism state during the study period to quantify their cumulative exposure. We fit sequential conditional mean models to estimate the effect of cumulative exposure on risk of four outcomes (chronic dieting, purging, binge eating, and overeating), controlling for individual-and state-level confounders via propensity scores. We then tested whether effects differed between girls/women and boys/men by including cumulative-exposure-by-gender-identity interaction terms and calculating the relative excess risk due to interaction (RERI).

Results:

In the full sample, each additional year of living in a high structural sexism state was associated with a 5% increased risk of purging (95% confidence interval (CI): 3%, 7%), an 8% increased risk of binge eating (95% CI: 6%, 10%), and a 9% increased risk of overeating (95% CI: 8%, 11%). Risk increases were larger on average for girls/women than for boys/men, and girls/women who had lived in a high structural sexism state for four or more years had excess risk of chronic dieting (RERI: 0.64, 95% CI: 0.18, 1.10), purging (RERI: 2.64, 95% CI: 1.24, 4.30), and binge eating (RERI: 2.21, 95% CI: 0.93, 3.50).

Conclusions:

Structural sexism may contribute to inequities in disordered eating between cisgender girls/women and boys/men. Future research should include transgender and gender diverse participants, explore intersectional effects, and identify underlying mechanisms to inform policy-oriented interventions.

1. Introduction

Eating disorders and related symptoms and behaviors (e.g., severe calorie restriction, purging, binge eating, etc.) are leading mental health issues among young people in the U.S. (Streatfeild et al., 2021) that are well-known to disproportionately affect those with marginalized gender identities (e.g., cisgender girls and women, transgender/gender diverse individuals) (Murnen et al., 2015; Nagata et al., 2020). To-date, empirical studies aiming to elucidate the determinants of these inequities have focused primarily on individual-level psychosocial factors such as thin-ideal internalization and body dissatisfaction, with recent work additionally considering potential genetic and/or hormonal causes (Smolak et al., 2012; Wang et al., 2013). However, these approaches have long received critique for ignoring the influence of broader sociocultural contexts; for locating risk within the attitudes and behaviors of affected individuals; and, regarding the research on biological determinants, for essentializing differences between gender groups (Piran, 2010; Springmann et al., 2020). From an epidemiologic perspective, thin-ideal internationalization, body dissatisfaction, and other such proximal disordered eating risk factors are unlikely to fully explain the observed inequities because they are themselves the downstream consequences of macro-level social processes and structural forces that ultimately shape population health patterns (Schwartz and Diez-Roux, 2001). As such, shifting research attention upstream may provide a more comprehensive understanding.

In this study, we examine structural sexism as one potential upstream determinant of the gendered social patterning of disordered eating (focusing specifically on the observed inequities among cisgender girls and women relative to cisgender boys and men, as explained below). The term structural sexism as used here refers to systematic gender inequality in power and resources embedded within a society’s institutions that primarily privileges cisgender boys and men (Homan, 2019); it encompasses targeted laws and policies (e.g., abortion bans), restrictive cultural norms and ideologies (e.g., hegemonic gender roles), and resultant socioeconomic injustices (e.g., gendered wage gaps) (Krieger, 2020). An emergent literature focused on understanding the population health impacts of discriminatory social systems suggests that structural sexism is an important contributor to gendered health inequities in the U.S. (King et al., 2020), a finding that is consistent with social epidemiologic and feminist theories of health. Drawing on these perspectives, we analyze the relationships between structural sexism, gender, and disordered eating using population-based data and rigorous epidemiologic methods for causal inference, providing novel information that can be used to inform action.

1.1. Structural sexism as a determinant of population health

Research on structural sexism and population health has increased considerably over the past several years (King et al., 2020). In U.S. contexts, this work has focused largely on the material dimensions of structural sexism (i.e., those that describe the gendered distribution of tangible resources, opportunities, etc.), using various composite indices of social, economic, and political gender inequality to demonstrate how these systems influence the health of predominantly cisgender samples. For example, leveraging the persistent variation in systematic gender inequality across geographies, studies have shown that cisgender women living in U.S. states characterized by a high degree of structural sexism (e.g., large wage gaps, few legal protections) experience more chronic health conditions, poorer self-rated health and physical functioning, and higher rates of violence victimization and premature mortality relative to cisgender women living elsewhere (e.g., Homan, 2019; Kawachi et al., 1999). Other studies have found similar effects, albeit smaller in magnitude, among cisgender men, a pattern described in the literature as reflecting structural sexism’s “universal harm” (Homan, 2019). Regarding mental health outcomes specifically, two studies have documented associations between state-level structural sexism and cisgender women’s risk of depressive symptoms and post-traumatic stress disorder (Chen et al., 2005; McLaughlin et al., 2011); there is also evidence to suggest that the wage gap partially accounts for gendered inequities in mood and anxiety disorders (Platt et al., 2016).

In explaining these findings, researchers have drawn on social epidemiologic theories of disease distribution that articulate the ways by which discriminatory social systems harm health and contribute to health inequities. Particularly relevant is Krieger’s (2001, 2012) ecosocial theory, which states that individuals “literally biologically embody exposures arising from our societal and ecological context, thereby producing population rates and distributions of health.” Ecosocial theory recognizes that social systems become embodied in both perceivable and non-perceivable ways via myriad interacting pathways, and that exposure accumulates over time in a way that reflects cumulative dis/advantage (i.e., the systematic growth of inequality with the passage of time) (Dannefer, 2003). Accordingly, structural sexism is thought to contribute to gendered health inequities by disproportionately exposing cisgender girls and women (as well as individuals belonging to other marginalized gender groups) to material deprivation, sexist discrimination, violence and/or unsafe living and working conditions, inadequate healthcare, and other adverse social experiences, all of which “get under the skin” to harm health across the life-course.

1.2. Structural sexism and disordered eating

A rich history of feminist scholarship provides an important framework for understanding the potential links between structural sexism and disordered eating more specifically. As summarized by Piran (2010), this work rejects an individualist framing of the observed gendered inequities, positioning them instead in relation to “structural factors of privilege and power related to gender.” For example, in her seminal text Unbearable Weight: Feminism, Western Culture, and the Body, influential feminist theorist Susan Bordo argued that U.S. girls’ and young women’s eating problems are ultimately a product of growing up in a patriarchal culture – particularly one that place paramount importance on physical appearance while simultaneously defining normative femininity in terms of thinness and bodily control (Bordo, 1993). Other theorists have implicated the material dimensions of structural sexism more specifically, suggesting that gender inequalities in social, economic, and political domains incentivize girls and women to use their appearance as a form of capital, thus increasing their risk of using dangerous disordered eating behaviors to attain the idealized beauty standards of the day (Hesse-Biber et al., 2006). Importantly, multicultural and intersectional feminist scholars have rightfully critiqued this work for focusing predominantly on the experiences of White, affluent, and cisgender girls and women, noting how beauty standards and their consequences are not only gendered, but racialized and classed as well; moreover, they have highlighted how individuals from (multiply) marginalized communities may also be using disordered eating behaviors to cope with broader experiences of discrimination, poverty, violence, and trauma (Katzman and Lee, 1997; Thompson, 1996).

Integrating and building upon these perspectives, the developmental theory of embodiment is a comprehensive feminist model of disordered eating that outlines potential mechanism(s) by which structural sexism could increase risk at the individual-level and drive inequities at the population-level (Piran, 2017). Briefly, the model states that exposure to adverse social experiences reflective of structural sexism (e.g., sexual objectification, constraining societal discourses of femininity vs. masculinity, sexist discrimination, etc.) jointly shape how people inhabit and relate to their bodies; over time, and disproportionately for people belonging to marginalized gender groups, these experiences can lead to and/or intensify “disrupted embodiment” (i.e., a persistent sense of disconnection from and/or feelings of negativity towards one’s body), in-turn increasing disordered eating risk. A wealth of qualitative data support these hypotheses (Piran, 2016, 2017), and, consistent with the tenets of ecosocial theory, indicate that structural sexism operates across the life-course to impact mental health in a cumulative fashion. There are also several quantitative studies that have documented cross-sectional associations between the various adverse social experiences identified by the developmental theory of embodiment and disordered eating among cisgender girls and women (Piran and Cormier, 2005; Piran and Thompson, 2008; Sabik and Tylka, 2006; Garaigordobil and Maganto, 2013). However, no quantitative study to our knowledge has yet to examine the role of structural sexism directly, nor how structural sexism may relate to the gendered social patterning of disordered eating at the population-level.

1.3. Current study

To help address these gaps, the current study aimed to evaluate whether structural sexism functions as an upstream determinant of gendered inequities in disordered eating among young people living in the U.S. We focused specifically on state-level structural sexism given the key role that states play in structuring relevant sociopolitical contexts within the U.S.; further, as noted above, state-level structural sexism has been linked to the gendered social patterning of other mental health outcomes (Chen et al., 2005; McLaughlin et al., 2011). Informed by ecosocial theory and the developmental theory of embodiment (Krieger, 2001; Piran, 2017), we hypothesized that state-level structural sexism would have a cumulative impact on disordered eating from childhood through young adulthood, such that each additional year spent in living in a structurally sexist state during this time would incrementally increase risk. Therefore, we used sequential conditional mean models to estimate the effect of cumulative exposure to state-level structural sexism on subsequent risk of disordered eating, and to test whether effects differed by gender identity. Finally, we note that our analyses focus on average differences between cisgender girls and women and cisgender boys and men, which we acknowledge reinforces normative gender binaries and ignores how structural sexism may differentially impact transgender/gender diverse people, as well as subgroups defined by race/ethnicity and other dimensions of identity. We thus consider our study to be a first step towards more inclusive and intersectional research on structural sexism and disordered eating, a topic that we return to in the Discussion.

2. Methods

2.1. Study data

Data were drawn from the Growing Up Today Study (GUTS), a U.S. national cohort of young people who are children of women from the Nurses’ Health Study II (Field et al., 2001). GUTS began in 1996 with 16, 875 participants aged 9–14 years (including siblings), with follow-up surveys having since been administered annually or biannually; further details are published elsewhere (Field et al., 2001). For the current study, we used data collected during the first nine waves of GUTS that span an 11-year window from 1996 to 2007, mirroring the years for which we had corresponding exposure data. We also used data collected during the most recent wave of GUTS that assessed our outcomes of interest, which occurred in 2016, for a secondary set of analyses considering a 20-year-post-baseline follow-up. Study waves are hereafter denoted T_k, where k = 1, 2 … 10; see Table 1 for details on wave timing. The Institutional Review Boards of the Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health approved GUTS, and the current analyses were exempt from further review by the Institutional Review Board of the University of Massachusetts Chan Medical School.

Table 1.

Overview of the Growing Up Today Study (GUTS) data collection.

Year of data collection	1996	1997	1998	1999	2000	2001	2003	2005	2007	2016
Study wave (Tk)	T 1	T 2	T 3	T 4	T 5	T 6	T 7	T 8	T 9	T 10
Exposurea
State of residence	X	X	X	X	X	X	X	X	Xb	Xb
Outcomes
Chronic dieting	Xc	X	X	X		X	X	X		X
Purging	Xc	X	X	X		X	X	X	X	X
Binge eating	Xc	X	X	X	X	X	X	X	X	X
Overeating	Xc	X	X	X	X	X	X	X	X	X
Covariates
Gender identity	X
Race/ethnicity	X
Age	X	X	X	X	X	X	X	X	X	X

Note: “X” indicates that the variable was assessed at that study wave.

^aWe used information on state of residence to assign participant values for the exposure.

^bAlthough state of residence was assessed in 2007 and 2016, we did not use this information as participants were not assigned an exposure value at these waves.

^cBaseline outcome information was only used to identify participants for exclusion (i.e., those reporting the outcome at baseline).

To be included in one or more of our four outcome-specific analytic samples, participants had to: (1) have information on file for their state of residence, (2) have lived in one of the 50 U.S. states or the District of Columbia (D.C.), (3) have provided information on the relevant disordered eating outcome at least once during the study period, (4) have complete information on other key variables, and (5) be free of the outcome at baseline. A flow diagram detailing analytic sample construction is presented in Supplementary Figure 1. After exclusions, 97,860 responses from 13,980 participants were included in the chronic dieting sample, 118,984 responses from 14,873 participants were included in the purging sample, 136,224 responses from 15,136 participants were included in the binge eating sample, and 134,784 responses from 14,976 participants were included in the overeating sample. Excluded participants differed from the analytic samples with respect to gender identity, race/ethnicity, and age; we accounted for these differences analytically using propensity score adjustment, as explained below.

2.2. Measures

2.2.1. Exposure

We measured state-level structural sexism using a set of indicators developed by the Institute for Women’s Policy Research (IWPR) that quantify the magnitude of gender inequality in all 50 U.S. states/D.C. in a given year across four domains: (1) political participation; (2) employment and earnings; (3) social and economic autonomy; and (4) reproductive rights; see Supplementary Table 1 for details on each indicator’s subcomponents (Institute for Women’s Policy Research, 2002). We recognize that these indicators have several limitations, including a reliance on data that often conflate gender identity with sex-assigned-at-birth and the fact that they combine measures of absolute and relative gender inequality (e.g., women’s median yearly income and the gendered wage gap, respectively), which can introduce interpretability challenges (Roberts, 2011; Vieraitis et al., 2015). That being said, the indicators are relatively comprehensive and are time-varying, allowing us to capture each state’s complex sociopolitical environment as experienced by participants over time. Ultimately, we consider their values to be proxies for an underlying latent construct of structural sexism, and we evaluate this to the extent possible via a series of sensitivity analyses described below.

We started by obtaining the state-specific values of each indicator for all available years (1996, 1998, 2000, 2002, and 2004) and used last observation carried forward imputation to define such values for each corresponding subsequent year (i.e., 1997, 1999, 2001, 2003, and 2005). As we were interested in the overall effect of state-level structural sexism on disordered eating risk rather than that of any one domain (and a factor analysis supported a single factor solution; see Supplementary Table 2), we combined the four indicators by standardizing their values and averaging them such that each state received a single structural sexism score for each year, where higher scores indicated higher levels of structural sexism. We then created a binary variable contrasting “high structural sexism states” (i.e., those in the top tertile of the structural sexism score) with “low structural sexism states” (i.e., those in the bottom two tertiles). Fig. 1 presents a series of maps showing how states were categorized over time.

Fig. 1.

Time-varying categorization of the 50 U.S. states and the District of Columbia as high or low structural sexism states.

To quantify participants’ cumulative exposure to state-level structural sexism, we merged this state categorization information to the GUTS database and summed the number of years participants lived in a high structural sexism state from baseline through a given study wave. Table 2 presents the distribution of cumulative exposure across the exposure window; values range from zero to the maximum possible exposed years at each wave, where one unit equals one year of living in a high structural sexism state. For example, if a participant lived in a low structural sexism state from T₁ to T₂ and a high structural sexism state from T₃ to T₄, their cumulative exposure value would be zero at T₁ and T₂, one at T₃, and two at T₄. We analyzed cumulative exposure to state-level structural sexism as a time-varying continuous variable, lagged one wave prior to outcome assessment to ensure temporality (e.g., cumulative exposure through T₄ in relation to disordered eating at T₅). In the case of the 20-year-post-baseline follow-up at T₁₀, we used participants’ final cumulative exposure value at T₈ (i.e., cumulative exposure through 2005, the final year for which we had data on state-level structural sexism) and considered this a test of possible “lingering” effects. We also defined a binary indicator contrasting participants in the top tertile of cumulative exposure by the end of the exposure window to participants in the lower two tertiles (corresponding to ≥4 and < 4 years of living in a high structural sexism state, respectively), allowing us to compare disordered eating risk between participants exposed to high versus low “doses” of state-level structural sexism.

Table 2.

Distribution of participants’ cumulative exposure to state-level structural sexism.

Tk	Theoretical rangea	Observed M (SD)
T1 (1996)	0–1	0.4 (0.5)
T2 (1997)	0–2	0.7 (1.0)
T3 (1998)	0–3	1.2 (1.4)
T4 (1999)	0–4	1.6 (1.8)
T5 (2000)	0–5	2.0 (2.3)
T6 (2001)	0–6	2.7 (3.2)
T7 (2003)	0–8	3.4 (4.0)
T8 (2005)	0–10	3.8 (4.5)

Note: T_k = study wave, M = mean, SD = standard deviation.

^aMinimum possible to maximum possible number of years that a participant could have lived in a high structural sexism state from T₁ through T_k.

2.2.2. Outcomes

We considered four outcomes – chronic dieting, purging, binge eating, and overeating – given documented differences in the magnitude of gendered inequities across restrictive-, compensatory-, and binge-type disordered eating behaviours (Murnen et al., 2015). These outcomes were measured using questions adapted from the CDC’s Youth Risk Behavior Surveillance System (see Table 1 for the timing of assessment in GUTS), all of which have good psychometric properties (Kann et al., 1996); our operationalizations were informed by previous research.

Chronic dieting and purging were measured with three questions that asked about past-year frequency of dieting, self-induced vomiting, and/or laxative use, with response options ranging from never to multiple times per week. For chronic dieting, we coded participants who reported at least monthly dieting as having the outcome. For purging, we coded participants who reported any frequency of self-induced vomiting and/or laxative use as having the outcome.

Two additional questions asked about the past-year frequency of overeating (defined as the consumption of a large amount of food in a short period of time) and concomitant feelings of loss-of-control. For binge eating, we coded participants who indicated at least monthly overeating with loss-of-control as having the outcome. We also considered overeating on its own, given that cisgender boys and men may be less likely than cisgender girls and women to endorse loss-of-control (Strother et al., 2012).

2.2.3. Effect measure modifier

We included gender identity as a potential effect measure modifier of the relationship between structural sexism and disordered eating. Unfortunately, the ascertainment of gender identity in early waves of GUTS is suboptimal with respect to inclusivity and validity (Lett and Everhart, 2022). Participants’ were identified by their mothers as either a “girl” or “boy” at enrollment; participants were allowed to switch between the two gendered survey forms at any point (an imperfect proxy for gender minority status), however, more accurate data on gender identity were not collected until 2010 (Reisner et al., 2014). As no participants in any of our analytic samples requested a survey switch during the primary study period and too few later identified as transgender/gender diverse to be included in our 20-year-post-baseline follow-up analyses (n < 15), we defined two gender identity categories: cisgender girl/woman, cisgender boy/man (hereafter: girl/woman, boy/man).

Of note, our conceptualization of gender identity rather than structural sexism as the effect measure modifier is consistent with epidemiologic approaches to identifying the drivers of health inequities, which typically involve assessing whether the direction and/or magnitude of a given exposure-outcome relationship differs across social groups (Ward et al., 2019). It also reflects our hypothesis that structural sexism – not gender identity – is the modifiable causal factor underlying the gendered social patterning of disordered eating.

2.2.4. Confounders

Guided by a directed acyclic graph (DAG; Supplementary Figure 2), we identified two sets of potential confounders: those measured at the individual-level and those measured at the state-level. Potential individual-level confounders were defined as common causes of cumulative exposure to state-level structural sexism (i.e., factors that influence one’s state of residence in a given year and over time) and disordered eating; these included race/ethnicity (time-invariant; conceptualized as an imperfect proxy for exposure to racism) and age in years (time-varying). Potential state-level confounders were defined as common causes of state-level structural sexism itself (i.e., factors that influence a state’s level of gender inequality in a given year and over time) and disordered eating; these included median household income in U.S. dollars and Gini income inequality ratio (both time-varying and obtained from U.S. Census data).

2.3. Statistical analysis

We used sequential conditional mean models (SCMMs) to analyze the relationship between cumulative exposure to state-level structural sexism and disordered eating risk. Briefly, SCMMs are a class of causal models that use propensity score adjustment and generalized estimating equations (GEEs) with an independence working correlation matrix to estimate the total effect of a time-varying exposure on a subsequent outcome (Keogh et al., 2017). Similar to marginal structural models (another commonly used method for analyzing exposure-outcome relationships that fluctuate over time), SCMMs can provide unbiased estimates in the presence of time-varying confounding under the assumptions of exchangeability, consistency, positivity, and correct model specification. SCMMs can additionally provide unbiased estimates in the presence of informative censoring (i.e., non-random missingness and/or loss-to-follow-up) provided that predictors of censoring are included in the propensity score model. Also of note, estimation via GEEs allowed us to obtain population-averaged effect estimates and to properly account for three levels of clustering in GUTS (i.e., repeated measures within individuals, individuals within family units, and family units within state of residence). For these reasons, SCMMs were ideally suited for handling the complex features of our data structure that would otherwise introduce bias if conventional analytic methods were used.

We started by estimating the propensity scores, defined as the probability of exposure to state-level structural sexism at T_k. conditional on exposure and confounder history, as well as all aforementioned predictors of censoring; this was done by fitting pooled logistic models and transforming the derived log-odds into predicted probabilities. We then fit three doubly robust (i.e., propensity score- and confounder-adjusted) modified Poisson SCMMs to estimate risk ratios (RRs) and 95% confidence intervals (CIs) for the effect of cumulative exposure to state-level structural sexism on disordered eating. SCMM #1 included the full sample to estimate overall effects. SCMM #2 stratified by gender identity to assess whether the magnitude of effects differed between girls/women and boys/men. To formally evaluate effect measure modification, SCMM #3 included a two-way interaction term between gender identity and cumulative exposure (binary operationalization only). We used the estimated beta coefficients from SCMM #3 to calculate the relative excess risk due to interaction (RERI), which tested whether girls/women in the top tertile of cumulative exposure by the end of the exposure window experienced a disproportionate increase in disordered eating risk (i.e., “excess risk”) (Ward et al., 2019; Vander-Weele, 2015). For each SCMM, we first included the nine waves of data from 1996 to 2007; we then re-fit the SCMM to include all 10 waves from 1996 to 2016.

We also conducted a series of sensitivity analyses. First, we considered each of the four IWPR indicators separately to assess whether structural sexism in any one domain drove our findings. Second, we evaluated the extent to which estimates were robust to using alternative exposure measures that distinguish between absolute and relative gender inequality, specifically, by using the empirically-derived measures described in Vieraitis et al. (2015). Third, we assessed the impact of adjusting for disordered eating at T_k–1; including prior outcome status in the adjustment set is recommended by Keogh et al. for improving the precision of estimates from SCMMs (Keogh et al., 2017), although we thought it unlikely that prior outcome functioned as a confounder in our case (i.e., there is no arrow from disordered eating to structural sexism in our DAG) and thus we did not use this approach in our main analysis.

All models were fit in R 4.1.0 using the “geepack” package (Halekoh et al., 2006). The Technical Appendix within the Supplementary Materials provides additional details regarding our analyses.

3. Results

3.1. Descriptive statistics

Table 3 presents baseline characteristics of each analytic sample. Just over half of the participants in each sample were girls/women. The majority were non-Hispanic White (93.6–93.7%); 0.8–0.9% were Asian, 0.6–0.7% were Black/African American, 1.4–1.5% were Hispanic/Latine, and 3.3% identified with an unlisted race/ethnicity. As for state-level covariates (which reflect participants’ geographic distribution across the U.S. in 1996), the average median household income was approximately $36,000 and the average Gini income inequality ratio was 0.6; these values are similar to their corresponding U.S. national average values in 1996 (U.S. Department of Commerce, 1997).

Table 3.

Baseline characteristics of the outcome-specific analytic samples.

	Chronic dieting (N = 13,980)	Purging (N = 14,873)	Binge eating (N = 15,136)	Overeating (N = 14,976)
	% or M (SD)	% or M (SD)	% or M (SD)	% or M (SD)
Individual-level covariates
Gender identity
Cisgender girls/women	55.2	55.4	55.0	55.0
Cisgender boys/men	44.8	44.6	45.0	45.0
Race/ethnicity
Asian	0.9	0.9	0.8	0.8
Black/African American	0.6	0.7	0.7	0.7
Hispanic/Latine	1.4	1.5	1.5	1.5
Non-Hispanic White	93.7	93.7	93.6	93.6
Unlisted	3.3	3.3	3.3	3.3
Age in years	11.5 (1.6)	11.5 (1.6)	11.5 (1.6)	11.5 (1.6)
State-level covariates
Median household income in $	35,954 (2695)	35,951 (2699)	35,948 (2699)	35,947 (2699)
Gini income inequality ratio	0.6 (0.0)	0.6 (0.0)	0.6 (0.0)	0.6 (0.0)

Note. M = mean, SD = standard deviation.

Fig. 2 presents the overall and gender identity-specific prevalence of disordered eating over the study period. In general, chronic dieting was the most commonly reported outcome and purging was the least commonly reported. Prevalence was higher among girls/women than among boys/men for all outcomes at each study wave, with the exception of overeating by the 20-year-post-baseline follow-up.

Fig. 2.

Prevalence of disordered eating across the study period.

3.2. Main analyses

Table 4 presents results from the first two sets of SCMMs. In models including the full sample and using data from 1996 to 2007, each additional year of living in a high structural sexism state was associated with a 5% increase in risk of purging (95% CI: 3%, 7%), an 8% increase in risk of binge eating (95% CI: 6%, 10%), and a 9% increase in risk of overeating (95% CI: 8%, 11%). Further, participants in the top tertile of cumulative exposure by the end of the exposure window (i.e., those who had lived in a high structural sexism state for four or more years) had 1.48 times the risk of purging (95% CI: 1.31, 1.67), 1.63 times the risk of binge eating (95% CI: 1.44, 1.85), and 1.72 times the risk of overeating (95% CI: 1.57, 1.89) compared to participants in the lower two tertiles (i.e., those who had lived in a high structural sexism state for less than four years). Effect estimates for chronic dieting were mostly non-significant. Models using data from 1996 to 2016 produced similar or only slightly attenuated estimates, indicating that the observed effects persisted through the 20-year-post-baseline follow-up.

Table 4.

Population-averaged effect estimates for the relationship between cumulative exposure to U.S. state-level structural sexism and disordered eating, overall and by gender identity.

	Chronic dieting	Purging	Binge eating	Overeating
	RR (95% CI)	RR (95% CI)	RR (95% CI)	RR (95% CI)
Study period: 1996–2007
Full sample
Cumulative exposure to structural sexisma	1.02 (1.00, 1.03)	1.05 (1.03, 1.07)	1.08 (1.06, 1.10)	1.09 (1.08, 1.11)
Top tertile of cumulative exposureb	1.11 (1.01, 1.22)	1.48 (1.31, 1.67)	1.63 (1.44, 1.85)	1.72 (1.57, 1.89)
Cisgender girls/women
Cumulative exposure to structural sexisma	1.02 (1.00, 1.04)	1.05 (1.03, 1.07)	1.07 (1.05, 1.10)	1.07 (1.06, 1.09)
Top tertile of cumulative exposureb	1.15 (1.03, 1.28)	1.44 (1.27, 1.63)	1.57 (1.38, 1.78)	1.58 (1.42, 1.76)
Cisgender boys/men
Cumulative exposure to structural sexisma	0.96 (0.92, 1.00)	1.01 (0.94, 1.10)	1.07 (1.01, 1.13)	1.12 (1.09, 1.15)
Top tertile of cumulative exposureb	0.81 (0.64, 1.02)	1.18 (0.77, 1.82)	1.50 (1.10, 2.04)	1.95 (1.65, 2.30)
Study period: 1996–2016
Full sample
Cumulative exposure to structural sexisma	1.02 (1.00, 1.03)	1.03 (1.01, 1.05)	1.08 (1.07, 1.10)	1.09 (1.07, 1.10)
Top tertile of cumulative exposureb	1.12 (1.03, 1.22)	1.38 (1.23, 1.56)	1.63 (1.45, 1.84)	1.71 (1.56, 1.86)
Cisgender girls/women
Cumulative exposure to structural sexisma	1.01 (1.00, 1.03)	1.03 (1.01, 1.05)	1.07 (1.05, 1.09)	1.06 (1.05, 1.08)
Top tertile of cumulative exposureb	1.12 (1.01, 1.24)	1.34 (1.18, 1.51)	1.52 (1.34, 1.73)	1.53 (1.38, 1.70)
Cisgender boys/men
Cumulative exposure to structural sexisma	1.00 (0.97, 1.04)	1.02 (0.95, 1.09)	1.13 (1.08, 1.18)	1.13 (1.10, 1.15)
Top tertile of cumulative exposureb	0.97 (0.79, 1.17)	1.13 (0.75, 1.71)	1.80 (1.37, 2.36)	2.03 (1.74, 2.37)

Note: RR = relative risk, CI = confidence interval.

Effect estimates are derived from modified Poisson sequential conditional mean models adjusted for individual-level age and race/ethnicity and state-level median household income and Gini income inequality ratio.

^aEffect estimates describe change in risk for each additional year of living in a high structural sexism state.

^bCorresponds to four or more years of living in a high structural sexism state (referent: participants in the lower two tertiles of cumulative exposure, i.e., those who lived in a high structural sexism state for less than four years).

Turning attention to the models stratified by gender identity, we found that each additional year of living in a high structural sexism state was associated with an increased risk of purging, binge eating, and overeating among girls/women. Additionally, girls/women who had lived in a high structural sexism state for four or more years had a significantly higher risk of all outcomes (including chronic dieting) compared to those who had lived in a high structural sexism state for less than four years. Among boys/men, cumulative exposure to state-level structural sexism was associated with an increased risk of binge eating and overeating only. Again, all observed effects persisted through the 20-year-post-baseline follow-up.

Table 5 presents results from the third set of SCMMs. In models using data from 1996 to 2007 and compared to boys/men who had lived in a high structural sexism state for less than four years, boys/men who had lived in a high structural sexism state for four or more years had significantly higher risks of binge eating and overeating, and girls/women had significantly higher risks of all outcomes regardless of cumulative exposure; however, effect estimates were largest for girls/women who had lived in a high structural sexism state for four or more years (chronic dieting RR: 3.69, 95% CI: 3.04, 4.34; purging RR: 10.17, 95% CI: 7.36, 12.98; binge eating RR: 7.92, 95% CI: 5.82, 10.01; overeating RR: 3.70, 95% CI: 3.08, 4.31).

Table 5.

Population-averaged effect estimates for the relationship between cumulative exposure to U.S. state-level structural sexism and disordered eating, with effect measure modification by gender identity.

	Chronic dieting		Purging		Binge eating		Overeating
	RR (95% CI)	RERI (95% CI)a	RR (95% CI)	RERI (95% CI)a	RR (95% CI)	RERI (95% CI)a	RR (95% CI)	RERI (95% CI)a
Study period: 1996–2007
Gender identity x tertile of cumulative exposure to structural sexismb
Cisgender girls/women, top tertile	3.69 (3.04, 4.34)	0.64 (0.18, 1.10)	10.17 (7.36, 12.98)	2.77 (1.24, 4.30)	7.92 (5.82, 10.01)	2.21 (0.93, 3.50)	3.70 (3.08, 4.31)	0.36 (−0.15, 0.87)
Cisgender girls/women, lower tertiles	3.23 (2.79, 3.67)		7.13 (5.36, 8.90)		5.10 (3.91, 6.28)		2.33 (2.03, 2.63)
Cisgender boys/men, top tertile	0.82 (0.61, 1.02)		1.27 (0.77, 1.78)		1.61 (1.15, 2.07)		2.01 (1.86, 2.34)
Study period: 1996–2016
Gender identity x tertile of cumulative exposure to structural sexismb
Cisgender girls/women, top tertile	3.40 (2.84, 3.96)	0.40 (0.00, 0.81)	9.19 (6.68, 11.70)	2.03 (0.70, 3.36)	7.01 (5.26, 8.75)	1.44 (0.36, 2.52)	3.27 (2.76, 3.78)	0.04 (−0.43, 0.50)
Cisgender girls/women, lower tertiles	3.04 (2.65, 3.43)		6.96 (5.26, 8.66)		4.66 (3.65, 5.68)		2.13 (1.88, 2.38)
Cisgender boys/men, top tertile	0.96 (0.76, 1.16)		1.20 (0.73, 1.66)		1.91 (1.42, 2.39)		2.11 (1.78, 2.43)

Note. RR = relative risk, CI = confidence interval, RERI = relative excess risk due to interaction.

Effect estimates are derived from modified Poisson sequential conditional mean models adjusted for individual-level age and race/ethnicity and state-level median household income and Gini income inequality ratio.

^aRERI values range from negative to positive infinity.

^bReferent: cisgender boys/men in the lower two tertiles of cumulative exposure.

The significant RERI values for chronic dieting, purging, and binge eating provide evidence of effect measure modification by gender identity for these outcomes. Here, a RERI value > 0 indicates that girls/women who had lived in a high structural sexism state for four or more years experienced a disproportionate (i.e., “excess”) increase in disordered eating risk, relative to the effect that such length of exposure had on disordered eating risk among boys/men. Taking purging as an example, if cumulative exposure to state-level structural sexism had the same effect on purging risk for girls/women and boys/men, we would have observed an RR of 7.40 for girls/women who had lived in a high structural sexism state for four or more years (where 7.40 is given by the sum of state-level structural sexism’s independent effect and girls’/women’s risk increase relative to boys/men in the absence of exposure; see the Technical Appendix for details). Instead, we observed an RR of 10.17, which was significantly higher than expected. With the exception of the significant RERI value for chronic dieting, all observed effect measure modification effects persisted through the 20-year-post-baseline follow-up.

3.3. Sensitivity analyses

Results from the sensitivity analyses are presented in the Supplementary Materials. All findings were robust to using alternative exposure operationalizations and measures; on average, cumulative exposure to state-level structural sexism in the reproductive rights domain and in terms of relative gender inequality was associated with the highest risk of disordered eating (Supplementary Tables 3–14). Effects were attenuated after adjusting for prior outcome, yet most remained statistically significant (Supplementary Tables 15–16).

4. Discussion

In this prospective cohort study using a large U.S.-based sample of cisgender people, we found that each additional year of living in a state characterized by a high degree of structural sexism increased disordered eating risk by 5–9%. However, the magnitude of effect differed by gender identity: girls and women had higher risk increases compared to boys and men on average, and they experienced excess risk of chronic dieting, purging, and binge eating when exposed to particularly high doses of state-level structural sexism. All observed effects were robust to adjustment for individual-and state-level confounders and persisted by the 20-year-post-baseline follow-up.

Our findings are consistent with the emergent literature on structural sexism and population health (Homan, 2019; King et al., 2020), which provides accumulating evidence for a link between state-level structural sexism and gendered mental health inequities in the U.S. (Chen et al., 2005; McLaughlin et al., 2011). They are also consistent with studies that have documented cross-sectional associations between perceived and/or internalized sexism, as well as related adverse social experiences (e.g., sexual objectification), and disordered eating among cisgender girls and women (Piran and Cormier, 2005; Piran and Thompson, 2008; Sabik and Tylka, 2006; Garaigordobil and Maganto, 2013; Hayes et al., 2021; Moradi and Huang, 2008). We are not aware of any studies that have directly examined the relationship between structural sexism or its downstream manifestations and disordered eating among cisgender boys and men; however, Homan (2019) found that structural sexism is associated with poorer self-rated health among adult cisgender men, and feminist theoretical work on disordered eating is increasingly considering the unique ways by which systematic gender inequality may increase risk for people of all gender identities (Springmann et al., 2020), as we elaborate upon below. We thus provide novel evidence that bridges and extends these literatures by showing how cumulative exposure to state-level structural sexism from childhood through young adulthood increases both cisgender girls’ and women’s and boys’ and men’s risk of disordered eating, albeit differentially to contribute to gendered inequities in risk at the population-level over time.

There are several possible causal explanations for these patterns. First, proximal disordered eating risk factors – especially those that disproportionately affect cisgender girls and women as compared to cisgender boys and men e.g., sexual objectification experiences, appearance expectations) – may be more prevalent in structurally sexist states; as such, long-term residence may increase the frequency with which these factors are encountered. Indeed, a large body of research has shown that rates of sexual harassment and violence (both of which are related to sexual objectification) are highest in states characterized by high levels of systematic gender inequality (Roberts, 2011), and there is some evidence to suggest that gender-related stereotypes and resultant beliefs about how girls’ and women’s bodies “should” look are patterned in ways that mirror the geography of structural sexism (Jager et al., 2002; Price et al., 2021). This explanation aligns with ecosocial theory’s notion of “pathways of embodiment” (Krieger, 2001, 2012), as well as with how the developmental theory of embodiment positions gendered disordered eating risk factors as “epiphenomena” of systematic gender inequality (Piran, 2010, 2017).

Second, drawing on the broader body of feminist scholarship on disordered eating, more structurally sexist states may also be more likely to enact social penalties (e.g., reduced status, power, and resources) to girls and women who fail to meet socially constructed and unrealistic thinness-oriented beauty standards (Fikkan and Rothblum, 2012). Moreover, in places where girls and women lack equal political power, access to social and economic capital, and full bodily autonomy, they may be incentivized to use beauty as a form of social currency, increasing their likelihood of using dangerous behaviors such as restriction and purging to meet the largely unattainable standards. Similar processes could help explain why structural sexism increased risk of overeating for boys and men, (e.g., more structurally sexist states may enact social penalties against those who fail to meet masculine body and eating norms) (Smolak et al., 2012; Griffiths and Yager, 2019).

Considering the differences in findings by gender identity and by outcome provides further insight into the complex mechanisms by which structural sexism may shape the social patterning of disordered eating. Specifically, we found that structural sexism was associated with an increased risk of all four outcomes among cisgender girls and women, yet only of overeating among cisgender boys and men; we also found that cisgender girls and women in the top tertile of cumulative exposure experienced excess risk of chronic dieting, purging, and binge eating such that structural sexism partially explained the observed inequities in these behaviors. Notably, these differences directly parallel gendered divergences in both the motivations for using and the social constructions of various forms of disordered eating. For example, chronic dieting and purging are typically used for weight loss purposes, especially as a means of attaining the thin-ideal, which is largely expected of girls and women (Smolak et al., 2012). Further, restrictive- and compensatory-type behaviors more broadly are conceptualized as feminine and even as part of “doing” womanhood (Smolak et al., 2012). Conversely, overeating may be a means of attaining a more muscular ideal that is increasingly expected of boys and men, and is overwhelmingly considered to be a masculine behavior (Smolak et al., 2012; Lavender et al., 2017). As such, our findings may indicate that structural sexism cultivates an environment in which people of all genders are pressured to engage in dangerous behaviors to embody and/or perform normative gender roles, which is consistent with recent research (Griffiths and Yager, 2019). They also suggest that whether or not structural sexism exerts “universal harm” is outcome-dependent. It is also worth noting that all statistically significant estimates for chronic dieting were small, which could reflect the fact that dieting is largely a normative behavior in the U.S. and that cultural messages encouraging the pursuit of weight loss via dieting are pervasive regardless of where one lives.

Of course, there are also non-causal explanations for our findings. As is true in all observational studies, unmeasured confounding may have resulted in exchangeability violations. Detection biases are another possibility. For example, cisgender boys and men may be less likely than cisgender girls and women to report chronic dieting or purging behaviors in more structurally sexist states due to processes similar to those described in the previous paragraph, which could have inflated our RERI estimates. More broadly, there are well-recognized challenges in using epidemiology’s potential outcomes framework to estimate the effects of complex social factors such as structural sexism (e.g., near-certain consistency violations) (Rehkopf et al., 2016); following the argument laid out by Prins et al. (2020), we thus suggest that our findings be thought of as approximating a possible “realized causal effect” (i.e., the effect that structural sexism had on disordered eating risk in the past), rather than a specific “intervention effect” (i.e., the effect that a hypothetical intervention to mitigate structural sexism would have on disordered eating risk in the future).

4.1. Strengths and limitations

Our study has several strengths, including the use of prospective data and rigorous epidemiologic methods for causal inference, and being the first to our knowledge to examine a structural determinant of gendered inequities in disordered eating. This shift towards directly analyzing structural sexism rather than its downstream proxies is important because it problematizes systems, not individuals, which holds implications for how these inequities will be conceptualized and addressed. We also outline the application of a novel approach to incorporating a life-course perspective into structural sexism and population health research more broadly, for which there have been explicit calls (Homan et al., 2021).

Our study also has several limitations. First, GUTS is not a nationally-representative sample – participants are predominantly cisgender, non-Hispanic White, and are all children of nurses with four-year college degrees. There are documented differences in disordered eating risk between cisgender and transgender/gender diverse populations, as well as by race/ethnicity and socioeconomic status (Nagata et al., 2020; Rodgers et al., 2018; Sonneville and Lipson, 2018); further, there is some evidence suggesting that healthcare providers (including nurses) represent a high-risk group for disordered eating (King et al., 2009), which could increase risk among their children. Thus, our findings may not generalize beyond the cohort.

Second, and as previously noted, the measurement of gender identity in GUTS does not adhere to current best practices. Relying on mothers’ report of their child as either a “girl” or “boy” at student enrollment may have led to the misclassification of transgender/gender diverse participants, as well as the erasure of those with non-binary identities. Our operationalization of gender identity additionally considered survey switch requests; however, this approach still upheld normative gender binaries and could have further contributed to erasure for participants whose gender identity differed from their mothers’ report and the limited array of gendered survey options.

Third, although we used a comprehensive measure of structural sexism that is increasing used in epidemiologic research on gendered mental health inequities (Chen et al., 2005; McLaughlin et al., 2011), it combined indicators of absolute and relative gender inequality, as well as indicators of the various domains in which gender inequality is perpetuated. Scholars from the fields of criminology and sociology have long noted that such indicators are likely to exert unique effects on population health (Roberts, 2011; Vieraitis et al., 2015); indeed, results from our sensitivity analyses suggest that there may be some meaningful differences regarding their effects on disordered eating risk. Interestingly, the use of distinct absolute and relative indicators has been used to test competing hypotheses of how intervening on structural sexism would impact gendered health inequities (e.g., “amelioration” vs. “backlash”) (Roberts, 2011; Vieraitis et al., 2015), suggesting that future research aiming to obtain more precise causal estimates should take these distinctions into careful consideration. Moreover, our measure did not capture aspects that may be especially relevant to disordered eating in both material (e.g., the prevalence of and policies regarding sexual violence, the relative influence of the food and fashion industries) and cultural dimensions (e.g., those that describe hegemonic gender role and institutional logics, including the acceptance and enforcement of gendered beauty ideals) (Piran, 2010). As such, we may have underestimated the extent to which structural sexism contributes to gendered inequities in disordered eating; developing and validating new measures that tap into these outcome-specific elements will be important.

Fourth, we did not have data on family socioeconomic status or the prevalence of mental health service, potential individual-and state-level confounders, respectively. We were also unable to adjust for past-year inter-state moves, a potential individual-level confounder. Such moves were infrequent (reported by ~1% of participants each wave) and thus were non-existent at some cumulative exposure levels, indicating a positivity violation and leading to issues with model convergence.

Finally, we did not consider potential risk heterogeneity within gender identity groups (e.g., by race/ethnicity, class, or sexual orientation), despite emerging research showing important intersectionality between these and other dimensions of identity and experience in relation to disordered eating (Burke et al., 2020, 2021; Beccia et al., 2019, 2021; Austin et al., 2013). We are currently conducting follow-up analyses that examine whether membership in groups jointly defined by gender identity and sexual orientation modifies the relationship between structural sexism and disordered eating risk, however, data sparseness precludes our ability to do the same with race/ethnicity and class. To increase breadth of applicability of this line of work, it will be imperative for subsequent studies to use more representative data sources and to explicitly incorporate an intersectional lens.

Other important areas of future research include: using measures that quantify structural sexism as manifest in different settings (e.g., schools, workplaces) and scales (e.g., county- or city-level); estimating the effect of structural sexism on additional relevant outcomes (e.g., muscularity-oriented behaviors, eating disorder diagnoses); further probing the life-course relationship between structural sexism and disordered eating risk (e.g., identifying potential critical/sensitive periods); considering the ways by which structural sexism intersects with other forms of structural inequity (e.g., structural racism) to shape the social patterning of disordered eating; and elucidating the mediators of these relationships.

5. Conclusions and implications

Given the considerable individual and societal costs of eating disorders and related behaviors (Streatfeild et al., 2021), advancing understanding of their structural drivers in the U.S. population overall and across and within gender identity and other social groups is essential to improving population mental health. In this study, we provide novel evidence for the role of structural sexism by demonstrating how cumulative exposure contributes to disproportionate risk increases for cisgender girls and women relative to cisgender boys and men over time. If the relationships we observed are causal, this suggests that improving gender equity at the state-level (e.g., through implementing laws, policies, and institutional practices that expand access to status, power, and other resources to people of all gender identities) could help reduce both the absolute population burden and inequitable social patterning of disordered eating – and likely, mental health problems more generally – in the U.S.