With over 1 billion users [1], Instagram is one of the most widely used social media platforms in the world. It is, first and foremost, a visual platform, centered around sharing user-generated photos and videos accompanied by captions. Instagram allows users to search for photos by “hashtag.” A hashtag is a word or phrase preceded by the pound (#) symbol, which, when typed in the caption of the photo, makes the photo searchable by that term. Importantly, Instagram users have the option to make their photos private or public. Public photos are visible to anyone—even those without an Instagram account. Without an Instagram account, site visitors can only scroll through approximately 30 photos before either logging in or refreshing the page and starting anew. Visitors without accounts are likewise prohibited from clicking on photos returned in a search to access poster usernames or photo captions.

In addition to its popularity for social networking, Instagram is a growing forum for discussing health-related topics. The medical community has become interested in these discussions, such as those regarding HIV, cancer, vaping, alcohol, and self-harm behavior [2-6]. As of September 2020, there are 20.7 million posts tagged with #mentalhealth on Instagram and 123 million tagged with #health [7,8]. Instagram provides insight into the beliefs and attitudes circulating amongst users and allows health care providers to understand the social influence on health-related behavior.

We aimed to investigate whether men and nonwhite individuals are underrepresented in Instagram photos tagged with #mentalhealth (versus photos tagged with #health) to understand better how gender and race-based representations present on this popular social media platform and to discuss the implications. The rationale for this project stems from decades of research—on gender, race, and mental health stigma—and is twofold. We are interested first in whether historical patterns of gender and race-based stigma make themselves manifest in the influential social media realm, and in anticipating potential consequences of underrepresentation on this platform in particular.

Several risk factors known to be positively associated with symptoms of depression and anxiety are experienced more frequently by racial minorities, chief among which is discrimination [9]. Race also affects the rate at which patients receive treatment for mental illness, compounding these risks. Compared to 49.1% of non-Hispanic white patients that receive treatment, African American or non-Hispanic Black patients are treated at a rate of 30.6%, Hispanic or Latino patients 32.9%, and non-Hispanic Asian patients 24.9% [10]. Many implicate greater stigma in treatment discrepancies, especially among Asian and African American individuals. For example, in a study of attitudes toward mental health treatment in a college-aged student population, 63% of African American students and 52% of white students perceived stigma toward seeking care [11]. These figures are even more concerning in the current global climate when minority communities face a disproportionate death toll from the COVID-19 pandemic [12] and are leading a historic civil rights movement that brings intense focus to past personal traumas and injustice [13]. All these factors have created a sense of urgency in supporting the mental health of Black communities.

Mental health stigma is of particular relevance to men as well. A large historical body of research on gender, stigma, and mental health has shown that men associate mental illness and help-seeking with a deviation from the masculine ideal [14]. The development of this phenomenon may date back to the mid-1900s when the word “depression” first appeared in the Diagnostic and Statistical Manual [14]. It has been noted that male patients were “conspicuously absent” from the process of developing diagnostic criteria for depression and that early antidepressant clinical trials focused largely on women [14]. As a result, the tearful female patient became almost synonymous with depression [14].

The idea of depression and other mental health concerns as a “woman’s problem” is alive and well today. The most troubling evidence that diagnostic criteria may overlook men is the high rate of suicide in men despite a low rate of depression [15]. It is thought that men cover up emotional turmoil for fear of stigma; men are less likely than women to report symptoms of mood disorders, and when they do, they report their symptoms as less severe compared to women’s self-reports [15,16]. Additionally, men are more likely than women to list stigma as a reason for not seeking care [17]. Relying on negative male stereotypes to explain a lack of help-seeking behavior creates what some have described as a culture of “victim-blaming” [15]. Victim-blaming ignores cultural, societal, economic, and personal factors that may in fact play a large role in influencing men’s behavior around mental health, including norms set by society, peers, and today, social media.

Given its ubiquity, social media may affect users’ beliefs and attitudes toward mental health, and because it is a visual platform, gender and race feature prominently in Instagram posts. The gender and race of individuals in photos may influence how viewers feel about certain topics. For example, the gender and race of individuals in photos tagged with #mentalhealth may influence user attitudes on mental health—their beliefs regarding who talks about it, and who cares about it. In a recent qualitative study about men’s discourse surrounding mental health, one participant expressed awareness that there is a lack of “manly men” discussing mental health problems in public, while women do so more freely, creating space for other women to do the same [14].

Described in this introduction is a vicious cycle; stigma may lead to underrepresentation on Instagram, and this underrepresentation may lead to even greater alienation of men from discussions about mental health unless directed interventions are made. With this background in mind, we first predicted that women would be featured more frequently than men in photos tagged with #mentalhealth. Second, we predicted that white subjects would be featured more frequently than nonwhite subjects in photos tagged with #mentalhealth. These hypotheses are based on the assumption that experiencing more stigma in the “real world” would make individuals less likely to associate themselves visually with the topic of mental health on social media.

The hashtag #mentalhealth was chosen as a topic that, unlike #depression or #anxiety, lacks epidemiologic bias. It is well established that more women are diagnosed with depression and anxiety than men [14]. Mental health is not a diagnosis but describes the totality of emotional, social, and psychological factors that impact behavior, and therefore as a concept, is superimposable on all individuals regardless of gender or race. #Mentalhealth was also chosen to facilitate simple control. Our analysis included an identical data collection process for photos tagged with #health. Because “mental health” falls under the umbrella of “health,” we considered the number of men and women in photos tagged with #health to more closely reflect baseline Instagram representation. We hypothesized that equal numbers of male and female and white and nonwhite subjects would be featured in photos tagged with #health.

In planning this project, investigators established that in order to investigate the research question of whether men and minorities are underrepresented in photos tagged with #mentalhealth as compared to #health, they would first have to generate a large pool of Instagram photos for each of these hashtags. The investigators would then look together at each photo, and each investigator would record the gender(s) and race(s) of subjects in these photos based on individual investigator’s interpretations. Because investigators predicted that they would disagree on gender or race for some subjects, data would need to be collected so that kappa agreement scores could be calculated.

This plan came to fruition in a multistep process with three investigators of different genders and racial backgrounds involved in data collection. The investigators met on nine separate days in July 2020 via teleconference. One investigator shared their screen and performed two internet searches in two separate tabs: one for “#mentalhealth Instagram” and another for “#health.” This search returned the respective links to access Instagram.com galleries of photos tagged with #mentalhealth and #health. Notably, the investigator was not logged into Instagram, such that only publicly available photos were visible. For both groups, photos were sorted such that the most recently posted appeared first, to avoid the bias of seeing the most “popular” photos. Photos featuring real human subjects and not meeting exclusion criteria were included for analysis. A detailed description of exclusion criteria is outlined in Multimedia Appendix 1.

One by one, eligible images were assigned a unique image ID, and investigators individually recorded the number of males, females, white, nonwhite, and racially unclassifiable subjects in each photo in separate excel sheets. During this process, investigators would be prompted to log in to Instagram, at which point the team would switch to data collection for the opposite hashtag after refreshing the page—this method avoids the need to log in. The popularity of these hashtags allowed this process to continue in a back-and-forth fashion, as dozens of new photos were added to each collection every few minutes.

On the first 8 data collection days, investigators recorded 25 photos from each hashtag. On the ninth data collection day, an additional 15 were added in each category if any data had to be removed after data collection. Investigators did not share their opinions about the gender or race of individuals in photos and did not share their data spreadsheets until data collection was complete.

Throughout this process, the authors used the term “gender” to describe the expressed gender identity of individuals in Instagram photos as investigators perceive it when forced to categorize as male or female. The term “white” was used among investigators to refer to individuals appearing to be of European descent. The term “nonwhite” refers to any individuals not meeting this definition, including Native and Indigenous populations.

Chi-squared analyses were used to determine the significance of differences between the total number of men versus women and white versus nonwhite subjects featured in each hashtag category. Investigators decided to analyze a minimum of 200 photos for each hashtag to have an 80% power to determine a 12% difference with alpha .05 or P value. In calculating chi-square analyses for race, individuals deemed to be “unclassifiable race” were excluded such that only a white versus nonwhite comparison was performed. Cohen kappa was used to calculate agreement between each pair of raters for each type of rating (gender or race) in each hashtag. Racially unclassifiable individuals were removed from this analysis, such that agreement focused on subjects whom raters deemed to be white or nonwhite. Chi-square values, P values, and kappa agreement scores were calculated using GraphPad Prism.

Out of respect for Instagram user privacy, included photos were not clicked on to reveal usernames or any other information about the posters. Only publicly available photos were included, and photos were not saved in any way. No identifying information was recorded or collected from individuals featured in photos. No one was contacted, and informed consent was waived. This research was exempted by the Institutional Review Board of Wayne State University (protocol #083519B3X).

Following data collection, ten mental health photos and six health photos were removed from all three investigators’ data sets due to transcription errors committed by at least one rater. For example, one rater may have marked a subject as both white and nonwhite, forgot to record gender, or missed a subject in a photo altogether. In order to calculate kappa scores accurately, these photos were removed from data analysis. As such, a total of 205 mental health photos and 209 health photos were analyzed. Raters categorized a total of 250 subjects in mental health photos and 261 subjects in health photos. Total numbers of males, females, white, nonwhite, and racially unclassifiable subjects observed by each investigator for #mentalhealth and #health are shown in Tables 1 and 2.