Indirect Measurement of Intersectionality Using Data from the Understanding America Study

Introduction

In this article, we show how we created a tool that social scientists across disciplines can use to study intersectionality and structural barriers. Intersectionality is the concept that an individual has multiple overlapping identities, such as sex and race and ethnicity, which can be subject to discrimination both individually and in combination. These identities are often associated with existing structural barriers, such as those encountered by Black people and women. For example, a Black woman has a merged identity as both a woman and a Black person that differs from her societally perceived identity as a member of either group singly.

We focus on people's perceptions about overall societal attitudes toward people in particular demographic groups rather than the perspectives of individuals about their own intersectional identities. We explore how intersectionality can amplify the discrimination experienced by certain groups. We also examine how discrimination, as measured by societal attitudes toward marginalized groups, can create structural barriers for those groups. Although the full breadth of the latter examination is beyond the scope of this article, social scientists can apply our measure of comparative sociopolitical power to their various fields of expertise to model the relationship between intersectionality and discrimination.

Defining “Structural Barriers”

Simms and others (2015, 4) define structural barriers as “obstacles that collectively affect a group disproportionately and perpetuate or maintain stark disparities in outcomes.” Hong and others (2021, 31) define structural barriers in the context of a job search as “the condition that no matter how good the person's qualifications may be, elements within the social and economic structures make it difficult for the person to obtain employment. These elements include secondary labor market; racial discrimination; immigrant status; gender discrimination; lack of jobs; transportation; neighborhood/location; and general structural factors.” Hong and others also examine how those factors affect the administration of income support programs, which is directly relevant to the Social Security Administration in its role of administering the Supplemental Security Income program.

History of the Study of Intersectionality

Among the origins of the concept of intersectionality is a 1976 case heard in U.S. District Court for the Eastern District of Missouri, DeGraffenreid v. General Motors Assembly Division. Five Black women who had been fired by General Motors (GM) brought a discrimination lawsuit against their former employer. The plaintiffs argued that they were discriminated against because they were both Black and women, not solely because they were Black and not solely because they were women: They acknowledged that GM hired Black men and White women. Ultimately, the judge denied that argument, writing that “the initial issue in this lawsuit is whether the plaintiffs are seeking relief from racial discrimination, or sex-based discrimination. The plaintiffs allege that they are suing on behalf of Black women, and that therefore this lawsuit attempts to combine two causes of action into a new special sub-category, namely, a combination of racial and sex-based discrimination.” The court decided that there was no protected class to be found at the intersection of the two identities and ruled in favor of GM. At the time this case was being adjudicated, a theory of intersectionality was arising organically among the Black feminist community (for example, Smith 1983).

Crenshaw (1989) coined the term intersectionality in a law review article revisiting DeGraffenreid v. GM to explore systemic racism in general and its effects against Black women in particular. Crenshaw argued that it was impossible to separate the identity of being Black from the identity of being a woman. Instead, the two identities create a new intersectional identity, in which the discrimination associated with being Black and the discrimination associated with being a woman are amplified by their coexistence. Crenshaw (1991) identified three forms of intersectionality: representational, political, and structural. Representational intersectionality refers to the way intersectional identities are portrayed in culture and media. Political intersectionality refers to the way that an intersectional identity can combine two or more marginalized groups for whom some political objectives may be at cross-purposes. Structural intersectionality refers to the way various institutions perpetuate or eliminate the barriers faced by people with marginalized intersecting identities. Intersectionality describes the effects of multiple existing structural barriers in combination (Hong and others 2021), as the DeGraffenreid plaintiffs attempted to argue: They faced the structural employment barriers that women faced coupled with the structural employment barriers that Black people faced. That combination amplified the structural barriers that they would have faced had they been either Black men or White women.

Although we aim to create a measure of all forms of intersectionality, we see our model of structural intersectionality as most useful to the Social Security Administration and other government agencies in efforts to prevent discrimination in their hiring and employee development policies¹ and in administering their programs. Since Crenshaw (1991), numerous studies have used intersectionality to describe the unique combinations of challenges faced by people with particular sex-and-race identities across various realms, including politics (Hancock 2007; Holvino 2010), education (McCall 2005; Jones 2003), health care (Kelly 2009; Viruell-Fuentes, Miranda, and Abdulrahim 2012), and economics (Ladson-Billings and Tate 1995). Hong and others (2021) focused on labor dynamics and used data from a small sample (388 respondents) to construct a Perceived Employment Barrier Scale. Yet all those studies tend to focus on one particular aspect of intersectionality, while our measure is meant to model multiple elements and provide a comprehensive quantitative measure of intersectionality that social scientists can use to examine empirically how intersectional identities affect access to social services, societal power, and government benefits.

In our research, we explore whether a measure of intersectional identities can be created using a novel indirect regression approach applied to survey data on societal perceptions of different groups' social and political power. We seek to understand how the simultaneity of race or ethnicity and sex affect different groups' social standings and power in society by creating a quantitative metric we call the Sociopolitical Power Scale (SPPS). Although our examination is purely methodological, we propose ways that the SPPS could be used in models measuring social groups' interactions with government agencies and programs.

Data and Methods

We use data from the Understanding America Study (UAS), a nationally representative survey fielded by the University of Southern California's Center for Economic and Social Research, to construct the SPPS. UAS survey 135, titled “Health Insurance, Politics, and Social Attitudes and Values” and fielded May–June 2018, included a Social Construction module containing a series of questions addressing perceptions of population groups' relative societal and political power.

The UAS is an internet-based panel survey administered to participants aged 18 or older. UAS surveys cover a wide array of topics, including demographic and socioeconomic characteristics, political affiliation, financial literacy, and personality type.² If needed, participants are provided a tablet and internet connection. UAS 135 had 4,679 respondents among the 6,154 UAS panel participants at the time, providing a 76 percent response rate.³

Demographic Information

Table 1 shows summary demographic characteristics of the UAS 135 respondents. Women outnumbered men, 57 percent to 43 percent. The majority of respondents (72 percent) were non-Hispanic White, while 8 percent were non-Hispanic Black and 11 percent were Hispanic (any race). Respondents' average years of education (14.5) included some years after a high school diploma, and the mean household income was almost $65,000.

In the following subsections, we provide a description of the method we used to create the SPPS, along with an analysis of that method and a discussion of the applicability of the scale and its possible uses and extensions.

SPPS Theory and Method

We aim to demonstrate the amplification of discrimination (or privilege) that Crenshaw (1989) identified so that it can be integrated into empirical social science research. The theoretical origins of the SPPS come from the psychosocial theory of social constructions, or the examination of the creation and endurance of stereotypes. Berger and Luckmann (1967) framed social constructions as the process by which beliefs and perceptions about groups of people become institutionalized such that the collective belief endures and becomes a dominant perception that is thus internalized by members of the groups that are the subject of these perceptions. For example, the societal perception of the experience being a Black woman reinforces the actual experience of being a Black woman. This in turn fortifies the societal perception of Black women, which differs from the societal perceptions both of Black people overall and of women overall. This exemplifies what Crenshaw (1989) calls the amplification of identities.

The SPPS combines the perceptions of societal power and political power into a scalar measure to help us better understand how those factors influence individuals' interactions with government agencies and programs. Although the possible combinations of intersectional identities may number in the hundreds, we simplify this analysis by focusing on the intersection of sex and race or ethnicity and limiting the latter to four groups: non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, and Hispanic (any race).⁴ We test whether the SPPS results for a single-characteristic identity (such as being a Black person) differ from an intersectional identity (such as being a Black woman).

Constructing the SPPS

We use the responses to three UAS 135 questions to collect data on the perceived societal and political power of 13 population groups: men; women; White, Black, Hispanic, and Asian people; residents of suburban, urban, and rural areas; immigrants with a visa; immigrants without a visa; aged people; and people with a disability. The survey questions listed the groups in random order. We tested the questions using exploratory factor analysis in the STATA statistical software and determined that they were scalable—that is, amenable to inclusion in a scale. We also confirmed that the questions were loaded on a limited number of common factors, meaning that they are related by an underlying concept.

The three questions are listed below:

• Political power. “Please rate the following groups of U.S. residents in terms of their political power, that is, how much politicians and lawmakers care about what the group wants.” Response options consist of very low (1), low (2), moderate (3), high (4), and very high (5).
• Societal power. “Please rate the following groups of U.S. residents in terms of their level of societal power. By societal power we mean the ability that members of those groups have to get things done within U.S. society.” Response options consist of very low (1), low (2), moderate (3), high (4), and very high (5).
• Social construction. “Thinking about those groups of U.S. residents in terms of how U.S. society generally views them, would you say the view is mostly negative, positive or somewhere in between?” Response options consist of negative (1), somewhat negative (2), neither positive nor negative (3), somewhat positive (4), and positive (5).

We used principal component analysis (PCA) of the SPPS' components—political power, societal power, and social construction—to ensure that they are different measures of an underlying concept and are thereby not biased by collinearity or correlation. PCA is an appropriate method of factor analysis because it looks for similarities in measures that have no prior theoretical structuring or grouping of observations (Bartholomew and others 2008; Jolliffe 2002). Chard, Rogofsky, and Yoong (2017) used PCA for their scalar measures and in constructing the weighted scale they used to analyze financial behavior and to create their Retirement Planning Index. We use PCA to test the independence of the political and societal power measures and to give insight into how they may be scaled together to create the SPPS. If the measures capture the same underlying concept, then we can expect a factor analysis to show that the components load on a small number of factors. If instead they capture different concepts, then we should see the opposite, loading on many factors with no single factor explaining most of the variance (Bartholomew and others 2008; Jolliffe 2002).⁵

Table 2 shows the correlations among the responses to each of the three survey questions with respect to women. All of the correlation coefficients of the three questions lie between 0.5 and 0.6, suggesting that there is a substantial correlation. However, the correlations are not above the level (0.8) at which adding the second or third question would be of no analytical value.

Next, we created a scree plot of the eigenvalues for the three components (survey questions) for each of the 13 population groups (Chart 1). The scree plots indicate whether the information in the three questions could be summarized in a single index variable to represent sociopolitical power, or whether adding a second variable (second component) would provide significantly more information. An eigenvalue lower than 1 is commonly a clear indicator that the given component does not contain sufficient additional information to warrant attention.⁶ The scree plots for the 13 population groups look very similar. In each case, the eigenvalues for the first component are close to 2, while the eigenvalues for the second and third components are well below 1. That result confirms that we can use a single index to represent the sociopolitical power of each of the groups.

Next, we analyze the PCA results for weighting of each factor (survey question), shown in Table 3, for each of the 13 groups we studied. In the example of women, the weights are similar: 0.578 for societal power, 0.570 for social construction, and 0.584 for political power. These weights indicate the relative contribution of the responses to each of these questions in the overall perceptions of the sociopolitical power of women as a group.

Across all 13 groups, the weights for each of the three questions are very similar. Within any group, the most dissimilar weights are those for the societal power question (0.595) and the social construction question (0.553) for people with a rural residence (a difference of 0.042, or about 7.6 percent of the weight for the social construction question). This result indicates that people perceive a difference between the societal power of rural residents (how powerful they are as a group) and their social construction (how society views them as a group).

Table 4 shows descriptive statistics on the SPPS scores. Possible SPPS scores range from 1 (lowest) to 5 (highest). The unweighted results support existing concepts of intersectionality. For example, men have a higher SPPS score (that is, more perceived sociopolitical power) than women, and White people have a significantly higher SPPS score than people of another race or ethnicity. Black people and women, the identities that intersect in Crenshaw (1989), have some of the lowest SPPS scores. The weighted results are fundamentally similar. For instance, the weighted mean SPPS score for White people is higher than the unweighted mean SPPS score, but only by 0.02.

Measuring Intersectional Identities Using the SPPS

The survey questions do not cover intersectional identities so we cannot directly compute SPPS scores for those identities as we can for the 13 population groups examined in UAS 135. Instead, we use an indirect approach that begins with conducting regression analyses to test whether there are differences between men and women, and between racial and ethnic groups, in the perceived sociopolitical power of their own group.⁷ To accomplish this, we first analyze differences between male and female respondents in the perceived sociopolitical power of their own racial or ethnic group. For instance, we examine whether Black women perceive lower sociopolitical power for Black people than Black men do and whether Hispanic women perceive lower sociopolitical power for Hispanic people than Hispanic men do. Next, we conduct similar analyses of female respondents' perceptions of women's sociopolitical power across racial and ethnic groups. For instance, we examine whether Black and Hispanic women perceive lower sociopolitical power for women than White women do.

Table 5 shows the results of our regression analysis of the relationship between sex and the perceived sociopolitical power for each respondent's own racial or ethnic group. Results for each racial and ethnic group are shown with and without control variables (age, education, and household income). In both cases, the dummy variable indicates the independent variable of interest: a female respondent.⁸ The coefficients for the SPPS score for Black people are negative, showing that Black women perceive lower sociopolitical power for Black people than Black men do. The p-values of a test comparing the results for Black respondents to those for White people are both 0.05 or less. Similarly, Hispanic women perceive lower sociopolitical power for Hispanic people than Hispanic men do. We do not find such differences between the views of Asian men and women, and White women perceive slightly higher sociopolitical power for White people than White men do.

Table 6 shows the results of a regression analysis relating Black, Asian, and Hispanic women's perceptions of the sociopolitical power of women overall against White women's perceptions of the sociopolitical power of women overall. This sample includes only female respondents. As in Table 5, results are shown with and without control variables (age, education, and household income). The coefficient for SPPS scores of White female respondents relative to those of Black women (without control variables) is positive and significant. This indicates that White women perceive higher sociopolitical power for women overall than Black women do.⁹ Hispanic women also perceive lower sociopolitical power for women than White women do. There are no significant differences between the views of White and Asian women. Tables 5 and 6 together show that Black women have lower perceptions of Black people's sociopolitical power than Black men have, and lower perceptions of women's sociopolitical power than White women have. Likewise, Hispanic women perceive lower sociopolitical power for Hispanic people than Hispanic men do, and lower sociopolitical power for women than White women do.

To summarize our results, we find that UAS 135 respondents, empaneled as a representative sample of the American population, have the following perceptions:

1. Sociopolitical power differs significantly by race, ethnicity, and sex. For example, Black and Hispanic people are generally perceived as having less sociopolitical power than White people and women are seen as having less sociopolitical power than men.
2. Black and Hispanic women perceive lower sociopolitical power for their own race or ethnicity than their male counterparts perceive.
3. Black and Hispanic women perceive lower sociopolitical power for women than White women perceive.

These findings support the concept of intersectionality and underscore the issues that were discussed in Black feminist literature as the theory of intersectionality was being developed. The concepts that we measure with the SPPS may support efforts to improve political efficacy for certain groups.¹⁰

Conclusion

Overall, we have accomplished our goal of using an indirect approach to quantitatively assess intersectionality using survey data that combine factual elements (race, ethnicity, and sex) with attitudinal elements (perceptions of political and societal power). We combine those elements to empirically model intersectional identities. Further, our quantitative measures support the idea of the amplification of discrimination and privilege that Crenshaw (1989) discussed. Our results also complement Hong and others (2021, 47), who found that “elements of race and gender discrimination are given significant attention as co-occurring structural barriers.”

Limitations

The sample size for the UAS 135 Social Construction module was not large enough to allow us to examine intersectional identities at more than two levels (sex; race or ethnicity). However, as the UAS sample size increases, we envision the possibility of applying this method to study a third layer of identity, such as age, disability status, or another characteristic.

A second limitation is that the data we capture are from a single point in time, but they are influenced by a mosaic of societal forces that have come together over the years to create those perceptions. Those intrinsic factors include historic barriers that helped to create the situation the DeGraffenreid plaintiffs called to light and that continue to shape political and societal views today. Despite those limitations, we envision researchers using our SPPS, along with additional intersectional dimensions such as age, place of residence, disability status, and educational attainment, to see if any of those variables further amplify or decrease disparity in the SPPS.

Future Research

The SPPS can be useful for research on a variety of topics, many of which are particularly relevant to Social Security researchers. For example, it can be combined with the diverse data collected by the UAS, ranging from respondent retirement preparedness to policy preferences and experiences with government agencies.

We would like to see the SPPS applied to study people with disabilities (and disability program beneficiaries in particular). We would also like to see the SPPS used to evaluate the public's experiences with the Social Security Administration, perhaps by expanding on previous research on people's preferred channels of receiving program information, to identify any potential structural barriers that limit use of any of those channels. The SPPS could also be used to study the myriad issues related to employment such as the declining availability of private pensions, and to examine wealth accumulation for retirement, building (for example) on work by Kijakazi, Smith, and Runes (2019).

We envision the SPPS being used to determine how the COVID-19 pandemic's effects were distributed among different groups, and incorporated into studies on topics such as homeownership and incarceration, where systemic racism is known to be a persistent historical factor.¹¹ Again, although the SPPS is a snapshot measure of cumulative systemic barriers, it can illuminate how those historical factors have affected different groups in modern America.