Uploaded by David Nathan

Bauer 2014 Guidelines for Quant Intersectional HealthResearch

advertisement
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/343140477
Harnessing the Power of Intersectionality: Guidelines for Quantitative
Intersectional Health Inequities Research
Technical Report · December 2014
DOI: 10.13140/RG.2.2.10403.48169
CITATIONS
READS
3
1,015
5 authors, including:
Greta Bauer
Lisa Bowleg
The University of Western Ontario
George Washington University
109 PUBLICATIONS 5,507 CITATIONS
90 PUBLICATIONS 4,639 CITATIONS
SEE PROFILE
Ayden I Scheim
Drexel University
82 PUBLICATIONS 2,147 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Health in Middlesex Men Matters (HiMMM) View project
CIHR Sex and Gender Science Chair: Improving Methods for SGBA+ View project
All content following this page was uploaded by Greta Bauer on 22 July 2020.
The user has requested enhancement of the downloaded file.
SEE PROFILE
Harnessing the Power of Intersectionality
Guidelines for Quantitative Intersectional Health Inequities Research
Greta Bauer  Lisa Bowleg  Setareh Rouhani  Ayden Scheim  Soraya Blot
Greta Bauer;1,2 Lisa Bowleg;3 Setareh Rouhani;4,5 Ayden Scheim;1 Soraya Blot1
1. Epidemiology & Biostatistics, Schulich School of Medicine and Dentistry, The University of Western Ontario,
London, Ontario, Canada; 2. Women’s Studies and Feminist Research, The University of Western Ontario, London,
Ontario, Canada; 3. Department of Psychology, The George Washington University, Washington, DC; 4. Institute of
Population Health, University of Ottawa, Ottawa, Ontario, Canada; 5. Institute for Intersectionality Research and
Policy, Simon Fraser University, Vancouver, British Columbia, Canada
SUGGESTED CITATION
Bauer G, Bowleg L, Rouhani S, Scheim A, Blot S. Harnessing the Power of Intersectionality: Guidelines for
Quantitative Intersectionality Health Inequities Research. London, Canada; 2014.
ACKNOWLEDGEMENTS
This work was supported by an operating grant from the Canadian Institutes of Health Research (FRN#
MOP-130489). Ayden Scheim’s work was supported by Trudeau Foundation and Vanier Canada
Graduate Scholarships. Soraya Blot’s work was supported by a master’s student award in CommunityBased Research in HIV/AIDS from the Canadian Institutes of Health Research and a University Without
Walls fellowship award from the Ontario HIV Treatment Network. The authors wish to thank Guangyong
Zou for early discussion regarding this topic.
This work is licensed under a Creative Commons Attribution 4.0 International License
SUMMARY
Health inequalities are by definition a quantitative concept: the absence of equality or parity in healthrelated outcomes. The presence of numeric inequality across social groups is often of interest with
regard to inequity (injustice), particularly where those social groups represent categories embedded in
a context of historical and social marginalization. Originating in Black feminist scholarship, an
intersectionality framework offers the potential to document health inequalities at intersections of
social identity or location, address interactions between macro-, meso- or micro-level causes in
creating and reinforcing such inequalities, and develop or evaluate interventions that work within the
intersectional contexts of communities. However, to date it has been applied primarily in qualitative
research. We present ten guidelines for researchers about the design and conduct of quantitative
intersectionality research addressing health inequalities. Important considerations include theoretical
conceptualization, study design, measurement, statistical analysis, research process, knowledge
translation, and interpretation of research results.
INTERSECTIONALITY & QUANTITATIVE RESEARCH
Health inequity is often studied through analysis of inequalities across social groups reflecting different
experiences of social power and socio-structural marginalization. Health inequalities (or disparities)
represent a lack of equality (or parity) in health-related phenomena such as incidence of disease or
access to health care services, and are thus by definition a quantitative concept. These inequalities are
often studied over one or more unidimensional domains, resulting in research on racial, socioeconomic,
or sexual orientation inequalities. Such approaches implicitly assume that these domains of social identity
or location can be studied separately, and foreclose the possibility and importance of simultaneous
membership in multiple groups. Commonly used phrases such as “women and minorities” attest to this.1
While unitary and multiple approaches ignore heterogeneity within groups, an intersectionality
approach considers the intersections of groups to, in and of themselves, be sub-groups of interest.2 Since
social identities mutually constitute each other, it is difficult to understand health inequalities related to
one social identity, without considering that identity’s intersection with other social identities.3 For
example, the effect of U.S. women’s education level on infant mortality varies by race/ethnicity.4 Health
inequalities thus reflect complex intersections; not the mere addition of marginalizations and subtraction
of privileges to obtain some net level of inequality.
Intersectionality traces its academic origins to Black feminist scholars’ criticism of the exclusion of Black
women from anti-racist and White middle class feminist activism.1,5,6 Over the past quarter century,
intersectionality has taken root as a research paradigm within feminist and critical race scholarship,2,7,8
and been introduced within fields such as psychology9 and public health.1,10 However, most
intersectionality research is qualitative.10,11 More intersectionality-informed quantitative health
inequalities research would generate greater exploration of health inequalities across a range of
intersectional identities or social locations, promote understanding of potentially interacting causal
processes, and help identify solutions to complex intersectional inequalities.10 However, this potential has
not yet been realized.
Within the small and growing body of quantitative intersectional research, most studies lack an explicit
rationale for how intersectionality informs the analytic methods chosen. Only a few preliminary
investigations of quantitative intersectionality methods exist.10,12,13
Moreover, most quantitative intersectional research has used secondary analysis of cross-sectional
population surveys not designed for intersectional analyses,13 resulting in findings that have been
primarily descriptive.. While this meets one objective of intersectionality by describing experiences for
those at historically marginalized intersectional locations, and contributes to research on health
inequalities by providing greater specificity to the groups experiencing adverse outcomes, it omits
information about the interlocking macro-, meso- and micro-level processes that lead to or reinforce the
maintenance of inequalities. These processes, sometimes called “determinants of population incidence
rate”,14 “solution-focused variables”,15 or “fundamental causes,16 drive inequalities across population
groups, and thus may also serve as targets for intervention. These solution-focused variables may either
vary in frequency or level across groups, or alternately, they may have a stronger effect within specific
communities by interacting with other determinants of health.
1
Table 1. Multidisciplinary Glossary of Terms for Quantitative Intersectionality Research
Term
Definition
Additive model
Non-intersectional theoretical or statistical model that posits the effects of multiple domains
(e.g. race/ethnicity, age, sexual orientation) as effects that can be added. For example,
the mental health of a bisexual Asian youth would be constructed as the overall effects of
being bisexual + Asian + youth. In statistical regression models, is equivalent to a main
effects analysis. As concepts such as “double or triple jeopardy” connote, this model
implies that the addition of each identity results in greater oppression. Reflects Hancock’s
(non-intersectional) multiple approach.2
Additive-scale
interaction
Scale of statistical interaction in multiplicative multivariable models wherein interactions
(e.g. intersectional positions) are assessed against a null hypothesis that absolute effects of
membership in each group (separately) are added. A significant interaction detects an
excess or deficit of cases in contrast to what would be expected. Most relevant scale for
public health and causal analyses.10
Additive-scale
regression
Statistical model in which effects are added to produce measures of absolute (rather than
relative) effects (e.g. linear regression models).
Group-level variable
A variable for quantitative analysis that is measured for groups such as states, schools, or
neighborhoods (e.g. neighborhood rates of violence, school policies on bullying)
Individual-level variable
A variable for quantitative analysis that is measured for individuals (e.g. ethnic identity,
personal annual income)
Interaction
A causal interaction refers to the ways that health outcomes are created or maintained
through multiple factors acting together synergistically or antagonistically. A statistical
interaction occurs when a different magnitude of effect (on health) is detected where
multiple factors co-occur than would be expected based on combining the effects of their
individual occurrence. This may be assessed in additive or multiplicative scales. Not all
statistical interactions reflect causal interactions.
Intracategorical
complexity
An intersectional approach to conducting analyses within specific intersectional subgroups.
Acknowledges that heterogeneity and intersectional complexities exist within any sociallydefined category (e.g. assessing differences in health outcomes of multiracial and
monoracial-identified people within the category of African-American).7
Intercategorical
complexity
An intersectional multi-group approach using existing analytical categories (e.g. gender,
SES) strategically to examine inequalities between social groups at cross-stratified
intersections. Assumes that the intersections that shape social life (e.g., racial/ethnic minority
woman) cannot be meaningfully reduced to their individual components (e.g. being
female).7
Intersectional approach
Research framework that approaches the question (e.g. health inequality) by examining
the impacts of intersectional positions (e.g. gender*race/ethnicity*SES) and processes (e.g.
gender, racial, and socioeconomic discrimination).2
Multiplicative model
Theoretical or statistical model that posits the effects of multiple domains (e.g.
race/ethnicity, age, sexual orientation) as effects that can only be understood in relation to
each other. For example, the mental health of a bisexual Asian youth would be constructed
as the unique effect of being bisexual and Asian and youth compared to groups at other
intersections. In statistical analysis is most often constructed using interaction terms rather
than just main effects. Reflects Hancock’s intersectional approach, though categories are
often still treated as fixed.2
Multiplicative-scale
interaction
Scale of statistical interaction in multiplicative multivariable models wherein interactions
(e.g. intersectional positions) are assessed against a null hypothesis that relative effects of
membership in each group (separately) are multiplied. Lack of significant interaction on this
scale may still reflect an excess or deficit of cases in contrast to what would be expected,
making this method less appropriate for public health or causal analysis.10
Multiplicative-scale
regression
Statistical model in which effects are multiplied to produce ratios (e.g. incidence,
prevalence or odds ratios). Any regression conducted in a logarithmic scale, such as
Poisson, logistic, or Cox.
Unitary approach
Non-intersectional approach to research that approaches the question (e.g. health
inequality) though examining a single primary identity or social location (e.g. gender).2
2
Intersectionality is uniquely positioned to highlight such solution-focused variables. Rooted in social
equity and justice, intersectionality simultaneously addresses the micro-level complexities of people’s
lived experiences at the intersection of social categories such as race, gender, socioeconomic status
(SES), and sexual orientation (to name just a few), and provides an in-depth understanding of meso- and
macro-level intersections (e.g., discrimination, prejudice, structural or policy barriers) that have a
deleterious impact on health.17 It encourages exploration of differences that are potentially driven by
social inequity with a focus on the unique ways these play out at different intersections over time. For
example, an intersectional approach to the study of reproductive services might consider how
interpersonal discrimination, structural barriers to access (e.g. cost, location, language), and historical
trauma related to 20th-century eugenics policies and practices may affect trust, willingness, and ability
to access services at different intersections of gender, race/ethnicity, SES, and age. We see links and
synergies between intersectionality and many of the approaches that guide health equity-focused
research, such as social determinants of health,18 sex- and gender-based analyses,19 biopsychosocial
approaches,20 socioecological models,21,22 lifecourse approaches,23 and multi-level social
epidemiology.24 We see the potential for intersectionality to complement these frameworks, and
challenge them to reach new and complex understandings about how to reduce or eliminate health
inequalities.
To harness the power of intersectionality to achieve these goals, an intersectionality framework can be
best employed in three key ways. First, it can inform descriptive research to measure health inequalities
and to monitor trends over time and across space, so that inequalities at intersectional positions can be
identified. Next, it can be used to conduct analytic research to identify individual- and group-level
causes of health inequalities, and how they interact. Lastly, it can be used to develop and implement
interventions at multiple levels, such as individual and population-level health promotion and public
policy, in ways that reflect intersectional effects. While these three levels of application correspond
directly to those that the U.S. Federal Collaboration on Health Disparities Research has identified as key
to eliminating health inequalities,25 we believe that an explicit intersectional approach at all levels
provides the greatest promise for reducing inequalities.
In the following section we present ten guidelines for quantitative intersectionality health inequality
research. We discuss each point briefly, making reference to key sources that may be useful for
additional information. These guidelines traverse theoretical conceptualization, study design,
measurement, statistical analysis, research team process, knowledge translation, and interpretation of
results.
3
TEN RECOMMENDATIONS FOR QUANTITATIVE INTERSECTIONAL RESEARCH
Ensure that researchers have the knowledge needed to apply an
intersectional framework to their health inequalities research
All research requires background theoretical and methodological knowledge. Given the status of
“intersectionality” as an academic buzzword,26 it is important to rely on intersectionality-focused
scholarship rather than work that superficially mentions intersectionality or related concepts without
explicating how intersectionality is being understood and used. Intersectionality may be best understood
as an “analytical sensibility” that does not reflect a consensus on methods.8 A brief reading list for a
health researcher might include methods papers by McCall7 and Hancock2 as well as recent work that
discusses approaches and methodological issues relevant to applying intersectionality to quantitative
or mixed-methods research.9-11,13, 27
Given the centrality of power relations to intersectionality, additional important knowledge includes
processes of marginalization, such as social exclusion, structural oppression, interpersonal discrimination,
and internalized oppression, as well as any relevant theory regarding the unitary or interactive effects of
social, psychological or biological mechanisms on the health outcome(s) under study. In addition to
theoretical expertise, community knowledge is important to understanding the realities of lived
experience and the historical and current social context of people’s lives, and may be brought into a
research project through community-based participatory research approaches.28,29 Lastly, strategies for
integrated knowledge translation may be helpful in building and using research partnerships to produce
knowledge that can transcend cultural, disciplinary, or social boundaries.30 This may increase policy
relevance and support the shifting of what is considered “expertise” to value organizational and
community perspectives.31
Consider advantages and limitations of intracategorical and
intercategorical approaches, to define study populations
A key consideration is whether study populations should consist of social groups identified with more
negative health experiences, or should be broader to allow for comparisons. Intersectionality calls on
researchers to examine heterogeneity at intersections within broader populations, but also within
specific sub-groups.7 Thus, an analysis of birth outcomes among Latina women would be considered
intersectional if age-, education-, immigration status-, and income-based heterogeneity within this
group were explored. The question of studying a sub-group or larger population maps onto McCall’s
distinction between intracategorical and intercategorical complexity.7 An intracategorical approach
can center the research within communities or identities experiencing marginalization, while
acknowledging the ways that experiences within groups may be constituted by identities or social
locations other than those that define the primary study population. This approach may align with
sampling methods (e.g. respondent-driven sampling or time-space sampling32) and community-based
processes that work well for research within marginalized (and sometimes hidden) populations, but less
well for comparison groups.
4
Despite these notable strengths, researchers should consider the risks of quantitative research that
samples marginalized communities alone. These include the possibility of inadvertently pathologizing, by
exclusively emphasizing health problems rather than strengths, or by highlighting experiences that
contribute to stigma or reinforce stereotypes, in the absence of comparisons (e.g. documenting what
appear to be high levels of substance use that may in actuality be similar to other groups’ use).
Intracategorical quantitative approaches may also foreclose the potential to document impacts of
inequality on overall population health or obscure potential benefits of privilege. Importantly,
examination of interindividual variation within a group may fail to identify drivers of inequalities between
that group and others if an exposure is fairly homogeneous within that group.33 For example, the full
effects of discrimination may not be detectable within a study consisting exclusively of racial/ethnic
minorities if experiences of racial discrimination are already high across the sample.
Calls for the prioritization of more marginalized groups’ experience do not preclude the study of
intercategorical complexity through the inclusion of less marginalized comparison groups. Including a
wide range of intersectional positions opens up the potential to describe inequalities across populations
and to detect additional factors driving inequalities. However, this broader focus may necessitate
careful consideration of how comparisons are conducted if those at the margins are to be kept at the
center of focus. Ultimately, the decision about whether to limit a study to a specific community or make
comparisons across a broader population should be based on the research question(s) and the
contrasts needed to identify effects of interest.
Decide whether the research will focus on intersections of social
identities, social locations, processes, social context, policies, or
intersections that cross these domains
Analysis of inequalities can be structured across intersections of social identity or social location that may
or may not be concordant.10 Some social locations (e.g. ethnicity, gender) are held as identities with
more strength than others (e.g. educational level), and at times identity and location may conflict (e.g.
more Americans consider themselves middle class than meet middle class income criteria). This is further
complicated by the fact that intersectional identities are not as fixed or stable as traditional categorical
approaches to intersectionality imply.27,34 Identities mutually constitute each other dynamically.34 For
example, a likely interpretation of a finding of difference between men and women would be that
gender influenced the inequality. However, gender is not a fixed social category, and there are multiple
differences within a gender category (e.g. hegemonic and subordinated masculinities).34 Thus,
researchers must also be cognizant of the shifting and fluid dynamics of social categories.8
While analysis of identities and social location situates inequalities within socio-demographic frameworks,
factors such as prejudice, discrimination, social contexts, and policies are potential causes of disparate
health, and may themselves interact. For example, experiences of employment discrimination may have
a different effect on anxiety depending on prior overall burden of discrimination, or the impact of a new
healthcare policy for undocumented immigrants may vary for those at different intersections of SES and
age. Likewise, these processes may mediate some of the effects of social identity or location on health,
or may serve as a risk modifier (moderator) that affects various groups differently. As an example, one
study found that transphobia had an observable impact on sexual risk behavior, but only for trans people
of color.35
Some questions researchers may want to ask include: 1) Are inequalities are best described across
identities or social locations?; 2) Which intersections are of primary interest?; 3) What processes may
explain any observed inequalities, and which are modifiable and thus potentially intervenable?, and; 4)
5
Might two or more processes interact synergistically or antagonistically to exacerbate or mitigate the
effects of each other on the outcome(s)?
Use individual-level data on social identities or locations to describe
inequalities, but consider individual- and group-level factors when
analyzing potentially causal processes
Causal processes that impact health may occur at micro, meso and macro levels,22 necessitating use
of both individual- and group-level variables. While individual-level data are appropriate for descriptive
studies of health inequalities, all causal processes are inherently multi-level. Thus, eliminating health
inequalities will almost certainly require multiple levels of intervention (e.g. policy, built environment, as
well as individual).36 Because research questions may be more limited, and not capture entire causal
processes, and since it is not always possible to obtain good multi-level data, it is important that
researchers working on individual-level studies think through potential causes of health outcomes that
act at other levels. For example, it is important to remember that individual and aggregate versions of
the same measure can capture different constructs,37 and that they may interact with each other (e.g.
the effects of experiencing homophobic violence may depend on the social context of living in a
jurisdiction where there are high versus low levels of homophobic violence). Thus, even if one completes
an analysis of individual-level effects of experiencing such violence, there may be unstudied effects that
exist at a group level. Moreover, a focus on multiple levels of intersection suggests that it is possible that
these individual- and group-level experiences may themselves interact to impact health. An
intersectional approach to health inequalities research necessitates retaining this broader context
through a careful interpretation of findings.38 In quantitative research, this requires acknowledging
unstudied variables, domains and levels of potentially causal factors.
Understand the sampling and measurement limitations of secondary data
sources for intersectionality analysis
While secondary data sources such as large population surveys can provide opportunities for describing
health outcomes across a wide range of intersections,10 they are also limited in that they were not
developed with the a priori conceptual knowledge needed to apply an intersectional framework.13 This
may allow only a rudimentary intersectional approach. Most secondary data sources consist of crosssectional surveys or examination data, and often do not include measures of discrimination, or include
measures that are not responsive to forms of discrimination specific to intersections of sexual orientation,
race and gender.13 Sociodemographic variables are frequently defined and measured using mutually
exclusive response options that limit analysis of differences between and within more complexly-defined
groups. Furthermore, sampling may or may not have been undertaken using techniques (e.g.
probabilistic sampling) to capture underrepresented populations, such as smaller racial/ethnic or sexual
orientation groups. Moreover, some populations (e.g. those living on reservations or in institutions such as
prisons) are often excluded from data collection. Thus, researchers may find that existing data sets either
do not contain representative data on populations of interest, or have insufficient sample sizes (i.e. low
statistical power) for meaningful intersectional analyses. Given these measurement and sampling
limitations, it may be useful to consider whether secondary data sources can be supplemented, for
6
example, by combining them with group-level measures from census or other data sources, or by using
qualitative data to inform the interpretation of findings; or whether primary data collection is needed.
Take an intersectional approach to survey measures
Primary data collection offers the potential to support more developed and sophisticated intersectional
analyses. In designing primary data collection – for an individual study or in modifying population surveys
for subsequent cycles – an intersectional approach has implications for measures collected. Given that
intersectional positions are not static, researchers may want to consider including measures that allow
an intersectional lifecourse approach, for example one that acknowledges the potential intersectional
effects of childhood poverty and current SES. Examination of intersectional cohort effects can also be
built into study measurements. For instance, in a study of mental health, one might examine potential
interactions between sexual orientation and era of coming out, given that the social and legal context
of homosexuality have changed dramatically over time. Next, including measures to capture multiple
domains of oppression such as stigma, prejudice and discrimination allows for an examination of the
ways these processes may have differential impacts for those at varying social locations. Individual
measures are often limited. For example, most measures of racism focus on interpersonal experiences of
racial discrimination, ignoring or excluding structural or systemic racism.39 Additional improvements can
be made in survey-based measures of discrimination, which are now often based on single axes of
discrimination, such as racism. This violates intersectionality’s central tenet that identities and
experiences are mutually constituted, and cannot be disaggregated by social identity or discrimination
experience (e.g. racism, sexism, or transphobia).11 Although measures exist to assess overall experiences
of discrimination across all axes (often with additional “check all that apply” options to capture all
identity factors and experiences to which it is attributed),11,40 we challenge researchers to create
measures that truly capture experiences of discrimination that may be unique to particular intersections.
Lastly, measures that can be aggregated to create meaningful group-level variables should be
considered. The inclusion of such reliable and valid measures would allow researchers to probe beneath
the complex factors that shape and influence the experiences of individuals and their relationship to
specific health outcomes and inequalities, and avoid speculation at the interpretation stage when such
measures are absent.
Structure intersectional groups of interest through data coding or
interactions to produce clear results
Investigators have two options with regard to defining intersectional groups in data coding and analysis:
inputting interaction terms in a regression model (e.g. race*sexual orientation*gender) or constructing
intersectional groups (e.g. Native American sexual minority women). If the objectives of the study are
purely descriptive, researchers can code their data to define different intersectional groups and allow
readers to make any comparisons that are of interest to them without specifying a main reference
group. However, when conducting analytical research using regression models, where comparisons are
required, where the adjusted risk estimates obtained are fictional, and where statistical power can be
best preserved by only including interactions/intersections that are significant, including interaction
terms in the equation might be warranted.41,42
7
Issues of statistical power and sample size that are important in categorizing groups for analysis43 may
create conflicts with intersectional approaches that prioritize the experiences of groups representing
small minorities. For instance, when comparing effects, researchers are advised to preserve statistical
power by avoiding using the smallest group as the category of reference,43 potentially reinforcing
conventions wherein more privileged groups serve as the default against which all others are compared.
However, in designing their analyses, researchers must recall that one of the core tenets of
intersectionality is the centering of marginalized groups.11 Using an intersectionality framework adds the
constraint of coding variables used in interactions so that effects are in the same direction, as this
simplifies the interpretation of relevant estimates.44 Using groups that experience the most adverse health
outcomes as reference points provides one option to address the issue of directionality while facilitating
the estimation of intersecting processes of privilege and oppression, provided these groups are relatively
large in the population or have been oversampled in the study.
Intersectional interaction effects can be presented as individual versus joint effects or as the
heterogeneous effects of one factor across levels of others.45 If one is studying the effects of interlocking
social positions such as ethnicity and immigration status, then describing the results as joint effects is
warranted. However, when analyzing intersections between processes and social positions or identities,
presenting the results as the heterogeneity of effects across different social categories may be more
useful. For instance, LeVasseur et al.41 examined bullying as a process potentially impacting suicide
attempts among youth by using a stratified analysis that quantified the magnitude of the effect of
bullying separately among different groups at the intersections of gender, sexual minority, and Latino
ethnic identities.
Use theory to guide analyses with regard to confounding and/or mediation
It is important is to ensure that the question answered by an analysis matches the research question,
though this is not always straight-forward with regard to the inclusion of variables in multivariable models.
Here, data-driven approaches to multivariable research, such as those still common within the field of
epidemiology, may limit intersectional applicability and it is thus necessary to use theory to guide
statistical analytic strategies.
For descriptive analyses seeking to identify health inequalities, crude effects are of primary importance
(i.e. differences in proportions, means, or number of outcomes between cross-stratified groups), and
potential intersectional factors should not be controlled for. Adjusted measures of risk estimate what the
impact of a factor would be were the groups under comparison equal with respect to the adjustment
variables, and thus are fictitious rather than real-world risks. They are therefore unsuitable for identifying
factual subpopulations experiencing the greatest burden of a disease or other health outcome. To
illustrate, controlling for personal income in an analysis of transgender identity and health care access
would standardize trans and non-trans groups to have the same income, a situation which does not exist
in reality. Moreover, income may mediate some of the effect, and adjusting for this would be doubly
inappropriate. In contrast, to produce unbiased estimates of the effects of potential causes of health
inequalities, control for potential confounders is necessary. Since data-driven approaches to
confounding based on change in coefficient estimates (typically  10%)43 may either introduce or
remove confounding, theory-driven approaches46 are preferred where possible. Moreover, data-driven
approaches to mediation and confounding are statistically indistinguishable; both involve comparing
the magnitude and statistical significance of coefficients that are unadjusted and adjusted for an
additional variable.47 Thus, even in exploratory or semi-theoretical research, it is crucial to use available
theory to distinguish potential mediators and confounders a priori.
8
Identifying factors that mediate the relationship between social identities or locations and health
outcomes can suggest targets for intervention to reduce health inequalities. However, unintentional
adjustment for mediators will lead to incorrect conclusions about the causal strength of a factor that
impacts an outcome through a mediated path (e.g. adjusting for income would actually assess the
effects of transgender status on health care access through pathways other than income). When health
inequalities are rooted in social inequity, processes of oppression will inevitably mediate relationships
between social identity or location and health outcomes, and should not be adjusted for other than in
the context of a mediation analysis.
Focus analysis on absolute rather than relative effects, including
additive-scale interactions
Using absolute rather than relative effects, for example risk differences rather that ratios, is important for
identifying levels of inequality and the potential public health impact on populations.48 Given the
centrality of interactions in quantitative intersectionality analysis, it is important to note that estimating
interactions on an additive scale is necessary to analyze absolute effects, and is more relevant than
multiplicative-scale interaction for both assessing public health impact and causal effects. 48,49 This is an
especially important distinction given that the language of intersectionality often refers to additive and
multiplicative models, wherein multiplicativity indicates intersectionality; additive-scale interactions can
and should be used to assess intersectional multiplicativity.10 In descriptive analyses, this can be done by
comparisons of risk differences across groups as per textbook examples.48 In regression analysis, it can
be done by including an interaction term in a linear regression, which is in the additive scale, or by taking
additional steps after including such a term in a log-scale regression (e.g. Poisson, Logistic) which is in
the multiplicative scale. Additive-scale interaction can be readily assessed from multiplicative-scale
models using measures such as the relative risk due to interaction (RERI), attributable proportion due to
interaction, or synergy index, along with their corresponding confidence intervals.50-52 While it is
sometimes technically easier to assess interactions in log-scale regression models in the multiplicative
scale (by just using the interaction term), the absence of multiplicative-scale interaction often indicates
precisely the presence of additive-scale interaction.51 Thus, use of multiplicative-scale interactions would
hinder the ability to identify intersectional effects on health inequalities.
Maintain an intersectional focus in the interpretation of results
Interpreting the results of a quantitative intersectionality study necessarily involves a revisit to the prior
methodological steps that have shaped the study’s design and analysis. Well-designed studies can
facilitate an understanding of the intersectional nature of health inequalities and allow researchers to
avoid speculation about intersectional effects. Researchers with primary data and relevant measures
would not need to speculate broadly about the role heterosexism and homophobia may play in a
hypothetical finding that Latino gay male youth show higher rates of eating disorders than their
heterosexual counterparts. However, even as we caution researchers not to speculate beyond their
data, an “intersectionality analytic sensibility”8 necessitates that researchers nonetheless be cognizant
of and attentive to the context of sociohistorical and structural inequality38 and dynamics of power that
have historically influenced health inequality. The absence of a finding of a difference between racial
groups on a certain health outcome does not negate the longstanding history of institutionalized and
interpersonal racial discrimination that racial minorities in the U.S. have experienced. Even when reliable
9
and valid measures of key intersectionality experiences exist, they may still offer limited understanding
of complex and multifaceted intersectionality-related health inequalities.
Collins’3 notion of a matrix of domination to describe the “penalty and privilege” of intersectional social
identities further complicates the interpretation of results. People are neither solely disadvantaged nor
advantaged all of the time and in every context. For example, an upper middle class Black male
professional may have class and gender privilege in an occupational context, but unlike his White
counterparts may experience the penalty of being more likely to be suspected of a crime, or to be
stopped and frisked by police. Thus, interpretations should be kept specific to the outcome(s) under
study.
A core caveat in interpreting quantitative results is that analysts must be cognizant of the limitations of
their study designs and types of conclusions that can be drawn. While intersectional approaches to
quantitative research allow for greater exploration of heterogeneity, homogeneity is still implicitly
assumed within these finer categories. Qualitative approaches, with their focus on rich, culturally and
contextually grounded data, are well-suited to adding nuance to the interpretation of quantitative
results. Mixed-method studies that capitalize on the strengths of both qualitative and quantitative
methods53 are likely to offer considerable advantage to intersectionality researchers challenged with
how to interpret quantitative results that document that an intersectional inequality exists, but are unable
to explain why.
MAKING PROGRESS ON ELIMINATING HEALTH INEQUITIES
We have provided recommendations for how researchers can better harness the power of
intersectionality in the service of eliminating health inequities, by applying an intersectional framework
to identify inequalities, causes, and effective intervention strategies. Applied together, these
approaches will generate research that identifies inequalities within more specific population subgroups, causes that interact to impact or maintain poor health, and strategies at the micro, meso and
macro levels. Such research is essential to intervene in ways that are both effective and appropriate to
the contexts of communities affected by health inequalities. Our recommendations should be
considered as food for thought rather than prescriptive. In the end, the design, conduct, and
interpretation of quantitative intersectionality studies are rooted not in particular statistical analytic
strategies, but in an “analytical sensibility” that acknowledges the roles of power and social injustice8 in
creating and maintaining health inequalities.
10
REFERENCES
1.
Bowleg L. The problem with the phrase women and minorities: Intersectionality – an important theoretical
framework for public health. Am J Public Health. 2012;102(7):1267-1273.
2.
Hancock A-M. When multiplication doesn’t equal quick addition: Examining intersectionality as a research
paradigm. Perspect Pol. 2007;5(1):63-79.
3.
Collins PH. Black Feminist Thought: Knowledge, Consciousness, and the Politics of Empowerment. New York:
Routledge; 1991.
4.
Braveman PA, Cubbin C, Egerter S, Williams DR, Pamuk E. Socioeconomic disparities in health in the United
States: What the patterns tell us. Am J Public Health. 2010;100:S186-S196.
5.
Collins PH. Some group matters: Intersectionality, situated standpoints, and Black feminist thought. In:
Fighting Words: Black Women and the Search for Justice. Minneapolis, MN: University of Minnesota Press;
1998:201-228.
6.
Crenshaw K. Demarginalizing the intersection of race and sex: A Black feminist critique of antidiscrimination
doctrine, feminist theory and antiracist politics. 1989 University of Chicago Legal Forum. 1989:139-167.
7.
McCall L. The complexity of intersectionality. Signs. 2005;30(3):1771-1800.
8.
Cho S, Crenshaw KW, McCall L. Toward a field of intersectionality studies: Theory, applications, and praxis.
Signs. 2013;38:785-810.
9.
Cole ER. Intersectionality and research in psychology. Am Psychol. 2009;64(3):170 180.
10.
Bauer GR. Incorporating intersectionality theory into population health research methodology: Challenges
and the potential to advance health equity. Soc Sci Med. 2014;110:10-17.
11.
Bowleg L. When black + lesbian + woman ≠ black lesbian woman: The methodological challenges of
qualitative and quantitative intersectionality research. Sex Roles. 2008;59:312-325.
12.
Dubrow JK. How can we account for intersectionality in quantitative analysis of survey data? Empirical
illustration for Central and Eastern Europe. ASK.Res Meth. 2008;17:85-100.
13.
Rouhani S. Intersectionality-informed Quantitative Research: A Primer. Vancouver, BC: The Institute for
Intersectionality Research & Policy, Simon Fraser University; 2014.
14.
Rose G. Sick individuals and sick populations. Int J Epidemiol. 1985;14:32-38.
15.
Lofters A, O’Campo P. Differences that matter. In: O’Campo P, Dunn JR, eds. Rethinking Social
Epidemiology: Towards a Science of Change. New York: Springer; 2012:93-108.
16.
Phelan JC, Link BG, Tehranifar P. Social conditions as fundamental causes of health inequalities. J Health Soc
Behav. 2010;51:S28-S40.
17.
Weber L, Parra-Medina D. Intersectionality and women's health: Charting a path to eliminating health
disparities. Adv Gender Res. 2003;7:181-230.
18.
Marmot M, Wilkinson R. Social Determinants of Health. Oxford: Oxford University Press; 2006.
19.
Krieger N. Genders, sexes, and health: What are the connections—And why does it matter? Int J Epidemiol.
2003;32:652-657.
20.
Engel G. The need for a new medical model: A challenge for biomedicine. Science. 1977;196:129-136.
21.
Krieger N. Epidemiology and the web of causation: Has anyone seen the spider? Soc Sci Med.
1994;39(7):887–903.
22.
Bronfenbrenner U. Toward an experimental ecology of human development. Am Psychol. 1977;32(7):513531.
23.
Pearlin LI, Schieman S, Fazio EM, Meersman SC. Stress, health, and the life course: Some conceptual
perspectives J Health Soc Behav. 2005;46(2):205- 219.
24.
Berkman LF, Kawachi I. (Eds.) Social Epidemiology. New York: Oxford University Press; 2000.
11
25.
Rashid JR, Spengler RF, Wagner RM, Melanson C, Skillen EL, Mays Jr RA, Heurtin-Roberts S, Long JA.
Eliminating health disparities through transdisciplinary research, cross-agency collaboration, and public
participation. Am J Public Health 2009;99(11):1955-1961.
26.
Davis K. Intersectionality as buzzword: A sociology of science perspective on what makes a feminist theory
successful. Feminist Theory. 2008;9:67-85.
27.
Nash JC. Re-thinking intersectionality. Feminist Rev. 2008;89:1-15. doi: 10.1177/0022146510383498
28.
Minkler M. Linking science and policy through community-based participatory research to study and
address health disparities. Am J Public Health. 2010;100(Suppl 1):S81-S87.
29.
Travers R, Pyne J, Bauer G, Munro L, Giambrone B, Hammond R, Scanlon K. ‘Community control’ in CBPR:
Challenges experienced and questions raised from the Trans PULSE Project. Action Res. 2013;11:403-422.
30.
Lapaige V. “Integrated knowledge translation” for globally oriented public health practitioners and
scientists: Framing together a sustainable transfrontier knowledge translation vision. J Multidiscip Healthc.
2010;3:33-47.
31.
Kothari A, Wathen CN. A critical second look at integrated knowledge translation. Health Policy.
2013;109:187-191.
32.
Magnani R, Sabin K, Saidel T, Heckathorn D. Review of sampling hard-to-reach and hidden populations for
HIV surveillance. AIDS. 2005;19(Suppl 2):S67-S72.
33.
Schwartz S, Carpenter KM. The right answer for the wrong question: Consequences of type III error for public
health research. Am J Public Health. 1999;89:1175-1180.
34.
Connell R. Gender, health and theory: Conceptualizing the issue, in local and world perspective. Soc Sci
Med. 2012;74:1675-1683.
35.
Marcellin RL, Bauer GR, Scheim AI. Intersecting impacts of transphobia and racism on HIV risk among trans
persons of colour in Ontario, Canada. Ethnicity Inequal Health Soc Care. 2013;6(4):97-107.
36.
Golden SD, Earp JAL. Social ecological approaches to individuals and their contexts. Health Educ Behav.
2012;39(3):364-372.
37.
Schwartz S. The fallacy of the ecological fallacy: The potential misuse of a concept and the consequences.
Am J Public Health. 1994;84(5):819-824.
38.
Cuadraz GH, Uttal L. Intersectionality and in-depth interviews: Methodological strategies for analyzing race,
class, and gender. Race Gender Class. 1999;6:156-181.
39.
Varcoe C, Browning A. What does intersectionality suggest about population measures of discrimination?
Oral presentation: International Conference on Intersectionality, Vancouver, BC; 2014.
40.
Hankivsky O. Women’s health, men’s health, and gender and health: Implications of intersectionality. Soc
Sci Med. 2012;74(11):1712-1720.
41.
LeVasseur MT, Kelvin EA, Grosskopf NA. Intersecting identities and the association between bullying and
suicide attempt among New York City youths: Results from the 2009 New York City Youth Risk Behavior
Survey. Am J Public Health. 2013;103(6):1082–1089.
42.
Veenstra G. Race, gender, class, and sexual orientation: Intersecting axes of inequality and self-rated health
in Canada. Int J Equity Health. 2011;10(3):1-11.
43.
Vittinghoff E, Glidden DV, Shiboski SC, McCulloch CE. Regression Methods in Biostatistics: Linear, Logistic,
Survival, and Repeated Measures Models. 2nd ed. New York, NY: Springer; 2012.
44.
Knol MJ, VanderWeele TJ, Groenwold RHH, Klungel OHM, Rovers MM, Grobbee DE. Estimating measures of
interaction on an additive scale for preventive exposures. Eur J Epidemiol. 2011;26(6):433-438.
45.
Knol MJ, VanderWeele TJ. Recommendations for presenting analyses of effect modification and interaction.
Int J Epidemiol. 2012;41:514-520.
46.
Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research. Epidemiol. 1999;10(1):37–48.
47.
MacKinnon DP, Krull JL, Lockwood CM. Equivalence of the mediation, confounding and suppression effect.
Prev Sci. 2000;1(4):173-181.
12
48.
Szklo M, Nieto FJ. Which is the relevant model? Additive versus multiplicative interaction. In: Epidemiology:
Beyond the Basics, 3rd ed. Burlington, MA: Jones & Bartlett; 2012: 205-207.
49.
Rothman KJ, Greenland S, Walker AM. Concepts of interaction. Am J Epidemiol. 1980;112(4):467-470.
50.
Skrondal A. Interaction as departure from additivity in case-control studies: A cautionary note. Am J
Epidemiol. 2003;158(3):251-258.
51.
Greenland S, Lash TL, Rothman KJ. Concepts of interaction. In: Rothman K, Greenland S, Nash T, eds.
Modern Epidemiology, 3rd ed. Philadelphia: Wolters Kluwer; 2008:71-83.
52.
Zou GY. On the estimation of additive interaction by use of the four-by-two table and beyond. Am J
Epidemiol. 2008;168:212-224.
53.
Creswell JW, Klassen AC, Plano Clark VL, Smith KC. Best practices for mixed methods research in the health
sciences. In: Rockville, MD: Office of Behavioral and Social Sciences Research of the National Institutes of
Health; 2011.
13
View publication stats
Download