MAIHDA
Also known as: Intersectional MAIHDA, Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy, Intersectional Multilevel Analysis
MAIHDA — Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy — is a quantitative method for studying intersectional inequalities. Introduced for intersectionality by Clare Evans and S. V. Subramanian in 2018, building on Juan Merlo's discriminatory-accuracy framework, it treats the many strata formed by crossing social categories (for example gender × race/ethnicity × education) as level-2 units in a multilevel model, then partitions outcome variation between and within those strata to assess how much intersectional position predicts the outcome.
Key highlights
- Handles many intersectional strata, including small ones, by shrinking unstable estimates through partial pooling.
- Separates additive main effects from genuinely interactive intersectional variation in a principled way.
- Provides a discriminatory-accuracy measure (the VPC) that tempers claims about how strongly categories define outcomes.
- Generalises to binary and count outcomes and accommodates further covariates and higher-level structures.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use MAIHDA when you have individual-level data and want to quantify inequalities across many intersectional strata without the instability of saturated interaction models, especially when some strata are small. It is well suited to health, education, and labour-market disparities studied through gender combined with race, class, and other axes. It is less appropriate with very few dimensions or large strata (where conventional interactions suffice), and it requires care in interpretation, since the model is descriptive of variance structure rather than causal.
Strengths & limitations
- Handles many intersectional strata, including small ones, by shrinking unstable estimates through partial pooling.
- Separates additive main effects from genuinely interactive intersectional variation in a principled way.
- Provides a discriminatory-accuracy measure (the VPC) that tempers claims about how strongly categories define outcomes.
- Generalises to binary and count outcomes and accommodates further covariates and higher-level structures.
- It is descriptive of variance and association, not a method for causal identification of discrimination.
- Results depend on which dimensions and how many categories are crossed to form strata.
- Random-effects normality and adequate stratum sample sizes are assumptions that can be violated in sparse data.
- A low VPC can be misread as evidence that intersectionality is unimportant, when it reflects within-stratum heterogeneity.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How is MAIHDA different from including interaction terms in a regression?
A saturated regression estimates a separate fixed coefficient for every category combination, which becomes noisy and unstable as strata multiply and shrink. MAIHDA treats the strata as random effects in a multilevel model, partially pooling them so small-stratum estimates borrow strength from the whole sample. This yields more precise, less overfit stratum estimates and a variance-partition measure of how much intersectional position discriminates outcomes, which a fixed-effects interaction model does not provide.
What does the variance partition coefficient tell us?
The VPC is the share of total outcome variance that lies between intersectional strata rather than within them. It answers a discriminatory-accuracy question: if you knew only someone's stratum, how well could you predict their outcome? A high VPC means intersectional position strongly structures the outcome; a low VPC means people within strata vary as much as strata vary from each other, which cautions against treating the categories as deterministic.
Does a low VPC mean intersectionality does not matter?
Not necessarily. A low VPC means that, on average, stratum membership explains little of the individual-level variation, but the strata can still differ in their mean outcomes in ways that are substantively and politically important. MAIHDA reports both the variance structure and the stratum-specific predictions, so analysts should examine which intersections are most disadvantaged rather than dismissing intersectionality on the VPC alone.
Sources
- 1.Evans, C. R., Williams, D. R., Onnela, J.-P., & Subramanian, S. V. (2018). A multilevel approach to modeling health inequalities at the intersection of multiple social identities. Social Science & Medicine, 203, 64–73.
- 2.Merlo, J. (2018). Multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA) within an intersectional framework. Social Science & Medicine, 203, 74–80.
- 3.Evans, C. R., Leckie, G., Subramanian, S. V., Bell, A., & Merlo, J. (2024). A tutorial for conducting intersectional multilevel analysis of individual heterogeneity and discriminatory accuracy (MAIHDA). SSM - Population Health, 26, 101664.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 22). MAIHDA. ScholarGate. https://scholargate.app/gender-studies/maihda-intersectional-analysis