Multilevel Analysis of Individual Heterogeneity and Discriminatory Accuracy (MAIHDA), as developed by Evans and colleagues, aims to map the frequencies of health outcomes across a broad range of pre-defined intersections of multiple social categories.

Another aim of MAIHDA is to determine measures of discriminatory accuracy for the selected intersections. Measures of discriminatory accuracy are suitable for assessing whether population-wide or target group-specific interventions should be used to positively influence the described characteristic.

The first step is to select social categories to be included in the analysis. In MAIHDA, both individual-level variables (e.g. a person's occupational status) and group-level variables (e.g. the gender inequality of a district) can be included. The PROGRESS Plus framework provides assistance in the selection of social categories. Subsequently, all selected categories are cross-classified and thus intersections are formed.

Multilevel regression models are used for data analysis. The intersections are used as random effects in these regression models. Analyses can be carried out with linear or logistic models. Furthermore, the method can be used to reliably determine means or proportions for the defined intersections, as well as their 95% confidence intervals.

Finally, the intra-class correlation coefficient (ICC) serves as a measure of discriminatory accuracy. This measure is used to assess whether population-wide or target group-specific interventions should be used. Low discriminatory accuracy supports population-wide measures, whereas high discriminatory accuracy supports target group-specific interventions.

MAIHDA can be used to describe the intersections of multiple categories of difference in quantitative data analyses in a theoretically informed way. Frequencies among groups with few numbers of observations can be reliably estimated using information from the entire data set.

Ultimately, the results are easy to interpret and thus well-suited to generate comprehensible statements that can be communicated to decision-makers.

Social categories have to be selected by the users. This can lead to selection bias based on users' preconceptions.

Important variables cannot be taken into account if the proportion of intersections with few numbers of observations is too large. In addition, knowledge of multilevel regression methods and Bayesian statistics, as well as large, representative data sets, is required.