I am using brms to model a real outcome, where the goal is to establish a model where coefficients can be used for prediction.

A number (i.e. 50+ ) of the columns in my dataset are indicator variables coded as “onehot” (i.e. 1 = presence, 0 = absence) for various potentially useful predictors.

I expect there to be correlation (and possibly 1-way or 2-way causation) for some of these predictors. For example, variables exposure1-exposure50 can each take values of 1 or 0, and subsets of these 50 may at some level be causally related.

*Section 12.7 in Regression and Other Stories (RAOS) “Models for regression coefficients”* has been informative, as well as *McElreath’s cautious admonitions about avoiding “Causal Salad”*, and I seek further depth in resources and examples that discuss strategies of how to include:

- many onehot predictors where correlation structures may be present
- many onehot predictors in conjunction with non-onehot predictors
- guidance with particular emphasis on inclusion of many such onehot variables, ideally using the
`brms`

package.

Thanks in advance.