Hi everyone,

I have a (multivariate) cumulative model (with a probit link) with the following variables:

IV1 `badge`

: categorical predictor with 3 levels: `nobadge`

, `twitter`

and `irma`

.

IV2 `context`

: categorical predictor with 2 levels: `earth`

and `cancer`

Interaction `badge:context`

with 6 levels accordingly

3 DVs with 5-point Likert scale data (sharing, sourcred, messcred)

```
library(brms)
d <- read.csv("example_data.csv")
m1 <- brm(
mvbind(sharing, sourcred, messcred) ~ 1 + badge + context + badge:context,
data = d,
family = cumulative("probit"),
prior = c(
prior(normal(0,4), class = Intercept),
prior(normal(0,4), class = b)
)
)
```

To test my 2 hypotheses I need to compare:

H1 `twitter`

vs. `nobadge`

(main effect)

H2 `irma_cancer`

vs. `nobadge_cancer`

(interaction effect)

To test H1 I would use sum contrasts for `context`

and dummy contrasts for `badge`

(`nobadge`

as the reference level) so that the the estimate `sharing_badgetwitter`

is the difference in sharing scores

between the levels twitter and nobadge while averaging over `context`

.

For the second hypothesis I wanted to use the `hypothesis()`

function from brms but I canâ€™t figure out how to create the right contrasts so that I compare only the relevant part of the interaction. So from the six levels of the interaction `badge:context`

Iâ€™m interested in the comparison of 'irma_cancer`vs.`

nobadge_cancer`.

**Question**: How can I create the desired contrasts for H2?

- Operating System: macOS 11.5.2
- brms Version: 2.13.5

Thank you for your suggestions!

Here is an example screenshot from the summary output & example data:

example_data.csv (27.5 KB)