Devil in the details: how can we avoid potential pitfalls of CATS regression when our data do not follow a Poisson distribution?