If the model and data both contain error components then these are added in quadrature - hence COMPARE can be used to compare two datasets for similarity.
If the data contain no error component then the user is given the options of assuming Poisson data errors (i.e. data error = sqrt (data value)), or unit data errors (i.e. all 1.0) (answer Y to POISS_ERR for the former).
Subtopics: