Dear Statalister,

I have a doubt about how to replicate into Stata syntax a procedure followed by authors in the following article: Gravel CA et al. Interrater reliability of pediatric point-of-care lung ultrasound findings. Am J Emerg Care (in press). I have a limited knowledge of multilevel modelling so I ask for your suggestions.

Shortly, each of two raters rated as normal/abnormal 12 lung zones in each patient of a sample of 71. They estimated a multilevel mixed-effects linear regression model with the normal/abnormal designation as the dependent variable (supposedly 0/1 coded) and patient-level and zone-level identifiers as nested random intercept effects. Each patient has 12 zones that are evaluated by 2 raters. So, zone-level would be nested within patient-level.

They calculated the intraclass correlation (ICC) as a measure of interrater reliability between the raters through the equation:

ICC=zone-level variance / [zone-level variance + patient-level variance + residual (rater-level) variance]

My questions are:
  1. Is the following Stata syntax appropriate for those calculations?
mixed DepVar rater || zone: || patient: , vce(robust)
  1. Taking into account that the dependent variable would be coded as 0/1, wouldn’t it be more appropriate the use of the melogit command?
melogit DepVar rater || zone: || patient: , vce(robust)

Any help would be deeply appreciated.

Inigo Gorostiza
Health Researcher