In this paper, we consider a linear regression model with intraclass correlation structure and study the behaviour of the precision diagnostic statistic COVRATIO as a function of the intraclass correlation coefficient ρ. Interestingly, it turns out that for the model with intercept, COVRATIO is a monotonically decreasing function of ρ. Thus if the model errors have larger correlation, then it is more stable with regard to deletion of observations. We have also considered the ratio of traces TRATIO of the least squares estimator of the parameter vector under data reduction and without da.ta reduction. This ratio is also decreasing in ρ implying that loss of precision when model errors have the same high correlation is smaller when an observation is deleted. Lastly, we have determined the linear parametric function (LPF) whose BLUE is affected the most by data reduction.
