Application of composite estimation in studies of animal population production with two-stage repeated sample designs

Abstract This paper reports results from two example data sets of a two-stage sampling design where sampling (in panels) both farms and animals within selected farms increases the efficiency of parameter estimation from measurements recorded over time. With such a design, not only are farms replaced from time-to-time but also animals subsampled within retained farms are subject to replacement. Three general categories of parameters estimated for the population (the set of animals belonging to the universe of farms of interest) were (1) the total at each measurement occasion; (2) the difference between means or totals on successive measurement occasions; (3) the total over a sequence of successive measurement periods. Whereas several responses at the farm level were highly correlated over time ( π 1 ), the corresponding animal responses were less correlated over time ( π 2 )—leading to only moderate gains in relative efficiency. Intraclass correlation values were too low in most cases to counteract the overall negative impact of π 2 . In general, sizeable gains in relative efficiency were observed for estimating change—confirming a previous result which showed this to be true provided that π 1 was high (irrespective of π 2 ).