Usefully combining a series of unreplicated cheesemaking experiments

Applied dairy research is characterized by experiments for which financial and physical constraints permit only a small number of experimental units. With few units it is difficult to replicate treatments, and without replica tion experimental error cannot be estimated. The statistical analysis and i nterpretation of such experiments is problematic. However, if there have be en several such experiments it may be possible to perform a combined analys is. Nine unreplicated experiments comparing effects of diet on the composit ion of cows' milk and on cheese characteristics were jointly analysed as an incomplete block design. This analysis method was contrasted with analyses of individual experiments. For cheese moisture, the key outcome measuremen t, the assessment of statistical significance concurred for the two methods in 13 out of 21 comparisons of treatments with the control. Sources of err or variation allowed for under the two methods were delineated. The combine d analysis paradigm provided stronger inference and a wider interpretation of results than could be achieved using analyses for individual experiments . Unequal replication of treatments and unequal concurrence of treatments w ithin experiments over the series gave rise to a wide range of SED. The cha llenge of presenting results with unequal SED was addressed graphically usi ng error bars. Attention to series design, in particular the apportioning o f replication and treatment concurrence across the series of experiments, w as shown to ameliorate presentation difficulties and, more importantly, to yield higher precision at no extra cost.