2.3. Appropriate worker and cage replicates for laboratory experiments involving adult workers
A minimum sample of 30 independent observations per treatment is relatively robust for conventional statistical analyses (e.g. Crawley, 2005); however, financial constraints and large effect sizes (e.g. difference among treatments for the variable(s) of interest; see statistics paper (Pirk et al. (2013)) will no doubt lower this limit, especially for experiments using groups of caged workers. Larger sample sizes (i.e. number of cages and workers per cage) reduce the probability of uncontrolled factors producing spurious insignificance or significance, and help to tease apart treatments with low effect size. Repeated sampling of individuals over time to observe development of parasite infection, for example, will also require larger samples.
Furthermore, it is important to consider biological relevance of the numbers of individuals in each cage. Unsurprisingly, isolated workers die much quicker than those maintained in groups, possibly due to timing of food consumption (Sitbon, 1967; Arnold, 1978), so experimenters must be aware of expected duration of survival. Possible individual and social behaviours that are of interest should also be considered (e.g. Beshers et al., 2001). For example, >75 workers were needed to consistently elicit clustering behaviour (Lecomte, 1950), whereas 50 workers and a queen were needed for the initiation of wax production (Hepburn, 1986).
A Monte Carlo simulation model incorporating average lifespan (and standard deviation) for treatments and controls has been created to determine percentage of cases where a significant difference is obtained between groups. Without preliminary trials to determine the magnitude of an effect elicited by an experimental treatment as well as the variation between cages in that effect, statistical power may be impossible to know in advance. In such cases, it is advisable to maintain as many cages per treatment (≥3) and individuals per cage (≥30) as possible. Examination of the literature for similar studies may also help choose sample size; however, caution should be exercised due to differences in experimental conditions. Refer to the BEEBOOK paper on statistical methods (Pirk et al., 2013) for further details on the Monte Carlo simulation and on selecting appropriate sample sizes.