Shedding patterns of various organisms, such as E. coli O157 and Salmonella, are dependent on the Agent-Host-Environment interaction. Because the animals’ environment is a key component, a greater number of intervention studies are being conducted within commercial settings to provide more meaningful inferential measures of effect. We have observed that in studies conducted in such settings, the outcome is highly clustered within the most closed level of organization, i.e., pen-level for feedlot studies of E. coli O157 or dairy-level for Salmonella. Consequently, substantial extra-binomial (or even bimodal) variation is frequently evident within the data. As a result, assumptions of many regression approaches, such as logistic regression, are violated and estimates of effect (and precision) are likely biased. To complicate matters further, many observations are expected to yield low (or even zero) prevalence given a bimodal distribution and consequently, do not provide informative data of the effectiveness of the intervention for reducing prevalence in a regimented manner. Other analytical approaches such a finite-mixture and zero-inflated models may be more appropriate for these situations, but new methods are also likely needed to address real-world distributions. In addition, sample-size estimation given observed population distributions is, therefore, likely to be anything but straightforward.