|
Abstract:
|
In an attempt to assess genotoxicity of different compounds, we need to analyze bacterial count data. A key-issue is the existence of suspicious data patterns which can arise either by data manipulation, or by negligence and poor laboratory conditions. In such cases, it is crucial to assess the likelihood of the degree of variability and replication of the bacterial counts.A potential approach is to model the sampling variation, using a count response model such as the Poisson distribution. The issue of under-dispersion demonstrated by the potentially manipulated data sets can be addressed using a two-parameter generalization that allows for this phenomenon, such as the Conway-Maxwell (COM) Poisson. An important metric in this analysis of bacterial counts would be the number of distinct observations. This, along with the coefficient of variation can be compared with the parametric count models and also with historical data.
The sampling distribution of these metrics will be constructed and empirical p-values will be evaluated. In an allegedly manipulated data set, it will be shown that, the degree of high replication and low variation is highly unlikely.
|
