ABSTRACT
The variation of concentration values over inspected units is described by a statistical distribution. Statistical estimation from “exact” values is often complemented with values reported to be within known intervals. A finite mixture of distributions could then be a flexible approach, but the number of parameters is increased, and the identifiability of their estimates can be challenging. A similar analysis for true zeros can be done if the data consist of the plate counts instead of reported cfu/g-values and their reported LOQs. The model choice should reflect the type of available data. More complicated models with more parameters require more data than simple models with few parameters. It is not possible to decide beforehand on the model without considering the data. Although a simple model may be “too simple”, “more realistic” models can fail estimations if attempted with insufficient information in the data.
