Count data distributions and their zero-modified equivalents as a framework for modelling microbial data with a...

Area of Science:

• Microbiology
• Statistical Modeling
• Food Safety

Background:

• Microbial count data often exhibit a high proportion of zero counts, posing challenges for traditional log-normal statistical assumptions.
• Accurate statistical treatment of zero-inflated microbial data is crucial for hygiene indicators and pathogen monitoring.

Purpose of the Study:

• To introduce and evaluate a Poisson-based distribution framework for zero-inflated microbial count data.
• To compare the performance of various zero-modified distributions against standard models for bacterial data.

Main Methods:

• Fitted Poisson, negative binomial, zero-inflated Poisson, hurdle Poisson, zero-inflated negative binomial, and hurdle negative binomial distributions.
• Utilized real-world zero-inflated bacterial data: total coliforms (n=590) and Escherichia coli (n=677) from beef carcasses.
• Assessed goodness-of-fit using chi-squared tests to evaluate model performance.

Main Results:

• The simple negative binomial distribution showed improvement over Poisson but slightly misestimated zero and low positive counts.
• Zero-modified Poisson models failed to handle data over-dispersion.
• The hurdle negative binomial distribution demonstrated a good fit for both coliform and E. coli data, comparable to the negative binomial model.

Conclusions:

• Zero-modified count distributions, particularly the hurdle negative binomial, can appropriately represent bacterial data with substantial zero counts.
• This framework provides a valuable alternative for statistical analysis and stochastic risk assessment in food safety.