Fitting a distribution to censored contamination data using Markov Chain Monte Carlo methods and samples selected...

Area of Science:

• Food safety and risk assessment
• Statistical modeling and data analysis
• Microbiology and contamination studies

Background:

• Statistical distributions are crucial for chemical and microbial contamination risk assessment, especially with censored data (e.g., nondetects).
• Current methods, like Markov Chain Monte Carlo, often assume independent and identically distributed (iid) data, which is frequently violated in food sampling.
• Non-random sampling designs, common in food commodity collection, can lead to biased statistical inferences.

Purpose of the Study:

• To develop a statistical framework for fitting distributions to microbiological data collected with unequal probabilities of selection.
• To address the limitations of existing methods that assume simple random sampling.
• To provide a more accurate risk assessment for food safety by accounting for sampling design biases.

Main Methods:

• Development of a weighted bootstrap estimation framework.
• Application of the framework to microbiological data, specifically using the Most Probable Number (MPN) technique.
• Comparison of results with methods that ignore unequal probability sampling designs.

Main Results:

• The weighted bootstrap method provides a robust approach for fitting distributions to non-randomly collected microbiological data.
• Ignoring unequal selection probabilities in sampling designs can lead to significant biases in statistical estimators.
• The proposed method demonstrates improved accuracy in risk assessment by accounting for sampling design.

Conclusions:

• A weighted bootstrap framework is effective for analyzing microbiological data from unequal probability samples.
• Accurate risk assessment in food safety requires statistical methods that acknowledge and correct for non-random sampling.
• This methodology enhances the reliability of inferences drawn from contamination data in practical settings.