The assessment of factors contributing to food fraud vulnerability: a Bayesian modelling approach on a food fraud database

The assessment of factors contributing to food fraud vulnerability: a Bayesian modelling approach on a food fraud database

Food fraud is defined as intentional behaviour designed to misrepresent or sabotage food items by third parties for economic gain. The growing incidences of food fraud over the last ten years lends support to the notion that traditional food safety intervention countermeasures have been largely inadequate. The increase in fraudulent incidents has further resulted in a change of focus in theory related to the investigation of food fraud from risk mitigation to vulnerability reduction. Food fraud is now classified as a separate category of food safety management systems in the literature. This shift in focus has also brought into question whether the traditional methods for combating food fraud also require a shift from common detection methods deployed by food safety managers to consideration of prevention and vulnerability reduction.

The food fraud vulnerability reduction literature has emerged from within the theoretical domain of Criminology and has relied on the Routine Activity Theory framework for identification of root causes for food fraud. The Routine Activity Theory suggests that there are three main factors to consider in the reduction of vulnerability to fraudulent activity. These are: opportunity; motivation; and countermeasures. Whilst this literature provides a framework from which to identify possible areas of vulnerability to food fraud, it does not provide a methodology that would allow food producers and processors to more accurately quantify and assess their vulnerability to food fraud. This study, therefore, takes up the call from researchers in this field to find better methods for the prevention of Food Fraud Vulnerability (FFV) factors by the development of a holistic model that can assess the level (degree) of vulnerability to food fraud for food products targeted at human consumption.

Embracing a pragmatism view, this research adopted a sequential exploratory mixed methods approach. Phase one of the research commenced with a qualitative Barrier Analysis approach to extract the main Food Fraud Vulnerability factors within specific categories of food fraud as detailed in the US Pharmaceutical Food Fraud Database (USP FFD). The second phase of the research adopted a quantitative method, applying a Bayesian Network BN) modelling approach using SPSS Modeler 18.2. The data for the research was extracted from 580 incidents of global food fraud recorded between 2000 and 2018 in the US Pharmaceutical Food Fraud Database (USP FFD). Four food product categories were used for this study from the sixteen different categories provided by the database. These were: seafood; meat; alcoholic beverages; and dairy. These food categories represented just over fifty percent of all incidents recorded in the database. Approximately 80 percent of the data was used to train and develop the BN model, with the remaining 20 percent used for testing the validity and accuracy of the final BN model.

The Barrier Analysis technique conducted in phase one of this research identified new Food Fraud Vulnerability dimensions such as the physical form of products, supply chain complexity/transparency, corruption level of the detection country, culture and religion, price spikes, the requirement for coordination of law enforcement agencies, extensiveness of traceability, and food safety, which were subsequently indexed into the appropriate vulnerability categories of: opportunity; motivation; and countermeasures. The Bayesian network analysis conducted in phase two of the research, produced three main findings. First, the study confirmed that the Tree Augmented Naïve BN model can provide a robust assessment of vulnerability to food fraud with an accuracy of 86%. Second, the variable country of origin was identified as having the greatest influence on FFV factors. Third, the variable food fraud incident type was identified as having the least influence on FFV factors.

The findings from this research make two main contributions to the food fraud literature. Firstly, the identification and indexing of new Food Fraud Vulnerability dimensions has expanded the current knowledge regarding the root causes of Food Fraud Vulnerability. Further, this study has also provided valid empirical support for the inclusion of these factors in future studies investigating issues relating to vulnerability to food fraud. Next, the model based on the Barrier Analysis technique and BN modelling approach has provided a methodology to more accurately assess the root causes of food fraud for a range of food product types. This model allows future researchers to achieve more impactful results and to better understand the inter-
relationships of food fraud variables through the ability to manipulate both input factors and output (FFV) factors within specified conditions.

The study also makes two practical contributions for food companies, authorities in border protection, policymakers and quality assurance agencies through the identification of the main factors that are most likely to increase the vulnerability to food fraud under different conditions and for different product types. This information will assist these groups to implement the most appropriate countermeasures to combat the areas of vulnerability, and will also assist in the determination of likely FFV factors for future incidents.