Household Food Waste Patterns Across Groups: A Clustering Analysis Based on Theory of Planned Behavior Constructs and Shopping Characteristics

Theory of planned behavior (TPB) constructs and shopping routines are strong predictors of food waste behavior, while socio-demographic factors show mixed and weaker associations. We analyzed survey data from a nationally representative sample of 1066 U.S. households, measuring self-reported food waste frequency across meals, food types, and disposal methods. We applied k-medoid clustering on 19 TPB constructs and 25 shopping characteristics to identify three distinct consumer segments. "Structured Planners" (Cluster 1) showed the most deliberate shopping habits and strongest engagement in food waste reduction. "Flexible Planners" (Cluster 2) shared similar waste outcomes but approached shopping with greater spontaneity, while "Younger Wasters" (Cluster 3) were younger, lower-income, and less educated, with casual shopping habits, lower ratings of TPB constructs, and the highest food waste frequency overall. These distinct behavioral profiles enable policymakers to directly identify and target specific demographic segments for tailored food waste interventions. Particularly, "Younger Wasters" reported a significantly higher food waste frequency at 6.7 times per week, while "Structured Planners" and "Flexible Planners" were statistically similar at approximately 4.6 and 4.4 times per week. Dinner is the meal resulting in the most food waste across all groups, and "Younger Wasters" reported the highest frequency of waste in protein, oil, and grain. Post-clustering ANOVA analysis tested the predictive power of TPB, shopping characteristics, and cluster membership on food waste frequency. Results show that "Younger Wasters", along with variables like attitude, store shopping frequency, and shopping behavior, are significantly positively associated with food waste frequency. This study demonstrates the potential of clustering analysis in exploring food waste determinants and suggests using clustered indices as proxies for respondents' overall traits.