Beyond household averages: daily volatility and stability patterns in food waste behavior

• Most households (75 %) exhibit substantial day-to-day variability in food discard. • Household baseline tendencies predict next-day discard more strongly than prior-day state. • Higher-stability drivers, including inedible parts, account for 37 % of total measured discard. • Single-time measures capture only one realization of a dynamic, multi-scale process. • Stability-based distinctions suggest different intervention orientations for recurring versus episodic discard. Most household food waste research relies on single-time retrospective surveys that implicitly treat waste as a stable household characteristic. We examined whether households exhibit consistent daily patterns or day-to-day variability through 14-day continuous tracking of 159 Florida households, measuring daily discard weights and recording 12 specific disposal reasons. We reframe household food waste as a temporally heterogeneous behavior characterized by stable and volatile patterns unobservable in cross-sectional designs. The findings reveal substantial behavioral heterogeneity: within the present quartile-based design, households in the lowest-waste quartile exhibit the strongest day-to-day stability, with an 82 % probability of remaining in the same waste state. The remaining 75 % show marked volatility, with no state exceeding 40–69 % persistence probability. Multinomial logit models find that household-level baseline averages are more strongly associated with next-day waste outcomes than prior-day state, though both contribute to daily transitions. To characterize patterns underlying this volatility, we apply hierarchical clustering of disposal reasons based on observed frequency and persistence, identifying three empirical stability groups. Higher-stability drivers (“inedible parts,” “old or exceeded time,” “blemish damage”) are jointly associated with approximately 37 % of total measured discard and show the largest same-day associations (0.36–0.69 lb) in panel regressions with household fixed effects. These findings suggest that continuous measurement reveals temporal dynamics unobservable in cross-sectional designs. The stability-based distinction suggests different intervention orientations: higher-stability recurring patterns may be more amenable to structural approaches, while episodic patterns may respond better to contextual prompts. Future research might examine whether these patterns generalize across broader populations and seasonal contexts.