Where's the beef? Consumer stockpiling of fresh meat during and following a disruption

Consumers' behaviors are an integral component affecting how supply chains must be managed (Baldi et al., 2024; Esper et al., 2020). The close relationship between supply chain issues and consumer behavior is particularly highlighted in the stockpiling literature (e.g. Ailawadi and Neslin, 1998). Also known as panic buying (Taylor, 2021), stockpiling is defined as “consumers buying large quantities of products” (Ahmadi et al., 2022, p. 57; Blattberg and Neslin, 1990), oftentimes in fear of something bad happening (Cambridge Dictionary, 2024). Stockpiling has implications for firms at various supply chain tiers, such as unavailability of goods (Modgil et al., 2022) and inflated prices (Dammeyer, 2020), leading to higher supply chain costs (Thomas, 2025) and dissatisfied consumers (Peinkofer et al., 2015). Recent research explores consumers' stockpiling behavior during disruptions such as natural disasters (Beatty et al., 2019) and pandemics (Pan et al., 2024), yet this literature is in its infancy, warranting deeper exploration (Pan et al., 2023).

We expand the disruption-oriented stockpiling literature with a focus on three core themes. First, this literature focuses on nonperishable products (e.g. Prentice et al., 2022) or both perishable and nonperishable products together (e.g. Papagiannidis et al., 2023), lacking a focus on perishables. While there is anecdotal evidence that consumers stockpile perishables (Kelly and McLaughlin, 2025) during widespread disruptions, an empirical examination is important given the nature of perishable products; the storage challenges may cause consumers not to engage in stockpiling of perishable products in the same way as they do nonperishables. Thus, there is a possibility that different inventory management approaches need to be applied during disruptions to perishables versus nonperishables. Accordingly, we focus on one type of perishable product: fresh meat. Second, we expand the study of locational determinants of stockpiling (e.g. Roos et al., 2025), with a focus on stockpiling by urban versus rural consumers. This focus is important practically and theoretically (Laudan, 1977) given (1) retailers can tailor inventory management decisions based on store location (Agrawal and Smith, 2013) and (2) there have been calls to understand how different consumers behave in relation to supply chain phenomena (Baldi et al., 2024; Esper and Peinkofer, 2017). Finally, we take a longitudinal approach and explore stockpiling during both the disruption and recovery phases. This enables stronger causal identification and allows us to better understand behavior given disruptions change over time (Ahmadi et al., 2022). Further, including the recovery phase is important given much of the product scarcity literature focuses on behavior only during a shortage (e.g. Kumar et al., 2023; Peinkofer et al., 2016), thus allowing our study to generate more detailed implications for retailers and suppliers.

Drawing on the shock response framework (SRF) (Argyres et al., 2019; Shin et al., 2025) and consumer psychology theory concerning scarcity (Bell, 1982; Lynn, 1991), we develop theory regarding expected stockpiling of fresh meat, addressing the following research question: How do consumers respond – in terms of stockpiling behavior – to disruptions and subsequent recovery of perishable product supply chains, specifically that of fresh meat, and does this behavior differ across urban versus rural households? We test our theory using a natural experiment (Rosenzweig and Wolpin, 2000) whereby meat production was disrupted in the United States (U.S.). Between April and June 2020, over 80 meatpacking plants experienced COVID-19 cases, resulting in slower operations and widespread closures (Cowley, 2020). This created potential (Gallagher and Kirkland, 2020) and actual (Gellerman, 2020) shortages of fresh meat at grocery stores. Using transaction data from NielsenIQ's Consumer Panel Dataset, we employ a difference-in-differences (DiD) design to compare the change in spend for fresh meat product categories to the change in spend for control product categories that were not affected – or were affected to a lesser degree (Angrist and Pischke, 2010) – by the meat supply chain disruption. We find behavior consistent with consumers stockpiling fresh meat during the disruption, with rural households stockpiling more than urban households. As the disruption subsided, behavior consistent with stockpiling continued.

This research makes several theoretical contributions. We offer the first use of the SRF to understand consumers' responses to supply chain disruptions such as shortages. Extant SRF applications concerning supply chain disruptions have focused mostly on B2B settings (Marzolf et al., 2026a; Miller et al., 2025; Shin et al., 2025), with limited application in a consumer context ^[1] (Marzolf et al., 2026b). By demonstrating that the SRF can explain consumer responses to shortages, we contribute theoretically by extending the SRF's explanatory power and consilience by showing that it can account for both a wider range of empirical phenomena as well as phenomena outside of those it was originally designed to explain (Tulodziecki, 2014). We further enrich the SRF by drawing on scarcity perspectives from commodity theory (Brock, 1998; Lynn, 1991) and regret theory (Bell, 1982; Loomes and Sugden, 1982) to better explain why shortages are expected to result in stockpiling. We contribute to the stockpiling literature by focusing on the three core themes previously addressed: expanding its focus to a perishable product type, exploring the boundary condition of urban-rural cohort, and taking a longitudinal perspective.

Finally, we generate several managerial insights, enabling retailers and suppliers to more accurately forecast demand during and following a disruption. This is important given the negative impacts of stockouts on consumer behavior (Amorim et al., 2025; Zinn and Liu, 2001) and satisfaction (Peinkofer et al., 2015). These insights are further important given the prevalence of supply chain disruptions (Gruchmann et al., 2024; Young, 2026), such as disruptions recently affecting eggs – a perishable product – which influenced retailers to implement purchase quantity limits in an attempt to lessen stockpiling (Archie, 2025).

A growing body of literature adopts a consumer-centric lens to inform supply chain management (SCM) (Baldi et al., 2024). Contributing to this discourse, we explore consumer stockpiling behavior to influence supply chain and inventory management strategy. The literature stream most relevant for this study is the stockpiling literature, which traditionally focuses on behavior during promotions (e.g. Ailawadi and Neslin, 1998; Huchzermeier et al., 2002). However, consumers' stockpiling behavior during disruptions or crises has emerged as an area of interest. Supplementary Materials Section I summarizes key studies of the empirical disruption-related stockpiling literature and their features that highlight how our work builds upon this literature.

As Table A (see Supplementary Materials Section I) highlights, this literature has explored stockpiling during events such as economic downturns (Pan et al., 2023), natural disasters (Pan et al., 2020), and pandemics (Papagiannidis et al., 2023). The studies demonstrate that stockpiling is heterogeneous across product types (Kassas and Nayga, 2021) and situational (Prentice et al., 2022) and consumer (Roos et al., 2025) characteristics. For example, Kassas and Nayga (2021) find the importance consumers place on stockpiling differs across product types, with the highest importance placed on hygiene and personal protection items (e.g. masks), and perishable foods being viewed as least important to stockpile. Regarding situational characteristics, researchers have explored how external factors such as government restrictions (Prentice et al., 2020) and media (Prentice et al., 2022) influence stockpiling. Further, consumer characteristics such as income (Pan et al., 2020) and personality traits (Roos et al., 2025) have been studied as predictors. Of these, a few studies have focused on locational determinants, with some finding stockpiling to be more common among individuals in more urban areas (Hori and Iwamoto, 2014; Roos et al., 2025), but another finding stockpiling of some items more common among those living farther away from stores (Rune and Keech, 2023).

In terms of methodological approach (see Table A, Supplementary Materials Section I), most studies exploring stockpiling during disruptions employ cross-sectional data (e.g. Omar et al., 2021), with the longitudinal studies focusing mostly on disruptions with a restricted geographic and temporal scope, such as hurricanes (Pan et al., 2020) or earthquakes (Hori and Iwamoto, 2014). The primary data source – especially in studies exploring stockpiling during widespread disruptions – is consumer surveys (Roos et al., 2025). Thus, the stockpiling measure is mostly perceptual and/or retrospective.

Our literature review revealed important gaps warranting further investigation. First, while some research explores stockpiling heterogeneity across product types (Kassas and Nayga, 2021), much of the research focuses on stockpiling of nonperishables – such as bottled water (Beatty et al., 2019) or toilet paper (Prentice et al., 2022) – or both perishable and nonperishable products with little to no distinction among product groups (Papagiannidis et al., 2023). Given heterogeneous scarcity effects across contexts (Shi et al., 2020), it is essential to explore behavior in response to a disruption-related shortage of a perishable product type. Second, considering initial insights regarding the importance of locational determinants (e.g. Roos et al., 2025), we explore the boundary condition of urban versus rural consumers in the context of a perishable product type during a widespread disruption. Third, given the typical cross-sectional focus (e.g. Omar et al., 2021) and that disruptions evolve (Ahmadi et al., 2022), we further explore the longitudinal nature of stockpiling behavior. We also address shortcomings of this primarily survey-based literature by capturing behavior with transaction data, thereby addressing recall bias associated with retrospective measures (Tax et al., 1998).

Our theorizing approach follows Cronin et al. (2021) as we extend a more general, abstract programmatic theory – the SRF (Argyres et al., 2019; Shin et al., 2025) – to create a unit theory (Wagner and Berger, 1985) that makes context-specific predictions. We do this by introducing elements from commodity theory (Brock, 1998; Lynn, 1991) and regret theory (Bell, 1982; Loomes and Sugden, 1982) into the SRF. To our knowledge, this research is the first to suggest incorporating principles from consumer psychology theories into the SRF to increase its explanatory power in consumer-centric SCM settings.

Argyres et al. (2019) define shocks as significant changes stemming from an actor's (e.g. firm's) external environment that are beyond any one actor's control. They identify four sources of shocks: (1) demand, (2) supply, (3) technology/innovation, and (4) regulation (Argyres et al., 2019), though as noted by Miller et al. (2025), shocks are often imbued with elements from multiple sources. For consumers, production disruptions caused by COVID-19 outbreaks at meatpacking plants and the resulting concerns about fresh meat supply represent a negative supply shock (Kulpa et al., 2026a, b). The crux of the SRF is that actors respond to shocks in ways that economize on four types of costs: (1) adjustment costs, (2) transaction costs, (3) opportunity costs for early response, and (4) opportunity costs for delayed response (Argyres et al., 2019). Marzolf et al. (2026a) further note that responses are influenced by a desire to reduce uncertainty imposed by the external environment.

The cost types require contextualization to each setting (Miller et al., 2025). In our context, adjustment costs capture costs consumers experience if they increase fresh meat purchases, such as filling storage space in freezers. Transaction costs are those incurred to find new or alternative sources of fresh meat, such as time spent researching and driving to alternative sellers. Opportunity costs for early response include extra near-term spending, purchasing fewer other products or services because a greater share of spend is being devoted to fresh meat, and potential for increased spoilage of fresh meat. Opportunity costs for delayed response include experiencing stockouts or paying higher prices than if fresh meat was purchased earlier.

While the SRF posits that individuals or organizations act in response to shocks to economize on these four costs and to reduce perceived uncertainty (Marzolf et al., 2026a), it provides little guidance regarding the magnitude of these costs to make predictions regarding stockpiling. Thus, we augment the SRF with theories from consumer psychology: commodity theory and regret theory. Commodity theory highlights that limited product availability increases perceived product attractiveness and value (Brock, 1998; Lynn, 1991), thereby leading to the expectation of consumers stockpiling during shortages. Framed in SRF's vernacular, commodity theory suggests greater attractiveness of products perceived as scarce will increase opportunity costs for delayed response. This is reinforced by regret theory, which highlights that consumers make decisions to avoid future regret (Bell, 1982; Loomes and Sugden, 1982); accordingly, consumers will value a product more if future availability is questionable. Regret theory further aligns with the assumption in SRF that responses to shocks are driven, in part, by the desire to reduce perceived uncertainty (Marzolf et al., 2026a). When these perspectives are introduced (Makadok et al., 2018) into the SRF, it makes a general prediction of consumer stockpiling in response to perceived or actual shortages.

However, it is less clear whether this prediction will hold for perishables such as fresh meat because, compared to nonperishables, consumers experience greater adjustment costs due to needing specialized storage space and higher opportunity costs for early response due to spoilage concerns. Despite these higher costs, we expect consumers' desire to avoid opportunity costs for delayed response (e.g. stockouts) will outweigh the increased costs consumers experience in other cost categories if they stockpile. Regret theory research provides evidence of the powerful role that regret-avoidance behaviors play in decision-making (Bell, 1982), which points to opportunity costs for delayed response looming predominantly in consumers' minds. This is supported by evidence that losses loom larger than gains (Kahneman and Tversky, 1979). This regret-avoidance mechanism should be augmented by commodity theory's principle that scarcity increases attractiveness (Lynn, 1991), which should further magnify perceived opportunity costs for delayed response. Moreover, given uncertainty about the ultimate impact that production slowdowns would have on meatpacking activity, the desire to reduce uncertainty in response to negative supply shocks should further drive stockpiling. The reason is that stockpiling fresh meat helps consumers buffer against uncertainty in future supply in the same manner that retailers and wholesalers increased inventories in response to supply chain disruptions in late 2021 and the first half of 2022 (Marzolf et al., 2026a). Accordingly, we hypothesize:

We next explore the moderating effect of whether a household is in an urban or rural area (i.e. urban-rural cohort) on stockpiling. Building from the SRF, we expect stronger stockpiling behavior among rural than urban households. First, rural households tend to have larger homes (U.S Energy Information Administration, 2023), lending to more storage space enabling greater ability to stockpile, especially perishables with more challenging storage requirements. Thus, rural consumers should experience lower adjustment costs. Second, rural consumers tend to earn less (Trovall, 2023), leading them to be more price sensitive than urban consumers (Horwich, 2024). Given consumers expect price increases when product availability is low (Lynn and Bogert, 1996), this suggests greater opportunity costs for delayed response relative to urban consumers, which should encourage stockpiling. Third, rural households often reside farther away from food stores (Rhone et al., 2019), with this lower access leading to greater repercussions if they delay the purchase and cannot secure the desired products later, further increasing opportunity costs for delayed response. Thus, to minimize regret (Bell, 1982), we expect rural households will be more likely to stockpile. This logic is supported by Shi et al. (2020, p. 388) who state, “consumers who have a greater need to avoid future regret choose to buy a product not because of its utility but, rather, because they are concerned that they won't be able to buy it in the future.” Accordingly, we hypothesize:

Finally, we hypothesize regarding short-term continued stockpiling as the disruption subsides. We expect consumers' purchases of fresh meat will remain above pre-shock levels even as disruptions subside due to consumers' desire to reduce perceived uncertainty about fresh meat availability. As noted by Omar et al. (2021), there is evidence that uncertainty drives stockpiling, likely due to a desire to be in control (Rune and Keech, 2023) and minimize potential regret (Bell, 1982) if a shortage recurs. In other words, the fact that fresh meat disruptions and subsequent shortages may occur again in the near future – something reasonable for consumers to perceive during the 2020 COVID-19 outbreak given uncertainty about how the disease would evolve – would encourage purchasing above pre-shock quantities to mitigate the negative effects (e.g. opportunity costs for delayed response) of shortages. We thus hypothesize:

We implement a DiD approach to establish causal identification by identifying (a) treatment products, which are product groups affected by the disruption, and (b) control products, which are product groups that were less affected (Angrist and Pischke, 2010). We consider two time periods after the treatment onset: during the disruption and during the recovery. Thus, we compare (1) the change in the outcome for the treatment products before and after the onset of the disruption to the change in the outcome for the control products before and after the onset of the disruption (Scott et al., 2021) to test H1 and H2, and (2) the change in the outcome for the treatment products before the disruption and during the recovery to the change in the outcome for the control products before the disruption and during the recovery to test H3. We used the onset of production disruptions caused by COVID-19 outbreaks in the meat supply chain in the U.S. in late spring/early summer 2020 as the disruption to test our hypotheses. This setting corresponds to what Merton (1987, p. 10–11) identifies as a location-related “strategic research site (SRS)” and a temporal-related “strategic research event (SRE),” which are part of the larger concept of “strategic research material (SRM)” and allow for the investigation of formerly “stubborn problems.”

We use several sources to determine the timeline regarding the disruption period (during which the meat supply chain was experiencing closures and production slowdowns) and the recovery period (as the disruption was subsiding). The USDA stated, “Partial plant closures and increased social distancing protocols were implemented at meatpacking plants across the country starting in late April 2020 through early June” 2020 (U.S. Department of Agriculture, 2021). The start was corroborated by popular press sources including an April 21st article stating meat processing plants in at least eight states had been forced to temporarily close due to COVID-19 cases (Gray Television, 2020) and an April 26th article reporting meat plants had faced outbreaks leading to closures (Gallagher and Kirkland, 2020). Google Trends (2020) also supported the timeline; conducting a search for “meat shortage” in the U.S. for February–August 2020 showed peak interest on April 28, and by June 7, the prominence of searches was down to 2% of the peak (see Supplementary Materials Section II, Figure A).

We accordingly determined the timeline (see Figure 1). The disruption period is April 19–June 6 (weeks 17–23 of 2020). The recovery period includes the same number of weeks as the disruption period (7 weeks) and is June 7–July 25 (weeks 24–30 of 2020). The pre-disruption phase includes all weeks in 2020 prior to the disruption (weeks 1–16) and the full 30-week timeframe (weeks 1–30) for 2019 to enable estimating product by week of year fixed effects.

Our first data source is the NielsenIQ Consumer Panel Dataset. Utilized in recent SCM research (Marzolf et al., 2026b; Pan et al., 2023; Sodero, 2022), it provides a panel of approximately 60,000 households in the U.S. It includes data regarding household characteristics, shopping trips (e.g. when, where, cost), and products purchased. Product-level data is available for fast-moving consumer goods, making it appropriate for understanding grocery shopping patterns. We accordingly gathered transaction data (item purchased, price paid, date) and household location (county). Second, we sourced Consumer Price Indices (CPI) from the U.S. Bureau of Labor Statistics (2024). Finally, we utilized data from the U.S. Department of Agriculture (2019, 2025) and the U.S. Census Bureau (2025a, b) to assign the measure for urban-rural cohorts.

Based on data coverage, the unit of analysis becomes the product category $×$ urban-rural cohort $×$ year $×$ week of year. The sample consists of 1,440 observations (eight product categories, three urban-rural cohorts, two years, 30 weeks per year). We identified four treatment product categories – fresh beef, fresh turkey, fresh pork, and fresh chicken – and four control product categories – refrigerated dairy milk, fresh lettuce, fresh carrots, and frankfurters (franks). While many closures affected beef and pork plants (Cowley, 2020), poultry products were also affected (McCarthy and Danley, 2020) but to a lesser degree (Durbin, 2020). Including poultry (i.e. chicken and turkey) in the treatment group may therefore make the estimated treatment effect more conservative (Angrist and Pischke, 2010) to the extent these categories experienced weaker disruption-related effects. Milk, lettuce, and carrots were chosen as controls as they are perishables with short shelf lives like fresh meat. Franks were included as they are a meat product, indicating similarity to the treatment categories, but their longer shelf life (FoodSafety.gov, 2023) means franks were less affected by the disruption to the meat supply chain. Aligned with Angrist and Pischke (2010)—who explain that the assignment of treatment and control categories is not always a binary 0/1 but rather a continuum of more versus less affected – these products are appropriate for a DiD analysis given the treated products were affected to a far greater degree by the disruption (i.e. the fresh meat production disruption) than the controls. Furthermore, as we empirically demonstrate (see Section 4.6), sales for these products exhibit parallel pre-treatment trends, providing supporting evidence for their use as comparison products. Moreover, identifying “pure” treatment and control categories is often not possible. For example, retail studies have used relatively arbitrary assignments of units into treatment and control groups (Akturk and Ketzenberg, 2022; Akturk et al., 2018; Gallino and Moreno, 2014), such as designating areas with customers living a median distance of more than 50 miles from a retail outlet as a control group but those with customers living closer than 50 miles as a treatment group (under the assumption that customers who are farther away are less likely to drive to the physical store) (Akturk et al., 2018).

We utilize three urban-rural cohorts. The U.S. Department of Agriculture (2025) assigns Rural-Urban Continuum codes, which range from 1 (most urban) to 9 (most rural), to each county in the U.S. We assign these codes to the categories of Urban, Mid, and Rural as shown in Supplementary Materials Table B (see Supplementary Materials Section III). Households in the purchase data are assigned to one category based on the commuting zone in which they live. There are 741 commuting zones in the U.S., which group counties to represent local economies in which people work and live (U.S. Department of Agriculture, 2019), providing a representation of the environment in which a household shops. This assignment is further explained in Supplementary Materials Section III.

We utilize i to index product category, j for urban-rural cohort, t for year, and k for week of year. The dependent variable (⁠$LnSpend$⁠) is the natural log of the dollars spent in a product category $×$ urban-rural cohort $×$ year $×$ week of year. The natural log transformation was applied because spend differs across product categories and to improve interpretability by reporting results in percentage form. We retrieved dollars spent at a shopping trip $×$ item level, linking each of the following: (1) the shopping trip to the household to determine urban-rural cohort, (2) the date of each trip to the year and week, and (3) each item to its product category.

The first predictor, $Disruption$⁠, represents the disruption period – the period of meat plant closures and production slowdowns due to COVID-19 cases – for the treatment products (i.e. fresh meat products). $Disruption$ is assigned a 1 during the disruption period (weeks 17–23 of 2020) for treatment products. This variable is assigned a 0 for control products during the full study and for treatment products during all other times of the study (outside of the disruption period).

The second predictor, $After$⁠, represents the recovery period (i.e. the period during which closures and slowdowns in the meat supply chain were improving) for the treatment products. Thus, $After$ is assigned a 1 during the after period (weeks 24–30 of 2020) for treatment products and a 0 for products in the control group during the full study and for the treatment products during all other times of the study (outside of the after period). The operationalization for the key predictors follows Scott et al. (2023).

The moderator corresponds to urban-rural cohorts. Urban cohort serves as the reference group. Therefore, our moderator of interest is $Rural$⁠, which is assigned 1 for the rural cohort and 0 otherwise. As we employ three urban-rural cohorts, $Mid$ (assigned 1 for the middle urban-rural cohort and 0 otherwise) is also included.

As the outcome is measured in dollars, we control for inflation with the natural log of the CPI for the product category $×$ year $×$ week of year. We label this variable $LnCPI$⁠. CPI data is available monthly at the product type level, and we map it to the respective weeks. CPI is sourced on a not seasonally adjusted basis and is adjusted to set January 2019 equal to 100. Like the dependent variable, the natural log transformation is applied to the CPI to interpret results as an elasticity. This aligns with studies examining pricing changes (Peltzman, 2000).

We include several fixed effects. Product $×$ week of year fixed effects (⁠$γik$⁠) account for product group seasonality. For example, sales of franks increased around Memorial Day (week 22) and the Fourth of July (week 27) in both years. Product $×$ urban-rural cohort fixed effects (⁠$λij$⁠) account for stable purchasing patterns within an urban-rural cohort and product category, such as one cohort purchasing more fresh meat overall than others. Urban-rural cohort $×$ year $×$ week of year fixed effects (⁠$ηjtk$⁠) control for factors impacting a cohort's purchasing patterns across all products at a given time. For example, urban consumers were typically more concerned about the COVID-19 pandemic and in favor of measures to stop the spread (Chauhan et al., 2021), leading to potentially different shopping habits at times. Urban-rural cohort $×$ year $×$ week of year fixed effects also control for year $×$ week of year fixed effects, thus accounting for average effects across all product groups during a week, such as general stockpiling at the start of COVID-19 (March 2020). Table 1 contains descriptive statistics.

To establish causal identification via DiD, some assumptions must be met. The treatment must be exogenous to other factors that could impact the outcome (Scott et al., 2021). Other factors that may impact the outcome of spend in a product category are sales/promotions and holidays (e.g. Memorial Day may influence meat purchases). Production disruptions due to COVID-19 cases in the meat supply chain are exogenous to these factors, as sales/promotions in the retail store would not influence the spread of cases in meat facilities upstream. Further, the treatment is exogenous to holidays as COVID-related issues in meat plants were declining by early June 2020 (U.S. Department of Agriculture, 2021) when an increase would have been expected following Memorial Day gatherings.

The parallel trend assumption states that in the absence of the treatment, the treatment group would have evolved in the same way as the control group (Angrist and Pischke, 2009). Accordingly, we examined treatment and control group trends – both visually and statistically – for parallel movement prior to the treatment onset (see Supplementary Materials Section IV, Figure B, Table C). These visual and statistical tests support the parallel pre-treatment trends assumption and provide evidence consistent with the use of these control products as a reasonable comparison group.

We employed a series of DiD models. H1 and H3 are tested by:

Where the variables are as described previously and $εijtk$ is the residual of $LnSpendijtk$⁠. H1 predicts that $β1$ is positive. H3 predicts that $β2$ is positive.

To test H2, we add interaction terms:

Where H2 predicts that $β4$ will be positive. We exclude first-order terms for $Ruralj$ and $Midj$ because the variance of the first-order terms is subsumed by product $×$ urban-rural cohort fixed effects (⁠$λij$⁠); we can include the two-way interaction terms of these variables with $Disruptionitk$ because this variable varies over time.

Results, estimated using panel data estimation with fixed effects in Stata 16.1, are presented in Table 2. Model 1 evaluates Equation (1), resulting in an unconditional DiD model. Model 2 adds $LnCPI$⁠. Concerning H1, $β1$ in Model 2 is statistically significant and positive (⁠$β1$ = 0.207, p < 0.01), indicating that the change in inflation-adjusted spend between the pre-disruption and disruption periods was 20.7% higher for the treatment product groups (fresh meat) as compared to the controls (the set of comparison products meeting parallel trends tests as detailed in Section 4.6); this increase in spend is consistent with stockpiling behavior. Thus, H1 is supported. Concerning H3, $β2$ is statistically significant and positive (⁠$β2$ = 0.178, p < 0.01) and indicates that the change in inflation-adjusted spend between the pre-disruption and after periods was 17.8% higher for the treatment product groups as compared to the controls. Therefore, H3 is supported.

Model 3 adds interactions (Equation 2). The coefficient of interest, $β4$⁠, compares behavior between the rural and urban cohorts and is positive and significant (⁠$β4$ = 0.102, p < 0.05). The change in inflation-adjusted spend between the pre-disruption and disruption periods for the treatment products compared to controls was 10.2% higher for the rural cohort than the urban cohort. These results are consistent with the rural cohort stockpiling fresh meat to a greater degree than the urban cohort, and H2 is supported.

Robustness tests are detailed in Supplementary Materials Section V (Tables D, E, F, G, H, I, J, K, L, M). Results are consistent with the main analysis.

We contribute theoretically by extending the SRF (Argyres et al., 2019) – which was brought to SCM by Shin et al. (2025) – by adding principles from commodity theory (Brock, 1998) and regret theory (Bell, 1982). Incorporating elements of these theories expands the SRF's explanatory power (Tulodziecki, 2014) by enabling it to explain why perceived and actual shortages result in consumer stockpiling. Doing so provides an explanation of why opportunity costs for delayed response (i.e. not stockpiling) will outweigh other costs associated with stockpiling (adjustment costs and opportunity costs for early action). That the SRF, with appropriate contextualization, can account for consumer stockpiling behavior is encouraging because this theory was developed in a substantially different context, with Argyres et al. (2019) focusing mainly on innovation shocks (while recognizing the existence of other shock types). Per Duerr and Fischer (2025), the ability of the SRF to explain phenomena well outside the scope of those it was originally designed to explain, termed consilience (Tulodziecki, 2014), gives researchers reason to engage in theory pursuit (Nyrup, 2015) by extending its application to other SCM settings to further understanding (Dellsén, 2021) regarding how consumers and organizations respond to shocks.

Considering domain-specific contributions, we extend the disruption-focused stockpiling literature by studying a perishable product type, fresh meat. Perishables have been largely overlooked despite their operational relevance. Thus, we contribute to consumer-centric SCM research (Esper et al., 2020) by demonstrating how a supply disruption shapes consumer purchasing behavior in a context where more challenging storage requirements and higher spoilage risk might constrain stockpiling. We find that despite these characteristics, consumers engage in behavior consistent with stockpiling of fresh meat during a shortage, thereby increasing the comprehensiveness of phenomena investigated (Dellsén, 2021). In extending the view of stockpiling from a phenomenon primarily related to nonperishable products, our research refines theoretical understanding of disruption-driven demand responses and highlights important implications for inventory management/positioning and demand forecasting.

We further refine understanding regarding the moderating impact of whether a consumer lives in a rural or urban area on stockpiling, building upon the literature which considers locational determinants of stockpiling (e.g. Roos et al., 2025). We found that rural households displayed behavior consistent with greater stockpiling than urban households in the context of a widespread disruption affecting fresh meat products. We theorized this was due to rural households experiencing lower adjustment costs given fewer space constraints (US Energy Information Administration, 2023) in conjunction with higher opportunity costs for delayed response because of greater price sensitivity (Horwich, 2024). Thus, we provide evidence for urban-rural cohort acting as a boundary condition of stockpiling of fresh meat products (Makadok et al., 2018).

Lastly, we contribute to the disruption-focused stockpiling and scarcity literature by taking a longitudinal perspective. This is important because (1) it allows for stronger causal identification and (2) supply disruptions – and related consumer behavior – evolve (Ahmadi et al., 2022), making it important to devise and test temporally-focused theory (Ancona et al., 2001). Additionally, much of the product scarcity literature examines consumer behavior during a scarcity (Kumar et al., 2023; Peinkofer et al., 2016). Our theory suggests that even as supply disruptions subside, if actors have reason to expect disruptions will recur in the near future, they will continue to purchase more than before the shock to mitigate uncertainty (Marzolf et al., 2026a). Our findings align with this theory, thus extending understanding regarding behavior as scarcity events subside.

This research offers practical insights for retailers and suppliers, which is important given the expectation for continued disruptions (Gruchmann et al., 2024; Young, 2026). We found behavior consistent with consumers stockpiling during the disruption to the fresh meat supply chain; thus, stockpiling is not unique to commonly studied nonperishables (e.g. Pan et al., 2020). This knowledge is important to enable retailers and suppliers to forecast demand more accurately with an impending or occurring disruption. For suppliers, this indicates the importance of avoiding or mitigating such disruptions given the repercussions of consumers' stockpiling behavior (e.g. Dammeyer, 2020). Our theory further provides guidance as to the features of products that may encourage stockpiling. Goods for which opportunity costs for delayed action outweigh other costs, such as adjustment costs and opportunity costs for early action, should be especially prone to stockpiling in response to perceived shortages. Items frequently consumed, such as toilet paper and, as we found, fresh meat, meet these criteria.

Finding that households in rural areas displayed behaviors consistent with greater stockpiling than households in urban areas likewise provides insights. Awareness that stores in rural areas may be more susceptible to stockpiling informs retailers regarding how quickly stockouts might occur at different stores during a disruption, as inventory will move more quickly in rural areas. For suppliers, similarly, they can better anticipate retailers' ordering behavior based on the geographic areas the retailers serve, with greater increases expected in rural areas. Such an understanding of consumer behavior will help both retailers and suppliers predict the challenges they will face throughout their network, enabling them to plan to mitigate the risks when a shortage occurs. This is important given stockout costs (Thomas, 2025) and that stockouts negatively affect purchase behavior (Zinn and Liu, 2001) and satisfaction (Peinkofer et al., 2015).

Further, finding that consumers still display behavior consistent with stockpiling after scarcity is reduced informs retailers and suppliers regarding recovery. We tested the ad-hoc interaction effects between urban-rural cohort and the after period (see Supplementary Materials Section VI, Table N). Aligned with the disruption period, the rural cohort displayed behavior consistent with stockpiling to a greater degree as the disruption subsided than the urban cohort. Like in the disruption period, this knowledge helps retailers and their suppliers better plan and forecast demand as they recover from the disruption. Continued stockpiling means demand will be heightened for some time, which prolongs recovery to normal inventory conditions. We also tested for prolonged stockpiling (see Supplementary Materials Section VI, Table O) and found that the effect was diminished by fall 2020, indicating that although behavior consistent with stockpiling continued during the recovery stage, the behavior did not occur indefinitely.

While this study is in the context of a supply chain disruption driving a shortage, understanding stockpiling behavior is important beyond this context. Recently, uncertainty around tariffs in the U.S. drove consumers to stockpile goods (Cavale, 2025). Further, the previously mentioned egg shortage in the U.S. sparked conversation around stockpiling (Archie, 2025). This highlights the relevance of stockpiling behavior and understanding its boundary conditions. Our results can help firms prepare for and respond to stockpiling caused by other drivers.

The first limitation is that we were unable to control for stockouts or purchase limits, which likely affected the volume of fresh meat purchased (Law, 2020). If stockouts or purchase limits reduced observed fresh meat purchases, then the estimated effects may understate consumer demand during the disruption. However, because the available data do not capture unrealized demand, the precise magnitude of this potential attenuation cannot be determined. Second, we cannot completely rule out the possibility that the observed effects represent increased consumption rather than stockpiling. This is a limitation in all archival data quantitative studies: perfect theoretical identification (Miller et al., 2023) is unattainable because regression coefficients do not embed theoretical mechanisms (Miller and Kulpa, 2022). Still, many factors suggest our stockpiling mechanism outperforms other explanations, such as consumers cooking more at home. Consumers cooking more at home is unlikely to fully explain why purchases of fresh meats spiked after COVID-19 outbreaks at meatpacking plants relative to our control products, especially given evidence of parallel purchase trends before treatment. Moreover, given increases in fresh meat prices during the disruption and after periods (US Bureau of Labor Statistics, 2024), it is unlikely that consumers intentionally shifted purchases to more expensive products merely for consumption.

Third, the timing for the disruption and after periods cannot be definitively determined. We reduce the threat that this poses by corroborating the start of the disruption across several sources (Gallagher and Kirkland, 2020; Google Trends, 2020; Gray Television, 2020) and conducting robustness tests. Finally, given unique characteristics of the control product categories as compared to the treated product categories (e.g. differences in freezability, storage duration, spoilage risk, centrality to meals, substitutability, etc.), the control categories may not perfectly capture fresh meat's untreated post-period trajectory. This concern is reduced, though not entirely eliminated, by the presence of parallel pre-treatment trends and by the robustness analyses (see Supplementary Materials Section V). To further lessen concerns regarding the control product categories, we also provide visual plots of spend by product category (see Supplementary Materials Section VII; Figures C, D, E, F, G, H, I, J).

Future research may explore stockpiling of other perishable product types – such as ones that cannot be frozen like fresh meat can – given the effect of scarcity is heterogeneous (Shi et al., 2020) and recent shortages of other perishable products (e.g. Archie, 2025). Additional predictors of stockpiling could also be explored in this context, including both consumer and retailer characteristics (e.g. demographics, retailer type). Also given the heterogenous effects of scarcity (Shi et al., 2020), there is an opportunity to further explore locational determinants of stockpiling behavior (e.g. Roos et al., 2025; Rune and Keech, 2023). Next, it would be fruitful to explore how to mitigate stockpiling behavior – such as through “nudging” via communication (Flygansvær et al., 2021) – in the context of perishables during a supply chain disruption. Researchers could also examine how other behaviors, such as the purchase of substitutes or changes in consumption, are affected by shortages of various types of perishables. Future research could also explore the duration of the stockpiling behavior following a shortage to further build upon our longitudinal approach. There is also an opportunity to explore behavior post-recovery to better understand the observed effects and confirm whether they are due to stockpiling or increased consumption. It would also be interesting to explore how stockpiling behavior is affected by media attention over time.

The supplementary material for this article can be found online