Pass or Fail: Economic Incentives to Reduce Salmonella Contamination in Ground Beef Sold to the National School Lunch Program

Pass or Fail: Economic Incentives to Reduce Salmonella Contamination in Ground Beef Sold to the... Abstract Ground beef sold to the USDA’s Agricultural Marketing Service (AMS) for distribution to the National School Lunch Program (NSLP) must meet stringent food-safety standards, specifically, a zero-tolerance standard for Salmonella. We use a unique data set containing information on Salmonella levels in order to examine the sequential decisions of ground-beef plants to become registered as AMS suppliers and then bid on contracts to supply the NSLP from 2006 to 2012. We find that plants exploit their competitive advantages in relatively high productivity and strong performance on Salmonella tests when choosing to bid on contracts in a given year. Furthermore, the incentives generated by the zero-tolerance standard for Salmonella are highly effective: ground beef supplied to the NSLP is 21–22 percentage points more likely to meet a zero-tolerance standard for Salmonella than ground beef tested as part of typical meat-plant inspections. The National School Lunch Program (NSLP) provides about 31 million subsidized or free meals to children in the United States each school day. Some of these meals are made from ground beef, chicken, and other foods purchased under contract by the U.S. Department of Agriculture’s (USDA) Agricultural Marketing Service (AMS). The safety of ground beef served in schools is particularly important because it is served to a vulnerable population (children) and it is easily contaminated with harmful pathogens such as Salmonella, which can cause foodborne illnesses that may have life-long health effects (Leirisalo-Repo et al. 1997).1 Since 2004, AMS has imposed strict, zero-tolerance standards for Salmonella content in the ground beef it purchases under contract for NSLP and enforces this standard with testing, rejecting shipments that fail to meet the standard. A growing body of research has focused on the use of food-safety standards as contractual requirements. Von Schlippenbach and Teichmann (2012) showed that retailers use private quality standards to improve their bargaining power and mitigate inefficiencies in upstream production. Hou, Grazia, and Malorgio (2015) argued that private standards are the prevailing mechanism that governs food safety when buyers have more market power than sellers. Other researchers have analyzed the effects of buyer standards for food safety in the context of sales to supermarkets in Latin America (Balsevich et al. 2003), fish exports from Kenya to the European Union (Henson, Brouder, and Mitullah 2000), mango exports from Peru (Lemeilleur 2013), and high-value fresh vegetables exported from Zimbabwe (Henson, Masakure, and Boselie 2005). See Henson and Humphrey (2009) for an overview of such research. Ollinger, Moore, and Chandran (2004) provided empirical evidence that meat plants with contracts to supply major restaurant chains made greater food-safety investments than other plants, and Ollinger and Moore (2008) found that meat plants supplying large-volume buyers had significantly better performance on Salmonella tests than other plants.2 Little prior research has explored the effects of food-safety standards on competitive behavior. In particular, no research has demonstrated that food-processing plants derive competitive advantage from their relative ability to meet food-safety standards. In addition, no research has demonstrated causal evidence that food-safety standards can improve food-safety outcomes.3 This article helps fill those voids. First, we provide empirical evidence that economic mechanisms and incentives drive plants to register as NSLP suppliers, and that plants seek to exploit their competitive advantages in food safety and relatively high productivity (volume of ground beef produced per employee) when bidding on contracts. Second, we demonstrate that food-safety standards can have a measurable impact on food-safety outcomes in a competitive market. The implications are particularly important for federal authorities overseeing the NSLP food purchasing programs and, more generally, policymakers considering food-safety regulations because they provide empirical evidence about the responses of food suppliers to food-safety standards and the effectiveness of a stringent standard for Salmonella content. Ground-beef sales to AMS for NSLP occur through a process of open bidding in which the lowest-cost bid is awarded a supply contract that requires adherence to a zero-tolerance standard for Salmonella.4 AMS purchases of ground beef for NSLP are a small share of the value of all ground beef sold in the United States—some $50 million annually. Yet these sales are an important component of certain ground-beef plants’ business: the mean NSLP supplier sold about 11% of its ground beef to NSLP, and the share ranged from 1% to 80%, USDA, 2009–2014; confidential FSIS administrative data. We use a unique dataset containing test results for the Salmonella sampling programs conducted by the USDA Food Safety and Inspection Service (FSIS) and third-party laboratories on behalf of AMS, and other public and administrative data to examine this contracting process and evaluate the performance of plants on tests for Salmonella in samples of ground beef from 2006 to 2012. The next sections provide background and discuss the data and empirical strategy. We then present results from several sets of empirical tests. In the first empirical analysis, we examine the relationship between plants’ past Salmonella test performance and the sequential decisions to (a) register as NSLP suppliers and (b) bid on contracts to supply NSLP in a given year. We then demonstrate that plants that bid on contracts to supply NSLP had improved performance on Salmonella tests relative to registered suppliers that did not ship ground beef to NSLP. Next, we show that the contamination rate of ground beef shipped to NSLP was substantially lower than the rate indicated by FSIS tests of ground beef from the same plants. Finally, we contrast the performance of ground beef suppliers to chicken suppliers, which do not face a stringent standard for Salmonella on sales to the NSLP, and discuss the implications of our findings. Background FSIS has been monitoring food safety in meat slaughter and processing plants since the passage of the Federal Meat Inspection Act in 1906. Current regulations mandate process controls for each plant, including Sanitation Standard Operating Procedures (SSOPs) and tasks required under Hazard Analysis and Critical Control Point (HACCP) plans. FSIS inspectors monitor compliance with all SSOPs and HACCP tasks and issue noncompliance reports. FSIS can temporarily halt a plant’s production if infractions are serious and corrective actions are not taken, but these instances are rare. Most noncompliance incidents are quickly resolved with no disruption of production.5 Since the implementation of the Pathogen Reduction/Hazard Analysis and Critical Control Point (PR/HACCP) Rule of 1996, FSIS has conducted random tests of cattle, hog, and poultry carcasses and ground meat and poultry for Salmonella at meat plants. FSIS samples plants on a weighted random basis under which poorer-performing plants and larger plants are sampled more frequently. The purpose of these tests is to verify that plants are maintaining food-safety process controls. FSIS requires that all plants achieve a certain performance standard on Salmonella tests—these are non-zero tolerances, set by FSIS at levels that most plants have historically been able to achieve. For ground beef, FSIS permits 5 out of 53 samples to test positive for Salmonella, a tolerance which has remained unchanged since the implementation of PR/HACCP. Plants that fail to meet this standard are retested, compelled to update food safety plans, and required to take corrective actions in order to continue operation and production. See Ollinger, Guthrie, and Bovay (2014) for further discussion. AMS, in contrast to FSIS, has no regulatory authority for food safety and functions much like a private buyer in its capacity as a purchaser of food for distribution to the NSLP, in that it sets standards to which only its own suppliers must comply. If a plant chooses to register with AMS and become eligible to supply the NSLP, that plant must demonstrate its ability to meet AMS specifications, pass an audit, adhere to FSIS food-safety standards, and meet other pre-approval criteria outlined by AMS in a document known as a Technical Requirements Schedule (TRS). AMS established zero-tolerance standards for Salmonella and E. coli O157:H7 in ground beef in 2004. Further, since 2008 AMS has required ground-beef plants to test for generic E. coli and other bacteria that indicate whether plants are maintaining good food-safety process controls, and also introduced some other requirements related to food safety in 2010. The AMS standards for ground beef purchases are outlined in USDA AMS (2012). See table 1 for a comparison of food-safety requirements for FSIS and AMS. Table 1. Key Differences between FSIS Regulations and AMS Standards Covering Food Safety Product Testing and Process Controls Process control  AMS tolerance  AMS testing frequency (pounds)  FSIS tolerance  FSIS testing frequency  Microbial testing          E coli O157:H7  0.0  2,000  0.0  Random: Less than once per year  Salmonella  0.0  10,000  0.113 (5 of 53 samples)  Random: Usually, less than once per year  Standard plate count  100,000/gram  10,000  No requirement  No requirement  Generic E. coli  500/gram  10,000  Done at establishment  Schedule in Ollinger and Mueller (2003)  Total coliforms  1,000/gram  10,000  No requirement  No requirement  Process controls  AMS standard    FSIS regulation    Removal of major lymph glands, thymus gland, and cartilage  Required    No requirement    Removal of white fibrous materials, e.g., elbow tendons  Required    No requirement    Removal of yellow elastin  Required    No requirement    Slaughter operation  AMS standard    FSIS regulation    Removal of spinal cord  Required    Required    Use of meat from non-ambulatory animals  Not allowed    Permitted with veterinarian consent    Processing interventions to control pathogens  At least two. One must be a critical control point    No requirement    Routine testing of E. coli types including E. coli O157:H7  Several E. coli types    Generic E. coli only    Process control  AMS tolerance  AMS testing frequency (pounds)  FSIS tolerance  FSIS testing frequency  Microbial testing          E coli O157:H7  0.0  2,000  0.0  Random: Less than once per year  Salmonella  0.0  10,000  0.113 (5 of 53 samples)  Random: Usually, less than once per year  Standard plate count  100,000/gram  10,000  No requirement  No requirement  Generic E. coli  500/gram  10,000  Done at establishment  Schedule in Ollinger and Mueller (2003)  Total coliforms  1,000/gram  10,000  No requirement  No requirement  Process controls  AMS standard    FSIS regulation    Removal of major lymph glands, thymus gland, and cartilage  Required    No requirement    Removal of white fibrous materials, e.g., elbow tendons  Required    No requirement    Removal of yellow elastin  Required    No requirement    Slaughter operation  AMS standard    FSIS regulation    Removal of spinal cord  Required    Required    Use of meat from non-ambulatory animals  Not allowed    Permitted with veterinarian consent    Processing interventions to control pathogens  At least two. One must be a critical control point    No requirement    Routine testing of E. coli types including E. coli O157:H7  Several E. coli types    Generic E. coli only    Notes: The AMS standards for E. coli O157:H7 and Salmonella were established in 2004. All requirements except these two standards were established in 2008. Sources: The AMS food-safety standards were outlined in USDA AMS (2012); the FSIS food-safety regulations were described in the Federal Register, and are available at: https://www.fsis.usda.gov/OPPDE/rdad/FRPubs/93-016F.pdf. Salmonella testing is an important feature of the AMS food-safety program for ground beef. AMS requires NSLP suppliers to have private laboratories submit Salmonella test results for each 10,000-pound lot they supply to NSLP. Once each lot is shown to meet the zero-tolerance standard, it may be shipped to warehouses owned by states or operated by contractors under contracts with states; products are then shipped to individual school districts upon demand. Ground-beef suppliers that fail to meet AMS standards for the NSLP cannot sell the meat to other USDA programs such as the Child and Adult Care Food Program and Summer Food Service Program. Moreover, these suppliers incur the sunk costs of having prepared ground beef for shipment to the NSLP and then having to destroy or reprocess the rejected product. Repeated failures may result in a plant losing the right to bid on AMS contracts. Data We created a unique data set of all plants that produced ground beef and whose products were tested for Salmonella by FSIS from 2006 to 2012, including public and administrative data from AMS and FSIS. The AMS data include information that identified which plants were registered with AMS to supply the NSLP and when these AMS-registered plants bid on contracts to supply the NSLP.6 AMS also shared its Salmonella test data with FSIS. The FSIS data include (a) the results of tests for Salmonella in ground beef conducted by FSIS for its Salmonella monitoring program, (b) types and numbers of animals slaughtered, (c) pounds of ground beef produced, (d) the date each plant began operations, (e) performance on sanitation and HACCP tasks, and (f) the types of further processing (i.e., processing ground beef into other products) done in the plant. The FSIS data also include information from Dun & Bradstreet on the number of plant-level employees and whether the plant was part of a firm that owned more than one plant. Data on Salmonella tests conducted by FSIS, as well as data on compliance with SSOP and HACCP tasks, are from confidential FSIS files. During the timeframe of our study, testing was done in batches or sample sets of 53, spread out over some period of days or weeks.7 FSIS did not sample each plant every year due to limited laboratory capacity, and collected partial sample sets for some plant-year observations. Plants producing less than 1,000 pounds per day were tested less frequently than larger plants and some were never tested. FSIS used a selection algorithm to determine which plants to test; this algorithm ensured that higher-risk plants were sampled at least once per year. Plants with Salmonella levels better than half the standard (about 80% of all ground beef plants) were considered low-risk unless other information available to FSIS, such as generic E. coli test results collected by plants, indicated otherwise. These low-risk plants were tested less than one time per year but were supposed to be tested at least once every two years. There are 1,291 plant-year observations of FSIS test results from 636 ground beef plants. These observations included 93 observations from 32 NSLP suppliers. The data on AMS Salmonella test results for ground beef shipped to NSLP covers 2007 to 2012. AMS does not conduct its own testing; rather, private laboratories test ground beef on behalf of NSLP suppliers for AMS review. AMS uses the Salmonella test data submitted by these private laboratories to evaluate whether a shipment of ground beef meets its standards, and shares the testing data with FSIS. Personnel from FSIS matched AMS test data to FSIS administrative data when plant identification information was available. The resulting matched AMS–FSIS data set included 95 observations for 25 unique plants covering the 2007 to 2012 period; in 31 instances, both AMS and FSIS Salmonella testing data were available for the same plant-year. Matches of FSIS data to AMS data could not be made in all cases because in some cases, AMS data did not have a FSIS identifier, and in other cases, FSIS did not conduct Salmonella testing for that plant in that year. Table 2 contrasts the performance on Salmonella tests of commercial-only plants (i.e., those plants that did not register to bid on NSLP contracts, which supplied processors, retailers, hospitals and other institutions, and other non-NSLP buyers) with inactive and active NSLP suppliers, and the performance of active NSLP suppliers on FSIS Salmonella tests with the performance of active NSLP suppliers on AMS Salmonella tests. Inactive and active NSLP suppliers were registered to bid on AMS contracts; they differed in that inactive NSLP suppliers did not bid on a contract to supply the NSLP during a given year, whereas active NSLP suppliers bid on at least one contract during that year. Inactive NSLP suppliers served the same types of buyers as commercial-only plants; active NSLP suppliers sold about 11% of their output to the NSLP. Table 2. Distribution of Plant Performance on Typical 53-sample Sets of Salmonella Tests Conducted in the FSIS and AMS Testing Programs, 2006–2012   FSIS tests   Third-party tests  Mean number of samples testing positive for Salmonella per 53-sample set  Commercial-only plants  Inactive NSLP suppliers  Active NSLP suppliers  Active NSLP suppliers (shipments to NSLP)    Share of plants  0=x  0.575  0.200  0.534  0.702  0<x≤1  0.0209  0.0286  0.0517  0.277  1<x≤2  0.174  0.229  0.155  0  2<x≤3  0.0968  0.171  0.172  0.0213  3<x≤4  0.0526  0.229  0.0517  0  4<x≤5  0.0301  0.0857  0.0172  0  5<x≤6  0.0192  0.0286  0.0172  0  6<x≤7  0.0109  0  0  0  7<x≤8  0.00334  0  0  0  8<x≤9  0.00668  0  0  0  9<x≤10  0.00334  0  0  0  10<x  0.00668  0.0286  0  0  Number of observations  1198  35  58  47    FSIS tests   Third-party tests  Mean number of samples testing positive for Salmonella per 53-sample set  Commercial-only plants  Inactive NSLP suppliers  Active NSLP suppliers  Active NSLP suppliers (shipments to NSLP)    Share of plants  0=x  0.575  0.200  0.534  0.702  0<x≤1  0.0209  0.0286  0.0517  0.277  1<x≤2  0.174  0.229  0.155  0  2<x≤3  0.0968  0.171  0.172  0.0213  3<x≤4  0.0526  0.229  0.0517  0  4<x≤5  0.0301  0.0857  0.0172  0  5<x≤6  0.0192  0.0286  0.0172  0  6<x≤7  0.0109  0  0  0  7<x≤8  0.00334  0  0  0  8<x≤9  0.00668  0  0  0  9<x≤10  0.00334  0  0  0  10<x  0.00668  0.0286  0  0  Number of observations  1198  35  58  47  Notes: The FSIS randomly samples meat for Salmonella in ground-beef plants, and over the study period, determined plants’ compliance status on the basis of the number of positive samples out of a set of 53 samples. For ground beef, the standard was that no more than 5 of 53 samples could test positive for Salmonella. The FSIS may have tested some plants more than once in a year or may have tested fewer than 53 samples. This table normalizes the share of a plant’s samples testing positive for Salmonella against a denominator of 53 samples. In the terminology used in this paper, commercial-only plants were not registered with the AMS as eligible NSLP suppliers. Inactive NSLP suppliers were registered but did not actively bid on NSLP contracts in a given year. Active NSLP suppliers actively bid on NSLP contracts in a given year. The AMS requires that suppliers submit the results of Salmonella tests conducted by third-party laboratories with shipments of ground beef for the NSLP. We report the distribution of plant-year test results for plants that shipped ground beef for the NSLP and were included in our analysis; the full set of regressors was available for the included plants. As seen in table 2, 57.5% of commercial-only plants had zero samples test positive for Salmonella, and the share of active NSLP suppliers with zero positive Salmonella samples was more than twice as high (53.4%) as that of inactive NSLP suppliers (20%). Table 2 also shows that the share of plants failing to meet the FSIS standard of 5 positive Salmonella samples out of 53 was more than twice as high for commercial-only (5.0%) and inactive NSLP suppliers (5.7%) as for active NSLP suppliers (1.7%). Finally, table 2 shows a striking difference between the Salmonella test results for ground beef shipped to NSLP and ground beef produced by active NSLP suppliers and tested by FSIS: 97.9% of active NSLP plants had performance equivalent to one or fewer samples out of a 53-sample set testing positive for Salmonella on shipments to NSLP, whereas only 58.6% of the active NSLP suppliers’ FSIS sample sets achieved the same level of performance. The patterns shown in table 2 lead to several questions about causality and incentives. For example, what motivates plants to register as NSLP suppliers and actively bid on NSLP contracts? What drives improved performance on Salmonella tests by active NSLP suppliers, relative to both inactive NSLP suppliers and commercial-only suppliers? Are plants that have better Salmonella performance more likely to bid on contracts to supply the NSLP, or does winning a contract to supply the NSLP force plants to improve their food-safety practices in order to improve their performance on Salmonella tests? In the sections that follow, we develop and test empirical models to explain producer behavior with respect to NSLP contracts, Salmonella standards, and Salmonella test performance. Empirical Models This section describes three empirical models that characterize the effects of food-safety standards imposed by AMS on the behavior of ground-beef suppliers to the NSLP. In the first model, we analyze the effects of past Salmonella test performance on ground-beef plants’ decisions to register as NSLP suppliers and actively supply the NSLP in a given year. The second model evaluates the performance of active NSLP suppliers on FSIS tests for Salmonella in ground beef, relative to inactive NSLP suppliers. The third model compares the performance of active NSLP suppliers on tests for Salmonella in ground beef shipped to the NSLP, conducted on behalf of AMS, to the performance of active NSLP suppliers on FSIS tests for Salmonella. Empirical results are presented in the next section. Empirical Model of Plant Decisions to Bid on AMS Contracts Figure 1 illustrates the sequential decisions plants must make when deciding whether to supply the NSLP. Plants must first decide whether to register as NSLP suppliers, which they are eligible to do if they comply with the AMS approval conditions in the TRS, and with the general FSIS food-safety regulations that apply to all plants. Registered NSLP suppliers must then choose whether to bid on contracts (i.e., become active NSLP suppliers). Because contracts are awarded to the lowest-cost bidder and all suppliers must meet a zero-tolerance standard for Salmonella and E. coli, not all plants are likely to find bidding on AMS contracts to be a profitable option. Some plants may have low costs of production but expect that meeting the zero-tolerance standard would prove too expensive; other plants may have perfect records with respect to Salmonella tests but find that their cost of production is too high to make supplying the NSLP profitable. Figure 1. View largeDownload slide Flow chart of plant registration and bidding decisions Note: In each stage, N includes only plant-year observations with lagged observations of the same plant, 2006–2012. Figure 1. View largeDownload slide Flow chart of plant registration and bidding decisions Note: In each stage, N includes only plant-year observations with lagged observations of the same plant, 2006–2012. We model the sequential AMS registration and bidding decision process using a sequential logit regression (Van Ophem and Schram 1997), as shown in equations (1) and (2) below. A sequential logit is most appropriate for modeling AMS contract bidding because bidding decisions are conditional on the decision to register as an AMS supplier. The population of plants available for the first transition in the sequential logit includes all FSIS-inspected ground beef plants.8 All of these plants sell ground beef to commercial and other institutional buyers and some are registered to sell to the NSLP; the dependent variable in the first transition distinguishes plants that register to supply the NSLP from commercial-only (i.e., non-NSLP-registered) plants. The population of plants for the second transition is restricted to the plants registered with AMS to supply the NSLP in a given year because only AMS-registered plants can bid on an AMS contract to supply the NSLP. We model the transition probability of becoming an AMS-registered NSLP supplier ( PAit^) as   PAit^=exp⁡αA+βASi,t-1+ϕAKi,t+ρAPostt1+exp⁡αA+βASi,t-1+ϕAKi,t+ρAPostt (1) and model the transition probability of actively bidding on AMS contracts in a given year ( PBit^), conditional on being a registered NSLP supplier, as   PBit^=( exp ⁡αA+βASi,t-1+γBi,t-1+ϕAKi,t+ρAPostt1+ exp ⁡αA+βASi,t-1+γBi,t-1+ϕAKi,t+ρAPostt|PAit=1) (2) where Si,t-1 is the performance of plant ion Salmonella tests in year t-1 expressed as the share of samples testing positive for Salmonella; B is a binary variable that indicates whether plants are active NSLP suppliers (i.e., bid on AMS contracts in a given year); K is a vector of plant characteristics; and Post is a binary variable with a value of 1 in 2008 and later years, included in one of the regression specifications. We include Post because AMS mandated additional food-safety requirements in 2008 (see table 1), which may have discouraged plants from bidding on contracts. We use lagged Salmonella test performance results, Si,t-1, as an explanatory variable because AMS requires that the ground beef it purchases for the NSLP meet a zero-tolerance standard. Past performance on Salmonella tests, which reflect the ability to meet the NSLP standard without incurring additional costs, may thus have a bearing on plants’ decisions to register as an NSLP supplier and to bid on NSLP contracts. For the second transition, we include the lagged binary variable Bi,t-1 to indicate whether plants bid on contracts to supply the NSLP in the previous year; plants with experience in bidding may have lower costs of preparing and managing bids and may have learned how to profitably meet NSLP requirements. Contractual requirements guide us in selecting the vector of plant characteristics ( K) and the post-2007 dummy as explanatory variables. Because contracts are awarded to the lowest-cost bidder, highly efficient plants may be more likely to register as NSLP suppliers; we use the log of ground beef output per employee as the measure of plant efficiency. Plant age may also be important because older plants may have more experienced workers and may also have managers with more experience in determining which contracts to compete for and how to submit successful bids. Larger plants (defined using the number of employees per plant) have greater plant capacity and may as a result be willing to compete for low-value contracts, such as those to supply the NSLP, to utilize this capacity. Two binary variables related to plant processing characteristics indicate whether plants (a) slaughter cattle and (b) further process ground beef into products such as pepperoni and jerky. Cattle slaughter plants enjoy substantial economies of scale (MacDonald and Ollinger 2005), suggesting that they continuously search for market outlets for their products in order to sell at their lowest cost. In contrast, plants that further process products have more market outlets for their products and may find other outlets more profitable than producing ground beef to fill NSLP contracts. Empirical Model of FSIS Salmonella Test Performance We have proposed that NSLP suppliers base their decisions to bid on AMS contracts, in part, on their performance on recent Salmonella tests. We now introduce a two-stage linear probability model (equations [3] and [4]) used to evaluate the relative performance of active and inactive NSLP suppliers on Salmonella tests, controlling for other factors. In this two-stage model, supplier type, Bi,t (whether a plant actively bids on contracts to supply NSLP), is endogenously determined by plants’ past performance on Salmonella tests and whether they bid on contracts to supply the NSLP in the previous year. The second stage of our linear probability model is given by equation (3) and the first stage by equation (4), as follows:   Sf,i,t*=α0+∑hδhKh,i,t+∑jρjRj,i,t+λPostt+ωBi,t+ξi,t (3)  Bi,t=α1+β1Bi,t-1+β2Si,t-1 (4) where food-safety test performance ( Sf,i,t*) is a binary variable indicating whether plant i met hypothetical tolerance level f in year t. In equation (3), food-safety test performance is a function of a vector of plant characteristics ( K), compliance with a vector of FSIS food-safety process control regulations ( R), and supplier type (B); Sf,i,t*=1if Si,t≤Tf and Sf,i,t*=0if Sf,i,t>Tf, where Si,t is the performance of plant i in year t on Salmonella tests and Tf is a hypothetical Salmonella tolerance that varies across regression specifications. We also include a binary variable ( Post) in equation (3) to indicate observations after 2007 when AMS increased the stringency of some food-safety requirements. In equation (4), as in equation (2), Bi,t-1 is a lagged binary variable indicating whether plants bid on AMS contracts in the previous year, and Si,t-1 is the lag of Salmonella test outcomes expressed as a continuous variable. These are used as instruments for the endogenous binary variable, Bi,t, which indicates whether plants bid on AMS contracts in the current year. Under the FSIS standard, five samples out of 53 are permitted to test positive for Salmonella. We test various tolerance levels, Tf, equal to the equivalents of 3, 2, and 1 samples testing positive for Salmonella for each 53-sample set, as well as a zero-tolerance standard, because (a) the AMS zero-tolerance requirement is much stricter than the FSIS standard of 5 positive samples, and, as table 2 shows, the vast majority of plants perform better than the FSIS standard, and (b) by varying tolerances, we can assess whether there is a particular threshold beyond which active NSLP suppliers consistently outperform inactive NSLP suppliers. There are several approaches to estimating econometric models with binary outcome variables and endogenous regressors. We use a two-stage linear probability model because it accommodates binary endogenous variables, and Wooldridge (2009) suggests that linear probability models perform well in predicting the values of independent variables that are near the averages of their samples.9Angrist and Pischke (2009) provide examples demonstrating the validity of linear probability models.10 The chief explanatory variable of interest in our model is the supplier type variable, Bi,t. The K variables from the bid decision model (equations [1] and [2]) are retained for this regression model, but are included for different reasons. If plants devote more effort to food-safety practices such as cleaning and sanitation, the quantity of ground beef produced per worker (productivity) may be lower. Ollinger and Moore (2008) found that larger plants had better food-safety performance in the chicken-slaughter industry but not in the ground-beef industry; including plant size (number of employees) as an explanatory variable allows us to revisit that conclusion. Muth, Fahimi, and Karns (2009) found that the vintage of plant capital is correlated with lower Salmonella levels in hog and chicken slaughter plants and further-processing plants, so we include plant age as a proxy for age of capital. We also include a binary variable indicating whether a plant uses meat from its own slaughter operations as meat inputs because vertically-integrated plants, which have greater control over inputs, may be able to attain better food-safety outcomes (Ollinger and Moore 2008). We also include three R variables, which reflect compliance with SSOP and HACCP tasks, as monitored and recorded by FSIS inspectors. Ollinger and Moore (2008) found that higher rates of compliance with SSOPs and HACCP tasks led to improved performance on Salmonella tests. We separately account for pre-operational SSOPs, which are cleaning and sanitation tasks performed at the beginning or end of the production day, and operational SSOPs, which are cleaning tasks performed during production. Empirical Model of Salmonella Test Performance for Products Sold to the NSLP Our third empirical model evaluates the performance of active NSLP ground beef suppliers on Salmonella tests of ground beef supplied to the NSLP, relative to ground beef randomly tested by FSIS as part of active NSLP suppliers’ general operations. We use a data set that combines FSIS Salmonella test results for active NSLP suppliers, by plant-year, with Salmonella test results, by plant-year, for ground beef tested for AMS. This combined data set enabled us to directly compare the performance of NSLP suppliers on tests for AMS of shipments of ground beef to the NSLP with the performance of NSLP suppliers on random FSIS tests. We ran a series of probit regressions (equation [5]) on the combined AMS–FSIS data set, where, again, the dependent variable, Sf,i,t*, is a binary variable indicating whether plant i met or performed better than each of three hypothetical tolerance thresholds ( Tf) on Salmonella tests conducted by FSIS or for AMS in year t:   Sf,i,t*=α0+∑hδhKh,i,t+∑jρjRj,i,t+λPostt+ωMi,t+ξi,t. (5) In equation (5), the binary variable M is defined as 1 if the test results were provided by third-party laboratories for shipments to AMS and 0 if the test results were from random testing conducted by FSIS. All other variables are defined as in equation (3). As before, we consider various tolerance levels in order to assess whether there is a particular threshold beyond which samples of shipments for NSLP consistently outperform the same plants’ FSIS test results. We do not include any endogenous variables in this empirical specification. Results We examine results for the three econometric models developed in the previous section. First, we evaluate the characteristics that may incentivize plants to bid on AMS contracts (equations [1] and [2]). Second, we examine the performance of Salmonella tests of active bidders, relative to inactive bidders (equations [3] and [4]). Finally, we compare the performance of active NSLP suppliers on Salmonella tests of ground beef shipped to the NSLP relative to their performance on random FSIS tests of ground beef that may be shipped either to commercial buyers or institutional buyers including AMS (equation [5]). Table 3 has definitions for all variables and the mean values of the observations used in the regressions. Table 3. Definitions and Mean Values of Selected Economic and Food-safety Variables that May Affect Contracting Choices, 2006–2012 Var.  Label  Definition  Active NSLP supplier  Inactive NSLP suppliera  Commercial- onlyb  Plant decisions to bid on AMS contracts        Si,t-1  Lag Share Positive Samples  Lag of share of Salmonella samples positive  0.015  0.041***  0.025  Bi,t-1  Lag Active NSLP Supplier  1 if plant bid on AMS contract during previous year; 0 otherwise  0.672  0.286***  0***  K1  Plant Employees  Employees per plant  342.4  257.5  136.7***  K2  Million Pounds per Employee  Million pounds of ground beef produced divided by number of plant employees  1.093  0.700  0.654***  K3  Age of Plant (years)  Current year minus year meat grant issued  15.59  19.00  29.58***  K4  Slaughters Cattle  1 if slaughters cattle; 0 otherwise  0.621  0.371***  0.124**  K5  Does Further Processing  1 if plant processes meat into products other than ground beef; 0 otherwise  0.0862  0.114  0.0876**  Postt  Year>2007  1 if year is after 2007; 0 otherwise  0.879  0.686**  0.591**    Observations  58  35  1,198  FSIS Salmonella test performance        S1*  3 positive Salmonella samples  1 if three or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.913  0.629***    S2*  2 positive Salmonella samples  1 if two or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.741  0.457***    S3*  1 positive Salmonella samples  1 if one or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.586  0.229***    S4*  0 positive Salmonella samples  1 if zero samples test positive for Salmonella; 0 otherwise  0.534  0.200***    R1  HACCP Compliance  Share of HACCP tasks in compliance with FSIS standards  0.998  0.995**    R2  Pre-operational SSOP Compliance  Share of pre-operational SSOP tasks in compliance with FSIS standards  0.964  0.972    R3  Operational SSOP Compliance  Share of operational SSOP tasks in compliance with FSIS standards  0.974  0.982*    R4  Multi-Plant  1 if plant is owned by a firm that owns other plants; 0 otherwise  0.138  0.0286*      Observationsc    58  35    Var.  Label  Definition  Active NSLP supplier  Inactive NSLP suppliera  Commercial- onlyb  Plant decisions to bid on AMS contracts        Si,t-1  Lag Share Positive Samples  Lag of share of Salmonella samples positive  0.015  0.041***  0.025  Bi,t-1  Lag Active NSLP Supplier  1 if plant bid on AMS contract during previous year; 0 otherwise  0.672  0.286***  0***  K1  Plant Employees  Employees per plant  342.4  257.5  136.7***  K2  Million Pounds per Employee  Million pounds of ground beef produced divided by number of plant employees  1.093  0.700  0.654***  K3  Age of Plant (years)  Current year minus year meat grant issued  15.59  19.00  29.58***  K4  Slaughters Cattle  1 if slaughters cattle; 0 otherwise  0.621  0.371***  0.124**  K5  Does Further Processing  1 if plant processes meat into products other than ground beef; 0 otherwise  0.0862  0.114  0.0876**  Postt  Year>2007  1 if year is after 2007; 0 otherwise  0.879  0.686**  0.591**    Observations  58  35  1,198  FSIS Salmonella test performance        S1*  3 positive Salmonella samples  1 if three or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.913  0.629***    S2*  2 positive Salmonella samples  1 if two or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.741  0.457***    S3*  1 positive Salmonella samples  1 if one or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.586  0.229***    S4*  0 positive Salmonella samples  1 if zero samples test positive for Salmonella; 0 otherwise  0.534  0.200***    R1  HACCP Compliance  Share of HACCP tasks in compliance with FSIS standards  0.998  0.995**    R2  Pre-operational SSOP Compliance  Share of pre-operational SSOP tasks in compliance with FSIS standards  0.964  0.972    R3  Operational SSOP Compliance  Share of operational SSOP tasks in compliance with FSIS standards  0.974  0.982*    R4  Multi-Plant  1 if plant is owned by a firm that owns other plants; 0 otherwise  0.138  0.0286*      Observationsc    58  35    Notes: Asterisks *, **, and *** denote 0.10, 0.05, and 0.01 levels of significance, for two-sided t-tests. The observations of commercial-only suppliers are used in the sequential logit regressions only (table 4), so we do not present means for the commercial-only suppliers for the covariates that are only used in tables 5 and 6. Superscript a indicates statistical significance of the difference in the mean values between inactive and active NSLP suppliers;.b indicates statistical significance of the difference in the mean values between approved (active and inactive) NSLP suppliers and commercial-only suppliers. Source: Authors’ calculations using FSIS data. The data are pooled time series and cross sectional data from FSIS and AMS. All FSIS data extend from 1998 to 2014; AMS registration data span from 2005 to 2012; Salmonella testing data for shipments to the AMS are available from 2007 to 2012. Our first two regressions include as explanatory and instrumental variables, respectively, whether a plant was an active AMS supplier during the previous year. Thus, we are able to examine the 2006–2012 period in our first and second empirical tests, and 2007–2012 in our third empirical test. We provide two or three specifications of each regression: the first specification uses only the plant-level characteristics of primary interest, while the second and third specifications include more plant-level characteristics as explanatory variables. We use plant-level clustered standard errors to correct for heteroskedasticity in all of our econometric models because plant technology and characteristics do not change much over time. For example, a plant that is large one year is likely to be large the next year. These conditions suggest that error terms may be correlated within plants; failure to control for these within-plant error correlations could lead to small standard errors and large t-statistics (Cameron and Miller 2015). Marginal Effects of Past Salmonella Test Performance on Sequential AMS Contracting Choices Our first empirical test is to evaluate the effects of past Salmonella test performance on plants’ sequential decisions to register as NSLP suppliers and bid on contracts to supply the NSLP. Data for the first-stage regression includes all ground-beef plants for which FSIS data were available over the 2006–2012 period (1,291 observations). The second-stage regression includes only those plants that have registered with AMS as an NSLP supplier in the current year (93 observations). Table 4 provides results for three specifications of the sequential logit regression. In the first stage of the regression (under all specifications), we find that plants’ past performance on Salmonella tests had no statistically significant effect on the decision to register as a NSLP supplier. This is not surprising because registrants are only required to meet the (relatively loose) FSIS food-safety standards. Results for stage 1 also show that larger plants (as measured by number of employees) and more efficient plants (number of pounds produced per employee), as well as those with vertical integration (slaughtering cattle) were more likely to register to supply NSLP; these results were in line with our hypotheses about plants with excess capacity and greater efficiency seeking to be awarded AMS contracts to supply the NSLP. All other proposed explanatory variables were statistically insignificant in the first-stage regressions. Table 4. Marginal Effects of Past Salmonella Tests on Ground-beef Plants’ Sequential Choices to Register and Bid on Contracts to Supply NSLP, 2006–2012   (1)   (2)   (3)     Stage 1:  Stage 2:  Stage 1:  Stage 2:  Stage 1:  Stage 2:    NSLP supplier  Active NSLP supplier  NSLP supplier  Active NSLP supplier  NSLP supplier  Active NSLP supplier  Lag Share Positive Samples  –0.0566  –6.31**  –0.00856  –4.48*  –0.00842  –4.12*  (0.0621)  (3.24)  (0.0409)  (2.38)  (0.0404)  (2.47)  Lag Active NSLP Supplier  –  0.290***  –  0.310**  –  0.284**  (0.0993)  (0.123)  (0.122)  Log (plant employees)  0.0169***  0.0250  0.0151***  –0.0980  0.0140***  –0.0845  (0.00433)  (0.0642)  (0.00412)  (0.0908)  (0.00417)  (0.102)  Log (million pounds per employee)  0.0169***  0.0214  0.0158***  –0.0554  0.0142***  –0.0441  (0.00400)  (0.0649)  (0.00403)  (0.0784)  (0.00410)  (0.0966)  Does Further Processing  –0.00660  –0.317***  –0.0147  –0.533***  –0.0142  –0.558***  (0.00822)  (0.125)  (0.00934)  (0.162)  (0.00927)  (0.166)  Slaughters Cattle  –  –  0.0264***  0.483**  0.0255***  0.465**  (0.0112)  (0.204)  (0.0107)  (0.206)  Post-2007  –  –  –  –  0.00855  0.270**  (0.00592)  (0.131)  Age of Plant  –  –  –  –  –0.000190  0.000180  (0.000240)  (0.00498)  Observations  1,291  93  1,291  93  1,291  93  Wald χ2  71.64***  84.06***  121.75***    (1)   (2)   (3)     Stage 1:  Stage 2:  Stage 1:  Stage 2:  Stage 1:  Stage 2:    NSLP supplier  Active NSLP supplier  NSLP supplier  Active NSLP supplier  NSLP supplier  Active NSLP supplier  Lag Share Positive Samples  –0.0566  –6.31**  –0.00856  –4.48*  –0.00842  –4.12*  (0.0621)  (3.24)  (0.0409)  (2.38)  (0.0404)  (2.47)  Lag Active NSLP Supplier  –  0.290***  –  0.310**  –  0.284**  (0.0993)  (0.123)  (0.122)  Log (plant employees)  0.0169***  0.0250  0.0151***  –0.0980  0.0140***  –0.0845  (0.00433)  (0.0642)  (0.00412)  (0.0908)  (0.00417)  (0.102)  Log (million pounds per employee)  0.0169***  0.0214  0.0158***  –0.0554  0.0142***  –0.0441  (0.00400)  (0.0649)  (0.00403)  (0.0784)  (0.00410)  (0.0966)  Does Further Processing  –0.00660  –0.317***  –0.0147  –0.533***  –0.0142  –0.558***  (0.00822)  (0.125)  (0.00934)  (0.162)  (0.00927)  (0.166)  Slaughters Cattle  –  –  0.0264***  0.483**  0.0255***  0.465**  (0.0112)  (0.204)  (0.0107)  (0.206)  Post-2007  –  –  –  –  0.00855  0.270**  (0.00592)  (0.131)  Age of Plant  –  –  –  –  –0.000190  0.000180  (0.000240)  (0.00498)  Observations  1,291  93  1,291  93  1,291  93  Wald χ2  71.64***  84.06***  121.75***  Notes: Stage 1 is the decision of whether to register as a NSLP supplier (the alternative is to be a commercial-only plant); Stage 2 is the decision whether to bid on AMS contracts in a given year. Plant-level clustered standard errors appear in parentheses. Asterisks *, **, and *** denote 0.10, 0.05, and 0.01 levels of significance, respectively. The second-stage regressions reveal the factors that affected plants’ decisions to bid on contracts in a given year, conditional on being registered as eligible NSLP suppliers. Plants with better performance on Salmonella tests in the previous year and plants that bid on contracts to supply the NSLP in the previous year were substantially more likely to bid on contracts to supply the NSLP. Table 4 reports marginal effects at the mean; thus, a 1 percentage-point increase in the share of samples testing positive for Salmonella in the previous year reduced the probability that plants (with characteristics at mean values) actively bid on AMS contracts by 4.1–6.3 percentage points.11 Plants that bid on contracts in the previous year were about 28–31 percentage points more likely to be active NSLP suppliers than other plants, ceteris paribus. Conditional on having registered to supply NSLP, plants that further processed ground beef into value-added products were between 32–56 percentage points less likely to bid on NSLP contracts. This result conforms with expectations: plants that further process ground beef have more outlets for beef, making bidding on contracts to supply the NSLP less important. In addition, plants that slaughtered cattle were 47–48 percentage points more likely to bid on NSLP contracts, conditional on having registered; plants were 27 percentage points more likely to bid after 2008, when new safety standards were required for all ground-beef plants. Neither plant productivity nor plant size affected the decision of registered NSLP suppliers to bid on contracts, suggesting that the costs of complying with the zero-tolerance standard for Salmonella outweighed their competitive advantages in productivity and size and need for greater access to markets. Performance of NSLP Suppliers on FSIS Tests for Salmonella The second empirical test examines the performance of active NSLP suppliers relative to inactive NSLP suppliers on Salmonella tests administered by FSIS. Results of two-stage linear probability regression models are presented in table 5. Again, the dependent variable in each regression is the probability that plants pass one of several hypothetical tolerance standards. The endogenous variable is a dummy variable identifying whether a plant is an active NSLP supplier. The Kleibergen-Paap rk LM statistic for underidentification, Hansen J statistic for overidentification, and Montiel-Pflueger F statistic (Montiel Olea and Plfueger 2013) for each variation of the regression indicate that the instruments are valid and strong.12 These regressions use 84 observations, 9 fewer than in the second stage of the first model, because FSIS Salmonella test results were not available for all plants in all years. As described above, the FSIS test results do not represent the Salmonella levels of ground beef shipped to NSLP; rather, they represent random samples collected by FSIS personnel during the plants’ regular operations. Table 5. Performance on FSIS Salmonella Tests of Active and Inactive Suppliers of Ground Beef to the National School Lunch Program, 2006–2012   3 positive Salmonella results per 53-sample set   2 positive Salmonella results per 53-sample set   1 positive Salmonella result per 53-sample set   0 positive Salmonella results per 53-sample set     (1)  (2)  (1)  (2)  (1)  (2)  (1)  (2)  Active NSLP supplier  0.316  0.337  0.279  0.306  0.471**  0.492**  0.473**  0.485**  (0.206)  (0.216)  (0.265)  (0.270)  (0.235)  (0.248)  (0.229)  (0.242)  Log (plant employees)  –0.176**  –0.164**  –0.117  –0.990  –0.152*  –0.137  –0.166**  –0.157*  (0.0761)  (0.0735)  (0.0784)  (0.798)  (0.804)  (0.0861)  (0.0794)  (0.0828)  Log (pounds per employee)  –0.104  –0.0875  –0.106  –0.0817  –0.131*  –0.109  –0.141*  –0.129  (0.0636)  (0.0700)  (0.0665)  (0.0789)  (0.764)  (0.0870)  (0.0732)  (0.0821)  HACCP Compliance  0.823  0.501  0.156  –0.278  –5.07  –5.42  –6.70  –6.89  (6.15)  (6.09)  (6.82)  (6.64)  (6.49)  (6.45)  (6.34)  (6.45)  Pre-operational SSOP Compliance  –1.93*  –1.71*  0.449  0.777  –0.0661  0.232  1.06  1.23  (0.998)  (0.998)  (1.59)  (1.57)  (1.76)  (1.71)  (1.49)  (1.49)  Operational SSOP Compliance  1.33*  1.49*  –0.661  –0.412  0.712  0.934  0.177  0.307  (0.704)  (0.886)  (1.21)  (1.43)  (1.43)  (1.50)  (1.34)  (1.42)  Slaughters Cattle  0.219**  0.220**  0.0986  0.101  0.229*  0.231*  0.253**  0.255**  (0.106)  (0.105)  (0.116)  (0.114)  (0.129)  (0.130)  (0.124)  (0.125)  Post-2007  0.0615  0.0546  0.209  0.200  0.0227  0.0150  0.0801  0.0757  (0.127)  (0.131)  (0.130)  (0.134)  (0.120)  (0.123)  (0.0922)  (0.0938)  Multi-plant  0.0464  0.0295  0.0729  0.0486  –0.118  –0.139  –0.0823  –0.0945  (0.114)  (0.0999)  (0.162)  (0.150)  (0.234)  (0.230)  (0.230)  (0.230)  Age of Plant  –  0.00178  –  0.00268  –  0.00243  –  0.00142  (0.00315)  (0.00339)  (0.00355)  (0.00348)  Constant  0.927  0.795  0.725  0.482  5.03  4.76  6.05  5.89  (5.61)  (5.28)  (6.24)  (5.91)  (6.62)  (6.19)  (6.46)  (6.16)  F statistic  4.54***  4.62***  2.08*  2.07*  2.26*  2.02*  4.31***  4.86***  Underidentification: Kleibergen-Paap rk LM statistic ( χ2(2) p-value)  7.54  6.08  7.54  6.08  7.54  6.08  7.54  6.08  (0.0230)  (0.0479)  (0.0230)  (0.0479)  (0.0230)  (0.0479)  (0.0230)  (0.0479)  Weak identification: Montiel-Pflueger effective F statistic  9.03  7.08  9.03  7.08  9.03  7.08  9.03  7.08  Overidentification: Hansen J statistic ( χ2(1) p-value)  1.32  1.27  0.221  0.189  0.192  0.213  0.105  0.113  (0.251)  (0.259)  (0.638)  (0.664)  (0.661)  (0.645)  (0.746)  (0.736)  Observations  84  84  84  84  84  84  84  84    3 positive Salmonella results per 53-sample set   2 positive Salmonella results per 53-sample set   1 positive Salmonella result per 53-sample set   0 positive Salmonella results per 53-sample set     (1)  (2)  (1)  (2)  (1)  (2)  (1)  (2)  Active NSLP supplier  0.316  0.337  0.279  0.306  0.471**  0.492**  0.473**  0.485**  (0.206)  (0.216)  (0.265)  (0.270)  (0.235)  (0.248)  (0.229)  (0.242)  Log (plant employees)  –0.176**  –0.164**  –0.117  –0.990  –0.152*  –0.137  –0.166**  –0.157*  (0.0761)  (0.0735)  (0.0784)  (0.798)  (0.804)  (0.0861)  (0.0794)  (0.0828)  Log (pounds per employee)  –0.104  –0.0875  –0.106  –0.0817  –0.131*  –0.109  –0.141*  –0.129  (0.0636)  (0.0700)  (0.0665)  (0.0789)  (0.764)  (0.0870)  (0.0732)  (0.0821)  HACCP Compliance  0.823  0.501  0.156  –0.278  –5.07  –5.42  –6.70  –6.89  (6.15)  (6.09)  (6.82)  (6.64)  (6.49)  (6.45)  (6.34)  (6.45)  Pre-operational SSOP Compliance  –1.93*  –1.71*  0.449  0.777  –0.0661  0.232  1.06  1.23  (0.998)  (0.998)  (1.59)  (1.57)  (1.76)  (1.71)  (1.49)  (1.49)  Operational SSOP Compliance  1.33*  1.49*  –0.661  –0.412  0.712  0.934  0.177  0.307  (0.704)  (0.886)  (1.21)  (1.43)  (1.43)  (1.50)  (1.34)  (1.42)  Slaughters Cattle  0.219**  0.220**  0.0986  0.101  0.229*  0.231*  0.253**  0.255**  (0.106)  (0.105)  (0.116)  (0.114)  (0.129)  (0.130)  (0.124)  (0.125)  Post-2007  0.0615  0.0546  0.209  0.200  0.0227  0.0150  0.0801  0.0757  (0.127)  (0.131)  (0.130)  (0.134)  (0.120)  (0.123)  (0.0922)  (0.0938)  Multi-plant  0.0464  0.0295  0.0729  0.0486  –0.118  –0.139  –0.0823  –0.0945  (0.114)  (0.0999)  (0.162)  (0.150)  (0.234)  (0.230)  (0.230)  (0.230)  Age of Plant  –  0.00178  –  0.00268  –  0.00243  –  0.00142  (0.00315)  (0.00339)  (0.00355)  (0.00348)  Constant  0.927  0.795  0.725  0.482  5.03  4.76  6.05  5.89  (5.61)  (5.28)  (6.24)  (5.91)  (6.62)  (6.19)  (6.46)  (6.16)  F statistic  4.54***  4.62***  2.08*  2.07*  2.26*  2.02*  4.31***  4.86***  Underidentification: Kleibergen-Paap rk LM statistic ( χ2(2) p-value)  7.54  6.08  7.54  6.08  7.54  6.08  7.54  6.08  (0.0230)  (0.0479)  (0.0230)  (0.0479)  (0.0230)  (0.0479)  (0.0230)  (0.0479)  Weak identification: Montiel-Pflueger effective F statistic  9.03  7.08  9.03  7.08  9.03  7.08  9.03  7.08  Overidentification: Hansen J statistic ( χ2(1) p-value)  1.32  1.27  0.221  0.189  0.192  0.213  0.105  0.113  (0.251)  (0.259)  (0.638)  (0.664)  (0.661)  (0.645)  (0.746)  (0.736)  Observations  84  84  84  84  84  84  84  84  Notes: Plant-level clustered standard errors appear in parentheses. Asterisks *, **, and *** denote 0.10, 0.05, and 0.01 levels of significance, respectively. The F statistic is the test for joint significance of all coefficients. In our terminology, active NSLP suppliers are approved to sell product to the NSLP and bid for contracts. Inactive NSLP suppliers also are approved to sell product to the NSLP but do not bid on contracts in a given year. The instrumental variables used in this regression for the Active NSLP Supplier dummy are lag active NSLP supplier and lag share positive samples. The empirical results in table 5 demonstrate robust evidence that active NSLP suppliers perform better than inactive NSLP suppliers on Salmonella tests administered by FSIS. In particular, as the hypothetical tolerance standard that we use as our dependent variable grows more stringent, active NSLP suppliers were more likely to achieve the standard, relative to inactive suppliers. The estimates suggest that, holding other factors constant, the probability that a plant met the most stringent hypothetical tolerance levels was 47–49 percentage points higher if the plant actively bid on AMS contracts to supply NSLP. However, the probability of meeting less-stringent tolerance levels was not statistically significantly better for active AMS suppliers. These results are consistent with the mean comparisons shown in table 2 under both econometric specifications. Some of the other explanatory variables were statistically significant in one or more regression specifications in table 5. Plants that slaughtered cattle were more likely to meet or perform better than the 3/53 threshold and the 1/53 threshold, and meet the zero-tolerance standard under at least one specification. These results are generally consistent with expectations.13 More efficient plants, as measured by the number of pounds produced per employee, and plant size were also less likely to meet the most stringent standards—a finding consistent with the sequential logit second-stage results shown in table 4. Effects of compliance with HACCP and SSOP tasks had mixed effects on Salmonella tests, but these effects were small and mostly statistically insignificant.14 Effect of AMS Standards on Salmonella Test Performance of Active NSLP Suppliers Finally, we examined the performance of active NSLP suppliers on Salmonella tests of ground beef shipped to NSLP and tested for AMS, relative to the performance of active NSLP suppliers on random tests for Salmonella conducted by FSIS. The data include the 95 observations of active NSLP suppliers for which we have complete data and cover 2007–2012.15 Results for two specifications of probit regressions for each of three performance thresholds—0, 1, and 2 positive Salmonella samples out of 53—are presented in table 6. Table 6. Performance on Salmonella tests of Active Suppliers of Ground Beef to the National School Lunch Program (Marginal Effects), 2007–2012   2 positive Salmonella results per 53-sample set   1 positive Salmonella result per 53-sample set   0 positive Salmonella results     (1)  (2)  (1)  (2)  (1)  (2)  Product shipped to NSLP  0.276***  0.273***  0.450***  0.451***  0.216**  0.217**  (0.0709)  (0.0677)  (0.0761)  (0.0752)  (0.0980)  (0.0973)  Log (plant employees)  –0.0439  –0.0417  –0.0639  –0.0572  –0.105*  –0.112  (0.0478)  (0.0563)  (0.0577)  (0.0704)  (0.0614)  (0.0708)  Log (pounds per employee)  –0.0407  –0.0359  –0.0689  –0.0597  –0.103**  –0.113*  (0.0390)  (0.0554)  (0.0478)  (0.0704)  (0.0509)  (0.0657)  HACCP Compliance  1.33  1.19  –0.0123  0.352  5.52  5.48  (1.71)  (1.71)  (4.09)  (3.47)  (3.82)  (3.83)  Pre-operational SSOP Compliance  1.23  1.27  1.44  1.60  –0.630  –0.752  (1.24)  (1.15)  (1.35)  (1.21)  (1.82)  (1.85)  Operational SSOP Compliance  –1.88  –1.82  –2.26  –2.24  –0.975  –1.05  (1.23)  (1.29)  (1.45)  (1.60)  (2.14)  (2.21)  Post-2007  0.0969  0.100  0.00416  –0.0107  –0.0446  –0.0354  (0.187)  (0.199)  (0.164)  (0.150)  (0.248)  (0.252)  Multi-plant    0.0371    –0.0558    0.0181  (0.0713)  (0.128)  (0.137)  Age of Plant    0.000373    0.00123    –0.000981  (0.00339)  (0.00431)  (0.00398)  Pseudo R2  0.255  0.258  0.340  0.343  0.0751  0.0756  Waldχ2 p-value  0.0001  0.0000  0.0000  0.0000  0.1271  0.2326  Observations  95  95  95  95  95  95    2 positive Salmonella results per 53-sample set   1 positive Salmonella result per 53-sample set   0 positive Salmonella results     (1)  (2)  (1)  (2)  (1)  (2)  Product shipped to NSLP  0.276***  0.273***  0.450***  0.451***  0.216**  0.217**  (0.0709)  (0.0677)  (0.0761)  (0.0752)  (0.0980)  (0.0973)  Log (plant employees)  –0.0439  –0.0417  –0.0639  –0.0572  –0.105*  –0.112  (0.0478)  (0.0563)  (0.0577)  (0.0704)  (0.0614)  (0.0708)  Log (pounds per employee)  –0.0407  –0.0359  –0.0689  –0.0597  –0.103**  –0.113*  (0.0390)  (0.0554)  (0.0478)  (0.0704)  (0.0509)  (0.0657)  HACCP Compliance  1.33  1.19  –0.0123  0.352  5.52  5.48  (1.71)  (1.71)  (4.09)  (3.47)  (3.82)  (3.83)  Pre-operational SSOP Compliance  1.23  1.27  1.44  1.60  –0.630  –0.752  (1.24)  (1.15)  (1.35)  (1.21)  (1.82)  (1.85)  Operational SSOP Compliance  –1.88  –1.82  –2.26  –2.24  –0.975  –1.05  (1.23)  (1.29)  (1.45)  (1.60)  (2.14)  (2.21)  Post-2007  0.0969  0.100  0.00416  –0.0107  –0.0446  –0.0354  (0.187)  (0.199)  (0.164)  (0.150)  (0.248)  (0.252)  Multi-plant    0.0371    –0.0558    0.0181  (0.0713)  (0.128)  (0.137)  Age of Plant    0.000373    0.00123    –0.000981  (0.00339)  (0.00431)  (0.00398)  Pseudo R2  0.255  0.258  0.340  0.343  0.0751  0.0756  Waldχ2 p-value  0.0001  0.0000  0.0000  0.0000  0.1271  0.2326  Observations  95  95  95  95  95  95  Note: Plant-level clustered standard errors appear in parentheses. Asterisks *, **, and *** denote 0.10, 0.05, and 0.01 levels of significance, respectively. Our empirical results demonstrate strong evidence that the Salmonella test results for ground beef shipped to NSLP were better than the FSIS Salmonella test results for ground beef. According to the estimates given in table 6, ground beef tested for AMS and shipped to NSLP was 21–22 percentage points more likely to meet the zero-tolerance standard for Salmonella and 45 percentage points more likely to meet or perform better than the 1-positive sample Salmonella threshold, relative to FSIS tests of random samples of ground beef. In addition, larger and more efficient plants were less likely to meet the zero-tolerance standard—a finding consistent with the results shown in tables 4 and 5. Coefficients on other explanatory variables are statistically insignificant. Implications for Public Policy There are several important implications for public policy. First, our analysis shows that the zero-tolerance standard for Salmonella in ground beef sold for distribution to schools has resulted in improved food-safety performance of ground-beef suppliers. This is immensely important because school children are relatively vulnerable to food-borne illness caused by Salmonella and other pathogens (Centers for Disease Control 2017b). More broadly, the analysis demonstrates that strict standards incentivized active AMS suppliers, which already had strong performance on Salmonella tests, to improve their performance on these tests. This shows that policy objectives such as improved food-safety outcomes, can be met while at the same time using a lowest bid-price selection criteria for choosing suppliers. Our empirical findings suggest that imposing stringent standards is more effective in ensuring improved food-safety outcomes than relying on reputation or other private economic incentives without enforceable standards. Ollinger et al. (2015) found that the Salmonella test performance of active suppliers of raw chicken to the NSLP was only slightly better than the Salmonella test performance of other groups of suppliers, and only at certain tolerance levels. Ollinger et al. (2015) attributed the modestly better performance of active AMS chicken suppliers on Salmonella tests to suppliers’ concerns about reputation, because active NSLP suppliers sell chicken to a highly visible market with few suppliers, and tracing any food-borne illness outbreak to its source is relatively easy in this context. The results presented in this article, in contrast, indicate that the stringent, enforceable, zero-tolerance standard for Salmonella in ground beef imposed by AMS has been highly effective at driving plants to improve their food-safety performance, at all tolerance levels. Taken together, it appears that stringent standards were more effective than private incentives (including concerns about reputation and the cost to producers associated with food-borne illness outbreaks) in driving markedly better performance on Salmonella tests.16 The empirical results also provide some evidence that performing HACCP and SSOP tasks may have limited effects on Salmonella test outcomes, a finding consistent with Ollinger et al. (2015). There are many possible explanations; HACCP is a preventative plan, and SSOP tasks are related to daily plant operations, but neither has a direct bearing on food-safety outcomes or mandates the testing of final products. Performing well on HACCP process control and sanitation tasks may enable a plant to avoid major food-safety failures but does not allow a plant to reduce Salmonella that is already present in meat inputs, since there is no pathogen “kill step” for uncooked ground beef. Regardless of the explanation, it appears that performing HACCP process control and sanitation tasks may not be sufficient for plants to meet very strict food-safety standards. Yet, mandatory sanitation and process controls are central features of the food-safety regulation program administered by FSIS. Conclusion The Agricultural Marketing Service of the USDA annually purchases about $50 million worth of ground beef to be distributed to school systems throughout the United States as part of the National School Lunch Program. Suppliers to the NSLP must be registered with AMS and are selected via a competitive bidding process, with the contract winner chosen on the basis of cost alone, which incentivizes low-cost producers to supply the NSLP. Furthermore, ground beef shipped to the NSLP must meet a zero-tolerance standard for E. coli O157:H7 and Salmonella. This article demonstrates that, under these conditions, from 2006 to 2012, relatively large and highly productive plants (as measured by the pounds of ground beef produced per employee) registered as NSLP suppliers. Furthermore, those plants with a history of supplying the NSLP with ground beef, and with stronger performance on previous Salmonella tests, were more likely to actively bid on NSLP contracts. In two additional empirical tests, we show that (a) active NSLP suppliers performed better on random FSIS tests for Salmonella than inactive NSLP suppliers, and (b) these better-performing active NSLP suppliers shipped ground beef to the NSLP that performed even better on Salmonella tests than the product produced by the same suppliers and randomly tested by FSIS and shipped to the NSLP or other institutional or commercial buyers. Active NSLP suppliers may have attained better Salmonella test performance on product destined for the NSLP in one of two ways. First, they may have taken more precautions, such as additional sanitation or process controls, for product destined to the NSLP, and shipped product only if they were reasonably confident that it would meet the zero-tolerance Salmonella standard. Alternatively, the suppliers could have tested product prior to shipment and shipped product that met AMS standards to the NSLP while shipping product that did not meet AMS standards to a different buyer. Regardless of how plants satisfied the AMS Salmonella standard, the empirical findings demonstrate that the stringent standard imposed by AMS on its suppliers, namely, the zero-tolerance standard for Salmonella, is an effective mechanism for improving the safety of ground beef served in schools. This is especially important because AMS awards contracts to the lowest-cost bidder, which could incentivize suppliers to reduce costs by reducing effort devoted to food safety. This article contributes four main findings to the literature on food safety and food-safety standards. First, our empirical results demonstrate that enforceable standards for food safety, particularly pathogen testing with outcome monitoring, can incentivize producers to improve food safety. Second, the results suggest that plants use performance on food-safety tests to their competitive advantage when food safety can be measured, and when meeting stringent food-safety standards allows them to fill certain contracts. Third, our empirical findings show that, from 2007 to 2012, AMS attracted ground-beef suppliers that shipped ground beef to AMS with lower Salmonella levels than those same producers shipped to the commercial market. These findings are consistent with AMS requirements to accept the lowest-cost bid and receive only ground beef meeting a zero-tolerance standard for Salmonella—two requirements that generate opposing incentives to invest in food safety. Fourth, we demonstrate that compliance with SSOP and HACCP tasks does not necessarily improve food-safety outcomes, which highlights the importance of product or outcome standards relative to process standards. It is important to remember that this article examined the performance of ground-beef suppliers to the NSLP on Salmonella tests and did not examine whether better performance on Salmonella tests led to reductions in foodborne illness. AMS sets more stringent standards for the ground beef it buys for the NSLP than those required by FSIS for general commerce because children are more vulnerable to foodborne illnesses than healthy adults (Young 2005). Data from the Centers for Disease Control and Prevention (CDC) indicate that there were no foodborne illness outbreaks due to Salmonella or E. coli O157:H7 in ground beef served in schools and colleges from 2005 to 2014, and recall data from FSIS indicate that no ground-beef products shipped to schools were recalled due to Salmonella or E. coli O157:H7 over from 2004 to 2013.17 By contrast, there were 21 outbreaks of Salmonella and 58 outbreaks of E. coli O157:H7 in ground beef sold commercially during this time span.18 Finally, the results presented in this article highlight the roles of two government agencies that set standards for food safety, and the rationale for these agencies to set standards differently. AMS is authorized to set standards for food safety for the meat it buys for distribution to the NSLP and a few other food-assistance programs. In this regard, it is similar to other buyers, which consider the costs of requiring food-safety standards and the benefits to the clients and customers they serve and the buyer themselves. For AMS, which serves school children and other vulnerable groups, reducing the risk of food-borne illness is a top priority and costs may be less important. In contrast, FSIS sets requirements that apply to all meat and poultry processors engaged in interstate commerce and must consider the costs and benefits of its regulations as they apply to all producers and consumer groups, including processors that may pre-cook ground beef, healthy adults, and various groups such as the elderly and young children who may be more vulnerable to food-borne illness. Strict standards may not be necessary as a national, universal requirement; imposing them would likely require suppliers to invest more in equipment and spend more on practices to promote food safety, thereby driving up prices.19 What emerges is a market in which FSIS sets minimum standards and within which AMS and other buyers set standards for their suppliers as they deem necessary for best serving their customers and clients. Footnotes 1 Salmonella is a genus of bacteria that can cause illness in humans if consumed. Although cooking kills Salmonella, the presence of Salmonella increases the risk of food-borne illness, particularly because of cross-contamination. The FSIS reports test results for non-typhoidal Salmonella, indicating the presence of one of several species of Salmonella. For the remainder of this article, we use Salmonella as shorthand for non-typhoidal Salmonella. The incidence rate of infections caused by Salmonella was higher in 2015 for children aged 5 to 9 than for any older group, according to data from CDC (2017b). This was also true for infections caused by Shigella and Cryptosporidium, and Shiga toxin-producing E. coli O157. The incidence rate of infections caused by both Shiga toxin-producing E. coli O157 and non-O157 was particularly elevated for all age groups under 20. 2 There is also a considerable body of literature on the empirical estimation of the costs of providing safe food (see, e.g., Antle 2000 and Ollinger and Moore 2009). Researchers have also estimated the total costs and benefits of food safety regulations (Crutchfield, et al. 1998) and the effects of product recalls on stock market values (Thomsen and McKenzie 2001; Pozo and Schroeder 2016), prices of branded products (Thomsen, Shiptsova, and Hamm 2006), and demand (Marsh, Schroeder, and Mintert 2004; Piggott and Marsh 2004; Bakhtavoryan, Capps, and Salin 2014). Several studies, including Muth, Wohlgenant, and Karns (2007) and Pouliot (2014), have analyzed the regulation of food safety in an industrial-organization context. 3 Sumner, Raven, and Givney (2004) showed that new Australian regulations of food safety in meat and poultry, implemented in the 1990s, were contemporaneous with a nationwide improvement in the results of tests of animal carcasses in meat plants for bacteria, although the article did not present causal evidence for this relationship. Minor and Parrett (2016, 2017) showed that certain U.S. food-safety regulations led to a decrease in the number of detected food-borne illnesses associated with the regulated products, but did not control for the efficacy of detection of illness. 4 The AMS specifies basic technical requirements for each NSLP contract and does not consider additional product-quality attributes (such as country of origin or animal-welfare certifications) when awarding contracts. 5 See Ollinger and Mueller (2003) for further discussion. 6 The data came from the AMS website, which identifies all registered plants for the current year and gives information about recent contract awards but does not give historical information. We started collecting data in 2009 at a time when some historical data was available, and collected more data as it became available. The recorded data included whether a plant bid on at least one contract during the year, the pounds of ground beef supplied to the NSLP, and the year(s) of award(s) of any contract(s). Separate information on the website identified all registered suppliers. 7 The FSIS has since replaced the sample-set framework with continuous testing in which plants may undergo periodic testing throughout the year, and the FSIS keeps running tallies to track compliance. 8 State-inspected ground-beef plants are not directly inspected by the FSIS but must meet FSIS standards and, therefore, are eligible to bid on NSLP contracts. However, the FSIS test data were available only for FSIS-inspected plants. 9 Other options are discussed by Lewbel, Dong, and Yang (2012). The special regressor model (Lewbel 2000; Dong and Lewbel 2015; Bontemps and Nauges 2015) is also appropriate for estimating binary choice regressions with a binary endogenous variable. This model requires a special regressor that is exogenous, continuously distributed, and conditionally independent of the error term (Baum et al. 2012). These conditions precluded use in estimating equation (3) because our dataset did not have a suitable special regressor. An instrumental variable probit regression is not appropriate because it is inconsistent for binary endogenous variables (Baum et al. 2012). 10 Two well-recognized drawbacks of the two-stage linear probability pointed out by Wooldridge (2009) and others are that model estimates can fall outside the unit interval of the dependent variable and the error term is heteroskedastic because the dependent variable is not continuous. 11 The mean share of samples testing positive on FSIS Salmonella tests, across all plant-year observations, was 2.0%. 12 In all cases, the Montiel-Pflueger F statistic rejects weak instruments with τ<30% and α=0.05. 13 Ollinger and Moore (2008) found that Salmonella levels decreased with plant size in cattle-, hog-, and chicken-slaughter plants but not ground-beef plants. 14 The largest statistically significant coefficient (–6.89) indicates that plants were 0.689 percentage points less likely to meet the zero-tolerance standard if their rate of compliance with operational SSOP tasks rose by 10 percentage points. The average rate of compliance with operational SSOP tasks for all plants was 98.5%, varying from 97.4% for inactive NSLP suppliers to 98.6% for commercial-only plants. Although HACCP and SSOP tasks are designed to reduce the risk of pathogen contamination, compliance with these checklist-like tasks does not ensure improved food-safety outcomes. 15 Recall that we do not have data on Salmonella testing for shipments to NSLP before 2007. 16 Reputation effects in food safety depend on the capacity of public health authorities to detect harmful pathogens in food products. Only rarely does a case of foodborne illness become attributed to its source by public health authorities, meaning that food-safety problems only rarely lead to economic costs for producers as long as contracts do not mandate pathogen standards, and harm to reputation is the only consequence of a food-safety problem. As detection technology improves and if more resources are used for the attribution of foodborne illnesses, then attribution of illnesses to producers would likely grow and reputation effects may become stronger. 17 The FSIS reports recalls of meat and poultry products at https://www.fsis.usda.gov/wps/portal/fsis/topics/recalls-and-public-health-alerts/recall-case-archive, which indicates the terminal location of recalled products. There were no recalls of ground beef destined to schools for E. coli O157:H7 or Salmonella from 2004 to 2013. The FSIS product recalls cover a period of time inclusive of the first and last dates for which pathogen contamination is detected by the FSIS. All meat shipped during recall periods must be recalled, and thus, the recalled meat may include meat that would not test positive for pathogen contamination. 18 Data came from the authors’ use of the Centers for Disease Control and Prevention (2017a) website foodborne illness tool. 19 Recall that our empirical results indicate that inactive NSLP suppliers did not bid on contracts to supply the NSLP, in part because they had weaker performance on Salmonella tests than active NSLP suppliers, suggesting that it would cost them more to meet NSLP standards than active NSLP suppliers. References Angrist J.D., Pischke J.-S.. 2009. Mostly Harmless Econometrics: An Empiricist’s Companion . Princeton, New Jersey: Princeton University Press. Antle J.M. 2000. No Such Thing as a Free Safe Lunch: The Cost of Food Safety Regulation in the Meat Industry. American Journal of Agricultural Economics  82 2: 310– 22. http://dx.doi.org/10.1111/0002-9092.00027 Google Scholar CrossRef Search ADS   Bakhtavoryan R., Capps O.Jr, Salin V.. 2014. The Impact of Food Safety Incidents Across Brands: The Case of the Peter Pan Peanut Butter Recall. Journal of Agricultural and Applied Economics  46 4: 559– 73. Balsevich F., Berdegue J.A., Flores L., Mainville D., Reardon T.. 2003. Supermarkets and Produce Quality and Safety Standards in Latin America. American Journal of Agricultural Economics  85 5: 1147– 54. Google Scholar CrossRef Search ADS   Baum C.F., Dong Y., Lewbel A., Yang T.. 2012. Binary Choice Models with Endogenous Regressors. Stata Conference in San Diego, California. Available at: http://www.stata.com/meeting/sandiego12/materials/sd12_baum.pdf. Bontemps C., Nauges C.. 2015. The Impact of Perceptions in Averting-decision Models: An Application of the Special Regressor Method to Drinking Water Choices. American Journal of Agricultural Economics  97 3: 1– 17. Cameron A.C., Miller D.L.. 2015. A Practitioner’s Guide to Cluster-Robust Inference. The Journal of Human Resources  50 2: 317– 72. Google Scholar CrossRef Search ADS   Centers for Disease Control and Prevention. 2017a. Available at: https://wwwn.cdc.gov/foodborneoutbreaks/. Centers for Disease Control and Prevention. 2017b. FoodNet 2015 Surveillance Report (Final Data). Available at: https://www.cdc.gov/foodnet/pdfs/FoodNet-Annual-Report-2015-508c.pdf. Crutchfield S.R., Buzby J.C., Roberts T., Ollinger M., Lin C.-T.J.. 1997. An Economic Assessment of Food Safety Regulations: The New Approach to Meat and Poultry Inspection. Washington DC: U.S. Department of Agriculture, Economic Research Service, Agricultural Economic Report No. 755. Dong Y., Lewbel A.. 2015. Simple Estimators for Binary Choice Models with Endogenous Regressors. Econometric Reviews  34 ( 1–2): 82– 105. http://dx.doi.org/10.1080/07474938.2014.944470 Google Scholar CrossRef Search ADS   Henson S., Brouder A.-M., Mitullah W.. 2000. Food Safety Requirements and Food Exports from Developing Countries: The Case of Fish Exports from Kenya to the European Union. American Journal of Agricultural Economics  82 5: 1159– 69. http://dx.doi.org/10.1111/0002-9092.00115 Google Scholar CrossRef Search ADS   Henson S., Masakure O., Boselie D.. 2005. Private Food Safety and Quality Standards for Fresh Produce Exporters: The Case of Hortico Agrisystems, Zimbabwe. Food Policy  30 4: 371– 84. http://dx.doi.org/10.1016/j.foodpol.2005.06.002 Google Scholar CrossRef Search ADS   Henson S., Humphrey J.. 2009. The Impacts of Private Food Safety Standards on the Food Chain and on Public Standard-Setting Processes. Paper prepared for FAO/WHO. Available at: http://www.fao.org/3/a-i1132e.pdf. Hou M.A., Grazia C., Malorgio G.. 2015. Food Safety Standards and International Supply Chain Organization: A Case Study of the Moroccan Fruit and Vegetable Exports. Food Control  55: 190– 9. Google Scholar CrossRef Search ADS   Leirisalo-Repo M., Helenius P., Hannu T., Lehtinen A., Kreula J., Taavitsainen M., Koskimies S.. 1997. Long Term Prognosis of Reactive Salmonella Arthritis. Annals of the Rheumatic Diseases  56 9: 516– 20. Google Scholar CrossRef Search ADS PubMed  Lemeilleur S. 2013. Smallholder Compliance with Private Standard Certification: The Case of GlobalGAP Adoption by Mango Producers in Peru. International Food and Agribusiness Management Review  16 4: 159– 80. Lewbel A. 2000. Semiparametric Qualitative Response Model Estimation with Unknown Heteroscedasticity or Instrumental Variables. Journal of Econometrics  97 1: 145– 77. http://dx.doi.org/10.1016/S0304-4076(00)00015-4 Google Scholar CrossRef Search ADS   Lewbel A., Dong Y., Yang T.T.. 2012. Comparing Features of Convenient Estimators for Binary Choice Models with Endogenous Regressors. Canadian Journal of Economics  45 3: 809– 29. http://dx.doi.org/10.1111/j.1540-5982.2012.01733.x Google Scholar CrossRef Search ADS   MacDonald J., Ollinger M.. 2005. Technology, Labor Wars, and Producer Dynamics: Explaining Consolidation in Beefpacking. American Journal of Agricultural Economics  87 4: 1020– 33. http://dx.doi.org/10.1111/j.1467-8276.2005.00785.x Google Scholar CrossRef Search ADS   Marsh T.L., Schroeder T.C., Mintert J.. 2004. Impacts of Meat Recalls on Consumer Demand in the USA. Applied Economics  36 9: 897– 909. http://dx.doi.org/10.1080/0003684042000233113 Google Scholar CrossRef Search ADS   Minor T., Parrett M.. 2016. A Retrospective Review of the Economic Impact of the Food and Drug Administration’s Proposed Egg Rule. Agricultural Economics  47 4: 457– 64. Google Scholar CrossRef Search ADS   Minor T., Parrett M.. 2017. The Economic Impact of the Food and Drug Administration’s Final Juice HACCP Rule. Food Policy  68: 206– 13. Google Scholar CrossRef Search ADS   Montiel Olea J.L., Pflueger C.E.. 2013. A Robust Test for Weak Instruments. Journal of Business and Economic Statistics  31 3: 358– 69. http://dx.doi.org/10.1080/00401706.2013.806694 Google Scholar CrossRef Search ADS   Muth M.K., Fahimi M., Karns S.A.. 2009. Analysis of Salmonella Control Performance in U.S. Young Chicken Slaughter and Pork Slaughter Establishments. Journal of Food Protection  72 1: 6– 13. http://dx.doi.org/10.4315/0362-028X-72.1.6 Google Scholar CrossRef Search ADS PubMed  Muth M., Wohlgenant M.K., Karns S.. 2007. Did the Pathogen Reduction and Hazard Analysis and Critical Control Points Regulation Cause Slaughter Plants to Exit? Review of Agricultural Economics  29 3: 596– 611. Google Scholar CrossRef Search ADS   Ollinger M., Bovay J., Benicio C., Guthrie J.. 2015. Economic Incentives to Supply Safe Chicken to the National School Lunch Program. Economic Research Report Number 202. Washington DC: U.S. Department of Agriculture, Economic Research Service. Available at: https://www.ers.usda.gov/webdocs/publications/45500/55537_err202.pdf?v=42332. Ollinger M., Guthrie J., Bovay J.. 2014. The Food Safety Performance of Ground Beef Suppliers to the National School Lunch Program. Washington DC: Department of Agriculture, Economic Research Service, Economic Research Report No. 180. Available at: https://www.ers.usda.gov/webdocs/publications/45326/50382_err180.pdf?v=41996. Ollinger M., Moore D.. 2008. The Economic Forces Driving Food Safety Quality in Meat and Poultry. Review of Agricultural Economics  30 2: 289– 310. http://dx.doi.org/10.1111/j.1467-9353.2008.00405.x Google Scholar CrossRef Search ADS   Ollinger M., Moore D.. 2009. The Direct and Indirect Costs of Food Safety Regulation. Review of Agricultural Economics  31 2: 247– 65. http://dx.doi.org/10.1111/j.1467-9353.2009.01436.x Google Scholar CrossRef Search ADS   Ollinger M., Moore D., Chandran R.. 2004. Meat and Poultry Plants’ Food Safety Investments: Survey Findings. Washington DC: U.S. Department of Agriculture, Economic Research Service, Technical Bulletin 1911. Available at: https://www.ers.usda.gov/webdocs/publications/47486/17469_tb1911.pdf?v=41029. Ollinger M., Mueller V.. 2003. Managing for Safer Food: The Economics of Sanitation and Process Controls in Meat and Poultry Plants. Washington DC: U.S. Department of Agriculture, Economic Research Service, AER-817. Available at: https://www.ers.usda.gov/webdocs/publications/41496/18901_aer817.pdf?v=41063. Piggott N.E., Marsh T.L.. 2004. Does Food Safety Information Impact U.S. Meat Demand? American Journal of Agricultural Economics  86 1: 154– 74. http://dx.doi.org/10.1111/j.0092-5853.2004.00569.x Google Scholar CrossRef Search ADS   Pouliot S., 2014. The Production of Safe Food According to Firm Size and Regulatory Exemption: Application to FSMA. Agribusiness  30 4: 493– 512. Google Scholar CrossRef Search ADS   Pozo V.F., Schroeder T.C.. 2016. Evaluating the Costs of Meat and Poultry Recalls to Food Firms Using Stock Returns. Food Policy  59: 66– 77. Google Scholar CrossRef Search ADS   Sumner J., Raven G., Givney R.. 2004. Have Changes to Meat and Poultry Food Safety Regulation in Australia Affected the Prevalence of Salmonella or of Salmonellosis? International Journal of Food Microbiology  92 2: 199– 205. Google Scholar CrossRef Search ADS PubMed  Thomsen M.R., McKenzie A.M.. 2001. Market Incentives for Safe Foods: An Examination of Shareholder Losses from Meat and Poultry Recalls. American Journal of Agricultural Economics  83 3: 526– 38. http://dx.doi.org/10.1111/0002-9092.00175 Google Scholar CrossRef Search ADS   Thomsen M.R., Shiptsova R., Hamm S.J.. 2006. Sales Responses to Recalls for Listeria monocytogenes: Evidence from Branded Ready-to-Eat Meats. Review of Agricultural Economics  28 4: 482– 93. http://dx.doi.org/10.1111/j.1467-9353.2006.00317.x Google Scholar CrossRef Search ADS   U.S. Department of Agriculture, Agricultural Marketing Service. Commodity Areas, Commodity Purchasing, Solicitations and Awards, Red Meat and Fish or Poultry and Eggs. 2009–2014. Available at: http://www.ams.usda.gov/AMSv1.0/ams.fetchTemplateData.do?template=TemplateJ&page=CPDCommodityPurchaseMeatFish (Accessed 2009–2014, link no longer valid). U.S. Department of Agriculture, Agricultural Marketing Service. 2012. Supplement 211 to AMS Master Solicitation: Purchase of Frozen Ground Beef Products for Distribution to Child Nutrition and Other Federal Food and Nutrition Programs. Available at: https://www.ams.usda.gov/sites/default/files/media/Supplement%20211%20Frozen%20Beef%20Products%20June%202012%20%28Oct%202012%29.pdf. Van Ophem H., Schram A.. 1997. Sequential and Multinomial Logit: A Nested Model. Empirical Economics  22 1: 131– 52. http://dx.doi.org/10.1007/BF01188174 Google Scholar CrossRef Search ADS   Von Schlippenbach V., Teichmann I.. 2012. The Strategic Use of Private Quality Standards in Food Supply Chains. American Journal of Agricultural Economics   94 5: 1189– 201. Google Scholar CrossRef Search ADS   Wooldridge J.M. 2009. Introductory Econometrics: A Modern Approach . Fourth Edition Mason, Ohio: South-Western Cengage Learning. Young R.W. 2005. Agricultural Marketing Service Management Controls to Ensure Compliance with Purchase Specification Requirements for Ground Beef. Washington DC, U.S. Department of Agriculture. Available at: http://i.usatoday.net/news/pdf/ams-standards.pdf. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association 2017. This work is written by US Government employees and is in the public domain in the US. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png American Journal of Agricultural Economics Oxford University Press

Pass or Fail: Economic Incentives to Reduce Salmonella Contamination in Ground Beef Sold to the National School Lunch Program

Loading next page...
 
/lp/ou_press/pass-or-fail-economic-incentives-to-reduce-salmonella-contamination-in-fOFeegXejA
Publisher
Oxford University Press
Copyright
Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association 2017. This work is written by US Government employees and is in the public domain in the US.
ISSN
0002-9092
eISSN
1467-8276
D.O.I.
10.1093/ajae/aax088
Publisher site
See Article on Publisher Site

Abstract

Abstract Ground beef sold to the USDA’s Agricultural Marketing Service (AMS) for distribution to the National School Lunch Program (NSLP) must meet stringent food-safety standards, specifically, a zero-tolerance standard for Salmonella. We use a unique data set containing information on Salmonella levels in order to examine the sequential decisions of ground-beef plants to become registered as AMS suppliers and then bid on contracts to supply the NSLP from 2006 to 2012. We find that plants exploit their competitive advantages in relatively high productivity and strong performance on Salmonella tests when choosing to bid on contracts in a given year. Furthermore, the incentives generated by the zero-tolerance standard for Salmonella are highly effective: ground beef supplied to the NSLP is 21–22 percentage points more likely to meet a zero-tolerance standard for Salmonella than ground beef tested as part of typical meat-plant inspections. The National School Lunch Program (NSLP) provides about 31 million subsidized or free meals to children in the United States each school day. Some of these meals are made from ground beef, chicken, and other foods purchased under contract by the U.S. Department of Agriculture’s (USDA) Agricultural Marketing Service (AMS). The safety of ground beef served in schools is particularly important because it is served to a vulnerable population (children) and it is easily contaminated with harmful pathogens such as Salmonella, which can cause foodborne illnesses that may have life-long health effects (Leirisalo-Repo et al. 1997).1 Since 2004, AMS has imposed strict, zero-tolerance standards for Salmonella content in the ground beef it purchases under contract for NSLP and enforces this standard with testing, rejecting shipments that fail to meet the standard. A growing body of research has focused on the use of food-safety standards as contractual requirements. Von Schlippenbach and Teichmann (2012) showed that retailers use private quality standards to improve their bargaining power and mitigate inefficiencies in upstream production. Hou, Grazia, and Malorgio (2015) argued that private standards are the prevailing mechanism that governs food safety when buyers have more market power than sellers. Other researchers have analyzed the effects of buyer standards for food safety in the context of sales to supermarkets in Latin America (Balsevich et al. 2003), fish exports from Kenya to the European Union (Henson, Brouder, and Mitullah 2000), mango exports from Peru (Lemeilleur 2013), and high-value fresh vegetables exported from Zimbabwe (Henson, Masakure, and Boselie 2005). See Henson and Humphrey (2009) for an overview of such research. Ollinger, Moore, and Chandran (2004) provided empirical evidence that meat plants with contracts to supply major restaurant chains made greater food-safety investments than other plants, and Ollinger and Moore (2008) found that meat plants supplying large-volume buyers had significantly better performance on Salmonella tests than other plants.2 Little prior research has explored the effects of food-safety standards on competitive behavior. In particular, no research has demonstrated that food-processing plants derive competitive advantage from their relative ability to meet food-safety standards. In addition, no research has demonstrated causal evidence that food-safety standards can improve food-safety outcomes.3 This article helps fill those voids. First, we provide empirical evidence that economic mechanisms and incentives drive plants to register as NSLP suppliers, and that plants seek to exploit their competitive advantages in food safety and relatively high productivity (volume of ground beef produced per employee) when bidding on contracts. Second, we demonstrate that food-safety standards can have a measurable impact on food-safety outcomes in a competitive market. The implications are particularly important for federal authorities overseeing the NSLP food purchasing programs and, more generally, policymakers considering food-safety regulations because they provide empirical evidence about the responses of food suppliers to food-safety standards and the effectiveness of a stringent standard for Salmonella content. Ground-beef sales to AMS for NSLP occur through a process of open bidding in which the lowest-cost bid is awarded a supply contract that requires adherence to a zero-tolerance standard for Salmonella.4 AMS purchases of ground beef for NSLP are a small share of the value of all ground beef sold in the United States—some $50 million annually. Yet these sales are an important component of certain ground-beef plants’ business: the mean NSLP supplier sold about 11% of its ground beef to NSLP, and the share ranged from 1% to 80%, USDA, 2009–2014; confidential FSIS administrative data. We use a unique dataset containing test results for the Salmonella sampling programs conducted by the USDA Food Safety and Inspection Service (FSIS) and third-party laboratories on behalf of AMS, and other public and administrative data to examine this contracting process and evaluate the performance of plants on tests for Salmonella in samples of ground beef from 2006 to 2012. The next sections provide background and discuss the data and empirical strategy. We then present results from several sets of empirical tests. In the first empirical analysis, we examine the relationship between plants’ past Salmonella test performance and the sequential decisions to (a) register as NSLP suppliers and (b) bid on contracts to supply NSLP in a given year. We then demonstrate that plants that bid on contracts to supply NSLP had improved performance on Salmonella tests relative to registered suppliers that did not ship ground beef to NSLP. Next, we show that the contamination rate of ground beef shipped to NSLP was substantially lower than the rate indicated by FSIS tests of ground beef from the same plants. Finally, we contrast the performance of ground beef suppliers to chicken suppliers, which do not face a stringent standard for Salmonella on sales to the NSLP, and discuss the implications of our findings. Background FSIS has been monitoring food safety in meat slaughter and processing plants since the passage of the Federal Meat Inspection Act in 1906. Current regulations mandate process controls for each plant, including Sanitation Standard Operating Procedures (SSOPs) and tasks required under Hazard Analysis and Critical Control Point (HACCP) plans. FSIS inspectors monitor compliance with all SSOPs and HACCP tasks and issue noncompliance reports. FSIS can temporarily halt a plant’s production if infractions are serious and corrective actions are not taken, but these instances are rare. Most noncompliance incidents are quickly resolved with no disruption of production.5 Since the implementation of the Pathogen Reduction/Hazard Analysis and Critical Control Point (PR/HACCP) Rule of 1996, FSIS has conducted random tests of cattle, hog, and poultry carcasses and ground meat and poultry for Salmonella at meat plants. FSIS samples plants on a weighted random basis under which poorer-performing plants and larger plants are sampled more frequently. The purpose of these tests is to verify that plants are maintaining food-safety process controls. FSIS requires that all plants achieve a certain performance standard on Salmonella tests—these are non-zero tolerances, set by FSIS at levels that most plants have historically been able to achieve. For ground beef, FSIS permits 5 out of 53 samples to test positive for Salmonella, a tolerance which has remained unchanged since the implementation of PR/HACCP. Plants that fail to meet this standard are retested, compelled to update food safety plans, and required to take corrective actions in order to continue operation and production. See Ollinger, Guthrie, and Bovay (2014) for further discussion. AMS, in contrast to FSIS, has no regulatory authority for food safety and functions much like a private buyer in its capacity as a purchaser of food for distribution to the NSLP, in that it sets standards to which only its own suppliers must comply. If a plant chooses to register with AMS and become eligible to supply the NSLP, that plant must demonstrate its ability to meet AMS specifications, pass an audit, adhere to FSIS food-safety standards, and meet other pre-approval criteria outlined by AMS in a document known as a Technical Requirements Schedule (TRS). AMS established zero-tolerance standards for Salmonella and E. coli O157:H7 in ground beef in 2004. Further, since 2008 AMS has required ground-beef plants to test for generic E. coli and other bacteria that indicate whether plants are maintaining good food-safety process controls, and also introduced some other requirements related to food safety in 2010. The AMS standards for ground beef purchases are outlined in USDA AMS (2012). See table 1 for a comparison of food-safety requirements for FSIS and AMS. Table 1. Key Differences between FSIS Regulations and AMS Standards Covering Food Safety Product Testing and Process Controls Process control  AMS tolerance  AMS testing frequency (pounds)  FSIS tolerance  FSIS testing frequency  Microbial testing          E coli O157:H7  0.0  2,000  0.0  Random: Less than once per year  Salmonella  0.0  10,000  0.113 (5 of 53 samples)  Random: Usually, less than once per year  Standard plate count  100,000/gram  10,000  No requirement  No requirement  Generic E. coli  500/gram  10,000  Done at establishment  Schedule in Ollinger and Mueller (2003)  Total coliforms  1,000/gram  10,000  No requirement  No requirement  Process controls  AMS standard    FSIS regulation    Removal of major lymph glands, thymus gland, and cartilage  Required    No requirement    Removal of white fibrous materials, e.g., elbow tendons  Required    No requirement    Removal of yellow elastin  Required    No requirement    Slaughter operation  AMS standard    FSIS regulation    Removal of spinal cord  Required    Required    Use of meat from non-ambulatory animals  Not allowed    Permitted with veterinarian consent    Processing interventions to control pathogens  At least two. One must be a critical control point    No requirement    Routine testing of E. coli types including E. coli O157:H7  Several E. coli types    Generic E. coli only    Process control  AMS tolerance  AMS testing frequency (pounds)  FSIS tolerance  FSIS testing frequency  Microbial testing          E coli O157:H7  0.0  2,000  0.0  Random: Less than once per year  Salmonella  0.0  10,000  0.113 (5 of 53 samples)  Random: Usually, less than once per year  Standard plate count  100,000/gram  10,000  No requirement  No requirement  Generic E. coli  500/gram  10,000  Done at establishment  Schedule in Ollinger and Mueller (2003)  Total coliforms  1,000/gram  10,000  No requirement  No requirement  Process controls  AMS standard    FSIS regulation    Removal of major lymph glands, thymus gland, and cartilage  Required    No requirement    Removal of white fibrous materials, e.g., elbow tendons  Required    No requirement    Removal of yellow elastin  Required    No requirement    Slaughter operation  AMS standard    FSIS regulation    Removal of spinal cord  Required    Required    Use of meat from non-ambulatory animals  Not allowed    Permitted with veterinarian consent    Processing interventions to control pathogens  At least two. One must be a critical control point    No requirement    Routine testing of E. coli types including E. coli O157:H7  Several E. coli types    Generic E. coli only    Notes: The AMS standards for E. coli O157:H7 and Salmonella were established in 2004. All requirements except these two standards were established in 2008. Sources: The AMS food-safety standards were outlined in USDA AMS (2012); the FSIS food-safety regulations were described in the Federal Register, and are available at: https://www.fsis.usda.gov/OPPDE/rdad/FRPubs/93-016F.pdf. Salmonella testing is an important feature of the AMS food-safety program for ground beef. AMS requires NSLP suppliers to have private laboratories submit Salmonella test results for each 10,000-pound lot they supply to NSLP. Once each lot is shown to meet the zero-tolerance standard, it may be shipped to warehouses owned by states or operated by contractors under contracts with states; products are then shipped to individual school districts upon demand. Ground-beef suppliers that fail to meet AMS standards for the NSLP cannot sell the meat to other USDA programs such as the Child and Adult Care Food Program and Summer Food Service Program. Moreover, these suppliers incur the sunk costs of having prepared ground beef for shipment to the NSLP and then having to destroy or reprocess the rejected product. Repeated failures may result in a plant losing the right to bid on AMS contracts. Data We created a unique data set of all plants that produced ground beef and whose products were tested for Salmonella by FSIS from 2006 to 2012, including public and administrative data from AMS and FSIS. The AMS data include information that identified which plants were registered with AMS to supply the NSLP and when these AMS-registered plants bid on contracts to supply the NSLP.6 AMS also shared its Salmonella test data with FSIS. The FSIS data include (a) the results of tests for Salmonella in ground beef conducted by FSIS for its Salmonella monitoring program, (b) types and numbers of animals slaughtered, (c) pounds of ground beef produced, (d) the date each plant began operations, (e) performance on sanitation and HACCP tasks, and (f) the types of further processing (i.e., processing ground beef into other products) done in the plant. The FSIS data also include information from Dun & Bradstreet on the number of plant-level employees and whether the plant was part of a firm that owned more than one plant. Data on Salmonella tests conducted by FSIS, as well as data on compliance with SSOP and HACCP tasks, are from confidential FSIS files. During the timeframe of our study, testing was done in batches or sample sets of 53, spread out over some period of days or weeks.7 FSIS did not sample each plant every year due to limited laboratory capacity, and collected partial sample sets for some plant-year observations. Plants producing less than 1,000 pounds per day were tested less frequently than larger plants and some were never tested. FSIS used a selection algorithm to determine which plants to test; this algorithm ensured that higher-risk plants were sampled at least once per year. Plants with Salmonella levels better than half the standard (about 80% of all ground beef plants) were considered low-risk unless other information available to FSIS, such as generic E. coli test results collected by plants, indicated otherwise. These low-risk plants were tested less than one time per year but were supposed to be tested at least once every two years. There are 1,291 plant-year observations of FSIS test results from 636 ground beef plants. These observations included 93 observations from 32 NSLP suppliers. The data on AMS Salmonella test results for ground beef shipped to NSLP covers 2007 to 2012. AMS does not conduct its own testing; rather, private laboratories test ground beef on behalf of NSLP suppliers for AMS review. AMS uses the Salmonella test data submitted by these private laboratories to evaluate whether a shipment of ground beef meets its standards, and shares the testing data with FSIS. Personnel from FSIS matched AMS test data to FSIS administrative data when plant identification information was available. The resulting matched AMS–FSIS data set included 95 observations for 25 unique plants covering the 2007 to 2012 period; in 31 instances, both AMS and FSIS Salmonella testing data were available for the same plant-year. Matches of FSIS data to AMS data could not be made in all cases because in some cases, AMS data did not have a FSIS identifier, and in other cases, FSIS did not conduct Salmonella testing for that plant in that year. Table 2 contrasts the performance on Salmonella tests of commercial-only plants (i.e., those plants that did not register to bid on NSLP contracts, which supplied processors, retailers, hospitals and other institutions, and other non-NSLP buyers) with inactive and active NSLP suppliers, and the performance of active NSLP suppliers on FSIS Salmonella tests with the performance of active NSLP suppliers on AMS Salmonella tests. Inactive and active NSLP suppliers were registered to bid on AMS contracts; they differed in that inactive NSLP suppliers did not bid on a contract to supply the NSLP during a given year, whereas active NSLP suppliers bid on at least one contract during that year. Inactive NSLP suppliers served the same types of buyers as commercial-only plants; active NSLP suppliers sold about 11% of their output to the NSLP. Table 2. Distribution of Plant Performance on Typical 53-sample Sets of Salmonella Tests Conducted in the FSIS and AMS Testing Programs, 2006–2012   FSIS tests   Third-party tests  Mean number of samples testing positive for Salmonella per 53-sample set  Commercial-only plants  Inactive NSLP suppliers  Active NSLP suppliers  Active NSLP suppliers (shipments to NSLP)    Share of plants  0=x  0.575  0.200  0.534  0.702  0<x≤1  0.0209  0.0286  0.0517  0.277  1<x≤2  0.174  0.229  0.155  0  2<x≤3  0.0968  0.171  0.172  0.0213  3<x≤4  0.0526  0.229  0.0517  0  4<x≤5  0.0301  0.0857  0.0172  0  5<x≤6  0.0192  0.0286  0.0172  0  6<x≤7  0.0109  0  0  0  7<x≤8  0.00334  0  0  0  8<x≤9  0.00668  0  0  0  9<x≤10  0.00334  0  0  0  10<x  0.00668  0.0286  0  0  Number of observations  1198  35  58  47    FSIS tests   Third-party tests  Mean number of samples testing positive for Salmonella per 53-sample set  Commercial-only plants  Inactive NSLP suppliers  Active NSLP suppliers  Active NSLP suppliers (shipments to NSLP)    Share of plants  0=x  0.575  0.200  0.534  0.702  0<x≤1  0.0209  0.0286  0.0517  0.277  1<x≤2  0.174  0.229  0.155  0  2<x≤3  0.0968  0.171  0.172  0.0213  3<x≤4  0.0526  0.229  0.0517  0  4<x≤5  0.0301  0.0857  0.0172  0  5<x≤6  0.0192  0.0286  0.0172  0  6<x≤7  0.0109  0  0  0  7<x≤8  0.00334  0  0  0  8<x≤9  0.00668  0  0  0  9<x≤10  0.00334  0  0  0  10<x  0.00668  0.0286  0  0  Number of observations  1198  35  58  47  Notes: The FSIS randomly samples meat for Salmonella in ground-beef plants, and over the study period, determined plants’ compliance status on the basis of the number of positive samples out of a set of 53 samples. For ground beef, the standard was that no more than 5 of 53 samples could test positive for Salmonella. The FSIS may have tested some plants more than once in a year or may have tested fewer than 53 samples. This table normalizes the share of a plant’s samples testing positive for Salmonella against a denominator of 53 samples. In the terminology used in this paper, commercial-only plants were not registered with the AMS as eligible NSLP suppliers. Inactive NSLP suppliers were registered but did not actively bid on NSLP contracts in a given year. Active NSLP suppliers actively bid on NSLP contracts in a given year. The AMS requires that suppliers submit the results of Salmonella tests conducted by third-party laboratories with shipments of ground beef for the NSLP. We report the distribution of plant-year test results for plants that shipped ground beef for the NSLP and were included in our analysis; the full set of regressors was available for the included plants. As seen in table 2, 57.5% of commercial-only plants had zero samples test positive for Salmonella, and the share of active NSLP suppliers with zero positive Salmonella samples was more than twice as high (53.4%) as that of inactive NSLP suppliers (20%). Table 2 also shows that the share of plants failing to meet the FSIS standard of 5 positive Salmonella samples out of 53 was more than twice as high for commercial-only (5.0%) and inactive NSLP suppliers (5.7%) as for active NSLP suppliers (1.7%). Finally, table 2 shows a striking difference between the Salmonella test results for ground beef shipped to NSLP and ground beef produced by active NSLP suppliers and tested by FSIS: 97.9% of active NSLP plants had performance equivalent to one or fewer samples out of a 53-sample set testing positive for Salmonella on shipments to NSLP, whereas only 58.6% of the active NSLP suppliers’ FSIS sample sets achieved the same level of performance. The patterns shown in table 2 lead to several questions about causality and incentives. For example, what motivates plants to register as NSLP suppliers and actively bid on NSLP contracts? What drives improved performance on Salmonella tests by active NSLP suppliers, relative to both inactive NSLP suppliers and commercial-only suppliers? Are plants that have better Salmonella performance more likely to bid on contracts to supply the NSLP, or does winning a contract to supply the NSLP force plants to improve their food-safety practices in order to improve their performance on Salmonella tests? In the sections that follow, we develop and test empirical models to explain producer behavior with respect to NSLP contracts, Salmonella standards, and Salmonella test performance. Empirical Models This section describes three empirical models that characterize the effects of food-safety standards imposed by AMS on the behavior of ground-beef suppliers to the NSLP. In the first model, we analyze the effects of past Salmonella test performance on ground-beef plants’ decisions to register as NSLP suppliers and actively supply the NSLP in a given year. The second model evaluates the performance of active NSLP suppliers on FSIS tests for Salmonella in ground beef, relative to inactive NSLP suppliers. The third model compares the performance of active NSLP suppliers on tests for Salmonella in ground beef shipped to the NSLP, conducted on behalf of AMS, to the performance of active NSLP suppliers on FSIS tests for Salmonella. Empirical results are presented in the next section. Empirical Model of Plant Decisions to Bid on AMS Contracts Figure 1 illustrates the sequential decisions plants must make when deciding whether to supply the NSLP. Plants must first decide whether to register as NSLP suppliers, which they are eligible to do if they comply with the AMS approval conditions in the TRS, and with the general FSIS food-safety regulations that apply to all plants. Registered NSLP suppliers must then choose whether to bid on contracts (i.e., become active NSLP suppliers). Because contracts are awarded to the lowest-cost bidder and all suppliers must meet a zero-tolerance standard for Salmonella and E. coli, not all plants are likely to find bidding on AMS contracts to be a profitable option. Some plants may have low costs of production but expect that meeting the zero-tolerance standard would prove too expensive; other plants may have perfect records with respect to Salmonella tests but find that their cost of production is too high to make supplying the NSLP profitable. Figure 1. View largeDownload slide Flow chart of plant registration and bidding decisions Note: In each stage, N includes only plant-year observations with lagged observations of the same plant, 2006–2012. Figure 1. View largeDownload slide Flow chart of plant registration and bidding decisions Note: In each stage, N includes only plant-year observations with lagged observations of the same plant, 2006–2012. We model the sequential AMS registration and bidding decision process using a sequential logit regression (Van Ophem and Schram 1997), as shown in equations (1) and (2) below. A sequential logit is most appropriate for modeling AMS contract bidding because bidding decisions are conditional on the decision to register as an AMS supplier. The population of plants available for the first transition in the sequential logit includes all FSIS-inspected ground beef plants.8 All of these plants sell ground beef to commercial and other institutional buyers and some are registered to sell to the NSLP; the dependent variable in the first transition distinguishes plants that register to supply the NSLP from commercial-only (i.e., non-NSLP-registered) plants. The population of plants for the second transition is restricted to the plants registered with AMS to supply the NSLP in a given year because only AMS-registered plants can bid on an AMS contract to supply the NSLP. We model the transition probability of becoming an AMS-registered NSLP supplier ( PAit^) as   PAit^=exp⁡αA+βASi,t-1+ϕAKi,t+ρAPostt1+exp⁡αA+βASi,t-1+ϕAKi,t+ρAPostt (1) and model the transition probability of actively bidding on AMS contracts in a given year ( PBit^), conditional on being a registered NSLP supplier, as   PBit^=( exp ⁡αA+βASi,t-1+γBi,t-1+ϕAKi,t+ρAPostt1+ exp ⁡αA+βASi,t-1+γBi,t-1+ϕAKi,t+ρAPostt|PAit=1) (2) where Si,t-1 is the performance of plant ion Salmonella tests in year t-1 expressed as the share of samples testing positive for Salmonella; B is a binary variable that indicates whether plants are active NSLP suppliers (i.e., bid on AMS contracts in a given year); K is a vector of plant characteristics; and Post is a binary variable with a value of 1 in 2008 and later years, included in one of the regression specifications. We include Post because AMS mandated additional food-safety requirements in 2008 (see table 1), which may have discouraged plants from bidding on contracts. We use lagged Salmonella test performance results, Si,t-1, as an explanatory variable because AMS requires that the ground beef it purchases for the NSLP meet a zero-tolerance standard. Past performance on Salmonella tests, which reflect the ability to meet the NSLP standard without incurring additional costs, may thus have a bearing on plants’ decisions to register as an NSLP supplier and to bid on NSLP contracts. For the second transition, we include the lagged binary variable Bi,t-1 to indicate whether plants bid on contracts to supply the NSLP in the previous year; plants with experience in bidding may have lower costs of preparing and managing bids and may have learned how to profitably meet NSLP requirements. Contractual requirements guide us in selecting the vector of plant characteristics ( K) and the post-2007 dummy as explanatory variables. Because contracts are awarded to the lowest-cost bidder, highly efficient plants may be more likely to register as NSLP suppliers; we use the log of ground beef output per employee as the measure of plant efficiency. Plant age may also be important because older plants may have more experienced workers and may also have managers with more experience in determining which contracts to compete for and how to submit successful bids. Larger plants (defined using the number of employees per plant) have greater plant capacity and may as a result be willing to compete for low-value contracts, such as those to supply the NSLP, to utilize this capacity. Two binary variables related to plant processing characteristics indicate whether plants (a) slaughter cattle and (b) further process ground beef into products such as pepperoni and jerky. Cattle slaughter plants enjoy substantial economies of scale (MacDonald and Ollinger 2005), suggesting that they continuously search for market outlets for their products in order to sell at their lowest cost. In contrast, plants that further process products have more market outlets for their products and may find other outlets more profitable than producing ground beef to fill NSLP contracts. Empirical Model of FSIS Salmonella Test Performance We have proposed that NSLP suppliers base their decisions to bid on AMS contracts, in part, on their performance on recent Salmonella tests. We now introduce a two-stage linear probability model (equations [3] and [4]) used to evaluate the relative performance of active and inactive NSLP suppliers on Salmonella tests, controlling for other factors. In this two-stage model, supplier type, Bi,t (whether a plant actively bids on contracts to supply NSLP), is endogenously determined by plants’ past performance on Salmonella tests and whether they bid on contracts to supply the NSLP in the previous year. The second stage of our linear probability model is given by equation (3) and the first stage by equation (4), as follows:   Sf,i,t*=α0+∑hδhKh,i,t+∑jρjRj,i,t+λPostt+ωBi,t+ξi,t (3)  Bi,t=α1+β1Bi,t-1+β2Si,t-1 (4) where food-safety test performance ( Sf,i,t*) is a binary variable indicating whether plant i met hypothetical tolerance level f in year t. In equation (3), food-safety test performance is a function of a vector of plant characteristics ( K), compliance with a vector of FSIS food-safety process control regulations ( R), and supplier type (B); Sf,i,t*=1if Si,t≤Tf and Sf,i,t*=0if Sf,i,t>Tf, where Si,t is the performance of plant i in year t on Salmonella tests and Tf is a hypothetical Salmonella tolerance that varies across regression specifications. We also include a binary variable ( Post) in equation (3) to indicate observations after 2007 when AMS increased the stringency of some food-safety requirements. In equation (4), as in equation (2), Bi,t-1 is a lagged binary variable indicating whether plants bid on AMS contracts in the previous year, and Si,t-1 is the lag of Salmonella test outcomes expressed as a continuous variable. These are used as instruments for the endogenous binary variable, Bi,t, which indicates whether plants bid on AMS contracts in the current year. Under the FSIS standard, five samples out of 53 are permitted to test positive for Salmonella. We test various tolerance levels, Tf, equal to the equivalents of 3, 2, and 1 samples testing positive for Salmonella for each 53-sample set, as well as a zero-tolerance standard, because (a) the AMS zero-tolerance requirement is much stricter than the FSIS standard of 5 positive samples, and, as table 2 shows, the vast majority of plants perform better than the FSIS standard, and (b) by varying tolerances, we can assess whether there is a particular threshold beyond which active NSLP suppliers consistently outperform inactive NSLP suppliers. There are several approaches to estimating econometric models with binary outcome variables and endogenous regressors. We use a two-stage linear probability model because it accommodates binary endogenous variables, and Wooldridge (2009) suggests that linear probability models perform well in predicting the values of independent variables that are near the averages of their samples.9Angrist and Pischke (2009) provide examples demonstrating the validity of linear probability models.10 The chief explanatory variable of interest in our model is the supplier type variable, Bi,t. The K variables from the bid decision model (equations [1] and [2]) are retained for this regression model, but are included for different reasons. If plants devote more effort to food-safety practices such as cleaning and sanitation, the quantity of ground beef produced per worker (productivity) may be lower. Ollinger and Moore (2008) found that larger plants had better food-safety performance in the chicken-slaughter industry but not in the ground-beef industry; including plant size (number of employees) as an explanatory variable allows us to revisit that conclusion. Muth, Fahimi, and Karns (2009) found that the vintage of plant capital is correlated with lower Salmonella levels in hog and chicken slaughter plants and further-processing plants, so we include plant age as a proxy for age of capital. We also include a binary variable indicating whether a plant uses meat from its own slaughter operations as meat inputs because vertically-integrated plants, which have greater control over inputs, may be able to attain better food-safety outcomes (Ollinger and Moore 2008). We also include three R variables, which reflect compliance with SSOP and HACCP tasks, as monitored and recorded by FSIS inspectors. Ollinger and Moore (2008) found that higher rates of compliance with SSOPs and HACCP tasks led to improved performance on Salmonella tests. We separately account for pre-operational SSOPs, which are cleaning and sanitation tasks performed at the beginning or end of the production day, and operational SSOPs, which are cleaning tasks performed during production. Empirical Model of Salmonella Test Performance for Products Sold to the NSLP Our third empirical model evaluates the performance of active NSLP ground beef suppliers on Salmonella tests of ground beef supplied to the NSLP, relative to ground beef randomly tested by FSIS as part of active NSLP suppliers’ general operations. We use a data set that combines FSIS Salmonella test results for active NSLP suppliers, by plant-year, with Salmonella test results, by plant-year, for ground beef tested for AMS. This combined data set enabled us to directly compare the performance of NSLP suppliers on tests for AMS of shipments of ground beef to the NSLP with the performance of NSLP suppliers on random FSIS tests. We ran a series of probit regressions (equation [5]) on the combined AMS–FSIS data set, where, again, the dependent variable, Sf,i,t*, is a binary variable indicating whether plant i met or performed better than each of three hypothetical tolerance thresholds ( Tf) on Salmonella tests conducted by FSIS or for AMS in year t:   Sf,i,t*=α0+∑hδhKh,i,t+∑jρjRj,i,t+λPostt+ωMi,t+ξi,t. (5) In equation (5), the binary variable M is defined as 1 if the test results were provided by third-party laboratories for shipments to AMS and 0 if the test results were from random testing conducted by FSIS. All other variables are defined as in equation (3). As before, we consider various tolerance levels in order to assess whether there is a particular threshold beyond which samples of shipments for NSLP consistently outperform the same plants’ FSIS test results. We do not include any endogenous variables in this empirical specification. Results We examine results for the three econometric models developed in the previous section. First, we evaluate the characteristics that may incentivize plants to bid on AMS contracts (equations [1] and [2]). Second, we examine the performance of Salmonella tests of active bidders, relative to inactive bidders (equations [3] and [4]). Finally, we compare the performance of active NSLP suppliers on Salmonella tests of ground beef shipped to the NSLP relative to their performance on random FSIS tests of ground beef that may be shipped either to commercial buyers or institutional buyers including AMS (equation [5]). Table 3 has definitions for all variables and the mean values of the observations used in the regressions. Table 3. Definitions and Mean Values of Selected Economic and Food-safety Variables that May Affect Contracting Choices, 2006–2012 Var.  Label  Definition  Active NSLP supplier  Inactive NSLP suppliera  Commercial- onlyb  Plant decisions to bid on AMS contracts        Si,t-1  Lag Share Positive Samples  Lag of share of Salmonella samples positive  0.015  0.041***  0.025  Bi,t-1  Lag Active NSLP Supplier  1 if plant bid on AMS contract during previous year; 0 otherwise  0.672  0.286***  0***  K1  Plant Employees  Employees per plant  342.4  257.5  136.7***  K2  Million Pounds per Employee  Million pounds of ground beef produced divided by number of plant employees  1.093  0.700  0.654***  K3  Age of Plant (years)  Current year minus year meat grant issued  15.59  19.00  29.58***  K4  Slaughters Cattle  1 if slaughters cattle; 0 otherwise  0.621  0.371***  0.124**  K5  Does Further Processing  1 if plant processes meat into products other than ground beef; 0 otherwise  0.0862  0.114  0.0876**  Postt  Year>2007  1 if year is after 2007; 0 otherwise  0.879  0.686**  0.591**    Observations  58  35  1,198  FSIS Salmonella test performance        S1*  3 positive Salmonella samples  1 if three or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.913  0.629***    S2*  2 positive Salmonella samples  1 if two or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.741  0.457***    S3*  1 positive Salmonella samples  1 if one or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.586  0.229***    S4*  0 positive Salmonella samples  1 if zero samples test positive for Salmonella; 0 otherwise  0.534  0.200***    R1  HACCP Compliance  Share of HACCP tasks in compliance with FSIS standards  0.998  0.995**    R2  Pre-operational SSOP Compliance  Share of pre-operational SSOP tasks in compliance with FSIS standards  0.964  0.972    R3  Operational SSOP Compliance  Share of operational SSOP tasks in compliance with FSIS standards  0.974  0.982*    R4  Multi-Plant  1 if plant is owned by a firm that owns other plants; 0 otherwise  0.138  0.0286*      Observationsc    58  35    Var.  Label  Definition  Active NSLP supplier  Inactive NSLP suppliera  Commercial- onlyb  Plant decisions to bid on AMS contracts        Si,t-1  Lag Share Positive Samples  Lag of share of Salmonella samples positive  0.015  0.041***  0.025  Bi,t-1  Lag Active NSLP Supplier  1 if plant bid on AMS contract during previous year; 0 otherwise  0.672  0.286***  0***  K1  Plant Employees  Employees per plant  342.4  257.5  136.7***  K2  Million Pounds per Employee  Million pounds of ground beef produced divided by number of plant employees  1.093  0.700  0.654***  K3  Age of Plant (years)  Current year minus year meat grant issued  15.59  19.00  29.58***  K4  Slaughters Cattle  1 if slaughters cattle; 0 otherwise  0.621  0.371***  0.124**  K5  Does Further Processing  1 if plant processes meat into products other than ground beef; 0 otherwise  0.0862  0.114  0.0876**  Postt  Year>2007  1 if year is after 2007; 0 otherwise  0.879  0.686**  0.591**    Observations  58  35  1,198  FSIS Salmonella test performance        S1*  3 positive Salmonella samples  1 if three or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.913  0.629***    S2*  2 positive Salmonella samples  1 if two or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.741  0.457***    S3*  1 positive Salmonella samples  1 if one or fewer samples per sample set test positive for Salmonella; 0 otherwise  0.586  0.229***    S4*  0 positive Salmonella samples  1 if zero samples test positive for Salmonella; 0 otherwise  0.534  0.200***    R1  HACCP Compliance  Share of HACCP tasks in compliance with FSIS standards  0.998  0.995**    R2  Pre-operational SSOP Compliance  Share of pre-operational SSOP tasks in compliance with FSIS standards  0.964  0.972    R3  Operational SSOP Compliance  Share of operational SSOP tasks in compliance with FSIS standards  0.974  0.982*    R4  Multi-Plant  1 if plant is owned by a firm that owns other plants; 0 otherwise  0.138  0.0286*      Observationsc    58  35    Notes: Asterisks *, **, and *** denote 0.10, 0.05, and 0.01 levels of significance, for two-sided t-tests. The observations of commercial-only suppliers are used in the sequential logit regressions only (table 4), so we do not present means for the commercial-only suppliers for the covariates that are only used in tables 5 and 6. Superscript a indicates statistical significance of the difference in the mean values between inactive and active NSLP suppliers;.b indicates statistical significance of the difference in the mean values between approved (active and inactive) NSLP suppliers and commercial-only suppliers. Source: Authors’ calculations using FSIS data. The data are pooled time series and cross sectional data from FSIS and AMS. All FSIS data extend from 1998 to 2014; AMS registration data span from 2005 to 2012; Salmonella testing data for shipments to the AMS are available from 2007 to 2012. Our first two regressions include as explanatory and instrumental variables, respectively, whether a plant was an active AMS supplier during the previous year. Thus, we are able to examine the 2006–2012 period in our first and second empirical tests, and 2007–2012 in our third empirical test. We provide two or three specifications of each regression: the first specification uses only the plant-level characteristics of primary interest, while the second and third specifications include more plant-level characteristics as explanatory variables. We use plant-level clustered standard errors to correct for heteroskedasticity in all of our econometric models because plant technology and characteristics do not change much over time. For example, a plant that is large one year is likely to be large the next year. These conditions suggest that error terms may be correlated within plants; failure to control for these within-plant error correlations could lead to small standard errors and large t-statistics (Cameron and Miller 2015). Marginal Effects of Past Salmonella Test Performance on Sequential AMS Contracting Choices Our first empirical test is to evaluate the effects of past Salmonella test performance on plants’ sequential decisions to register as NSLP suppliers and bid on contracts to supply the NSLP. Data for the first-stage regression includes all ground-beef plants for which FSIS data were available over the 2006–2012 period (1,291 observations). The second-stage regression includes only those plants that have registered with AMS as an NSLP supplier in the current year (93 observations). Table 4 provides results for three specifications of the sequential logit regression. In the first stage of the regression (under all specifications), we find that plants’ past performance on Salmonella tests had no statistically significant effect on the decision to register as a NSLP supplier. This is not surprising because registrants are only required to meet the (relatively loose) FSIS food-safety standards. Results for stage 1 also show that larger plants (as measured by number of employees) and more efficient plants (number of pounds produced per employee), as well as those with vertical integration (slaughtering cattle) were more likely to register to supply NSLP; these results were in line with our hypotheses about plants with excess capacity and greater efficiency seeking to be awarded AMS contracts to supply the NSLP. All other proposed explanatory variables were statistically insignificant in the first-stage regressions. Table 4. Marginal Effects of Past Salmonella Tests on Ground-beef Plants’ Sequential Choices to Register and Bid on Contracts to Supply NSLP, 2006–2012   (1)   (2)   (3)     Stage 1:  Stage 2:  Stage 1:  Stage 2:  Stage 1:  Stage 2:    NSLP supplier  Active NSLP supplier  NSLP supplier  Active NSLP supplier  NSLP supplier  Active NSLP supplier  Lag Share Positive Samples  –0.0566  –6.31**  –0.00856  –4.48*  –0.00842  –4.12*  (0.0621)  (3.24)  (0.0409)  (2.38)  (0.0404)  (2.47)  Lag Active NSLP Supplier  –  0.290***  –  0.310**  –  0.284**  (0.0993)  (0.123)  (0.122)  Log (plant employees)  0.0169***  0.0250  0.0151***  –0.0980  0.0140***  –0.0845  (0.00433)  (0.0642)  (0.00412)  (0.0908)  (0.00417)  (0.102)  Log (million pounds per employee)  0.0169***  0.0214  0.0158***  –0.0554  0.0142***  –0.0441  (0.00400)  (0.0649)  (0.00403)  (0.0784)  (0.00410)  (0.0966)  Does Further Processing  –0.00660  –0.317***  –0.0147  –0.533***  –0.0142  –0.558***  (0.00822)  (0.125)  (0.00934)  (0.162)  (0.00927)  (0.166)  Slaughters Cattle  –  –  0.0264***  0.483**  0.0255***  0.465**  (0.0112)  (0.204)  (0.0107)  (0.206)  Post-2007  –  –  –  –  0.00855  0.270**  (0.00592)  (0.131)  Age of Plant  –  –  –  –  –0.000190  0.000180  (0.000240)  (0.00498)  Observations  1,291  93  1,291  93  1,291  93  Wald χ2  71.64***  84.06***  121.75***    (1)   (2)   (3)     Stage 1:  Stage 2:  Stage 1:  Stage 2:  Stage 1:  Stage 2:    NSLP supplier  Active NSLP supplier  NSLP supplier  Active NSLP supplier  NSLP supplier  Active NSLP supplier  Lag Share Positive Samples  –0.0566  –6.31**  –0.00856  –4.48*  –0.00842  –4.12*  (0.0621)  (3.24)  (0.0409)  (2.38)  (0.0404)  (2.47)  Lag Active NSLP Supplier  –  0.290***  –  0.310**  –  0.284**  (0.0993)  (0.123)  (0.122)  Log (plant employees)  0.0169***  0.0250  0.0151***  –0.0980  0.0140***  –0.0845  (0.00433)  (0.0642)  (0.00412)  (0.0908)  (0.00417)  (0.102)  Log (million pounds per employee)  0.0169***  0.0214  0.0158***  –0.0554  0.0142***  –0.0441  (0.00400)  (0.0649)  (0.00403)  (0.0784)  (0.00410)  (0.0966)  Does Further Processing  –0.00660  –0.317***  –0.0147  –0.533***  –0.0142  –0.558***  (0.00822)  (0.125)  (0.00934)  (0.162)  (0.00927)  (0.166)  Slaughters Cattle  –  –  0.0264***  0.483**  0.0255***  0.465**  (0.0112)  (0.204)  (0.0107)  (0.206)  Post-2007  –  –  –  –  0.00855  0.270**  (0.00592)  (0.131)  Age of Plant  –  –  –  –  –0.000190  0.000180  (0.000240)  (0.00498)  Observations  1,291  93  1,291  93  1,291  93  Wald χ2  71.64***  84.06***  121.75***  Notes: Stage 1 is the decision of whether to register as a NSLP supplier (the alternative is to be a commercial-only plant); Stage 2 is the decision whether to bid on AMS contracts in a given year. Plant-level clustered standard errors appear in parentheses. Asterisks *, **, and *** denote 0.10, 0.05, and 0.01 levels of significance, respectively. The second-stage regressions reveal the factors that affected plants’ decisions to bid on contracts in a given year, conditional on being registered as eligible NSLP suppliers. Plants with better performance on Salmonella tests in the previous year and plants that bid on contracts to supply the NSLP in the previous year were substantially more likely to bid on contracts to supply the NSLP. Table 4 reports marginal effects at the mean; thus, a 1 percentage-point increase in the share of samples testing positive for Salmonella in the previous year reduced the probability that plants (with characteristics at mean values) actively bid on AMS contracts by 4.1–6.3 percentage points.11 Plants that bid on contracts in the previous year were about 28–31 percentage points more likely to be active NSLP suppliers than other plants, ceteris paribus. Conditional on having registered to supply NSLP, plants that further processed ground beef into value-added products were between 32–56 percentage points less likely to bid on NSLP contracts. This result conforms with expectations: plants that further process ground beef have more outlets for beef, making bidding on contracts to supply the NSLP less important. In addition, plants that slaughtered cattle were 47–48 percentage points more likely to bid on NSLP contracts, conditional on having registered; plants were 27 percentage points more likely to bid after 2008, when new safety standards were required for all ground-beef plants. Neither plant productivity nor plant size affected the decision of registered NSLP suppliers to bid on contracts, suggesting that the costs of complying with the zero-tolerance standard for Salmonella outweighed their competitive advantages in productivity and size and need for greater access to markets. Performance of NSLP Suppliers on FSIS Tests for Salmonella The second empirical test examines the performance of active NSLP suppliers relative to inactive NSLP suppliers on Salmonella tests administered by FSIS. Results of two-stage linear probability regression models are presented in table 5. Again, the dependent variable in each regression is the probability that plants pass one of several hypothetical tolerance standards. The endogenous variable is a dummy variable identifying whether a plant is an active NSLP supplier. The Kleibergen-Paap rk LM statistic for underidentification, Hansen J statistic for overidentification, and Montiel-Pflueger F statistic (Montiel Olea and Plfueger 2013) for each variation of the regression indicate that the instruments are valid and strong.12 These regressions use 84 observations, 9 fewer than in the second stage of the first model, because FSIS Salmonella test results were not available for all plants in all years. As described above, the FSIS test results do not represent the Salmonella levels of ground beef shipped to NSLP; rather, they represent random samples collected by FSIS personnel during the plants’ regular operations. Table 5. Performance on FSIS Salmonella Tests of Active and Inactive Suppliers of Ground Beef to the National School Lunch Program, 2006–2012   3 positive Salmonella results per 53-sample set   2 positive Salmonella results per 53-sample set   1 positive Salmonella result per 53-sample set   0 positive Salmonella results per 53-sample set     (1)  (2)  (1)  (2)  (1)  (2)  (1)  (2)  Active NSLP supplier  0.316  0.337  0.279  0.306  0.471**  0.492**  0.473**  0.485**  (0.206)  (0.216)  (0.265)  (0.270)  (0.235)  (0.248)  (0.229)  (0.242)  Log (plant employees)  –0.176**  –0.164**  –0.117  –0.990  –0.152*  –0.137  –0.166**  –0.157*  (0.0761)  (0.0735)  (0.0784)  (0.798)  (0.804)  (0.0861)  (0.0794)  (0.0828)  Log (pounds per employee)  –0.104  –0.0875  –0.106  –0.0817  –0.131*  –0.109  –0.141*  –0.129  (0.0636)  (0.0700)  (0.0665)  (0.0789)  (0.764)  (0.0870)  (0.0732)  (0.0821)  HACCP Compliance  0.823  0.501  0.156  –0.278  –5.07  –5.42  –6.70  –6.89  (6.15)  (6.09)  (6.82)  (6.64)  (6.49)  (6.45)  (6.34)  (6.45)  Pre-operational SSOP Compliance  –1.93*  –1.71*  0.449  0.777  –0.0661  0.232  1.06  1.23  (0.998)  (0.998)  (1.59)  (1.57)  (1.76)  (1.71)  (1.49)  (1.49)  Operational SSOP Compliance  1.33*  1.49*  –0.661  –0.412  0.712  0.934  0.177  0.307  (0.704)  (0.886)  (1.21)  (1.43)  (1.43)  (1.50)  (1.34)  (1.42)  Slaughters Cattle  0.219**  0.220**  0.0986  0.101  0.229*  0.231*  0.253**  0.255**  (0.106)  (0.105)  (0.116)  (0.114)  (0.129)  (0.130)  (0.124)  (0.125)  Post-2007  0.0615  0.0546  0.209  0.200  0.0227  0.0150  0.0801  0.0757  (0.127)  (0.131)  (0.130)  (0.134)  (0.120)  (0.123)  (0.0922)  (0.0938)  Multi-plant  0.0464  0.0295  0.0729  0.0486  –0.118  –0.139  –0.0823  –0.0945  (0.114)  (0.0999)  (0.162)  (0.150)  (0.234)  (0.230)  (0.230)  (0.230)  Age of Plant  –  0.00178  –  0.00268  –  0.00243  –  0.00142  (0.00315)  (0.00339)  (0.00355)  (0.00348)  Constant  0.927  0.795  0.725  0.482  5.03  4.76  6.05  5.89  (5.61)  (5.28)  (6.24)  (5.91)  (6.62)  (6.19)  (6.46)  (6.16)  F statistic  4.54***  4.62***  2.08*  2.07*  2.26*  2.02*  4.31***  4.86***  Underidentification: Kleibergen-Paap rk LM statistic ( χ2(2) p-value)  7.54  6.08  7.54  6.08  7.54  6.08  7.54  6.08  (0.0230)  (0.0479)  (0.0230)  (0.0479)  (0.0230)  (0.0479)  (0.0230)  (0.0479)  Weak identification: Montiel-Pflueger effective F statistic  9.03  7.08  9.03  7.08  9.03  7.08  9.03  7.08  Overidentification: Hansen J statistic ( χ2(1) p-value)  1.32  1.27  0.221  0.189  0.192  0.213  0.105  0.113  (0.251)  (0.259)  (0.638)  (0.664)  (0.661)  (0.645)  (0.746)  (0.736)  Observations  84  84  84  84  84  84  84  84    3 positive Salmonella results per 53-sample set   2 positive Salmonella results per 53-sample set   1 positive Salmonella result per 53-sample set   0 positive Salmonella results per 53-sample set     (1)  (2)  (1)  (2)  (1)  (2)  (1)  (2)  Active NSLP supplier  0.316  0.337  0.279  0.306  0.471**  0.492**  0.473**  0.485**  (0.206)  (0.216)  (0.265)  (0.270)  (0.235)  (0.248)  (0.229)  (0.242)  Log (plant employees)  –0.176**  –0.164**  –0.117  –0.990  –0.152*  –0.137  –0.166**  –0.157*  (0.0761)  (0.0735)  (0.0784)  (0.798)  (0.804)  (0.0861)  (0.0794)  (0.0828)  Log (pounds per employee)  –0.104  –0.0875  –0.106  –0.0817  –0.131*  –0.109  –0.141*  –0.129  (0.0636)  (0.0700)  (0.0665)  (0.0789)  (0.764)  (0.0870)  (0.0732)  (0.0821)  HACCP Compliance  0.823  0.501  0.156  –0.278  –5.07  –5.42  –6.70  –6.89  (6.15)  (6.09)  (6.82)  (6.64)  (6.49)  (6.45)  (6.34)  (6.45)  Pre-operational SSOP Compliance  –1.93*  –1.71*  0.449  0.777  –0.0661  0.232  1.06  1.23  (0.998)  (0.998)  (1.59)  (1.57)  (1.76)  (1.71)  (1.49)  (1.49)  Operational SSOP Compliance  1.33*  1.49*  –0.661  –0.412  0.712  0.934  0.177  0.307  (0.704)  (0.886)  (1.21)  (1.43)  (1.43)  (1.50)  (1.34)  (1.42)  Slaughters Cattle  0.219**  0.220**  0.0986  0.101  0.229*  0.231*  0.253**  0.255**  (0.106)  (0.105)  (0.116)  (0.114)  (0.129)  (0.130)  (0.124)  (0.125)  Post-2007  0.0615  0.0546  0.209  0.200  0.0227  0.0150  0.0801  0.0757  (0.127)  (0.131)  (0.130)  (0.134)  (0.120)  (0.123)  (0.0922)  (0.0938)  Multi-plant  0.0464  0.0295  0.0729  0.0486  –0.118  –0.139  –0.0823  –0.0945  (0.114)  (0.0999)  (0.162)  (0.150)  (0.234)  (0.230)  (0.230)  (0.230)  Age of Plant  –  0.00178  –  0.00268  –  0.00243  –  0.00142  (0.00315)  (0.00339)  (0.00355)  (0.00348)  Constant  0.927  0.795  0.725  0.482  5.03  4.76  6.05  5.89  (5.61)  (5.28)  (6.24)  (5.91)  (6.62)  (6.19)  (6.46)  (6.16)  F statistic  4.54***  4.62***  2.08*  2.07*  2.26*  2.02*  4.31***  4.86***  Underidentification: Kleibergen-Paap rk LM statistic ( χ2(2) p-value)  7.54  6.08  7.54  6.08  7.54  6.08  7.54  6.08  (0.0230)  (0.0479)  (0.0230)  (0.0479)  (0.0230)  (0.0479)  (0.0230)  (0.0479)  Weak identification: Montiel-Pflueger effective F statistic  9.03  7.08  9.03  7.08  9.03  7.08  9.03  7.08  Overidentification: Hansen J statistic ( χ2(1) p-value)  1.32  1.27  0.221  0.189  0.192  0.213  0.105  0.113  (0.251)  (0.259)  (0.638)  (0.664)  (0.661)  (0.645)  (0.746)  (0.736)  Observations  84  84  84  84  84  84  84  84  Notes: Plant-level clustered standard errors appear in parentheses. Asterisks *, **, and *** denote 0.10, 0.05, and 0.01 levels of significance, respectively. The F statistic is the test for joint significance of all coefficients. In our terminology, active NSLP suppliers are approved to sell product to the NSLP and bid for contracts. Inactive NSLP suppliers also are approved to sell product to the NSLP but do not bid on contracts in a given year. The instrumental variables used in this regression for the Active NSLP Supplier dummy are lag active NSLP supplier and lag share positive samples. The empirical results in table 5 demonstrate robust evidence that active NSLP suppliers perform better than inactive NSLP suppliers on Salmonella tests administered by FSIS. In particular, as the hypothetical tolerance standard that we use as our dependent variable grows more stringent, active NSLP suppliers were more likely to achieve the standard, relative to inactive suppliers. The estimates suggest that, holding other factors constant, the probability that a plant met the most stringent hypothetical tolerance levels was 47–49 percentage points higher if the plant actively bid on AMS contracts to supply NSLP. However, the probability of meeting less-stringent tolerance levels was not statistically significantly better for active AMS suppliers. These results are consistent with the mean comparisons shown in table 2 under both econometric specifications. Some of the other explanatory variables were statistically significant in one or more regression specifications in table 5. Plants that slaughtered cattle were more likely to meet or perform better than the 3/53 threshold and the 1/53 threshold, and meet the zero-tolerance standard under at least one specification. These results are generally consistent with expectations.13 More efficient plants, as measured by the number of pounds produced per employee, and plant size were also less likely to meet the most stringent standards—a finding consistent with the sequential logit second-stage results shown in table 4. Effects of compliance with HACCP and SSOP tasks had mixed effects on Salmonella tests, but these effects were small and mostly statistically insignificant.14 Effect of AMS Standards on Salmonella Test Performance of Active NSLP Suppliers Finally, we examined the performance of active NSLP suppliers on Salmonella tests of ground beef shipped to NSLP and tested for AMS, relative to the performance of active NSLP suppliers on random tests for Salmonella conducted by FSIS. The data include the 95 observations of active NSLP suppliers for which we have complete data and cover 2007–2012.15 Results for two specifications of probit regressions for each of three performance thresholds—0, 1, and 2 positive Salmonella samples out of 53—are presented in table 6. Table 6. Performance on Salmonella tests of Active Suppliers of Ground Beef to the National School Lunch Program (Marginal Effects), 2007–2012   2 positive Salmonella results per 53-sample set   1 positive Salmonella result per 53-sample set   0 positive Salmonella results     (1)  (2)  (1)  (2)  (1)  (2)  Product shipped to NSLP  0.276***  0.273***  0.450***  0.451***  0.216**  0.217**  (0.0709)  (0.0677)  (0.0761)  (0.0752)  (0.0980)  (0.0973)  Log (plant employees)  –0.0439  –0.0417  –0.0639  –0.0572  –0.105*  –0.112  (0.0478)  (0.0563)  (0.0577)  (0.0704)  (0.0614)  (0.0708)  Log (pounds per employee)  –0.0407  –0.0359  –0.0689  –0.0597  –0.103**  –0.113*  (0.0390)  (0.0554)  (0.0478)  (0.0704)  (0.0509)  (0.0657)  HACCP Compliance  1.33  1.19  –0.0123  0.352  5.52  5.48  (1.71)  (1.71)  (4.09)  (3.47)  (3.82)  (3.83)  Pre-operational SSOP Compliance  1.23  1.27  1.44  1.60  –0.630  –0.752  (1.24)  (1.15)  (1.35)  (1.21)  (1.82)  (1.85)  Operational SSOP Compliance  –1.88  –1.82  –2.26  –2.24  –0.975  –1.05  (1.23)  (1.29)  (1.45)  (1.60)  (2.14)  (2.21)  Post-2007  0.0969  0.100  0.00416  –0.0107  –0.0446  –0.0354  (0.187)  (0.199)  (0.164)  (0.150)  (0.248)  (0.252)  Multi-plant    0.0371    –0.0558    0.0181  (0.0713)  (0.128)  (0.137)  Age of Plant    0.000373    0.00123    –0.000981  (0.00339)  (0.00431)  (0.00398)  Pseudo R2  0.255  0.258  0.340  0.343  0.0751  0.0756  Waldχ2 p-value  0.0001  0.0000  0.0000  0.0000  0.1271  0.2326  Observations  95  95  95  95  95  95    2 positive Salmonella results per 53-sample set   1 positive Salmonella result per 53-sample set   0 positive Salmonella results     (1)  (2)  (1)  (2)  (1)  (2)  Product shipped to NSLP  0.276***  0.273***  0.450***  0.451***  0.216**  0.217**  (0.0709)  (0.0677)  (0.0761)  (0.0752)  (0.0980)  (0.0973)  Log (plant employees)  –0.0439  –0.0417  –0.0639  –0.0572  –0.105*  –0.112  (0.0478)  (0.0563)  (0.0577)  (0.0704)  (0.0614)  (0.0708)  Log (pounds per employee)  –0.0407  –0.0359  –0.0689  –0.0597  –0.103**  –0.113*  (0.0390)  (0.0554)  (0.0478)  (0.0704)  (0.0509)  (0.0657)  HACCP Compliance  1.33  1.19  –0.0123  0.352  5.52  5.48  (1.71)  (1.71)  (4.09)  (3.47)  (3.82)  (3.83)  Pre-operational SSOP Compliance  1.23  1.27  1.44  1.60  –0.630  –0.752  (1.24)  (1.15)  (1.35)  (1.21)  (1.82)  (1.85)  Operational SSOP Compliance  –1.88  –1.82  –2.26  –2.24  –0.975  –1.05  (1.23)  (1.29)  (1.45)  (1.60)  (2.14)  (2.21)  Post-2007  0.0969  0.100  0.00416  –0.0107  –0.0446  –0.0354  (0.187)  (0.199)  (0.164)  (0.150)  (0.248)  (0.252)  Multi-plant    0.0371    –0.0558    0.0181  (0.0713)  (0.128)  (0.137)  Age of Plant    0.000373    0.00123    –0.000981  (0.00339)  (0.00431)  (0.00398)  Pseudo R2  0.255  0.258  0.340  0.343  0.0751  0.0756  Waldχ2 p-value  0.0001  0.0000  0.0000  0.0000  0.1271  0.2326  Observations  95  95  95  95  95  95  Note: Plant-level clustered standard errors appear in parentheses. Asterisks *, **, and *** denote 0.10, 0.05, and 0.01 levels of significance, respectively. Our empirical results demonstrate strong evidence that the Salmonella test results for ground beef shipped to NSLP were better than the FSIS Salmonella test results for ground beef. According to the estimates given in table 6, ground beef tested for AMS and shipped to NSLP was 21–22 percentage points more likely to meet the zero-tolerance standard for Salmonella and 45 percentage points more likely to meet or perform better than the 1-positive sample Salmonella threshold, relative to FSIS tests of random samples of ground beef. In addition, larger and more efficient plants were less likely to meet the zero-tolerance standard—a finding consistent with the results shown in tables 4 and 5. Coefficients on other explanatory variables are statistically insignificant. Implications for Public Policy There are several important implications for public policy. First, our analysis shows that the zero-tolerance standard for Salmonella in ground beef sold for distribution to schools has resulted in improved food-safety performance of ground-beef suppliers. This is immensely important because school children are relatively vulnerable to food-borne illness caused by Salmonella and other pathogens (Centers for Disease Control 2017b). More broadly, the analysis demonstrates that strict standards incentivized active AMS suppliers, which already had strong performance on Salmonella tests, to improve their performance on these tests. This shows that policy objectives such as improved food-safety outcomes, can be met while at the same time using a lowest bid-price selection criteria for choosing suppliers. Our empirical findings suggest that imposing stringent standards is more effective in ensuring improved food-safety outcomes than relying on reputation or other private economic incentives without enforceable standards. Ollinger et al. (2015) found that the Salmonella test performance of active suppliers of raw chicken to the NSLP was only slightly better than the Salmonella test performance of other groups of suppliers, and only at certain tolerance levels. Ollinger et al. (2015) attributed the modestly better performance of active AMS chicken suppliers on Salmonella tests to suppliers’ concerns about reputation, because active NSLP suppliers sell chicken to a highly visible market with few suppliers, and tracing any food-borne illness outbreak to its source is relatively easy in this context. The results presented in this article, in contrast, indicate that the stringent, enforceable, zero-tolerance standard for Salmonella in ground beef imposed by AMS has been highly effective at driving plants to improve their food-safety performance, at all tolerance levels. Taken together, it appears that stringent standards were more effective than private incentives (including concerns about reputation and the cost to producers associated with food-borne illness outbreaks) in driving markedly better performance on Salmonella tests.16 The empirical results also provide some evidence that performing HACCP and SSOP tasks may have limited effects on Salmonella test outcomes, a finding consistent with Ollinger et al. (2015). There are many possible explanations; HACCP is a preventative plan, and SSOP tasks are related to daily plant operations, but neither has a direct bearing on food-safety outcomes or mandates the testing of final products. Performing well on HACCP process control and sanitation tasks may enable a plant to avoid major food-safety failures but does not allow a plant to reduce Salmonella that is already present in meat inputs, since there is no pathogen “kill step” for uncooked ground beef. Regardless of the explanation, it appears that performing HACCP process control and sanitation tasks may not be sufficient for plants to meet very strict food-safety standards. Yet, mandatory sanitation and process controls are central features of the food-safety regulation program administered by FSIS. Conclusion The Agricultural Marketing Service of the USDA annually purchases about $50 million worth of ground beef to be distributed to school systems throughout the United States as part of the National School Lunch Program. Suppliers to the NSLP must be registered with AMS and are selected via a competitive bidding process, with the contract winner chosen on the basis of cost alone, which incentivizes low-cost producers to supply the NSLP. Furthermore, ground beef shipped to the NSLP must meet a zero-tolerance standard for E. coli O157:H7 and Salmonella. This article demonstrates that, under these conditions, from 2006 to 2012, relatively large and highly productive plants (as measured by the pounds of ground beef produced per employee) registered as NSLP suppliers. Furthermore, those plants with a history of supplying the NSLP with ground beef, and with stronger performance on previous Salmonella tests, were more likely to actively bid on NSLP contracts. In two additional empirical tests, we show that (a) active NSLP suppliers performed better on random FSIS tests for Salmonella than inactive NSLP suppliers, and (b) these better-performing active NSLP suppliers shipped ground beef to the NSLP that performed even better on Salmonella tests than the product produced by the same suppliers and randomly tested by FSIS and shipped to the NSLP or other institutional or commercial buyers. Active NSLP suppliers may have attained better Salmonella test performance on product destined for the NSLP in one of two ways. First, they may have taken more precautions, such as additional sanitation or process controls, for product destined to the NSLP, and shipped product only if they were reasonably confident that it would meet the zero-tolerance Salmonella standard. Alternatively, the suppliers could have tested product prior to shipment and shipped product that met AMS standards to the NSLP while shipping product that did not meet AMS standards to a different buyer. Regardless of how plants satisfied the AMS Salmonella standard, the empirical findings demonstrate that the stringent standard imposed by AMS on its suppliers, namely, the zero-tolerance standard for Salmonella, is an effective mechanism for improving the safety of ground beef served in schools. This is especially important because AMS awards contracts to the lowest-cost bidder, which could incentivize suppliers to reduce costs by reducing effort devoted to food safety. This article contributes four main findings to the literature on food safety and food-safety standards. First, our empirical results demonstrate that enforceable standards for food safety, particularly pathogen testing with outcome monitoring, can incentivize producers to improve food safety. Second, the results suggest that plants use performance on food-safety tests to their competitive advantage when food safety can be measured, and when meeting stringent food-safety standards allows them to fill certain contracts. Third, our empirical findings show that, from 2007 to 2012, AMS attracted ground-beef suppliers that shipped ground beef to AMS with lower Salmonella levels than those same producers shipped to the commercial market. These findings are consistent with AMS requirements to accept the lowest-cost bid and receive only ground beef meeting a zero-tolerance standard for Salmonella—two requirements that generate opposing incentives to invest in food safety. Fourth, we demonstrate that compliance with SSOP and HACCP tasks does not necessarily improve food-safety outcomes, which highlights the importance of product or outcome standards relative to process standards. It is important to remember that this article examined the performance of ground-beef suppliers to the NSLP on Salmonella tests and did not examine whether better performance on Salmonella tests led to reductions in foodborne illness. AMS sets more stringent standards for the ground beef it buys for the NSLP than those required by FSIS for general commerce because children are more vulnerable to foodborne illnesses than healthy adults (Young 2005). Data from the Centers for Disease Control and Prevention (CDC) indicate that there were no foodborne illness outbreaks due to Salmonella or E. coli O157:H7 in ground beef served in schools and colleges from 2005 to 2014, and recall data from FSIS indicate that no ground-beef products shipped to schools were recalled due to Salmonella or E. coli O157:H7 over from 2004 to 2013.17 By contrast, there were 21 outbreaks of Salmonella and 58 outbreaks of E. coli O157:H7 in ground beef sold commercially during this time span.18 Finally, the results presented in this article highlight the roles of two government agencies that set standards for food safety, and the rationale for these agencies to set standards differently. AMS is authorized to set standards for food safety for the meat it buys for distribution to the NSLP and a few other food-assistance programs. In this regard, it is similar to other buyers, which consider the costs of requiring food-safety standards and the benefits to the clients and customers they serve and the buyer themselves. For AMS, which serves school children and other vulnerable groups, reducing the risk of food-borne illness is a top priority and costs may be less important. In contrast, FSIS sets requirements that apply to all meat and poultry processors engaged in interstate commerce and must consider the costs and benefits of its regulations as they apply to all producers and consumer groups, including processors that may pre-cook ground beef, healthy adults, and various groups such as the elderly and young children who may be more vulnerable to food-borne illness. Strict standards may not be necessary as a national, universal requirement; imposing them would likely require suppliers to invest more in equipment and spend more on practices to promote food safety, thereby driving up prices.19 What emerges is a market in which FSIS sets minimum standards and within which AMS and other buyers set standards for their suppliers as they deem necessary for best serving their customers and clients. Footnotes 1 Salmonella is a genus of bacteria that can cause illness in humans if consumed. Although cooking kills Salmonella, the presence of Salmonella increases the risk of food-borne illness, particularly because of cross-contamination. The FSIS reports test results for non-typhoidal Salmonella, indicating the presence of one of several species of Salmonella. For the remainder of this article, we use Salmonella as shorthand for non-typhoidal Salmonella. The incidence rate of infections caused by Salmonella was higher in 2015 for children aged 5 to 9 than for any older group, according to data from CDC (2017b). This was also true for infections caused by Shigella and Cryptosporidium, and Shiga toxin-producing E. coli O157. The incidence rate of infections caused by both Shiga toxin-producing E. coli O157 and non-O157 was particularly elevated for all age groups under 20. 2 There is also a considerable body of literature on the empirical estimation of the costs of providing safe food (see, e.g., Antle 2000 and Ollinger and Moore 2009). Researchers have also estimated the total costs and benefits of food safety regulations (Crutchfield, et al. 1998) and the effects of product recalls on stock market values (Thomsen and McKenzie 2001; Pozo and Schroeder 2016), prices of branded products (Thomsen, Shiptsova, and Hamm 2006), and demand (Marsh, Schroeder, and Mintert 2004; Piggott and Marsh 2004; Bakhtavoryan, Capps, and Salin 2014). Several studies, including Muth, Wohlgenant, and Karns (2007) and Pouliot (2014), have analyzed the regulation of food safety in an industrial-organization context. 3 Sumner, Raven, and Givney (2004) showed that new Australian regulations of food safety in meat and poultry, implemented in the 1990s, were contemporaneous with a nationwide improvement in the results of tests of animal carcasses in meat plants for bacteria, although the article did not present causal evidence for this relationship. Minor and Parrett (2016, 2017) showed that certain U.S. food-safety regulations led to a decrease in the number of detected food-borne illnesses associated with the regulated products, but did not control for the efficacy of detection of illness. 4 The AMS specifies basic technical requirements for each NSLP contract and does not consider additional product-quality attributes (such as country of origin or animal-welfare certifications) when awarding contracts. 5 See Ollinger and Mueller (2003) for further discussion. 6 The data came from the AMS website, which identifies all registered plants for the current year and gives information about recent contract awards but does not give historical information. We started collecting data in 2009 at a time when some historical data was available, and collected more data as it became available. The recorded data included whether a plant bid on at least one contract during the year, the pounds of ground beef supplied to the NSLP, and the year(s) of award(s) of any contract(s). Separate information on the website identified all registered suppliers. 7 The FSIS has since replaced the sample-set framework with continuous testing in which plants may undergo periodic testing throughout the year, and the FSIS keeps running tallies to track compliance. 8 State-inspected ground-beef plants are not directly inspected by the FSIS but must meet FSIS standards and, therefore, are eligible to bid on NSLP contracts. However, the FSIS test data were available only for FSIS-inspected plants. 9 Other options are discussed by Lewbel, Dong, and Yang (2012). The special regressor model (Lewbel 2000; Dong and Lewbel 2015; Bontemps and Nauges 2015) is also appropriate for estimating binary choice regressions with a binary endogenous variable. This model requires a special regressor that is exogenous, continuously distributed, and conditionally independent of the error term (Baum et al. 2012). These conditions precluded use in estimating equation (3) because our dataset did not have a suitable special regressor. An instrumental variable probit regression is not appropriate because it is inconsistent for binary endogenous variables (Baum et al. 2012). 10 Two well-recognized drawbacks of the two-stage linear probability pointed out by Wooldridge (2009) and others are that model estimates can fall outside the unit interval of the dependent variable and the error term is heteroskedastic because the dependent variable is not continuous. 11 The mean share of samples testing positive on FSIS Salmonella tests, across all plant-year observations, was 2.0%. 12 In all cases, the Montiel-Pflueger F statistic rejects weak instruments with τ<30% and α=0.05. 13 Ollinger and Moore (2008) found that Salmonella levels decreased with plant size in cattle-, hog-, and chicken-slaughter plants but not ground-beef plants. 14 The largest statistically significant coefficient (–6.89) indicates that plants were 0.689 percentage points less likely to meet the zero-tolerance standard if their rate of compliance with operational SSOP tasks rose by 10 percentage points. The average rate of compliance with operational SSOP tasks for all plants was 98.5%, varying from 97.4% for inactive NSLP suppliers to 98.6% for commercial-only plants. Although HACCP and SSOP tasks are designed to reduce the risk of pathogen contamination, compliance with these checklist-like tasks does not ensure improved food-safety outcomes. 15 Recall that we do not have data on Salmonella testing for shipments to NSLP before 2007. 16 Reputation effects in food safety depend on the capacity of public health authorities to detect harmful pathogens in food products. Only rarely does a case of foodborne illness become attributed to its source by public health authorities, meaning that food-safety problems only rarely lead to economic costs for producers as long as contracts do not mandate pathogen standards, and harm to reputation is the only consequence of a food-safety problem. As detection technology improves and if more resources are used for the attribution of foodborne illnesses, then attribution of illnesses to producers would likely grow and reputation effects may become stronger. 17 The FSIS reports recalls of meat and poultry products at https://www.fsis.usda.gov/wps/portal/fsis/topics/recalls-and-public-health-alerts/recall-case-archive, which indicates the terminal location of recalled products. There were no recalls of ground beef destined to schools for E. coli O157:H7 or Salmonella from 2004 to 2013. The FSIS product recalls cover a period of time inclusive of the first and last dates for which pathogen contamination is detected by the FSIS. All meat shipped during recall periods must be recalled, and thus, the recalled meat may include meat that would not test positive for pathogen contamination. 18 Data came from the authors’ use of the Centers for Disease Control and Prevention (2017a) website foodborne illness tool. 19 Recall that our empirical results indicate that inactive NSLP suppliers did not bid on contracts to supply the NSLP, in part because they had weaker performance on Salmonella tests than active NSLP suppliers, suggesting that it would cost them more to meet NSLP standards than active NSLP suppliers. References Angrist J.D., Pischke J.-S.. 2009. Mostly Harmless Econometrics: An Empiricist’s Companion . Princeton, New Jersey: Princeton University Press. Antle J.M. 2000. No Such Thing as a Free Safe Lunch: The Cost of Food Safety Regulation in the Meat Industry. American Journal of Agricultural Economics  82 2: 310– 22. http://dx.doi.org/10.1111/0002-9092.00027 Google Scholar CrossRef Search ADS   Bakhtavoryan R., Capps O.Jr, Salin V.. 2014. The Impact of Food Safety Incidents Across Brands: The Case of the Peter Pan Peanut Butter Recall. Journal of Agricultural and Applied Economics  46 4: 559– 73. Balsevich F., Berdegue J.A., Flores L., Mainville D., Reardon T.. 2003. Supermarkets and Produce Quality and Safety Standards in Latin America. American Journal of Agricultural Economics  85 5: 1147– 54. Google Scholar CrossRef Search ADS   Baum C.F., Dong Y., Lewbel A., Yang T.. 2012. Binary Choice Models with Endogenous Regressors. Stata Conference in San Diego, California. Available at: http://www.stata.com/meeting/sandiego12/materials/sd12_baum.pdf. Bontemps C., Nauges C.. 2015. The Impact of Perceptions in Averting-decision Models: An Application of the Special Regressor Method to Drinking Water Choices. American Journal of Agricultural Economics  97 3: 1– 17. Cameron A.C., Miller D.L.. 2015. A Practitioner’s Guide to Cluster-Robust Inference. The Journal of Human Resources  50 2: 317– 72. Google Scholar CrossRef Search ADS   Centers for Disease Control and Prevention. 2017a. Available at: https://wwwn.cdc.gov/foodborneoutbreaks/. Centers for Disease Control and Prevention. 2017b. FoodNet 2015 Surveillance Report (Final Data). Available at: https://www.cdc.gov/foodnet/pdfs/FoodNet-Annual-Report-2015-508c.pdf. Crutchfield S.R., Buzby J.C., Roberts T., Ollinger M., Lin C.-T.J.. 1997. An Economic Assessment of Food Safety Regulations: The New Approach to Meat and Poultry Inspection. Washington DC: U.S. Department of Agriculture, Economic Research Service, Agricultural Economic Report No. 755. Dong Y., Lewbel A.. 2015. Simple Estimators for Binary Choice Models with Endogenous Regressors. Econometric Reviews  34 ( 1–2): 82– 105. http://dx.doi.org/10.1080/07474938.2014.944470 Google Scholar CrossRef Search ADS   Henson S., Brouder A.-M., Mitullah W.. 2000. Food Safety Requirements and Food Exports from Developing Countries: The Case of Fish Exports from Kenya to the European Union. American Journal of Agricultural Economics  82 5: 1159– 69. http://dx.doi.org/10.1111/0002-9092.00115 Google Scholar CrossRef Search ADS   Henson S., Masakure O., Boselie D.. 2005. Private Food Safety and Quality Standards for Fresh Produce Exporters: The Case of Hortico Agrisystems, Zimbabwe. Food Policy  30 4: 371– 84. http://dx.doi.org/10.1016/j.foodpol.2005.06.002 Google Scholar CrossRef Search ADS   Henson S., Humphrey J.. 2009. The Impacts of Private Food Safety Standards on the Food Chain and on Public Standard-Setting Processes. Paper prepared for FAO/WHO. Available at: http://www.fao.org/3/a-i1132e.pdf. Hou M.A., Grazia C., Malorgio G.. 2015. Food Safety Standards and International Supply Chain Organization: A Case Study of the Moroccan Fruit and Vegetable Exports. Food Control  55: 190– 9. Google Scholar CrossRef Search ADS   Leirisalo-Repo M., Helenius P., Hannu T., Lehtinen A., Kreula J., Taavitsainen M., Koskimies S.. 1997. Long Term Prognosis of Reactive Salmonella Arthritis. Annals of the Rheumatic Diseases  56 9: 516– 20. Google Scholar CrossRef Search ADS PubMed  Lemeilleur S. 2013. Smallholder Compliance with Private Standard Certification: The Case of GlobalGAP Adoption by Mango Producers in Peru. International Food and Agribusiness Management Review  16 4: 159– 80. Lewbel A. 2000. Semiparametric Qualitative Response Model Estimation with Unknown Heteroscedasticity or Instrumental Variables. Journal of Econometrics  97 1: 145– 77. http://dx.doi.org/10.1016/S0304-4076(00)00015-4 Google Scholar CrossRef Search ADS   Lewbel A., Dong Y., Yang T.T.. 2012. Comparing Features of Convenient Estimators for Binary Choice Models with Endogenous Regressors. Canadian Journal of Economics  45 3: 809– 29. http://dx.doi.org/10.1111/j.1540-5982.2012.01733.x Google Scholar CrossRef Search ADS   MacDonald J., Ollinger M.. 2005. Technology, Labor Wars, and Producer Dynamics: Explaining Consolidation in Beefpacking. American Journal of Agricultural Economics  87 4: 1020– 33. http://dx.doi.org/10.1111/j.1467-8276.2005.00785.x Google Scholar CrossRef Search ADS   Marsh T.L., Schroeder T.C., Mintert J.. 2004. Impacts of Meat Recalls on Consumer Demand in the USA. Applied Economics  36 9: 897– 909. http://dx.doi.org/10.1080/0003684042000233113 Google Scholar CrossRef Search ADS   Minor T., Parrett M.. 2016. A Retrospective Review of the Economic Impact of the Food and Drug Administration’s Proposed Egg Rule. Agricultural Economics  47 4: 457– 64. Google Scholar CrossRef Search ADS   Minor T., Parrett M.. 2017. The Economic Impact of the Food and Drug Administration’s Final Juice HACCP Rule. Food Policy  68: 206– 13. Google Scholar CrossRef Search ADS   Montiel Olea J.L., Pflueger C.E.. 2013. A Robust Test for Weak Instruments. Journal of Business and Economic Statistics  31 3: 358– 69. http://dx.doi.org/10.1080/00401706.2013.806694 Google Scholar CrossRef Search ADS   Muth M.K., Fahimi M., Karns S.A.. 2009. Analysis of Salmonella Control Performance in U.S. Young Chicken Slaughter and Pork Slaughter Establishments. Journal of Food Protection  72 1: 6– 13. http://dx.doi.org/10.4315/0362-028X-72.1.6 Google Scholar CrossRef Search ADS PubMed  Muth M., Wohlgenant M.K., Karns S.. 2007. Did the Pathogen Reduction and Hazard Analysis and Critical Control Points Regulation Cause Slaughter Plants to Exit? Review of Agricultural Economics  29 3: 596– 611. Google Scholar CrossRef Search ADS   Ollinger M., Bovay J., Benicio C., Guthrie J.. 2015. Economic Incentives to Supply Safe Chicken to the National School Lunch Program. Economic Research Report Number 202. Washington DC: U.S. Department of Agriculture, Economic Research Service. Available at: https://www.ers.usda.gov/webdocs/publications/45500/55537_err202.pdf?v=42332. Ollinger M., Guthrie J., Bovay J.. 2014. The Food Safety Performance of Ground Beef Suppliers to the National School Lunch Program. Washington DC: Department of Agriculture, Economic Research Service, Economic Research Report No. 180. Available at: https://www.ers.usda.gov/webdocs/publications/45326/50382_err180.pdf?v=41996. Ollinger M., Moore D.. 2008. The Economic Forces Driving Food Safety Quality in Meat and Poultry. Review of Agricultural Economics  30 2: 289– 310. http://dx.doi.org/10.1111/j.1467-9353.2008.00405.x Google Scholar CrossRef Search ADS   Ollinger M., Moore D.. 2009. The Direct and Indirect Costs of Food Safety Regulation. Review of Agricultural Economics  31 2: 247– 65. http://dx.doi.org/10.1111/j.1467-9353.2009.01436.x Google Scholar CrossRef Search ADS   Ollinger M., Moore D., Chandran R.. 2004. Meat and Poultry Plants’ Food Safety Investments: Survey Findings. Washington DC: U.S. Department of Agriculture, Economic Research Service, Technical Bulletin 1911. Available at: https://www.ers.usda.gov/webdocs/publications/47486/17469_tb1911.pdf?v=41029. Ollinger M., Mueller V.. 2003. Managing for Safer Food: The Economics of Sanitation and Process Controls in Meat and Poultry Plants. Washington DC: U.S. Department of Agriculture, Economic Research Service, AER-817. Available at: https://www.ers.usda.gov/webdocs/publications/41496/18901_aer817.pdf?v=41063. Piggott N.E., Marsh T.L.. 2004. Does Food Safety Information Impact U.S. Meat Demand? American Journal of Agricultural Economics  86 1: 154– 74. http://dx.doi.org/10.1111/j.0092-5853.2004.00569.x Google Scholar CrossRef Search ADS   Pouliot S., 2014. The Production of Safe Food According to Firm Size and Regulatory Exemption: Application to FSMA. Agribusiness  30 4: 493– 512. Google Scholar CrossRef Search ADS   Pozo V.F., Schroeder T.C.. 2016. Evaluating the Costs of Meat and Poultry Recalls to Food Firms Using Stock Returns. Food Policy  59: 66– 77. Google Scholar CrossRef Search ADS   Sumner J., Raven G., Givney R.. 2004. Have Changes to Meat and Poultry Food Safety Regulation in Australia Affected the Prevalence of Salmonella or of Salmonellosis? International Journal of Food Microbiology  92 2: 199– 205. Google Scholar CrossRef Search ADS PubMed  Thomsen M.R., McKenzie A.M.. 2001. Market Incentives for Safe Foods: An Examination of Shareholder Losses from Meat and Poultry Recalls. American Journal of Agricultural Economics  83 3: 526– 38. http://dx.doi.org/10.1111/0002-9092.00175 Google Scholar CrossRef Search ADS   Thomsen M.R., Shiptsova R., Hamm S.J.. 2006. Sales Responses to Recalls for Listeria monocytogenes: Evidence from Branded Ready-to-Eat Meats. Review of Agricultural Economics  28 4: 482– 93. http://dx.doi.org/10.1111/j.1467-9353.2006.00317.x Google Scholar CrossRef Search ADS   U.S. Department of Agriculture, Agricultural Marketing Service. Commodity Areas, Commodity Purchasing, Solicitations and Awards, Red Meat and Fish or Poultry and Eggs. 2009–2014. Available at: http://www.ams.usda.gov/AMSv1.0/ams.fetchTemplateData.do?template=TemplateJ&page=CPDCommodityPurchaseMeatFish (Accessed 2009–2014, link no longer valid). U.S. Department of Agriculture, Agricultural Marketing Service. 2012. Supplement 211 to AMS Master Solicitation: Purchase of Frozen Ground Beef Products for Distribution to Child Nutrition and Other Federal Food and Nutrition Programs. Available at: https://www.ams.usda.gov/sites/default/files/media/Supplement%20211%20Frozen%20Beef%20Products%20June%202012%20%28Oct%202012%29.pdf. Van Ophem H., Schram A.. 1997. Sequential and Multinomial Logit: A Nested Model. Empirical Economics  22 1: 131– 52. http://dx.doi.org/10.1007/BF01188174 Google Scholar CrossRef Search ADS   Von Schlippenbach V., Teichmann I.. 2012. The Strategic Use of Private Quality Standards in Food Supply Chains. American Journal of Agricultural Economics   94 5: 1189– 201. Google Scholar CrossRef Search ADS   Wooldridge J.M. 2009. Introductory Econometrics: A Modern Approach . Fourth Edition Mason, Ohio: South-Western Cengage Learning. Young R.W. 2005. Agricultural Marketing Service Management Controls to Ensure Compliance with Purchase Specification Requirements for Ground Beef. Washington DC, U.S. Department of Agriculture. Available at: http://i.usatoday.net/news/pdf/ams-standards.pdf. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association 2017. This work is written by US Government employees and is in the public domain in the US.

Journal

American Journal of Agricultural EconomicsOxford University Press

Published: Mar 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 12 million articles from more than
10,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Unlimited reading

Read as many articles as you need. Full articles with original layout, charts and figures. Read online, from anywhere.

Stay up to date

Keep up with your field with Personalized Recommendations and Follow Journals to get automatic updates.

Organize your research

It’s easy to organize your research with our built-in tools.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve Freelancer

DeepDyve Pro

Price
FREE
$49/month

$360/year
Save searches from
Google Scholar,
PubMed
Create lists to
organize your research
Export lists, citations
Read DeepDyve articles
Abstract access only
Unlimited access to over
18 million full-text articles
Print
20 pages/month
PDF Discount
20% off