The Impact of Market Mechanisms and HACCP Regulation on Food Safety Quality. Michael Ollinger

Size: px
Start display at page:

Download "The Impact of Market Mechanisms and HACCP Regulation on Food Safety Quality. Michael Ollinger"

Transcription

1 The Impact of Market Mechansms and HACCP Regulaton on Food Safety Qualty. Mchael Ollnger Mchael Ollnger s an economst at the Economc Research Servce of the Unted States Department of Agrculture at 1800 M Street, NW Room 3064, Washngton DC, (202) (phone), (202) (fax), and E-mal ollnger@ers.usda.gov. Selected paper prepared for presentaton at the Amercan Agrcultural Economcs Assocaton Annual Meetng, Denver, Colorado, August 1-4, Copyrght 2003 s the property of the U.S. Department of Agrculture. All rghts reserved. Readers may make verbatm copes of ths document for non-commercal purposes by any means, provded that ths copyrght notce appears on all such copes. * The judgments and conclusons heren are those of the author and do not necessarly reflect those of the U.S. Department of Agrculture. The author s responsble for any errors. 1

2 ABSTRACT The Impact of Market Mechansms and HACCP Regulaton on Food Safety Qualty. Economsts have long debated the relatve effectveness of markets and regulatons n reachng socally desrable outcomes. Ths emprcal study of meat and poultry food safety regulaton suggests that market mechansms and flexble regulatory nstruments, e.g. HACCP systems, have a greater mpact on food safety qualty than less flexble regulatory nstruments. Subject code: 9 (Food Safety and Nutrton). 2

3 The Impact of Market Mechansms and HACCP Regulaton on Food Safety Qualty. Economsts have long debated the relatve effectveness of market mechansms and regulatory nstruments n reachng socally desrable outcomes. Antle (1996) and many other economsts reluctantly argue that food safety regulaton may be justfed because buyers cannot drectly evaluate pathogen levels before or after purchase and thrd partes cannot unambguously certfy qualty because the costs of acqurng pathogen nformaton s prohbtve. In the absence of these tradtonal market mechansms, large restaurant and grocery store chans have mposed ther own food safety qualty control programs on supplers (Ollnger and Mueller, 2003). These buyers have strong ncentves to mpose standards because they can spread the assocated costs over mllons of pounds of product sales and an outbreak of a foodborne llness attrbuted to them could cause them to lose ther reputatons, resultng n a catastrophc drop n profts. Large buyers that are able to mpose sophstcated qualty control programs on supplers account for only a small fracton of the meat and poultry market and even they may not know the pathogen levels of the meat products they sell. Snce buyers that do not requre pathogen testng and some that do requre t do not know the pathogen levels of the meat they buy, food safety qualty remans an mportant publc health concern of the U.S. government and has led the Food Safety Inspecton Servce (FSIS) of USDA to promulgate the Pathogen Reducton Hazard Analyss Crtcal Control Pont Regulaton (PR/HACCP) rule n Ths regulaton mandates that all plants must establsh and mplement standard santaton operatng procedures (SSOPs) and a HACCP process control program, slaughter plants must adhere to generc E. col standards, and slaughter and ground meat plants must comply wth Salmonella standards. The SSOPs are cleanng, santaton, and process control tasks that plants must perform n order to be n 1

4 complance wth FSIS regulatons. Plants develop and mplement ther own HACCP programs but are subject to FSIS oversght. The SSOPs mandated n the PR/HACCP rule reflected a contnuaton of a prevous regulatory regme of process standards n whch FSIS mandated and montored specfc tasks. If plants completed the tasks, then they were n regulatory complance. Under the HACCP programs, FSIS took a dfferent approach n that plants specfed and mplemented ther own program. FSIS, however, ensured that plants compled wth ther HACCP plans. Antle (2001) classfed SSOPs and HACCP programs as desgn standards because they requre certan tasks to be performed. Pontng out that HACCP programs are much more flexble than SSOPs snce they are talored to meet the unque crcumstances of each plant, Coglanese and Lazer (2002) label HACCP programs as a management-based regulaton. The generc E. col and Salmonella standards mandated under PR/HACCP marked a sgnfcant departure from pror regulatory practces at FSIS. Under these performance standards, plants must meet mcrobologcal tolerances. As long as the tests ndcate that the plant s wthn tolerances, then t s consdered to be n regulatory complance. Plants could use whatever approach they deemed necessary to meet the standard. Antle (2001) and many other economsts favor performance standards over desgn standards and management-based regulatons because, due to ther flexblty, they are consdered to be a less costly way to comply wth a regulaton. In a smlar way, the flexblty offered by a management based regulaton should make ths preferable to a strcter desgn standard that nhbts nnovaton. The purpose of ths paper s to examne the mpacts of large buyers, process regulatory standards, and management-based regulatory standards on food safety performance. We use the number of postve Salmonella test samples as a share of all samples examned by FSIS as a 2

5 measure of food safety performance. The analyss requres the use of unque datasets from FSIS and the Economc Research Servce of USDA. The FSIS datasets provde nformaton about the plant-level number of meat and poultry samples testng postve for Salmonella, the number of samples tested for Salmonella, and plant performance of SSOP and HACCP tasks. FSIS data also contans nformaton about anmals slaughtered and some producton data. The ERS data provdes nformaton on plant-level food safety technologes, whether plants have large customers that nfluence ther producton practces, and plant characterstcs. Meat and Poultry Food Safety and the Regulatory Envronment FSIS promulgated the fnal PR/HACCP rule on July 25, The prncpal element of the PR/HACCP rule was the requrement to develop and mplement HACCP process control programs for each FSIS-defned product produced by the plant. HACCP programs are based on these seven crtera: (1) assessng all hazards, (2) fndng all crtcal ponts, (3) settng crtcal lmts for each crtcal control pont (CCP), (4) developng procedures to montor each CCP, (5) determnng correctve actons, (6) mplementng a recordkeepng system, and (7) establshng verfcaton procedures (Unnevehr and Jensen,1996). There were other mportant components of the PR/HACCP rule. The regulaton requred meat and poultry establshments to develop, mplement, and take responsblty for performng SSOPs. Second, t mandated that slaughter plants conduct generc E. col mcrobal tests n order to verfy that fecal contamnaton was under control. Fnally, n order to verfy that ther HACCP systems were controllng pathogens, the PR/HACCP rule establshed Salmonella performance standards for slaughter and ground meat and poultry plants. 3

6 PR/HACCP was phased n over several years. All plants had to comply wth the HACCP requrement by January 31, Large plants were requred to mplement ther programs by January 31, 1998, followed by small and very small plants over the next two years. All plants had to have ther SSOPs n place by January of Pathogen testng marked a sharp departure from prevous regulatory practces at FSIS because t requred plants to meet standards wthout specfyng how to reach the objectve. Under the generc E. col standard, plant testng requrements depended on producton volume. Cattle slaughter plants, for example, have to take one sample per 300 carcasses, whle broler plants are requred to take one sample per 22,000 brds. Plants falng to meet the generc E. col standard had to dscover and correct the cause of the falure or face ncreased FSIS scrutny of facltes, products, and plant complance wth ther HACCP plan and SSOPs. FSIS may also perform more product testng. If plant performance s deemed unsatsfactory, FSIS can remove ts nspectors. FSIS uses Salmonella tests as a measure of overall plant food safety process control, and can deem a falure to meet the standard as one of the bases for declarng a product to be adulterated. Under ths program, FSIS randomly selects plants for testng and then tests plants products (ground meat or carcasses) for Salmonella over a seres of days. A plant that exceeds the maxmum allowed number of postve Salmonella samples must complete a second round of tests after t modfes ts food safety process control program. If the plant fals that round, then, t agan modfes ts process control program and undergoes another round of testng. Falure to pass on the thrd attempt consttutes falure to mantan santary condtons and mantan an adequate HACCP plan and wll cause FSIS to suspend nspecton servces. The suspenson 4

7 remans n effect untl the plant submts a detaled acton plan to correct the HACCP plan and outlnes the other measures taken by the plant to reduce the prevalence of pathogens. It has been rare for plants to fal Salmonella complance testng. Only about 100 out of the approxmately 2,050 slaughter and grndng plants tested up to 1999 faled to pass the frst test and only 22 of these 100 plants faled ther frst two tests. Falure to comply after two tests would have led to ncreased enforcement revew, but 19 of these plants passed the thrd test and contnued producton. These 19 plants ncluded 1 for ground turkey and 7 for ground beef, and 4 hog slaughter, 6 broler slaughter and 2 cow and bull slaughter plants. Supreme Beef and one other ground beef plant and one cow/bull slaughter plant faled three tests, and FSIS suspended them, meanng that plants retaned the rght to nspecton servces f the suspenson was lfted. Economc Framework We hypothesze that plant characterstcs, food safety technology, performance of regulatory tasks, and the nfluence of large buyers affect Salmonella test performance. Plant characterstcs descrbe the sze, scope, and complexty of a plant s processng technology. Food safety technologes nclude the equpment and procedures desgned and adopted by plants for the purpose of controllng pathogens. These technologes ether kll pathogens drectly by heat or chemcal treatments or reduce the opportunty for product contamnaton through changes n operatons or more ntensve cleanng. Plants are requred to comply wth Standard Santaton Operatng Procedures (SSOPs), a Hazard Analyss Crtcal Control Pont (HACCP) program, and mantan facltes accordng to FSIS specfcatons. Plants wrte ther own SSOP and HACCP plans and are expected to perform all the functons specfed n ther plans. FSIS montors plants and notes when plants 5

8 fal to comply. Snce the SSOPs and HACCP plans specfy practces that are desgned to control pathogens, devatons from the plan may result n a hgher ncdence of pathogens n meat and poultry products. Major buyers of a company s products exert control over producton practces n a varety of ways rangng from specfyng technologes to refusng orders that may be of questonable qualty. We express food safety performance as functon of plant characterstcs (CHAR), food safety technology (FSTECH), regulatory complance (REG), and major buyers on food safety performance (MARKET) n equaton 1. PERF = (CHAR, FSTECH, REG, MARKET) (1) where PERF s a number from zero to one ndcatng the share of samples testng postve for Salmonella n FSIS Salmonella testng program descrbed above. Emprcal Specfcaton The dependent varable, PERF, s the number of postve test samples as a share of all test samples, makng t an ndex bounded below by zero (plants wth no samples testng postve for Salmonella) and bounded above by one (plants wth all samples testng postve for Salmonella). In practce, many plants had no reported postve cases of Salmonella but no plant had only Salmonella postve cases, makng the real dstrbuton bounded from below but not from above. For purposes of the analyss we assume that PERF has a normal dstrbuton centered around and truncated at zero, meanng that, f t were possble, some values would be less than zero. Plants that fall n ths truncated regon are those plants that are partcularly concerned 6

9 about ther reputatons for food safety and want to ensure that none of ther products are contamnated. Tobn (1958) was the frst to consder regressons wth censored dependent varables. He specfed a dependent varable wth a dstrbuton centered at zero that contaned a theoretcally possble latent varable (y*). Greene (1993) gves the followng general formulaton of the censored regresson model, also known as the Tobt model: y = β ' + ε * x (2) * y = 0 f y 0 * y > 0 f y > 0 Applyng equaton 2 to equaton 1 means that the dstrbuton for y * s the percent samples testng postve for Salmonella. A postve coeffcent on an ndependent varable (a plant characterstcs, technology, regulatory, or market varable n equaton 1) means that the ndependent varable (x ) encourages postve Salmonella test samples and a negatve value means that x decreases the lkelhood of a postve Salmonella test sample. Margnal effects, as llustrated n equaton 3, ndcate by how much a change n an ndependent varable affects the lkelhood of testng postve for Salmonella. E [ y * x ]/ x = β (3) Typcally, margnal effects, whch show how margnal changes n the ndependent varables affect the dependent varable, can be obtaned by takng a dervatve of the regresson equaton, as shown n equaton 3. However, snce Salmonella test performance occurs only n the postve porton of the dstrbuton, an alternatve specfcaton, equaton 6, s requred. As always, the sgn and sze of the coeffcent ndcates how much the ndependent varable affects the dependent varable. Greene s (1993) dervaton of the margnal effects follows: 7

10 E y x ] = Φ( β ' x / σ )( β ' x + σλ ), (4) [ where and the margnal effect of the ndependent varables on y s: λ = φ( β ' x / σ ) / Φ( β ' x / σ ), (5) E[ y x ]/ x = βφ( β ' x / σ ). (6) Note, σ s the standard devaton, φ s the probablty densty functon of the standard normal dstrbuton, and Φ s the cumulatve densty functon of the standard normal dstrbuton. Below, we emprcally represent our model. PERF g = q h= p = β + γ CHAR + φ FSTECH + α REG + χ MARKET + ε 0 g g h h j j k k g = 1 h= 1 j= 1 k = 1 j n k = m (7) where the varables have been defned above and PERF (share of samples testng postve for Salmonella) = 0, f PERF * 0, PERF (share of samples testng postve for Salmonella) > 0, f PERF * > 0. The mean of PERF *, a theoretcally normally dstrbuted dependent varable, s less than the mean of PERF because PERF cannot be less than zero. The ndependent varables represented by CHAR are plant characterstcs, such as plant sze and whether the plant slaughters anmals. The mpact of these varables on the share of samples testng postve for Salmonella depends on whether the characterstc adds to the complexty of plant operatons or encourages food safety nvestments. Greater complexty due to an ncrease n plant operatng procedures makes the control of pathogens more costly due to hgher management costs. Greater economes of scale, on the other hand, may lower the cost of food safety nvestments. Larger 8

11 sze plants, for example, can reduce per unt food safety costs by spreadng expendtures for new food safety equpment over a larger producton volume. FSTECH represents fve types of food safety technologes: santaton procedures (not SSOPs), operatng practces, product and envronmental testng, equpment, and, for cattle slaughter plants, dehdng practces. A negatve relatonshp should exst f a technology s effectve n reducng the number of postve Salmonella samples. REG s the nfluence of varous FSIS regulatory nstruments, e.g. SSOPs and HACCP plans. Better performance of SSOPs or a plant s HACCP program should lead to fewer postve Salmonella test samples. MARKET s a dummy varable for dfferent types of market mechansms k. These mechansms nclude relatonshps between plants and large buyers, lke McDonalds, lnkages between plants and export markets, and whether the plant produces a product under a brand name. MARKET should negatvely affect the number of postve Salmonella test samples. Varable Defntons We defned the dependent varable, PERF, above. There are several plant characterstcs varables. Large plants may be able to acheve economes of scale n food safety technology, makng t easer for them to control pathogens, but they are also more complex than small plants, causng greater dffculty n controllng pathogens (Havelaar, Nauta, and Jansen, 2004). As a measure of plant sze, we use the log of pounds of plant output, Log (POUNDS). We also control for mult-speces plants because they are more complex than sngle-speces plants, perhaps, makng them more dffcult to manage and less able to control pathogens. Emprcally, 9

12 we use a dummy varable ndcatng whether the plant slaughters more than one anmal speces, e.g. cattle and hogs, (MULTI-SPECIES). Plants that slaughter anmals and grnd and cook meat or poultry are much more complex than sngle process plants; thus, we controlled for them. We are uncertan of the sgns because they may have offsettng ncentves. For example, a plant that both slaughters anmals and grnds meat may be more cautous about controllng pathogens than a plant that smply slaughters anmals because grnders may get feedback drectly from consumers about food safety qualty whle a carcass producer may not. To control for grndng operatons n a slaughter plant, we defne GROUND as a dummy varable equal to one f the plant grnds meat and zero otherwse. For plants that cook or further process meat or poultry, we defne COOKED as dummy varable equal to one f the plant cooks or further processes meat or poultry and zero otherwse. Fnally, for ground beef plants we defne SLAUGHTER as a dummy varable equal to one f the plant slaughters anmals and zero otherwse. The food safety technology varables for cattle slaughter are DEHIDE_TECH, an ndex value between zero and one descrbng the sophstcaton of plant dehdng technology, DEHIDE_KNIFE, a dummy varable ndcatng whether the plant santze knves or rotate knves after slaughterng each anmal. One of the food safety technology varables for brolers s dummy varable for whether the plant uses a modern evscerator (EVISCERATOR). One of the food safety technology varables for hogs s STEAM_VAC (a dummy varable equal one for plants that use steam vacuum machnes for hogs). For broler, hogs, and ground beef an addtonal food safety technology varable s a dummy varable for plants wth plant food safety operatng procedures rated to be n the 90 th percentle on an ndex ratng the ntensty of food safety producton practces (see the data secton for further dscusson). A fnal food safety technology varable for poultry and hog slaughter s a dummy varable ndcatng whether a plant 10

13 removes the anmal from feed pror to slaughter (NOFEED). A regulatory varable for all plants s the number of nspected HACCP tasks not n complance wth HACCP plan requrements as a share of all HACCP tasks (HACCP_NOCOMPLY). For all ndustres except cattle slaughter, another regulatory varable s the number of SSOP tasks not n complance wth the SSOP plan as a share of all SSOP tasks (SSOP_NOCOMPLY). The market nfluence varables for poultry and hog slaughter and ground beef (CUSTOMER_SPECS) s a dummy varable equal to one for plants subject to requrements from a domestc customers and zero otherwse whle the market nfluence varable for cattle slaughter s one for plants that export and zero otherwse (EXPORT). We ncluded only one market mechansm varable n each regresson. The choce depended on whch type of market would lkely have the greatest nfluence on producton practces. Export markets may be more mportant for cattle slaughter plants than for the other plants because cattle carcasses are far removed from consumers and beef exports amount to about 10 percent of output. Both hog slaughter and broler plants have extensve further processng operatons wthn ther plants, thus, these plants and ground meat producers sell most of ther products to domestc buyers that sell to consumers. The large domestc buyers have substantal nfluence over ther supplers due to ther large volume purchases. Data Data comes from the Food Safety Inspecton Servce (FSIS) and Economc Research Servce (ERS). One of the FSIS datasets provdes data on the two peces of nformaton needed to 11

14 compute the share of samples testng postve for Salmonella. Among other data, ths dataset ncludes the total number of test samples and the number of test samples found to be postve for Salmonella by FSIS personnel n randomly selected plants n the year For cattle slaughter, we also used 2001 data because FSIS tested only 40 plants producng carcasses n Another of the FSIS datasets, also obtaned n a personal communcaton, contans data on the number of SSOP and HACCP tasks out of complance wth the plant SSOP or HACCP plan. Each plant nspected by FSIS s requred to have SSOP and HACCP plans that specfy certan key ponts that should be montored and mantaned. Snce the number of SSOP and HACCP tasks vares by plant, we dvded by the number of tasks not n complance wth SSOP and HACCP standards by total number of tasks to create varables for the shares of tasks out of complance wth SSOP and HACCP requrements. The FSIS data also contans the number of faclty requrements establshed by FSIS that are out of complance wth FSIS specfcatons. However, ths varable was later dropped because t had lttle mpact on Salmonella test performance. Some observers of FSIS nspecton actvtes beleve that varance exsts n the way nspectors measure the performance of food safety tasks. Whle ths may be the case, we have no reason to beleve that there s a systematc bas n these data. Thus, t appears unlkely that random reportng dfferences wll affect statstcal results. The fnal FSIS dataset used n the analyss s the Enhanced Facltes Database (EFD) for Ths dataset contans detaled nformaton on the numbers and types of anmals slaughtered, SIC codes, pounds of meat or poultry produced, whether a plant produced meat or poultry, and categorcal data on process types for each plant nspected by FSIS. 12

15 The Economc Research Servce has a unque dataset contanng nformaton on plant characterstcs, market mechansms and food safety technologes. The data were obtaned n a survey contanng approxmately 40 questons dealng wth food safety technology, 15 questons on aspects of the costs of the PR/HACCP regulaton, varous plant characterstcs, and the types of markets plants serve. The 40 food safety responses were used to create fve types of food safety ndces: food safety equpment, food safety tests, dehdng, santaton, and food safety operatng practces. The ndex values more ntensve food safety actvtes hgher than less ntensve ones. Thus, a plant wth the hghest food safety equpment ratng (one) would have all the types of equpment dealt wth by the survey whle a plant wth a zero ratng would have none of the equpment. Plants wth a partal number of peces of equpment would have ntermedate values. See Ollnger et al, (2004) for a complete descrpton of how the ndex was created and about the ERS survey. Results We ran separate regressons for the beef and hog carcasses, brolers, and ground beef products tested by FSIS. We used only year 2000 data because the ERS and one FSIS dataset contaned data only from that year. The other FSIS datasets contanng regulatory and performance nformaton had multple years. We had to combne year 2000 data wth 2001 data for cattle slaughter because we only had 40 observatons n the year Tables 1-3 contan the parameter estmates for margnal effects of plant characterstcs, food safety technology, regulatory effort and market mechansms of food safety performance,.e. the share of meat or poultry samples testng postve for Salmonella. All models were adjusted 13

16 for multplcatve heteroskedastcty wth the followng specfcaton: σ 2 = σ 2 exp (γ Z) where Z s a vector of varables that affect the dsturbance term, σ. In our case, Z = Log of OUTPUT and γ = [ln σ 2, β]. A lkelhood test shows that the heteroskedastc correcton s sgnfcant for each model. Regresson results show that the jont lkelhood of the model (equaton 7) s sgnfcant n each case. Plant sze s sgnfcant and postve n hog slaughter and ground beef and sgnfcant and negatve for cattle slaughter. These dfferences are lkely due to the range of food safety technologes avalable for dfferent types of processng operatons. Cattle slaughter plants have a varety of technologes that control pathogens ether through heat or chemcal solutons. These technologes become less costly per unt of output as plant sze ncreases. Ground beef plants, on the other hand, have few food safety technologes other than santaton and operatng procedures to control pathogens. Snce t s lkely that smaller economes of scale occur n santaton and operatng procedures than n equpment, t may be that economes of scale enable large cattle slaughter to produce carcasses wth lower pathogen counts than ther smaller slaughter plant counterparts but such economes of scale do not exst n ground beef. Results show that mult-speces poultry and hog slaughter and ground beef plants are more lkely to have meat or poultry samples test postve for Salmonella, perhaps because Salmonella grow readly and greater complexty makes food safety more dffcult to control. Cattle slaughter plants that grnd meat and ground beef plants that slaughter anmals have a lower lkelhood of havng samples test postve for Salmonella, perhaps, because cattle slaughter plants wth grndng operatons take extra precautons to avod pathogen contamnaton whle poultry slaughter plants do not. Why mght cattle slaughter plants be more cautous about contamnatng ground meat 14

17 than poultry slaughter plants? Ground beef s used to make hamburgers and a wde range of other products that may be eaten wthout reachng the temperature needed to kll pathogens, makng contamnated ground beef a potental source for severe gastrontestnal problems and a cause for numerous product recalls that have cost the ndustry mllons of dollars. Ground poultry, on the other hand, s rarely eaten partally cooked and rarely recalled. The other plant characterstc, whether the plant produces cooked products, was not sgnfcant. There were also other characterstcs that we consdered, ncludng whether a cattle slaughter plant butchered cows and bulls, but they were not sgnfcant and were not ncluded n our model. Numerous food safety technology varables were examned, but only a few were sgnfcant or nearly sgnfcant. Even though cattle slaughter has a wde array of types of food safety equpment, none of them sgnfcantly affected postve Salmonella test results and only dehdng technology had much of an mpact. Even wthn dehdng, only knfe santaton and, to a lesser extent, hand and glove santaton (not shown) had a measurable mpact on Salmonella. Poultry slaughter plants usng an evscerator that completely removes the vscera from the brd and hog slaughter plants employng steam vacuum machnes to control pathogens after slaughter had fewer samples testng postve for Salmonella. Food safety technology ndexes were created for food safety equpment, testng, plant operatons, dehdng (cattle slaughter only), santaton, and overall (see data secton for dscusson). Only hog slaughter and ground beef plants wth a food safety operatng procedures ndex value n the 90 th percentle had sgnfcantly fewer samples testng postve for Salmonella. The plant operatons ndex ncludes responses to questons dealng wth how the plant handles acceptable product that needs to be reworked, employees tranng, product refrgeraton, and 15

18 others. Results also show that a hgher ndex value for dehdng technology reduced the number of samples testng postve for Salmonella. The dehdng ndex ncludes responses to questons dealng wth knfe and glove santaton, whether ar vacuums are used at the kll and hde removal staton, etc. Results show that equpment, testng, and santaton technologes play lttle role n enhancng food safety and that plants must perform multple operatng tasks to beneft from mproved food safety. Judgng from these results, t appears that technology plays a small role n controllng Salmonella. Why mght ths be the case? It appears that plants are choosng food safety technologes that enable them reach ther food safety goals. Plants that use fewer new technologes are lkely the ones that need them the least and plants that use more have the greatest dffculty n controllng pathogens (Haavalar, Nauta, and Jansen). Thus, plants wth a more technologcally sophstcated faclty lkely have a greater need for the technology. Poultry and hog plants often remove anmals from feed pror to slaughter n order to reduce fecal matter. Results show that broler plants removng anmals from feed had fewer samples testng postve for Salmonella whle hog plants had more samples testng postve. The PR/HACCP rule contans three types of standards: process standards n whch the regulatng agency (FSIS) specfes tasks that must be performed to be n complance, management based regulatons (HACCP plans) that are establshed by the plant n consultaton wth the regulatng agency (FSIS), and performance standards n whch the regulatng agency (FSIS) specfes a food safety performance target (a maxmum number of samples testng postve for Salmonella). Regulators establsh process standards and requre management-based regulatons n order to mprove plant regulatory performance (reduce pathogens n plants). Thus, both the process and management-based regulatons should reduce the number of samples testng 16

19 postve for Salmonella. Results show that as the share of HACCP tasks not n complance wth management plans rse, the number of samples testng postve for Salmonella also rses n each of the ndustres and sgnfcantly so n brolers and cattle slaughter. Tables 1-3 also show that the number of samples testng postve for Salmonella rses as the share of tasks not n complance wth SSOPs rses n ground beef and hog slaughter. A sgnfcant relatonshp exsts n ground beef. The tables also show that a negatve relatonshp exsts between SSOPs and the number of samples testng postve for Salmonella n poultry and cattle slaughter. It appears that ths s due to collnearty snce the relatonshp becomes postve f the varable for HACCP noncomplance s removed. Overall, the results suggest that both process-based and management-based regulatons mprove food safety performance. Notce the coeffcents on the share of HACCP tasks not n complance wth regulatons s larger than the coeffcents on the share of SSOP task not n complance n all cases, even ground beef. These fndngs suggest that the management-based regulatory approach (HACCP plans) have a greater mpact on food safety performance than the regulator-based process standard (SSOPs). A 10 percent decrease n noncomplant HACCP tasks would reduce postve Salmonella test samples by about 5 percent n all ndustres except cattle slaughter, n whch case the reducton would be about 3 percent. By contrast, the changes n noncomplant SSOP tasks vares from a decrease of about 4 percent n ground beef to an ncrease of about 1 percent n cattle slaughter. Market mechansms (customer specfcatons and export markets) negatvely affect the share of samples testng postve for Salmonella n all ndustres except ground beef, whch has a very small postve relatonshp. The relatonshp s sgnfcant and negatve n brolers and 17

20 cattle slaughter. The mpact of market mechansms n brolers s about a 6 percent reducton n postve Salmonella test samples and n cattle slaughter t s about 1 percent. For hog carcasses, there s about a 3 percent reducton. These results suggest that prvate markets play a vtal role n reducng harmful pathogens n meat and poultry products. Concluson Ths paper examned the mpact of plant characterstcs, food safety technology, two types of regulatons, and market mechansms on food safety performance, as measured by the number of product samples testng postve for Salmonella n an FSIS testng program. Plant characterstcs plant sze, varety of speces of anmals slaughtered n the plant, and whether slaughter plants had grndng operatons had an mportant mpact on Salmonella testng results. Specfc food safety technologes, however, provded lttle advantage to ther users. We attrbute ths fndng to the ncentves of frms to obtan more sophstcated technologes or employ more rgorous standards both of whch are costly. We suggest that plants wth the most sophstcated technologes and technques are the ones that most need them to avod product contamnaton. Plants wth less need for sophstcated technologes perform have no need to obtan them and so they do not. The most mportant fndng n the paper s the lnkage between the PR/HACCP process standard (SSOPs) and the PR/HACCP managed regulatory approach (HACCP plans) to the PR/HACCP performance standard (Salmonella results). We found that both process and management-based regulatons enhanced food safety performance. The more flexble management-based HACCP plans, whch are developed by the plants and ncludes tasks that are 18

21 montored by FSIS, had the greatest mpact n reducng Salmonella levels. We cannot say whether process or management-based regulatons are superor because the two approaches covered dfferent types of tasks that provde dfferent opportuntes for controllng pathogens. However, we can say that both regulatons play an mportant role n controllng pathogens. Another mportant result of ths emprcal research s the fndng that market mechansms reduce the lkelhood of product contamnaton. Ths should not be too surprsng. Large buyers, such as fast food restaurants, demand hgh food safety qualty because they could lose ther reputaton for servng wholesome food f they serve food wth low food safety qualty. These and the other fndngs lead to overall concluson that market mechansms and flexble regulatons have exerted powerful forces n mproved pathogen control over meat and poultry. 19

22 References Antle, J.M. Effcent Food Safety Regulaton n the Food Manufacturng Sector, Amercan Journal of Agrcultural Economcs. 78 (Decemeber 1996) pp Antle, J.M. Economc Analyss of Food Safety, B. Gardner and G. Rausser (eds), Handbook of Economcs. Vol. 1 Chapter 19. Amsterdam: Elsever Scence B.V., Coglanese, Gary and Davd Lazer, Management-Based Regulaton: Prescrbng Prvate Management to Acheve Publc Goals. AEI-Brookngs Jont Center for Regulatory Studes, workng Paper (November, 2002). Golan, Else, Tanya Roberts, Elsabete Salay, Jule Caswell, Mchael Ollnger, and Danna Moore. Food Safety Innovaton n the Unted States: Evdence from the Meat Industry, Agrcultural Economc Report 831, U.S. Department of Agrculture, Economc Research Servce, Haavelar, Are H., Maartan J. Nauta, and Jaap T. Jansen. Fne-tunng Food Safety Objectves and rsk assessment, Internatonal Journal of Food Mcrobology (2004) Ollnger, Mchael and Valere Mueller. Managng for Safer Food: The Economcs of Santaton and Process Controls n Meat and Poultry Plants. Agrcultural Economc Report 817. U.S. Department of Agrculture, Economc Research Servce,

23 Table 1: Margnal Effects of Plant and Food Safety Technology, Customer Specfcatons, and Food Safety Effort on Salmonella Performance n Cattle Slaughter Cattle Varable Parameter Estmates Parameter Estmates Mean CONSTANT ** * - (0.023) (0.024) Log (POUNDS) ** ** (0.0013) (0.0015) MULTI-SPECIES (0.005) (0.005) GROUND ** ** 0.35 (0.006) (0.006) DEHIDE_TECH ** (0.0065) DEHIDE_KNIFE (0.004) HACCP_ NOCOMPLY ** *** (0.099) (0.103 SSOP_NOCOMPLY (0.085) (0.092) EXPORTS (0.005) *** (0.005) Observatons Χ 2 27 *** 26 *** 21

24 Table 2: Margnal Effects of Plant and Food Safety Technology, Customer Specfcatons, and Food Safety Effort on Salmonella Performance n Brolers and Hog Slaughter Brolers Hogs Varable Parameter Estmates Mean Parameter Estmates Mean CONSTANT * ** - (0.131) (0.110) Log (POUNDS) ** (0.007) (0.006) MULTI-SPECIES *** *** 0.67 (0.026) (0.029) GROUND *** (0.030) (0.027) COOKED (0.052) (0.026 EVISCERATOR (0.028) STEAM_VAC *** 0.12 (0.039) OPERATIONS_TEK_HI * 0.12 (0.039) (0.034) NOFEED *** * 0.16 (0.039) (0.029) HACCP_NOCOMPLY ** (0.215) (0.570) SSOP_NOCOMPLY (0.148) (0.197) CUSTOMER_SPECS *** (0.022) (0.023) 0.33 Observatons Χ 2 58 *** 22 ** 22

25 Table 3: Margnal Effects of Plant and Food Safety Technology, Customer Specfcatons, and Food Safety Effort on Salmonella Performance n Ground Beef Varable Parameter Estmates Mean CONSTANT ** (0.050) Log (POUNDS) ** (0.003) MULTI-SPECIES ** (0.025) SLAUGHTER ** (0.020) COOKED (0.011) OPERATIONS_TEK_HI ** (0.017) HACCP_NOCOMPLY (0.409) SSOP_NOCOMPLY ** (0.180) CUSTOMER_SPECS (0.011) Observatons 117 Χ 2 22 *** 23