Valuing lives saved from safer food--a cautionary tale revisited.

Prioritizing risk-reduction strategies to maximize net benefits requires information on the monetary value people assign to safer food (Roberts 1986). The challenge is that individual behavior reflects more than just unobserved preferences for risk reduction. Behavior also reflects each person's unobserved and potentially unique food risk and ability to reduce this risk privately, through consumption patterns, food preparation, cleaning efforts, and so on. Risk is endogenous (Shogren and Crocker 1991). People with a low valuation of collective risk reduction may seem to tolerate greater risk, but they may instead have already used unobservable skills to reduce the risk themselves. Food selection alone is unlikely to reveal perfectly these two characteristics--risk preferences and risk reduction technology--because multidimensional heterogeneity exists in the population. In such cases, we propose using a General Method of Moments (GMM) framework to separate risk reduction activities from risk preferences (Shogren and Stamland 2002, 2006).

Our promotion of GMM presumes that people can make food choices based on the differentiated risks of different foods and gauge the appropriate responses to these risks. The open question, however, is whether people actually do differentiate low-probability food risks and respond accordingly and consistently. In experimental auctions designed to value safer food, Hayes et al. (1995) found evidence of surrogate bidding (also called embedding, part-whole bias, or insensitivity to scope).

Surrogate bidding exists when behavior-either risk perceptions or values--for a specific good reflects general preferences for a phenomenon rather than for the specific good in question, or when perceptions and values are insensitive to changes in the quantity or quality of the good. Hayes et al. (1995) compared the bidding behavior from food-borne pathogen treatments to the bids in a treatment that combines the risks of all the pathogens. They observed surrogate bidding for reduced health risk; bids for a cluster of pathogens were indistinguishable from bids for specific pathogens.

Using a contingent valuation survey, Hammitt and Graham (1999) reproduced the Hayes et al. (1995) study, and found the same insensitivity to probability. In addition, Bateman et al. (1997) found similar results for "parts" versus a "whole" restaurant meal. Using an incentive compatible mechanism to auction off parts of a restaurant meal (e.g., appetizer, main course, desert), they observed the sum of the parts exceeded the value for the whole meal. The results support the idea that values for specific food items can reflect general preferences for food.

Multidimensional heterogeneity and surrogate bidding undercut the traditional approaches people use to estimate the value of safer food. First, one should separate out unobservable risk preferences from risk-reduction skill assuming differentiated responses to risk; second, one should test for whether the perceptions of risk and the responses to risk are rational to begin with. We use the GMM method to address the first task of identification of risk preferences, and we can use internal consistency checks to address the question of differentiated responses and reactions to risk. Here we describe the initial results on rational risk valuation for safer food using data from Wave I of the University of Wyoming (UW) Food Web Diary project. We designed the UW survey to provide the data needed to use a GMM framework to obtain improved valuation estimates in the face of multidimensional individual heterogeneity. (1)

The UW survey captures the idea that each respondent has an idiosyncratic mortality risk from each risk source, an idiosyncratic ability to reduce each risk, and his or her own value of statistical life. We allow consumers to be heterogeneous in several observable and unobservable dimensions: tastes, budgets, base-level risks, abilities to reduce risk, and willingness to pay for risk reduction. Given this heterogeneity, rational people should choose different consumption levels based on different levels of risk. We test for rationality by asking several questions in the diary about risk perceptions and responses to risk. We create two tests of internal consistency: identical risk perception questions for three meats and an identical risk reduction asked in two ways. Overall, even for a relatively familiar commodity-related food, our results do not contradict the cautionary tale told by those concerned with surrogate behavior, such as Kahneman and Knetsch (1992). On average, we observe that the same question generated different responses (identical risk but different responses), and that different questions generated the same answer (insensitivity to different risks posed by pork, beef, and chicken).

Design of Survey

The UW survey design consisted of five parts: general health status; risk-taking behavior, knowledge of food safety and risk-reduction actions; awareness of effectiveness to reduce the risks of food-borne illnesses; and risk perceptions of food-borne illnesses. We also elicited socio-demographic information. First, we asked questions to learn about the general condition of the subjects' health: Q1. Are you (or your partner) currently pregnant? Q2: Is there anybody smoking in the family? Q3. Are you exposed on a regular basis (daily/weekly) to second-hand smoke? Q4. Is there exposure to second-hand smoke inside your household? Q5. Do you consume more than two alcoholic beverages per day? Q12. How would you rate your physical health compared to others your age and gender? Q18. Does anybody in your family have specific dietary needs due to a medical condition?

Second, we asked questions about the respondents' general effort to self-protect: Q6. Imagine you will be in a vehicle 10 times. How many times would you say you would: (a) wear a seatbelt as a driver or a passenger, and (b) drive more than 5 mph over the speed limit? Q7. Do you have a smoke detector in the house? Q8. Do you change your batteries in your smoke detectors at least once every year? Q9. Do you have a carbon monoxide detector in the house? Q10. Do you have a first-aid kit in the house? Q11. Do you have a fire extinguisher in the house?

Third, we asked one multipart question about food preparation: Q33. Out of 10 meal preparations for the relevant food, how frequently do you:

1. Wash your hands with hot soapy water before handling food?

2. Wash your hands with hot soapy water after handling raw meat products?

3. Wash utensils and surfaces immediately with hot soapy water where meals are prepared?

4. Wash utensils and surfaces with hot soapy water after preparing each food item and before you go onto the next food item?

5. Wash vegetables and fruit?

6. Cook meat products to a safe temperature recommended by health experts?

7. Refrigerate leftovers within two hours of preparation?

8. Use nonexpired food?

9. Use a separate cutting board for raw meat products and other nonmeat items?

10. Use a meat thermometer?

11. Use plastic or other nonporous cutting boards?

12. Separate raw meat, poultry, and seafood from other foods in your shopping cart?

13. Separate raw meat, poultry, and seafood from other foods in your refrigerator?

14. Place cooked food on a plate that previously held raw meat, poultry, or seafood?

15. Cook eggs until the yolk and white are firm?

16. Use recipes in which eggs remain raw or only partially cooked?

17. Cook fish until it is opaque and flakes easily with a fork?

18. Cover microwave food, stirring and rotating it?

19. Bring sauces, soup, and gravy to a boil when reheating?

20. Defrost food at room temperature?

Fourth, we asked four specific questions about the subjects' perception of the effectiveness of the four-key risk-reduction methods for food-borne pathogens: washing, separating, cooking, and prompt storage for one of the three meats beef, pork, or chicken. For each question, the respondent checked off one of the intervals from (91-100%); (81-90%), ..., (1-10%). Q43. How effective is washing one's hands, utensils, and food before a meal in reducing the risk of food-borne illness from this food? Q44. How effective is separating raw meat/poultry/seafood from other foods and using a different cutting board for raw meat products in reducing the risk of foodborne illness from this food for a meal? Q45. How effective is cooking food to proper temperatures in reducing the risk of food-borne illness from this food for a meal? Q46. How effective is prompt storage and refrigeration in reducing the risk of food-borne illness from this food for a meal?

Next, we wanted to understand perceptions toward the changed effectiveness if the respondent cut in half (or doubled, depending on survey) the frequency of washing, separating, cooking, and prompt storage, again for the three meats. For each question, the respondent circled an integer value ranging from 5 (large decrease) to -5 (large increase). For example, subjects were asked: Q48. If you cut in half the number of times you washed your hands, utensils, and food before a meal, how much would your risk for each type change?

Fifth, we are interested in each person's risk perception about food-borne illness. We asked two questions: Q21. Please mark the point that you think best represents how frequently A TYPICAL AMERICAN can be expected to suffer a food-borne illness in any given year. How Often Frequency (In Terms of

Number of People) Twice every week (1 in 10 people) Once every month (1 in 100 people) Once every year (1 in 1,000 people) Once every 10 years (1 in 10,000 people) Once every 100 years (1 in 100,000 people) Once every 1,000 years (1 in 1,000,000 people) Once every 10,000 years (1 in 10,000,000 people) Once every 100,000 years (1 in 100,000,000 people)

Q22. Please mark the point that you think best represents how frequently YOU can be expected to suffer a food-borne illness in any given year. How Often Frequency (In Terms of

Number of People) Twice every week (1 in 10 people) Once every month (1 in 100 people) Once every year (1 in 1,000 people) Once every 10 years (1 in 10,000 people) Once every 100 years (1 in 100,000 people) Once every 1,000 years (1 in 1,000,000 people) Once every 10,000 years (1 in 10,000,000 people) Once every 100,000 years (1 in 100,000,000 people)

Implementation of Survey

Knowledge Networks (KN) implemented the survey, which was funded by the U.S. Department of Agriculture through the Economic Research Service. The study design consisted of three waves of data collection with two interventions between the three waves. Each wave of data collection lasted for fourteen days. In these fourteen days, respondents were instructed to collect their household's grocery shopping receipts. In Wave I, 1,274 surveys were fielded with 923 completed--a 72% response rate. After the first wave of data collection, respondents were sent the first intervention survey. Respondents were instructed to visit a website about the ten least-wanted bacteria present in foods (see http://www.fightbac.org/10least.cfm). In Wave II, 905 surveys were fielded, with 800 completed--a 88% response rate. The second wave took place next. After the second wave, respondents were sent the second intervention survey. They were instructed to visit a different website about food safety (see http://www.thebody.com/fda/fsebac.html). Finally, in Wave III, 774 surveys were fielded, with 703 completed--a 91% response rate.

The sample of the study was restricted to the overlapping panelists between KN's web-enabled panel and the National Shopper Lab (NSL) panel. This sample design was intended to allow analysis of UPC data collected at grocery stores where respondents used their NSL card to shop. (2) The survey instrument averaged approximately 19 minutes. Each respondent was awarded 5,000 bonus points (an equivalent of $5) for their participation in each survey and for the collection of their household's grocery store receipts.

Wave I Results

We created six indexes that summarize a subject's responses to related sets of questions and three measures of risk perception. First, we created a health index, HI. Several questions pertain to the respondents' health and susceptibility to illness. We converted the answers to numeric values. For instance, if the answer to question 1 was "yes," we defined q1 - 1, whereas q1 = -1 if the answer was "no." If the question was not answered, we defined q1 = 0. The corresponding was done with questions 2-5, 12, and 18. Based on these numerical responses, the health index, HI, was:

HI = q1 + q2 + q3 + q4 + q5 + q12 + q18. (1)

The questions were formulated so "yes" implied poor health or higher susceptibility to illness. Higher values of the health index indicate greater sensitivity to health risk.

Second, the survey respondents reported their height and weight in questions 13 and 14, which we combine into the body mass index, BMI:

BMI = weight/height (2) (2)

where the weight is measured in kilograms and the height in meters.

Third, we created a measure of the respondent's effort in self-protection through various activities that reduce risk. We had two questions that already required quantitative responses, namely the two sub-questions in question 6 that asked how many times, out of ten times, the respondent wore a seat belt and adhered to the speed limit when driving or riding in a car. The risk index, RI, was calculated as

RI = q6_wear/10 - q6_drive/10 + q7 + q8 + q9 + q10 + q11 + q20 + Q24 (3)

where Q24 = 1 if the response to question 24 was greater than 0 and Q24 = 0 otherwise. The first two terms in the expressions on the right-hand side of (3) indicate that we use the fraction of times the respondent wore a seatbelt and adhered to the speed limit, as components of the index.

Fourth, we created a food preparation index from the responses to question 33. This question inquired about numerous activities carried out before, during, and after food preparation that may have influenced the heath risks posed by the prepared food. The food preparation index (FPI) was calculated as the sum of the reported number of times, out of ten meal preparations, the specific activity was carried out:

FPI = q33a + q33b + ... + q33t. (4)

Fifth, the survey asks how effective various activities were perceived to be in reducing food-borne health risks posed by three categories of food: beef, pork, and chicken. We created from the responses the following self-protection effectiveness index (SPEI):

SPEI = q43_beef + q43_pork + q43_chicken + ... + q46_beef + q46_pork + q46_chicken. (5)

Questions 43-46 are formulated so that the lowest response, 1, indicated that the activity is very effective. Larger values of SPEI thus corresponded to less effectiveness of self-protection.

Sixth, questions 48 through 51 asked about the change in health risk that would result if the respondent changes the activity level for specific activities. From these responses, we created an index that measures the perceived incremental risk reduction (IRR) from doubling, or cutting in half (two treatments), the number of times the risk reduction activity is carried out. The variable amount equals 1 if the "cut in half" treatment was given, and amount = 2 if the "double" treatment was given. The responses to the questions range from 1 = "large decrease," through 6 = "no change," to 11 = "large increase" in risk due to the halving/doubling of the activity We combined the responses to the questions into the index IRR as follows:

IRRI = ((q48_beef-6) + (q48_pork-6) + ... + (q51_chicken-6)) x (2 x amount - 1). (6)

The index is thus calculated so that it takes on negative values if doubling (halving) the activity level is associated with decreased (increased) risk, and positive values if doubling (halving) the activity level is associated with increased (decreased) risk.

Finally questions 21 and 22 provide three risk perception variables: (3)

(7) Personalrisk = q21

(8) Baserisk = q22

(9) Relativerisk = Personalrisk/Baserisk

Table 1 provides descriptive statistics about the respondents and their responses to the questions as summarized by the indexes defined above. Table 2 reports the significant correlations between the variables.

Several key points arise in the correlation matrix. First, a strong relationship exists between the health index and the body mass index. This is to be expected since both indexes are based on questions about facts, and there likely is a factual relationship between health and body mass. Second, a significant correlation exists between some pairings of the indexes. The risk index (RI) is correlated with the food preparation index (FPI). Respondents with high levels of self-protection in food preparation also report more caution in other activities, such as driving. They also report that self-protection is more effective (SPEI). Also, respondents who believe self-protection is more effective (low SPEI) believe in a greater reduction of risk associated with an increase in self-protection (low IRRI). Given the strong similarity of the underlying perceptions asked about in the set of Q43-46 (SPEI) and Q48-51 (IRRI), we find it troublesome that the correlation between the two indexes is no more than 0.115. The adjusted-[R.sup.2] is only 1.2% in a regression between SPEI and IRRI.

Third, and perhaps most notably, no evidence of correlation exists between either of the two health status indexes, HI and BMI, and any of the other indexes. HI and BMI are based on questions with strong factual basis, such as a person's weight, height, smoking, and pregnancy. The risk index, RI, and to a lesser degree the food preparation index, FPI, are based in part on factual information. Some of the component questions have a weaker factual basis, such as the questions about how often, out of ten times, a person engages in a certain behavior. The "facts" underlying such questions may be poorly remembered and may suffer from reporting errors and perhaps reporting biases. Finally, the factual underpinnings for SPEI and IRRI are probably most complex of all, and are most susceptible to reporting errors and biases.

The possibility of reporting errors and biases is hard to avoid in survey studies and we were prepared for their presence here. Our goal was to qualify the survey responses with hard data for two purposes: (1) to confront the survey responses with factual data with the goal of removing some of the reporting errors and biases, and (2) to combine the survey data and the behavioral data (because neither data source contains all the relevant information) to identify each person's willingness to pay for mortality risk reduction. But both of these aims require a connection between the survey data and the behavioral data. We find it disconcerting that the picture emerging from table 2 seems to be this: there are internal correlations between the factually oriented indexes and between the more subjective indexes, but there is little correlation between the two groups of indexes. There seems to be a disconnection between exposures to health risk and (reported) health risk reduction activities.

Fourth, this disconnect story is strengthened by noting the significant correlation between household income and both HI and BMI. Again, we observe correlation between factual data, which appears to reflect socioeconomic patterns in health and obesity. Household income is also significantly related to the general risk reduction index, RI, which in part is based on specific factual questions. A weak relationship exists, not significant at the 1% level, between the FPI and SPEI indexes.

Fifth, the respondents' relativerisk is not significantly correlated with any of the indexes, and neither is the personalrisk. Interestingly, baserisk, the respondents' perception of the risk exposure of other individuals, is significantly correlated (at the 1% level) with the individual's health index, HI, and the household income with correlation coefficients of 0.108 and -0.123. The lack of correlation between personalrisk or relativerisk with any of the other variables confirms the picture of a disconnect in the survey responses. To us, respondents seem to have vague notions of their risk exposure; these notions are so vague that they correlate with health status, reported behavior, or effectiveness of risk reduction.

While the evidence so far indicates a disconnection in the survey responses, the last and perhaps most consequential result we report concerns a correlation between the responses to certain questions that seems "too high." Table 3 reports the correlations between the responses to the various components of Q43. The correlations are similar for Q44-46, and for Q48-51 (correlations are calculated treatment by treatment). The lowest correlation observed in any of these correlation matrixes is 0.773 while most correlations actually are above 0.9, in which the responses to Q48-51 fare worst with correlations ranging between 0.93 and 0.99 (with only one exception--a correlation of 0.88).

This strong correlation in the presence of the otherwise low to nonexistent correlations is baffling. We are at a loss to explain this finding, other than that the survey respondents seem to be guessing at numbers they perceive only vaguely, at best. Their guesses have a strong internal consistency between what they perceive to be similar questions, but there is no external consistency to several things we expected to be related: health status, reported behavior, reported self-protection activity, and worst of all, the responses to differently worded questions about similar underlying realities. The strong internal correlations between the responses about risks posed by beef, pork, and chicken suggest surrogate behavior. These results contrast the lack of correlation between similar responses summarized by SPEI and IRRI.

As a final check, we examine the correlations between the responses to matched pairs of questions that form part of SPEI and IRRI. Question 43 is matched with question 48, 44 with 49, etc., and the matching is done treatment by treatment, i.e., conditional on amount. The contrast still exists--too much correlation across different food groups, 0.7; too little correlation across the same questions asked in different ways, 0.177. Moreover, there is no tendency for matched sub-questions, like q43_beef and q48_beef, to be more correlated than nonmatched sub-questions, like q43_beef and q48_chicken. The strongest observed correlation, of -0.177 (negative due to differing treatments), is between q45_chicken and q50 beef.

This strong internal consistency, and poor near-internal and external consistency, means these data are unsuitable for either the proposed GMM approach or for any other method. The internal consistency means that not enough variation exists between a person's responses to the risks posed by beef, pork, and chicken that could be correlated to behavior. The lack of near-internal consistency causes us to doubt that the responses represent strongly felt perceptions. Rather, the responses appear to be "best guesses." Finally, the lack of external consistency indicates that the responses are near orthogonal to other facts and behaviors, making them uninformative and unable to shed further light on the value of risk reduction.

In summary, the evidence from Wave I of the UW survey revealed that responses contained poor information about how people perceive food safety risks and how they will respond. This implies these data cannot support reasonable VSL estimates. In the end, one can question whether this result is due to the theory of endogenous risk, the survey design, or the rationality of the respondents. We conclude on a constructive note. Future survey work that tries to collect data at the level of detail needed to run a GMM-style model will have to find a set of different behaviors that really are distinct in the average person's mind, even after accounting for endogenous risk. The challenge is to identify a framework of comparable activities that are sufficiently distinct to generate both different behavioral reactions and decided perceptions that are tied to the true nature of the activities. What we have learned is that the behaviors in the risk questions have to be sufficiently differentiated, almost a day and night distinction, to make inferences about the value of statistical life given multidimensional heterogeneity

Thanks to the USDA/ERS for supporting this research, and Tanya Roberts and Fred Kuchler for their support.

References

Bateman, I., A. Munro, B. Rhodes, C. Starmer, and R. Sugden. 1997. "Does Part-Whole Bias Exist? An Experimental Investigation." Economic Journal 107:322-32.

Hammitt, J., and J. Graham. 1999. "Willingness to Pay for Health Protection: Inadequate Sensitivity to Probability?" Journal of Risk and Uncertainty 18:33-62.

Hayes, D., J. Shogren, S. Shin, and J. Kliebenstein. 1995. "Valuing Food Safety in Experimental Auction Markets." American Journal of Agricultural Economics 77:40-53.

Kahneman, D., and J. Knetsch. 1992. "Valuing Public Goods: The Purchase of Moral Satisfaction." Journal of Environmental Economics and Management 22:57-70.

Roberts, T. 1986. "The Economic Losses Due to Selected Foodborne Diseases." Proceedings of the Ninetieth Annual Meeting of the United States Animal Health Association, October 1986, pp. 336-53.

Shogren, J., and T. Crocker. 1991. "Risk, Self-Protection, and Ex Ante Economic Value." Journal of Environmental Economics and Management 20:1-15.

Shogren, J., and T. Stamland. 2002. "Skill and the Value of Life." Journal of Political Economy 110:1168-97.

--. 2006. "Consistent Estimation of the Value of Statistical Life." Resource and Energy Economics 28:262-81.

(1) In GMM terms, the goal is to obtain a model in which all parameters are identified. This model then yields an asymptotically unbiased estimate of the key parameter--the value of statistical life. This approach has the potential of restoring greater accuracy to the estimates of the value of statistical life by avoiding the bias inherent in single-equation approaches. This advance can come at considerable cost. Keeping the model tractable and its data requirements reasonable requires some strong assumptions. But we believe these assumptions do not invalidate the main strength of the model, which is to strip away the presumption--implicit in previous estimation approaches--that all people are identical.

(2) Due to the fact that NSL participants primarily resided on the east coast and Midwest of the United States, only KN panelists in the following states were selected for the survey: Connecticut, New York, New Jersey, Pennsylvania, Ohio, Indiana, Michigan, Maryland, Washington DC, Virginia, North Carolina, South Carolina, and Florida.

(3) We use the 'frequency value' of the response in decimal notation, for instance q21 = 0.001 if the response was '1 in 1000 people.'

Jason F. Shogren is the Stroock Professor of Natural Resource Conservation and Management, University of Wyoming, and the King Carl XVI Gustaf Professor of Environmental Sciences, Umea University; Tommy Stamland is Associate Professor of Finance, Norwegian School of Economics and Business Administration.

This article was presented in a principal paper session at the AAEA annual meeting (Portland, OR, July 2007). The articles in these sessions are not subjected to the journal's standard refereeing process. Table 1. Summary Statistics Variable Mean S.D. Min. Max. Health, HI -1.4 2.7 -5.0 7.0 Body mass, BMI 28.0 6.2 15.9 60.5 Risk, RI 4.3 2.6 -4.0 9.1 Food preparation, FPI 137.9 23.6 13.0 200.0 Self-protection 22.9 14.3 12.0 120.0

effect, SPEI Inc. risk reduction, IRRI -24.1 28.4 -60.0 60.0 Household income 64,913.2 42,804.2 2,500.0 200,000.0 Table 2. Correlation Matrix

HI BMI RI FPI Health, HI 1 Body mass, BMI 0.178 1 Risk, RI 1 Food prep., FPI 0.188 1 Self-prot., SPEI -0.25 Inc. risk red., IRRI -0.08 -0.087 Income -0.201 -0.091 0.158 -0.073

SPEI IRRI Income Health, HI Body mass, BMI Risk, RI Food prep., FPI Self-prot., SPEI 1 Inc. risk red., IRRI 0.115 1 Income 0.064 1 Note: Entries in bold are significant at the 1% level. Correlations that are not significant at the 10% level are not reported. Table 3. Correlation Matrix Between Responses to Question 43

q43_beef q43_pork q43_chicken q43_beef 1 q43_pork 0.92 1 q43_chicken 0.87 0.89 1 Note: Entries in bold are significant at the 1% level. Correlations that are not significant at the 10% level are not reported.