Willingness to pay for food safety: sensitivity to duration and severity of illness.

There is a mismatch between the incidence and economic valuation of food-borne illness. The vast majority of cases result from exposure to microbial pathogens (e.g., E. coli O157, Salmonella) and consist of short-term illnesses that last from one to a few days. Yet research on valuation of health risk has been dominated by the study of mortality risk. Economic Research Service and other estimates of the value of reducing morbidity risk have been forced to rely on cost-of-illness measures that include productivity and medical costs but exclude pain and suffering, lost leisure time, disruption of daily activities, and other components. To help fill this gap, we provide stated-preference estimates of the value of reducing risk of foodborne illness, focusing on short-term morbidity.

We design and conduct a stated-preference survey to estimate willingness to pay (WTP) to reduce the risk of food-borne illness of specified severity and duration. The following sections describe the survey instrument, sample, and results. We find that the value per statistical case avoided is larger for risk to children ($23,600-$30,500) than to adults ($8,300-$16,400) and is surprisingly insensitive to duration (one to seven days) and severity of illness (mild discomfort to hospitalization). Estimated WTP is larger for women, Blacks, Hispanics, and respondents who observe safe food-handling practices or perceive their risk to be higher than average and larger for risks transmitted on chicken than on ground beef or packaged deli meat.

Survey Instrument and Sample

The survey instrument is organized as follows. First, respondents are asked about their experience with food-borne illness and their perception of how common it is in the United States. Second, they complete a tutorial designed to help them practice making trade-offs between food price and safety. The tutorial introduces a visual aid to help communicate risk. The visual aid contains red and white areas representing 10,000 meals, where the fraction colored red equals the probability of illness (Corso, Hammitt, and Graham 2001).

Third, respondents are asked to consider buying food for a meal that only they will eat, randomly selected from (chicken, ground beef, packaged deli meat). After answering questions about their typical consumption frequency and serving size (respondents who do not eat the selected food are asked about another), respondents are told their baseline probability of illness (2 x [10.sup.-4], 4 x 10-4 per meal), the symptoms associated with illness (mild, moderate, severe),1 duration (one, three, seven days), conditional mortality risk (0, [10.sup.-4], [10.sup.-3]), and informed that they could reduce their risk to 1 x [10.sup.-4] per meal by purchasing a safer but more expensive type of food. The baseline and reduction in probability are communicated using the visual aid described above. The risk reduction is described as produced by a stringent safety program established and monitored by the U.S. government that does not use chemicals or irradiation (which some respondents may believe would present other risks). WTP to reduce the probability of illness is elicited using double-bounded, dichotomous-choice questions (Hanemann, Loomis, and Kanninen 1991) with initial bids ranging from $0.04 to $4.00 per meal and follow-up bids equal to twice or half the initial bid as appropriate. After the valuation questions, respondents answer follow-up questions about their food-handling practices, acceptance of the hypothetical scenario, and personal characteristics.

Each respondent values two risk reductions that differ by reduction in probability of illness, severity and duration of symptoms, conditional mortality risk, and food (chicken, ground beef, packaged deli meat). Attributes are randomly assigned using a full factorial design. Respondents living in a household with a child aged two to eighteen years value one risk to themselves and one to a randomly selected child in the household (in random order); other respondents value two risks to themselves.

To test for framing effects, respondents complete a version of the survey with risks and costs expressed per meal (as described above) or per month (converted from per-meal values using respondent-reported consumption frequency). If WTP is proportional to probability reduction, estimated values per case will be the same.

The survey was fielded to 6,368 randomly selected members of a demographically representative panel maintained by Knowledge Networks. Households were recruited to the panel using random-digit dialing and provided free Internet access and hardware as a participation incentive. In total, 3,902 interviews were completed in several waves between August and October 2004, yielding a response rate of 61%. We exclude 136 respondents who do not eat any of the three foods (N = 107) or declined to answer the WTP questions (N = 29), leaving 3,766 respondents for analysis.

Results

This section describes respondent characteristics and how estimated WTP varies with risk and respondent characteristics.

Respondent Characteristics

Descriptive statistics are reported in the first column of table 1. Statistics for the subsamples of respondents living in households with and without children are similar to the full sample except age (mean 37 and 48, respectively), married (68% and 48%), household size (3.6 and 2.2), and college degree (21% and 27%).

On average, respondents estimate that 33% of the U.S. population contracts food-borne illness in a year, roughly compatible with an official estimate of seventy-six million cases per year (Mead et al. 1999). Respondents are significantly more likely to eat chicken and ground beef than packaged deli meat, which results in 40%, 35%, and 25% of respondents answering their first question about chicken, ground beef, and packaged deli meat, respectively. Most respondents report taking precautions when preparing food: 62% report consistent hand washing and 67% report taking one or more recommended steps to ensure that food is fully cooked or otherwise safe to eat. In questions relating to acceptance of the hypothetical scenario, 48%, 39%, and 13 % of respondents perceive their risk of foodborne illness to be similar to, smaller than, and larger than that presented in the survey, respectively.

Effects of Risk and Respondent Characteristics on WTP

We model WTP as a function of the severity and duration of illness, reduction in probability, and respondent characteristics. Regression models are estimated assuming a log-normal error term and using maximum-likelihood methods (Alberini 1995). We estimate separate models for WTP to reduce own risk for respondents with and without children in the household and for WTP to reduce a child's risk. Results are in table 1. We first describe Models 1, 3, and 5, which include only risk characteristics and then Models 2, 4, and 6, which add respondent characteristics.

For respondents in households without children, WTP to reduce own risk increases with reduction in probability of illness, symptom severity, duration, and conditional mortality risk (Model 1). Estimated WTP is 35% and 47% larger for moderate and severe than for mild symptoms (2) (the difference between WTP for moderate and severe symptoms is not statistically significant). WTP is only modestly sensitive to duration; compared with one day, it is 29% larger for seven and 6% larger (not significant) for three days. Estimated effects of conditional mortality risk are not significant, but the point estimates imply a value per statistical life (VSL) of $9-25 million, (3) which is somewhat larger than conventional estimates (e.g., $7 million, Viscusi and Aldy 2003; $5.4 million, Kochi, Hubbell, and Kramer 2006). WTP is significantly greater for the larger reduction in probability of illness but the proportionate increase (1.85) is significantly smaller than the threefold increase required by conventional theory (consistent with most stated-preference studies: Hammitt 2000; Hammitt and Graham 1999). In contrast, estimated WTP per month is 5.85 times larger than per meal. This ratio is nearly equal to the average frequency of consumption (5.68), which suggests that estimated values of risk reduction are not sensitive to the alternative framing.

For respondents living in households without children, estimated WTP to reduce own risk is less sensitive to the risk characteristics. In Model 3, the estimated coefficients of the risk reduction, duration, severity, and mortality-risk variables are smaller than the corresponding estimates for households without children (Model 1) (except the coefficient on severe symptoms is negligibly larger). Standard errors are larger, in part because of the smaller sample size, and only the coefficients on severe symptoms and the per-month framing differ significantly from zero. Similarly, estimated WTP to reduce risk to a child (Model 5) is insensitive to duration, severity, and mortality risk, with estimated coefficients of these variables smaller than the corresponding estimates in Model 1 (except the coefficient on three-day duration is negligibly larger). The estimated coefficient on risk reduction is significantly greater than zero but also significantly smaller than required for WTP to be proportional to risk reduction.

Models 2, 4, and 6 supplement the basic specification with variables that describe respondent characteristics. Coefficients of the risk-characteristic variables are not substantially altered. All three models suggest that estimated WTP is larger for respondents who are female, Black, Hispanic, or have no college degree. The estimated effects of income and marital status are insignificant and there is modest evidence that WTP increases with respondent age and decreases with age of the child to be protected. Respondents' perceptions of risk and confidence in protection mechanisms show significant effects: Estimated WTP is generally larger when own or child's risk is perceived to be larger than average and smaller when risk is perceived to be smaller than average (though only four of six relevant coefficients are significant). Estimated WTP is smaller for respondents who are not confident that the hypothetical safety system would be effective. Among respondents in households without children, estimated WTP is larger when trust in the private sector is low and smaller when trust in government is low; in other words, WTP increases with the perceived need for and efficacy of a government program. Estimated WTP to reduce own risk appears larger for respondents who wash their hands or practice other safe food-handling practices (perhaps reflecting greater concern for food safety) and smaller for those with primary responsibility for preparing meals in their households (four of six relevant coefficients are significant). There is evidence of an order effect, with WTP estimated from the second valuation question significantly larger than that from the first, except when the second question is about risk to a child. Finally, estimated WTP to reduce own risk, but not a child's risk, is significantly larger for chicken than for ground beef and packaged deli meat.

Value per Statistical Case of Food-Borne Illness

Estimated WTP per statistical case avoided is reported in table 2. It is calculated by predicting median WTP for the full-sample-mean respondent for each severity-duration combination and dividing by the risk reduction. (4) For households without children, WTP to reduce own risk varies between $8,300 and $16,100 per case, increasing with both severity and duration (with standard errors of $700 to $1,200).

For households with children, WTP to reduce own risk is of similar magnitude but less sensitive to risk characteristics. The values per statistical case are larger for one- and three-day episodes and smaller for seven-day episodes than for respondents in households without children. The range of values is correspondingly smaller, between $10,800 and $16,400 per case (with standard errors of $900 to $1,500).

WTP to reduce risk to a child is much greater than to reduce own risk but insensitive to severity and duration of illness. The value per statistical case ranges between $23,600 and $30,500 (with standard errors of $6,600 to $8,300, which are much larger than for adults). The estimated value per seven-day case is implausibly smaller than for shorter episodes because the estimated coefficient on seven-day duration is less than (though not significantly different from) zero. Controlling for severity and duration, the value per statistical case for a child is between 1.7 and 2.6 times as large as for an adult for households with children.

Conclusion

Our stated-preference estimates suggest that WTP to reduce risk of short-term morbidity from food-borne pathogens is on the order of $10,000 per statistical case avoided for adults and twice as large for children. Estimated WTP is surprisingly insensitive to severity and duration of illness, especially for children. This insensitivity is unlikely to reflect respondents' unfamiliarity with these attributes or inattention to details of the scenarios given the significant association of WTP with stated risk reduction, a much more difficult attribute to grasp.

New Estimates of the Demand for Food Safety (Tanya Roberts, USDA, Organizer)

References

Alberini, A. 1995. "Efficiency vs. Bias of Willingness-to-Pay Estimates: Bivariate and Interval-Data Models." Journal of Environmental Economics and Management 29:169-80.

Corso, P.S., J.K. Hammitt, and J.D. Graham. 2001. "Valuing Mortality-Risk Reduction: Using Visual Aids to Improve the Validity of Contingent Valuation." Journal of Risk and Uncertainty 23:165-84.

Hammitt, J.K. 2000. "Evaluating Contingent Valuation of Environmental Health Risks: The Proportionality Test." Association of Environmental and Resource Economists Newsletter 20(1):14-19.

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

Hanemann, W.M., J. Loomis, and B. Kanninen. 1991. "Statistical Efficiency of Double-Bounded Dichotomous Choice Contingent Valuation." American Journal of Agricultural Economics 73:1255-61.

Kochi, I., B. Hubbell, and R. Kramer. 2006. "An Empirical Bayes Approach to Combining and Comparing Estimates of the Value of a Statistical Life for Environmental Policy Analysis." Environmental and Resource Economics 34:385-206.

Mead, P.S., L. Slutsker, V. Dietz, L.F. McCaig, J.S. Bresee, C. Shapiro, P.M. Griffin, and R.V. Tauxe. 1999. "Food-Related Illness and Death in the United States." Emerging Infectious Diseases 5:607-25.

Viscusi, W.K., and J.E. Aldy. 2003. "The Value of a Statistical Life: A Critical Review of Market Estimates Throughout the World." Journal of Risk and Uncertainty 27:5-76.

(1) Mild: "You will have an upset stomach and will feel tired, but these symptoms will not prevent you from going to work or from doing most of your regular activities." Moderate: "You will have an upset stomach, fever, and will need to lie down most of the time. You will be tired and will not feel like eating or drinking much. Occasionally, you will have painful cramps in your stomach.

In addition, you will have some diarrhea and will need to stay close to a bathroom. While you are sick, you will not be able to go to work or do most of your regular activities." Severe: "You will have to be admitted to a hospital. You will have painful cramps in your stomach, fever, and will need to spend most of your time lying in bed. You will need to vomit and will have severe diarrhea that will leave you seriously dehydrated. Because you will be unable to eat or drink much, you will need to have intravenous tubes put in your arm to provide nourishment."

(2) I.e., exp(0.297) - 1 = 35%; exp(0.387) 1 = 47%.

(3) Estimates derived by adjusting values per case reported in table 2 for mortality risk. For example, estimated value per mild one day case is $8.300 with no mortality risk and exp(0.121) larger (i.e., $9,400) with conditional mortality risk of [10.sup.-4]. Implied VSL = ($9,400-$8,400)/([10.sup.-4]) = $10 million.\

(4) The estimated value per statistical case is calculated for each of the two probability reductions and then averaged.

James K. Hammitt is Professor and Kevin Haninger is Research Associate at the Center for Risk Analysis, Harvard University.

This work was supported by the Economic Research Service of the United States Department of Agriculture. We thank Nicole Ballenger, Steve Crutchfield, Joni Hersch, Fred Kuchler. Amanda Lee, Katherine Ralston, Tanya Roberts, Milton Weinstein, and Kip Viscusi for helpful discussions.

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. Descriptive Statistics and Regression Estimates Variable Mean

(Std. Dev.)

Full Sample

(N = 3,766) Intercept Risk reduction = 3 x

[10.sup.-4] Moderate symptoms Severe symptoms 3 days illness 7 days illness Conditional mortality

risk = [10.sup.-4] Conditional mortality

risk = [10.sup.-3] Monthly version 0.283

(0.450) Age (years) 44.738

(16.306) Child age (years) 8.462

(4.885) Male 0.475

(0.499) Male child 0.530

(0.499) Black, non-Hispanic 0.110

(0.313) Hispanic 0.121

(0.326) Other race, non-Hispanic 0.038

(0.191) Married 0.542

(0.498) Household size (number) 2.596

(1.357) Log household income 10.442

(0.944) College degree 0.247

(0.431) Perceived prevalence 32.717

(percent) (23.227) Perceive own risk 0.134

to be higher (0.340) Perceive own risk 0.389

to be lower (0.488) Perceive child's risk 0.155

to be higher (0.362) Perceive child's risk 0.344

to be lower (0.475) Prior foodborne illness 0.379

(0.485) Not confident in 0.159

safety system (0.365) Somewhat confident in 0.490

safety system (0.500) Low trust in government 0.041

(0.199) Low trust in private sector 0.061

(0.240) Ground beef 0.346

(0.476) Packaged deli meat 0.248

(0.432) Monthly consumption 5.679

frequency (number) (6.310) Monthly consumption 5.923

frequency--child (number) (6.426) Responsibility for preparing 2.356

meals (0-4) (1.472) Wash hands 0.617

(0.486) Safe food practices 0.674

(0.469) Second risk Residual geometric

standard deviation Log likelihood

WTP to Reduce Risk to Self

Households without

Children (N = 4,934)

Model 1 Model 2 Intercept 0.058 -0.827

(0.114) (0.535) Risk reduction = 3 x 0.616 *** 0.666 ***

[10.sup.-4] (0.081) (0.078) Moderate symptoms 0.297 *** 0.292 ***

(0.098) (0.094) Severe symptoms 0.387 *** 0.406 ***

(0.098) (0.094) 3 days illness 0.058 0.055

(0.097) (0.093) 7 days illness 0.251 *** 0.257 ***

(0.097) (0.094) Conditional mortality 0.121 0.143

risk = [10.sup.-4] (0.097) (0.093) Conditional mortality 0.032 0.056

risk = [10.sup.-3] (0.098) (0.095) Monthly version 1.767 *** 1.779 ***

(0.089) (0.087) Age (years) 0.010 ***

(0.003) Child age (years) Male -0.420 ***

(0.083) Male child Black, non-Hispanic 0.862 ***

(0.136) Hispanic 0.616 ***

(0.133) Other race, non-Hispanic -0.103

(0.209) Married -0.135

(0.088) Household size (number) 0.028

(0.040) Log household income 0.031

(0.046) College degree -0.355 ***

(0.092) Perceived prevalence 0.013 ***

(percent) (0.002) Perceive own risk 0.392 ***

to be higher (0.126) Perceive own risk -0.276 ***

to be lower (0.085) Perceive child's risk

to be higher Perceive child's risk

to be lower Prior foodborne illness -0.207 **

(0.082) Not confident in -1.041 ***

safety system (0.117) Somewhat confident in -0.041

safety system (0.087) Low trust in government -0.706 ***

(0.204) Low trust in private sector 0.485 ***

(0.178) Ground beef -0.376 ***

(0.092) Packaged deli meat -0.321 **

-0.130 Monthly consumption 0.009

frequency (number) (0.006) Monthly consumption

frequency--child (number) Responsibility for preparing -0.098 ***

meals (0-4) (0.031) Wash hands 0.423 ***

(0.080) Safe food practices 0.220 *

(0.118) Second risk 0.187 **

(0.077) Residual geometric 2.310 2.179

standard deviation (0.050) (0.047) Log likelihood -5,501.9 5,262.2

WTP to Reduce Risk to Self

Households with

Children N = 1,1600

Model 3 Model 4 Intercept 0.623 ** -0.380

(0.246) (1.115) Risk reduction = 3 x 0.202 0.280 *

[10.sup.-4] (0.162) (0.158) Moderate symptoms 0.214 0.190

(0.199) (0.193) Severe symptoms 0.389 ** 0.417 **

(0.198) (0.192) 3 days illness -0.058 -0.062

(0.196) (0.189) 7 days illness -0.055 -0.044

(0.200) (0.195) Conditional mortality -0.052 -0.121

risk = [10.sup.-4] (0.199) -0.193 Conditional mortality 0.005 -0.124

risk = [10.sup.-3] (0.197) (0.192) Monthly version 1.681 *** 1.723 ***

(0.179) (0.174) Age (years) 0.007

(0.008) Child age (years) Male -0.251

(0.180) Male child Black, non-Hispanic 1.210 ***

(0.291) Hispanic 0.723 ***

(0.238) Other race, non-Hispanic -0.029

(0.401) Married 0.045

(0.191) Household size (number) 0.013

(0.060) Log household income 0.040

(0.097) College degree -0.478 **

(0.205) Perceived prevalence 0.001

(percent) (0.004) Perceive own risk 0.649 ***

to be higher (0.255) Perceive own risk -0.259

to be lower (0.174) Perceive child's risk

to be higher Perceive child's risk

to be lower Prior foodborne illness 0.021

(0.164) Not confident in -0.629 ***

safety system (0.239) Somewhat confident in -0.298 *

safety system (0.177) Low trust in government 0.120

(0.448) Low trust in private sector -0.073

(0.385) Ground beef -0.460 **

(0.194) Packaged deli meat -0.577 **

(0.289) Monthly consumption 0.017

frequency (number) (0.014) Monthly consumption

frequency--child (number) Responsibility for preparing -0.012

meals (0-4) (0.067) Wash hands 0.528 ***

(0.164) Safe food practices 0.082

(0.262) Second risk 0.584 ***

(0.160) Residual geometric 2.257 2.134

standard deviation (0.101) (0.095) Log likelihood -1,279.2 -1,231.0

WTP to Reduce Risk

to Child (=1,149)

Model 5 Model 6 Intercept 1.185 *** 2.835 **

(0.245) (1.192) Risk reduction = 3 x 0.453 *** 0.459 ***

[10.sup.-4] (0.166) (0.162) Moderate symptoms 0.041 0.119

(0.206) (0.201) Severe symptoms 0.034 0.123

(0.201) (0.197) 3 days illness 0.081 0.080

(0.203) (0.197) 7 days illness -0.114 -0.053

(0.204) (0.199) Conditional mortality -0.085 -0.087

risk = [10.sup.-4] (0.202) -0.195 Conditional mortality -0.272 -0.228

risk = [10.sup.-3] (0.203) (0.198) Monthly version 1.637 *** 1.746 ***

(0.182) (0.178) Age (years) 0.017 **

(0.009) Child age (years) -0.032 *

(0.018) Male -0.539 ***

(0.189) Male child -0.010

(0.161) Black, non-Hispanic 0.833 ***

(0.286) Hispanic 0.552 **

(0.241) Other race, non-Hispanic -0.219

(0.382) Married -0.047

(0.200) Household size (number) 0.020

(0.062) Log household income -0.151

(0.103) College degree -0.461 **

(0.211) Perceived prevalence 0.002

(percent) (0.004) Perceive own risk

to be higher Perceive own risk

to be lower Perceive child's risk 0.341

to be higher (0.252) Perceive child's risk -0.646 ***

to be lower (0.179) Prior foodborne illness -0.015

(0.167) Not confident in -0.516 **

safety system (0.242) Somewhat confident in -0.211

safety system (0.182) Low trust in government -0.183

(0.459) Low trust in private sector -0.213

(0.393) Ground beef 0.256

(0.201) Packaged deli meat -0.352

(0.287) Monthly consumption

frequency (number) Monthly consumption 0.028 **

frequency--child (number) (0.013) Responsibility for preparing 0.016

meals (0-4) (0.071) Wash hands 0.080

(0.169) Safe food practices -0.400

(0.272) Second risk -0.166

(0.161) Residual geometric 2.205 2.082

standard deviation (0.104) (0.098) Log likelihood -1,136.2 -1,095.5 Notes: Variables are 0-1 and individual characteristics are for respondent except as noted. Mean and standard deviation of child characteristics are for households with children. Regression standard errors are in parentheses. Triple asterisk (***), double asterisk. (**), and single asterisk (*) denote statistical significance at 1, 5, and 105, respectively, using likelihood-ratio tests. Table 2. Estimated Value per Statistical Case of Foodborne Illness (US$)

Adult Case

Households

without Children Severity Duration Median Std. Err. Mild 1 day $8,300 $700 Moderate 1 day $11,100 $900 Severe 1 day $12,500 $1,000 Mild 3 days $8,800 $700 Moderate 3 days $11,700 $1,000 Severe 3 days $13,200 $1,100 Mild 7 days $10,800 $900 Moderate 7 days $14,400 $1,100 Severe 7 days $16,100 $1,200

Households

with Children Severity Median Std. Err. Mild $10,800 $900 Moderate $13,100 $1,100 Severe $16,400 $1,300 Mild $10,200 $900 Moderate $12,300 $1,100 Severe $15,400 $1,300 Mild $10,400 $1,000 Moderate $12,500 $1,300 Severe $15,700 $1,500

Child Case Severity Median Std. Err. Mild $24,900 $7,000 Moderate $28,000 $7,700 Severe $28,100 $7,600 Mild $27,000 $7,400 Moderate $30,400 $8,300 Severe $30,500 $8,300 Mild $23,600 $6,600 Moderate $26,500 $7,500 Severe $26,700 $7,300 Note: Estimates are based on predicted median WTP for full-sample-mean Respondent using Models 2, 4, and 6 (table 1).