Jump to ContentJump to Main Navigation
Rational Risk PolicyThe 1996 Arne Ryde Memorial Lectures$

W. Kip Viscusi

Print publication date: 1998

Print ISBN-13: 9780198293637

Published to Oxford Scholarship Online: November 2003

DOI: 10.1093/0198293631.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2018. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in OSO for personal use (for details see http://www.oxfordscholarship.com/page/privacy-policy). Subscriber: null; date: 17 August 2018

Hazard Warnings and Risk Information

Hazard Warnings and Risk Information

Chapter:
(p.26) 3 Hazard Warnings and Risk Information
Source:
Rational Risk Policy
Author(s):

W. Kip Viscusi (Contributor Webpage)

Publisher:
Oxford University Press
DOI:10.1093/0198293631.003.0003

Abstract and Keywords

Hazard warnings can potentially play a constructive policy role by providing risk information when people are not informed of the risk. Cigarette warnings are a principal example of how the public can be informed adequately of the risk and can form risk beliefs comparable to the actual product risk. Warnings can potentially alter discrete decisions, such as whether or not to use a product or take a job, as well as precautionary behaviour with respect to a product or a job. In Bayesian learning contexts, effective warnings are those that provide accurate new information in an effective manner.

Keywords:   Bayesian learning, hazard warnings, risk beliefs, risk information, smoking

If people do not have accurate risk perceptions, the obvious direct solution is to provide information to remedy this market failure. So long as people can learn and act upon their new knowledge, there is the prospect of addressing the source of the inadequacy.

This seemingly straightforward approach has long been resisted by government officials. The widely held view among regulators has been that information programs are ineffective in altering behavior.1 As a result, more direct intervention is needed through regulatory action. This pessimistic assessment of the efficacy of information programs is often driven by self‐interest, as it leads to an outcome that fosters the agenda of regulators wishing to expand their domain.

Many risk information efforts have not had success since they simply reminded people of what they already knew. The information did not alter their risk judgments. This chapter explores the potential role of risk information and the economic mechanisms through which this information operates.

Rationale for Hazard Warnings

The 1970s was the era in which government policy focused on direct regulation of risk and the environment. The 1980s marked the emergence of hazard warnings as a prominent policy alternative. Recognition of the importance of personal risk protection and risk‐taking decisions has led to increasing emphasis on right‐to‐know policies, efforts that promote an informed citizenry, and recognition of individuals' ability and right to make choices to protect their own lives. The contexts in which risk information has become instrumental (p.27) range from information on job‐related chemical exposures and established efforts such as cigarette warnings, to more novel risk communication efforts such as the provision of information to assist public debates regarding hazardous waste cleanups.

There are a variety of rationales for using hazard warnings as the chosen form of regulatory intervention. If the market failure is a lack of information, then warnings can address this limitation directly. In terms of the model in Chapter 2, warning information that conveys an accurate risk probability in a sufficiently convincing way will lead individuals to have sound risk beliefs.

More specifically, let qi be the person's initial assessed risk of outcome i and let pi be the risk conveyed by the warning. Unless pi exceeds qi, the warning will not increase risk beliefs. However, what appears to be most influential in practice is the degree of informational content associated with the warning. Warnings that have high values of ξ relative to the weight γ on prior beliefs will be more effective in shifting risk beliefs. The credibility of the risk message is of enormous practical consequence.

Warnings policies also play an essential role in situations in which there are decentralized decisions. Often it is not possible for regulators to monitor our actions that are related to risk. For example, government officials cannot go into people's homes to ensure that households are handling dangerous chemicals properly. They can, however, provide consumers with information that ideally will assist them in making sensible decisions on their own.

A particularly disturbing example of hazard warnings for decentralized decisions is the unprecedented set of warnings currently provided by the Hertz car rental dealership at the Miami International Airport.2 Because of the rash of car hijackings and foreign tourist murders in the Miami area, Hertz provides all those renting cars with a map indicating the safe highways with a ‘Sun Symbol,’ a 12‐page flier with a series of 18 different warnings in six different languages called ‘Safety Tips,’3 and a written waiver that must be signed by the car renter acknowledging that Hertz has provided this safety information. Clearly, Hertz is unable to monitor where drivers take their (p.28) cars and how they choose to drive. However, through this set of warnings, the firm has communicated the distinctive nature of the risks that threaten drivers in this area and has provided a comprehensive set of behavioral rules for decreasing the risk. Direct regulation would not have been feasible, given the decentralized nature of driver decisions.

Apart from these rationales, there is also a political impetus for informational policies in that they provide an intermediate policy option. Policies that either must permit an unfettered use of a product or ban the product altogether represent extreme policy alternatives. In situations in which there is heterogeneity in the risk level or in risk preferences, banning a product may deprive many consumers of a beneficial new product in the effort to address problems associated with a minority of consumers. As a practical matter, information often plays a constructive role in giving policymakers an intermediate policy option when there is insufficient evidence to warrant direct regulation, but enough concern about a potential risk to alert the public of the need for care.

Risk Beliefs and Discrete Choices

Rational Risk Policy Risk Beliefs and Discrete Choices

Risk communication policies operate in two general contexts. One potential role is to affect individual beliefs concerning the riskiness of an activity or a product and thus to influence the discrete decision of whether or not to engage in such a pursuit. Hazard warnings consequently influence our choices of which jobs to take, which products to consume, and which activities to pursue. However, there is a second role of warnings as well, which is to influence our behavior within these activities. Workplace warnings may urge the use of protective equipment, and the warnings for consumer products may advise the use of bicycle helmets or care while skiing or scuba diving.4

The two types of evidence with respect to the impact of warnings involve structured experiments and natural experiments. Table 3.1 reports the results of a structured experiment in which we gave four (p.29)

Table 3.1. Means of Variables for Each Labeling Group

Risk variable

Sodium

Lachrymator

Asbestos

TNT

bicarbonate (n = 31)

(n = 106)

(n = 102)

(n = 96)

Baseline risk

0.12

0.10

0.09

0.10

Risk after receiving warning

0.06

0.18

0.26

0.31

Risk premium required

0.00

$3,032

$4,733

$8,150

after receiving warning

of increased risk

(1995 dollar rate)a

Fraction of workers

0.00

0.23

0.65

0.73

wishing to quit after warning

Fraction of workers

0.90

0.58

0.11

0.07

willing to take the job

again after warning

(a) The risk premium figures are conditional upon facing an increased risk and being willing to accept a finite risk premium.

Source: adapted from Viscusi and O'Connor (1984).

different groups of chemical workers a hazard warning for a chemical. The survey informed each worker receiving the warning that the labelled chemical would replace the chemicals with which the individual currently worked. The four possible labels were sodium bicarbonate (i.e., household baking soda), chloroacetophenome (a lachrymator that makes workers cry and is an eye irritant), asbestos, and TNT.

Table 3.1 summarizes the effect of each warning on each worker group. The worker's baseline risk assessment appears in the first row of the table. Workers assessed the job risk probability using a linear injury frequency scale for which the average US injury rate was the anchor. The range of the average assessed prior risks for the four subsamples was from an equivalent annual probability of job injury of 0.09 to 0.12. After receiving this labeling information, workers who received the sodium bicarbonate label decreased their assessed risk assessment from 0.12 to 0.06. Being told that household baking (p.30) soda would replace the chemicals with which they now worked had a favorable effect on risk beliefs. The other warnings of hazards increased workers' assessed risks where these risk perceptions after the information acquisition ranged from 0.18 to 0.31.

The usual market model since the time of Adam Smith is that workers will require a compensating differential to incur the added risk, which is the observed result. This risk premium ranges from $3,032 for the lachrymator to $8,150 for TNT (1995 dollars).

If workers are not paid a wage increase, they should respond in some manner. The results indicate that, although no workers would quit the safe sodium bicarbonate job without a wage increase, 65 percent would quit the job with asbestos exposures and 73 percent would quit the job with TNT exposures. Similarly, workers' willingness to take the job again without a wage increase would diminish to 11 percent for asbestos and 7 percent for TNT exposures. Hazard warnings consequently create a multiplicity of market responses arising from increasing risk perceptions. These responses include workers' willingness to stay on the job and the terms of tradeoff they demand for remaining at that position.

The importance of informational content in driving risk perception differences is exemplified in the results for the hazard warnings considered earlier in this chapter. Both the prior risk assessment qi and the risk level implied by the warning pi are consequential. Using information on the individual's prior and posterior risk assessments, it is possible to estimate the relative precision ξ/γ of the information content of the warning relative to the worker's prior risk beliefs. More specifically, break a variant of equation (2.1) into component terms, or

Perceived product risk P i = γ q i γ + ξ + ξ p i γ + ξ .
(3.1)
In a regression of the perceived product risk (the dependent variable) on the initial assessed risk of the job qi, one can write (3.1) in terms of the observable probability values—the posterior risk dependent variable pi and the prior risk qi—and a constant term αi, a coefficient βi, and a random error term εi, or
P i = α i + β iqi + ε i .
(3.2)
The coefficient of qi, which we will call βi, is the relative information weight attached to the prior, γ/(γ + ξ). The value of the implied risk (p.31) pi is not observed. As a result, the estimation captures the value of the second term in (3.1) through a constant term αi which varies by chemical and reflects the combined influence of the various components ξpi/(γ + ξ).

The estimates of the parameters αi and βi provide the key information regarding the character of the warnings effort since the underlying characteristics of the warning can be expressed in terms of these estimated values. The parameters estimated directly give the fraction of the information weight placed on the γ prior, or βi, and the value of the job risk probability weighted by its relative informational content, or αi. In particular, one can estimate the risk level pi implied by the warning, where

p i = α i / ( β i 1 ) .
(3.3)
Similarly, one can derive the relative weight placed on the information in the hazard warning as compared with the person's prior beliefs, or
ξ i / γ = ( 1 / β i ) 1 .
(3.4)

As the results in Table 3.2 indicate, the estimated implied risks pi from the warning were fairly high for all chemicals other than sodium bicarbonate. However, the relative informational content differed starkly and was often influential. The relative informational content of the warning compared with the prior was low for the unfamiliar chemical that was a lachrymator and 31 times as influential as the prior beliefs for the TNT warning—a highly familiar explosive. The mechanisms of the learning process—in particular, the informational content as well as the implied risk value—determine the ultimate economic effects of warnings. The asbestos and TNT warnings were particularly effective, not simply because they conveyed the

Table 3.2 Estimated Characteristics of the Hazard Warnings

Implied warning risk

Relative weight on

pi

warning information

ξi

Sodium bicarbonate

0.04

3.7

Lachrymator

0.24

1.3

Asbestos

0.29

6.4

TNT

0.32

31.4

(p.32) highest risk level, but also because they had a higher informational content. In these instances, the relative informational content for asbestos was 6.4 and for TNT was 31.4, whereas the informational content for the lachrymator/chloroacetophenome warning, which had much vaguer properties, was only 1.3.

The Cigarette Warnings Experience

Rational Risk Policy The Cigarette Warnings Experience

Perhaps the most highly publicized large scale individual risk is that posed by cigarettes. In the United States, this product has been the subject of three decades of on‐product labeling, annual reports by the US Surgeon General regarding smoking hazards, and bans on various forms of cigarette advertising. Canada and many Western European nations have also adopted vigorous warning efforts pertaining to smoking, along with advertising restrictions and limitations on the character of cigarette packaging.

Concern that smoking is a risky activity is not misplaced. As

                   Hazard Warnings and Risk Information

Fig. 3.1 Comparison of Different Lifetime Risks

(p.33) Figure 3.1 indicates, the risks of smoking are substantial and dwarf most other hazards we face. The annual risk of death for a smoker is 1/150, which is double the annual risk of death from cancer from all causes and more than an order of magnitude greater than risks such as those posed by motor vehicle accidents and work accidents.

A particularly noteworthy entry in Figure 3.1 is the risk of being killed by an asteroid. One consensus scientific estimate is that there is a probability of 1/500,000 that the Earth will be struck by a doomsday rock, posing an annual risk of death to any given individual of 1/2,000,000.5 Not surprisingly, there has been a divergence of scientific opinion regarding the magnitude of the risk. Indeed, scientists at Cornell University, who sought to obtain government funding for their research on rockets that will destroy such an asteroid, estimated the risk to be as high as 1/6,000.6 If this estimate is to be believed, the annual risk of being killed by an asteroid is greater than the risk of being killed at work. One clearly desirable policy option is to invest in refining such widely disparate risk estimates before diverting substantial resources to this protective task.

A more solidly established target of government concern is smoking. Table 3.3 summarizes the different eras of hazard warnings for cigarettes in the United States. These warnings were on‐product labels on the side of cigarette packages, which were also required to be included in cigarette print and billboard advertising. These product‐specific warnings have been accompanied by other informational efforts as well, such as a ban on television and radio ads.

In conjunction with a variety of public education efforts, these various information campaigns have contributed to a dampening of the growth of cigarette consumption in the United States. As is shown in Figure 3.2, cigarette consumption was on the rise through the first half of this century, after which it plateaued and has subsequently decreased. It is also noteworthy that the major dip in smoking in the early 1950s occurred after the release of an adverse Mayo Clinic report on cigarette smoking, which received substantial publicity.

Although these statistics reflect an impressive reversal of the upward trend in smoking, the even greater influence has been with respect to changes in the types of cigarette smoked. Figure 3.3 sketches the trend in per capita cigarette consumption as well as the trend in tar‐adjusted per capita cigarette consumption. For (p.34)

Table 3.3. Cigarette Warning Content Summaries

Warning period

Warning contenta

1965

‘Caution: Cigarette Smoking May Be Hazardous to Your

Health.’

1969

‘Warning: The Surgeon General Has Determined That

Cigarette Smoking Is Dangerous to Your Health.’

1984

1. ‘SURGEON GENERAL'S WARNING: Smoking Causes

Lung Cancer, Heart Disease, Emphysema, and May

Complicate Pregnancy.’

2. ‘SURGEON GENERAL'S WARNING: Quitting Smoking

Now Greatly Reduces Serious Risks to Your Health.’

3. ‘SURGEON GENERAL'S WARNING: Smoking by

Pregnant Women May Result in Fetal Injury, Premature

Birth, and Low Birth Weight.’

4. ‘SURGEON GENERAL'S WARNING: Cigarette Smoke

Contains Carbon Monoxide.’

(a) All warnings wording is specified by legislation. See 15 U. S. C. §§1331–41 (1982).

concreteness, the starting point for each series has been normalized relative to 1944. Per capita cigarette consumption displayed an upward drift that stabilized in the 1960s and has been declining since the mid‐1970s. In contrast, the tar‐adjusted cigarette consumption has dropped by more than two‐thirds over the past half century. The smoking literature has focused primarily on the effect of information on the discrete choice of whether to smoke and the quantity of cigarettes smoked. There has been comparatively little attention paid to the safety precautions adopted while smoking—the choice of the cigarette tar level. There has been a much greater shift in the kind of cigarettes people smoke than in the number of cigarettes. Thus, hazard warnings affect not only the decision to engage in the risky activity but also the nature of the risks one takes within the context of these activities.

These changes in smoking activity have also been accompanied by shifts on public perceptions of smoking. Reflective of these changes in beliefs is the stark change in public attitudes toward environmental (p.35)

                   Hazard Warnings and Risk Information

Fig. 3.2 Trends in US Total Per Capita Cigarette Consumption, 1900–1996

Source: based on data from the Tobacco Institute (1996), p. 6.

                   Hazard Warnings and Risk Information

Fig. 3.3 US Per Capita Cigarette Consumption and Tar‐Adjusted Consumption, 1944–1994

(p.36) tobacco smoke. In 1977, only 16 percent of the public favored no smoking in public places, whereas by 1988, 60 percent of the public favored a complete ban on smoking in all public places.7

Whereas a half century ago there was not as much public consciousness of the risks of smoking, current individual risk perceptions are substantial. Table 3.4 indicates various estimates of the risk of smoking based on the state of information in 1985 and 1991. These estimates, which I prepared using reports by the US Surgeon General, indicate that in 1985 the lung cancer mortality risk to smokers was 0.05–0.10, the total mortality risk was just over triple that amount, and the total mortality risk to society including environmental tobacco smoke, fetal deaths, and fires was a bit greater.

Table 3.4. Actual Smoking Risk Ranges, 1985 and 1991

Survey year

Lung cancer

Total

Total

mortality risk

mortality risk

mortality risk

to smoker

to smoker

to society

1985

0.05–0.10

0.16–0.32

0.21–0.42

1991

0.06–0.13

0.18–0.36

0.23–0.46

Evidence I have analyzed based on a national survey for 1985 suggests that the public generally assessed the lung cancer risk of smoking to be 0.43, with smokers assessing the risk at 0.37. In each case, the lung cancer risk component is substantially overassessed, with the total risk assessment even exceeding the total mortality risk to smokers. Equally noteworthy is that only 5 percent of the public and 9 percent of the smokers assessed the lung cancer probability as being below 0.05 and just under 10 percent of the total population and 15 percent of the smoking population estimated the risk to be under 0.10. The extent of underassessment of the risk seems to be particularly small.

My more recent survey evidence reported in Table 3.5 indicates that for the full sample of respondents the lung cancer fatality risk assessment is 0.38, where this value is 0.31 for current smokers. The assessed total smoking mortality risk is greater, with a value of 0.54 (p.37)

Table 3.5. Smoking Fatality Risk Perceptionsa

Sample

Mean (standard error of the mean)

Lung cancer fatality risk

Total smoking mortality risk

Full sample

0.38

0.54

(0.02)

(0.07)

Current smokers

0.31

0.47

(0.04)

(0.05)

Current nonsmokers

0.40

0.56

(0.02)

(0.03)

Former

0.36

0.50

(0.03)

(0.04)

Never

0.42

0.59

(0.03)

(0.03)

(a) Sample size = 206.

for the full sample and 0.47 for current smokers. Each of these values exceeds the upper bound of the scientific assessments of the total smoking mortality risk to smokers.

The overassessment of smoking risks does not appear to be a size‐related bias. Smoking risks are sufficiently large so as not to qualify as a small overassessed risk. The substantial publicity given to smoking hazards appears largely responsible. Since the character of the anti‐smoking efforts is to note that smoking is risky rather than to convey a particular risk probability, the main effect of the smoking risk information campaigns has been to raise risk perceptions rather than to foster convergence to the true estimated risk level.

The variation in smoking risk perceptions across the population is of considerable interest as well. If p is the prior risk assessment, q 1 is the risk implied by the individual's smoking experience, q 2 is the risk implied by hazard warnings, and the informational content parameters are specified as before, then,

Perceived risk P = γ p + ξ 1 q 1 + ξ 2 q 2 γ + ξ 1 + ξ 2 .
(3.5)
If warning risk information q 2 suggests a higher risk probability than the value of (γp + ξ1 q 1)/(γ + ξ1), then it will increase risk beliefs. (p.38) Thus, on an empirical basis, one might expect anti‐smoking warnings to raise risk beliefs. One possible exception is the warning proposed in 1995 for public comment by the US Food and Drug Administration but which was not adopted. The proposal was to indicate that one out of three smokers die from smoking. That risk level is considerably below current risk beliefs and might depress risk beliefs.

Smokers are expected to have lower risk beliefs wholly apart from their self‐selected status. If smoking activity conveys a risk level q 1 to smokers that is below the warning risk q 2, then it will dilute the effect of warnings in the weighted informational average in equation (3.5).

Younger smokers are particularly likely to assess high risks since such a substantial fraction of their risk information has come in the post‐warnings era. The effect of warnings information is to increase risk beliefs, so that

P / ξ 2 > 0 .
(3.6)
whenever q 2 > p, q 1. However, the extent of this effect dampens over time, or
2 P / ξ 2 2 < 0 .
(3.7)
There should consequently be a dampening of the effect of warnings with age.

The lung cancer risk perception results in Table 3.6 bear out these various predictions. Nonsmokers have higher risk perceptions than current or former smokers. This pattern is borne out for every age group in Table 3.6. Moreover, current smokers always have the lowest risk perceptions, followed by former smokers, and then non‐smokers in every case except the youngest age group—for which the difference between current and former smokers was not statistically significant. Risk perceptions rise as the fraction of information received in the post‐warnings era increases. The youngest age group (age 16–21) has the highest risk beliefs, or a perceived lung cancer risk of 0.49. These risk perception amounts decrease for the older age groups to a value of 0.42 for those age 22–45 and an almost identical value for those age 46+.

Empirical evidence also indicates that these risk perceptions have a significant negative effect on the discrete smoking decisions which does not differ significantly by age. Indeed, the average US excise tax on cigarettes of 31 percent of the retail price per pack decreases (p.39)

Table 3.6. Variations in Lung Cancer Risk Perceptions With Age and Smoking Status

Mean risk, by groupa

Current

Former

Nonsmoker

All

smoker

smoker

respondents

Age group

in age group

Age 16–21

0.445

0.429

0.511

0.490

(0.043)

(0.037)

(0.017)

(0.015)

Age 22–45

0.382

0.390

0.454

0.417

(0.011)

(0.013)

(0.010)

(0.006)

Age 46+

0.328

0.421

0.456

0.418

(0.017)

(0.015)

(0.011)

(0.008)

All ages

0.368

0.408

0.464

0.426

(0.009)

(0.010)

(0.007)

(0.005)

(a) Standard error of mean in parentheses.

smoking demand by less than half as much as the effect of smoking risk perceptions.8 In the current era of hazard warnings and extensive smoking risk information, individuals not only appear to be aware of the risk but seem to overassess it. These risk perceptions in turn affect smoking behavior.

This evidence alone does not imply that smoking decisions are fully rational. Nor does evidence of overassessment of the risk imply that current hazard communication efforts have been excessively zealous. It is difficult to achieve pinpoint optimality with respect to hazard warnings. What is clear is that the combination of societal efforts to convey the importance of smoking risks has had a dramatic effect in influencing this risky behavior and has led to a situation in which risks seem to be reasonably well understood.

Generalizing from the Cigarette Experience

Rational Risk Policy Generalizing from the Cigarette Experience

Policymakers' natural inclination has been to attempt to imitate the (p.40) success of cigarette warnings in other policy contexts. One such effort took place in the state of California with respect to hazardous chemicals in food. The warning contemplated by regulators pertained to many product risks that exceeded a lifetime cancer risk threshold of 1/100,000 over a 70‐year life time of exposure. This risk level involved an annual risk of 1/7,000,000, which is below the scientific consensus estimate that we will be killed by an asteroid.

Notwithstanding the negligible level of risks, the proposed wording was: ‘WARNING: The State of California has determined that this product causes cancer.’ Officials patterned this warning after the 1969 cigarette warning (see Table 3.3). Not surprisingly, test subjects viewing this warning considered any product bearing such a label to be almost as risky as cigarettes. Indeed, the average person assessing this warning rated the risk as equivalent to 0.4 pack of cigarettes—a consumption activity that poses enormously greater risks.9

The market response to hazard warnings is evident not only for cigarettes, but also for other products. Consumption of products containing artificial sweeteners such as saccharin declined after the advent of saccharin warnings. Even in situations in which the recipient group for the warning is a learned intermediary bringing to bear advanced expertise, there is an evident impact of hazard warnings. A widely used pharmaceutical product in the United States was tetracycline, which continues to be the drug of choice for ailments such as Rocky Mountain spotted fever and lyme disease. The unfortunate consequence of this drug is that it causes tooth staining for children under the age of 9, who are forming their permanent teeth. In 1963 companies introduced information pertaining to the tooth staining hazard into the hazard warning section of the patient package insert. As is shown in Figure 3.4, tetracycline use has continued to increase for the population group aged 9 and above that was not affected by the warning, whereas the targeted warning group greatly diminished its use of the product. In this as in other instances, well designed warning policies have played an instrumental role in influencing potentially risky decisions. (p.41)

                   Hazard Warnings and Risk Information

Fig. 3.4 Use of Tetracycline, as Measured by Mentions (Prescriptions and Renewals), 1960–1975

Warnings and Precautions

Rational Risk Policy Warnings and Precautions

In some cases, the task of warnings is to influence not the choice of an activity, but rather the precautions one takes within that activity. Warnings that increase the perceived probability of injury from product misuse or which provide information on how to decrease this risk can be effective. However, even effective warnings that lead to accurate risk beliefs may not alter behavior if, for example, the individual considers the precaution to be unduly onerous. Table 3.7 summarizes how different precautionary behaviors respond to the presence of hazard warnings for bleach. The most prominent hazard of bleach is that mixing bleach with ammonia or ammonia‐based products will form chloramine gas, which is the leading non‐suicidal cause of poisonings among adults in the United States and in Europe. The products tested include a label in which the warning information was eliminated, labels for the Clorox brand of bleach and the Brite brand of bleach that are nationally marketed in the United States, and a ‘Test’ label designed by our warnings research project team to convey the information in an effective manner.

The presence of a warning increases the fraction of respondents who would not mix bleach with toilet bowl cleaner even if the toilet was badly stained by 24 percent, and it leads an additional 16 percent of respondents to avoid adding bleach to ammonia‐based cleaners (p.42)

Table 3.7. Effects of Labels on Precaution‐Taking

% of respondents taking precaution

Maximum incremental

No

Clorox

Brite

Test

effect

Precaution

warning (n = 51)

(n = 59)

(n = 42)

(n = 44)

1. Do not mix bleach

16

23

36

40

24

with toilet bowl

cleaner (if toilet

is badly stained)

2. Do not add bleach

69

68

69

84

16

to ammonia‐based

cleaners (for

particularly dirty

jobs)

3. Store bleach in

43

63

50

76

33

childproof

location

Source: Viscusi and Magat (1987), table 4.2.

even for particularly dirty jobs. Warning labels also increase the fraction of respondents who would store the product in a childproof location by 33 percent.

Table 3.8 reports analogous results for drain openers. In addition to the no warning option and a test label, a third label patterned after the nationally marketed Drano and Red Devil Lye also was included in the experiment, where this label includes such a concentration of warnings information that it is the most effective of the labels tested. Once again, hazard warnings have a significant effect on precautionary behavior, increasing the fraction who wear rubber gloves by 19 percent and the storage of the product in a childproof location by 20 percent for households with children under age 5.

In this and other instances, labels are not fully effective. Not all individuals will take the recommended precaution. However, the failure of some people to wear rubber gloves, for example, should not necessarily be viewed as an inadequacy in individual decisions. The same survey also elicited information with respect to individual disutility associated with the precautionary behavior. For quite reasonable (p.43)

Table 3.8. Effect of Warnings on Precautions Taken With Drain Openers

% of sample taking precaution

Maximum incremental

Drano

Test

No

effect (%)

Red Devil

label

warning

Precaution

Lye label

Wear rubber gloves

82

73

63

19

Store in childproof location:

Households with

90

83

70

20

children under 5

Households with no

63

61

48

15

children under 5

Source: Viscusi and Magat (1987), tables 4.3 and 4.6, and calculations by the author.

values of the risk probabilities and the disutilities, it would not necessarily be irrational to forgo the types of precaution listed. Men, for example, are particularly adverse to wearing rubber gloves when performing household chemical chores and are less likely to respond to the urging of the warning.10

Principles for Hazard Warnings Policy

Rational Risk Policy Principles for Hazard Warnings Policy

These results in no way imply that warnings can remedy all informational inadequacies. However, a well designed information effort can play a constructive role in fostering sounder risk decisions.

Perhaps the main prerequisite for an effective warnings policy is that it provide new information in a convincing manner. In terms of the informational formulation in Chapter 2, for information to be new the risk conveyed by the warning must be different from the risk probability that the individual holds initially. Otherwise, risk beliefs are not altered. The information also must be convincing, which means that the informational content term ξ must be high relative to (p.44) the value of the informational content of associated with prior risk beliefs γ.

An essential concern in designing effective warnings policies is to recognize that individuals have cognitive limitations. Individuals must receive the warnings message, process it, and then act upon it. The structure, format, and content of warnings consequently play an instrumental role in the effect warnings policies have. Boxing the warnings and exhibiting them in a bolder print size may matter up to a point, but the effects of such nuances appear to be diminishing once a reasonable degree of readability has been achieved.

In thinking about hazard warnings, one should conceptualize the entire hazard communication system. Focusing on only a single risk of the product while neglecting others will not promote a reasonable response to risk. Lift trucks, for example, pose roughly three dozen types of fatality risk to the driver or others. Operating motor vehicles such as cars also poses a multiplicity of hazards. On‐product labels are often not sufficient. Television ads, advice of a physician, required certified training programs, and licensing of drivers also may be integral parts of an effective hazard communication system.

Perhaps somewhat surprisingly, standardization is often desirable. Contrary to the economists' usual urging that there be a diversity of responses, in this instance uniformity in the warnings vocabulary and in the warnings approach is a desirable feature. International standardization is also desirable for countries in which international trade or a diversity of worker nationalities is consequential. By having a common format, individuals will be more readily able to locate hazard warning information. Moreover, standardizing various warning symbols, such as that for poisoning, as well as human hazard signal words, such as ‘danger’, ‘warning’, and ‘caution’, will give these urgings a comparable meaning which recipients could interpret more reliably.

A final principle for effective warnings is that overwarning is dangerous. Excessive warning distorts relative product comparisons and threatens the credibility of the information provider. The task of information policy is to promote correct risk‐taking decisions, not to distort the underlying probabilities and discourage efficient risk‐taking behavior.

Notes:

(1) See Adler and Pittle (1984) and Wagenaar (1992).

(2) These warnings appear to have been stimulated by liability concerns so as to avoid lawsuits after car‐jackings.

(3) For example, safety tip number 13 is: ‘If your car is bumped from behind in a secluded or dark area, do not pull over and stop. Drive to the nearest public area and call for police assistance.’

(4) It is noteworthy that at the time of these lectures there was a large scale public information campaign to encourage children to use helmets in Scandinavia. In Lund, Sweden, which was the site of these lectures, the McDonald's restaurant placemats featured a cartoon in which Ronald McDonald urged children to adopt sensible behavior while riding bikes by wearing a bicycle helmet.

(5) The Economist (Sept. 11, 1993), p. 13.

(6) New York Times (June 18, 1991), p. B5.

(7) These results are from Gallup poll surveys for the respective years, where the complete results are reported in table 3‐3 of Viscusi (1992b).

(8) More specifically, the cigarette tax effect is the same as a lung cancer risk perception of 0.17 (using demand elasticity of –0.4), and the average lung cancer risk perception is 0.43. See Viscusi (1992b), especially p. 109.

(9) In practice, the implementation of this regulation has not led to severe consequences. For example, many companies have reformulated their products. One such reformulated product is Liquid Paper. One consumer product that has received the warning is sand for children's sandboxes, which could not be reformulated to eliminate all potential hazardous chemicals. The state of California also proposed comparable warnings for all products associated with birth defects, where a strict application of the initial proposed warnings would have led to on‐product warnings for all products containing caffeine, including coffee, tea, and cola beverages.

(10) As Wagenaar (1992) has noted, what people say they will do is not always what they actually do. In Magat and Viscusi (1992) we corroborated the experimental precautions with a survey on actual precautionary behavior and found a close correspondence for our consumer survey instrument.