3.1 Options

If your options have such a wide range, you might immediately ask: What is my rule of decision? How should you choose among the alternatives? As we have seen, Congress did not say. From the fabric of current regulatory law and practice, you might wonder about these possibilities:

1. (i) A safety-based, cost-blind standard, which would require the technology most likely to improve safety, at the highest level of stringency. This approach would, in a sense, simplify the regulatory inquiry. The question would be: What technology would be best, in terms of reducing relevant risks? Note that under this approach, the statute would be “technology-forcing.” It would authorize the agency to require companies to install technology that may not now exist.

2. (ii) Same as (i), but subject to a constraint of technological feasibility. Under this approach, regulators would not consider costs, but they would have to ensure that the technology is “feasible” to do what they are requiring. They would not consider costs in any way; they would not balance costs against benefits; they would not even entertain the question whether the regulation is economically feasible (bracketing the question of what that means).

3. (iii) Same as (i), but subject to a constraint of economic as well as technological feasibility. Regulators might be constrained not only by what is technologically feasible but also by economic feasibility. This idea is opaque, but it usually refers to significant adverse effects on businesses – in the extreme cases, to a large number of business failures. Agencies have enjoyed a degree of flexibility in understanding the idea of feasibility, but it does not refer to cost-benefit balancing (Masur & Posner, Reference Masur and Posner2010). A regulation might be feasible even if its costs greatly exceed its benefits.

4. (iv) Same as (i), but subject to a constraint of “best available technology.” The agency might conclude that it is forbidden to force technological innovation, and that it is both permitted and required to require the best technology that is now available for general use. This is a standard idea in federal regulation in the USA. It is not so different from (iii), but it is generally regarded as less stringent. It requires the agency to use what is now “available,” not to go to the point of economic and technological feasibility.

5. (v) Same as (iv), but also subject to a constraint of cost-effectiveness. Under this approach, regulators would ask whether one or another approach is more cost-effective. The idea of cost-effectiveness can be understood in different ways. In its usual version, it asks: What is the cheapest way of achieving a specific goal? If two methods are feasible and would both save 50 lives, the question would be which is least expensive. In some forms, regulators are allowed to have a degree of flexibility with respect to the goal, as in: What is the cheapest way of achieving substantial safety benefits? If an expensive but feasible regulation would save 51 lives while an inexpensive one would save 49 lives, regulators might choose the latter. This version veers in the direction of cost-benefit balancing.

6. (vi) Some form of cost-benefit balancing, alongside a requirement to maximize net benefits. There is, of course, a large literature on that form of balancing, and it raises many questions and doubts. But it is the standard method used within the executive branch of the federal government, and must be used unless Congress requires otherwise (Sunstein, Reference Sunstein2018). A requirement to maximize net benefits can be understood as supplemental to cost-benefit balancing. Two or more approaches might have benefits in excess of costs. If so, the question would be which has higher net benefits (or lower net costs).

7. (vii) A “least burdensome alternative” standard, which would require adoption of the approach that imposes the lowest costs. This formulation can be found in some corners of federal regulatory law. In principle, it is disconnected from an analysis of feasibility and from cost-benefit balancing, and close or perhaps identical to a freestanding requirement of cost-effectiveness (in the usual version described in [v] above). But some people think that it requires agencies to describe their goals at a level of abstraction (“significant savings in terms of lives”), and to choose the least burdensome way of doing that. On this understanding, it looks like the more flexible version of cost-effectiveness described in (v) above.

Remarkably, the text of the Act seems to give the Department no guidance on how to think about the choice among these possibilities. It appears to be close to a blank check. If executive branch officials are effectively given the authority to choose the rule of decision for important regulations – with no constraints on content – might there be a legitimate non-delegation problem (Sunstein, Reference Sunstein2008b)? Under longstanding law, the answer is highly likely to be “no” (Whitman v. American Trucking Association 2001), because the Court has not invoked the non-delegation doctrine to strike down an act of Congress in 80 years (and counting; Schecter Poultry Corp. v. U.S. 1935). Nonetheless, some justices have recently expressed interest in the doctrine (Gundy v. U.S. 2019), and the fact that the question is worth asking attests to the extraordinary breadth of the statutory grant of authority.

The Court has long required Congress to offer an “intelligible principle,” which would be both necessary and sufficient for validation (Gundy v. U.S. 2019). It is not easy to identify such a principle here.Footnote ³ If the Secretary is permitted to choose among the seven approaches sketched above, the non-delegation challenge would seem to be serious. Perhaps the best response to that challenge would invoke context and purpose, and suggest that something like (vi) is mandatory (Gundy v. U.S. 2019; Michigan v. EPA 2015).

Since 1981, American presidents have required officials (i) to identify a market failure, (ii) to show (to the extent permitted by law) that all regulations pass some kind of cost-benefit test, and (iii) to show that the chosen approach maximizes net benefits (Executive Order 13563 2011; Executive Order 12866 1993; Executive Order 12291 1981). For present purposes, relevant guidance comes from Executive Orders 12866 and 13563; since Executive Order 13563 incorporates Executive Order 12866, I shall sometimes refer to the former as shorthand. At first glance, the Act unquestionably “permits” the Department to use those ideas as the rule of decision (Executive Order 13563 2011, p. 3821). And because Executive Orders 12866 and 13563 require the Department to use a cost-benefit test and to maximize net benefits if it is permitted to do so, the essential task for the Department of Transportation seems straightforward: ensure that the benefits justify the costs, and maximize net benefits.

3.2 Market failure and experience goods

But what is the market failure? Consumers can demand cameras if they like. We should expect the motor vehicle market to provide an assortment of diverse offerings, in which consumers who are willing to pay for cameras (or some other technology) end up with them. The mix of offerings would change over time, depending on consumer preferences and perhaps decreasing costs. Why should government require all motor vehicles to come with cameras (or some other technology)? Why should consumers be compelled to buy them? Why would such a forced exchange be a good idea?

There are several imaginable answers. The first and most obvious points to externalities: drivers without cameras impose risks on others, including property holders, pedestrians, animals, and other drivers. In principle, the argument is secure. Insofar as backup maneuvers cause deaths and injuries to pedestrians or cyclists who are not family members, there are clear externalities. The question is their magnitude. It is also the case that drivers should of course be concerned about risks that they impose on themselves, and any crash imposes risks on them, not merely on third parties. Drivers’ concern for their own safety (and that of, say, their children) means that their consumption decisions should ensure that they take the externalities partly (or sufficiently?) on-board. This possibility raises some difficult conceptual and empirical issues, which remain to be fully answered. We should conclude that there are externalities here, but they might not be large.

The second market failure points to an inadequate information: consumers might not have a full sense of the safety benefits of cameras. Fortunately, deaths and serious injuries from backover crashes are relatively rare events (though backover accidents are not so rare). Are consumers able to think well about the benefits of reducing the very low risk that deaths and serious injuries will occur? Can they compare the costs of rearview cameras against the benefits? To do so, it would be important to know something both about the expected probabilities and the likely outcomes; to say the least, it is challenging to know those things. If consumers suffer from unrealistic optimism, the problem will be compounded (Sharot, Reference Sharot2011). Statistical knowledge, even if it exists, might not be enough if consumers also think: “I am a good driver, and the risk is essentially zero for me personally.” In short, ordinary consumers might well be inadequate decision makers when it comes to low-probability, high-consequence risks. Behavioral market failures of this kind would need to be shown, not merely asserted. But it is reasonable to think that with respect to very low-probability risks, drivers are making suboptimal tradeoffs.

A third and more speculative market failure, related to the second, points to the potential mismatch between (i) what consumers ask for or are offered and (ii) what will prove to be valuable to them or increase their welfare. To what extent do or will consumers want products or devices that do not yet exist? It is true that in many contexts, companies are able to answer that question with premarket tests. But in the period before the rear visibility regulation was finalized, the profit incentive to do so for rearview cameras was apparently modest, because consumers were not demanding or flocking to them. It is also true that companies that provide cameras should be expected to advertise their availability, while also providing information about their benefits (and also, perhaps, counteracting unrealistic optimism). But such information might not be an adequate corrective, especially in view of the very wide range of attributes that consumers consider in deciding what kind of car to buy.

All of these points are plausible. But the largest problem, I suggest, is that it is not easy to appreciate the benefits of cameras unless one has spent some time driving with them. Rear visibility is best seen as an experience good. If there is a mismatch between ex ante consumer demand and ex post consumer welfare, that is why. Drivers who have driven without cameras, and who are used to navigating without them, are unlikely to have a sufficient understanding of what it is like to drive with them. The point is not only about safety; it is also about ease and convenience. This is a market failure, but it is not a standard one. It involves limited foresight. It points to the difference between the welfare or utility expected at the time of decision (“decision utility”) and the welfare or utility actually enjoyed (“experienced utility”). It points to information that is best or only obtained through experience, and possibly also to changes in preferences and tastes. I speculate that the difference between decision utility and experienced utility is substantial in this context.

We lack conclusive data, but I will provide suggestive evidence in support of that speculation below. It is true that over time, markets should usually be expected to overcome the problem of limited foresight. People learn that electric shavers have significant advantages over straight edge razors, that air travel is better than train travel (at least over long distances), that large cell phones have advantages over small ones, that air conditioning really is great. Over time, products that deliver benefits, in terms of experienced well-being, will be rewarded in markets; information spreads. But in some cases, information spreads relatively slowly, and people can be locked into a suboptimal equilibrium for a long time. Regulation can be a justified response.

3.3 A glimpse inside government

If we stipulate that there is a market failure, or if we emphasize that the law requires action even if there is not, the numbers might make NHTSA’s task straightforward. Suppose, for example, that cameras would cost $400 million and create $600 million in benefits; that sensors would cost $100 million and create $80 million in benefits; and that additional mirrors would cost $50 million and create $25 million in benefits. With such numbers, the argument for cameras would seem conclusive, and NHTSA would have no discretion under Executive Order 13563. The principal qualification involves “different types of motor vehicles.” Suppose that trucks accounted for $100 million of the total cost of cameras – but $500 million of the total benefits. If so, Executive Order 13563 would appear to require NHTSA to make relevant distinctions, by requiring cameras in trucks but sensors or additional mirrors in other smaller vehicles.

The basic point is that if the numbers work out in certain ways, Executive Order 13563 might make NHTSA’s task fairly straightforward – and sharply constrain its discretion. At least this would be so if the technical experts, within the Department, were able to generate numbers of that relatively precise kind.

I can report that in 2009, officials from NHTSA asked for a White House meeting with me and offered some numbers on three proposals: mirrors only; sensors; and cameras. They sought preliminary guidance on the likely views of the Office of Information and Regulatory Affairs, which oversees federal regulations (Sunstein, Reference Sunstein2013). Their early analysis suggested roughly the following. (i) Mirrors would cost very little, but would do almost nothing about the problem. The cost-benefit ratio would be very bad. (ii) Sensors would cost much more, but would not do a great deal. The cost-benefit ratio would also be bad, although better than for mirrors. (iii) Cameras would cost the most by far. They would also prevent a significant number of premature deaths, and indeed were the only approach that would do so. For them, the cost-benefit ratio was the best of the available options. As I recall, the ratio was something like $15 million per life saved – well above the standard figure, which is now in the vicinity of $9 million (U.S. Department of Transportation, 2016), but well below the corresponding figures for mirrors and sensors.

I was aware that under U.S. Office of Management and Budget (“OMB”) Guidance and for reasons to be explained shortly, what matters is net benefits, not the cost-benefit ratio. NHTSA’s figures showed that all three options had net costs, not net benefits, and that mirrors had the lowest net costs and that cameras had the highest. An analysis of the cost-benefit ratio, favoring cameras, led in exactly the wrong direction; it suggested that the worst option was the best. It would therefore be standard for the Office of Information and Regulatory Affairs (“OIRA”) to argue vigorously in favor of mirrors, and perhaps to insist on selection of that option. Nonetheless, and over the objection of some members of my staff, I gave a strong signal that NHTSA should look carefully at cameras, and that we would be likely to be receptive to that approach. For better or for worse, my judgment on that count was a product of three considerations: (i) cameras were the only approach that would significantly dent the problem and perhaps were required by law, (ii) very young children were at risk, and (iii) there were likely to be non-quantifiable factors, on which we did not have an adequate handle, that would tip the balance.

3.4 The proposed rule

3.4.1 Numbers

As the analysis was formalized in the proposed rule, the official numbers did not make things easy. According to NHTSA’s estimates, there are 292 annual fatalities and 18,000 annual injuries from backover crashes (75 Fed. Reg. 76,186 2010). About 44 % of the fatalities involve children under the age of five; gruesomely, many of those “involve parents (or caregivers) accidentally backing over children” (75 Fed. Reg. 76,186 2010, p. 76,187). At the proposal stage, cameras would have cost $1.9 billion to $2.7 billion; sensors $300 million to $1.2 billion; and mirrors $600 million (75 Fed. Reg. 76,186 2010, p. 76,236). The monetized benefits, using a 3 % discount rate, would have been between $780 million and $920 million for cameras and around $47 million for sensors (and much less for mirrors; 75 Fed. Reg. 76,186 2010, p. 76,237).Footnote ⁴ Those benefits include deaths and injuries prevented, and also the prevention of property damage.

To have a sense of what we are talking about, a statistical life was valued, at the time, at around $6.1 million (it is now over $9 million), which means that in terms of statistical life equivalents, the rule would prevent between 130 and 160 deaths. Because much of the total monetized benefit comes from the prevention of accidents and property damage, the number of actual deaths prevented would of course be significantly lower. (See the discussion of the final rule below for some details.)

Consistent with the analysis during that early meeting, cameras would have been the only effective response to the problem of backover crashes, with the other two contributing very little – and again cameras would have had, by far, the best cost-effectiveness ratio. There is no question that in terms of cost per life saved, cameras were the preferred option. At the same time, all three options would have negative net benefits, which appears to mean that they would be worse than doing nothing at all – and of the three options, cameras would have, by far, the highest net costs. What seems, on one view, to be unquestionably the best option is, on another view, unquestionably the worst. Congress did not think at all about this prospect, and it might be doubted whether it was equipped to do so. By contrast, the executive branch certainly did.

Under Executive Order 13563, the issue would seem to be at an end, at least on these numbers. (As we shall see, the word “seem” is important here.) NHTSA should do nothing, because no approach would have net benefits, and if it ended up doing something (as the Act seems to require), additional mirrors would be preferable, because they would impose the lowest net costs. In this regard that the decisive question is not the cost-benefit ratio (on which cameras look like the best optionFootnote ⁵) but instead the net benefits or costs (U.S. Office of Management and Budget, 2003). The reason is that the latter figure provides valuable information about the social welfare effects, as the former does not. A rule that costs $1 billion but that has $1.5 billion in benefits has a benefit-cost ratio of 1.5 to 1, which is not nearly as impressive as a rule that costs $2 and that has $1000 in net benefits, for a ratio of 500 to 1. But in welfare terms, it is much better to deliver $500 million in net benefits than to deliver merely $998.

The analysis is symmetrical for rules that produce net losses. A rule that costs $100 and has benefits of $1 is much better or less bad than a rule that costs $500 million and has benefits of $100 million. I have noted that under OMB guidance (U.S. OMB, 2003), the task of the Department is to produce the highest net benefits or the lowest net costs, and on those counts, mirrors would be best. The whole point of cost-benefit analysis is to provide information about the effects on social welfare, and on this count, the net figure is what matters, not the ratio.

Under the Act and Executive Order 13563, there is a further question, which is whether to make distinctions among vehicles. It should go without saying that a great deal depends on the costs and benefits of doing so, and to know that, we would have to produce a set of numbers. We could imagine an analysis that would show that it would be best to require cameras on some vehicles and mirrors on others. Here as well, the Department had a great deal of information, some of which is captured in Table 1.

Table 1

Rear visibility proposal and alternatives discounted at 3 % (millions of 2007$) in decreasing order of installation costs and monetized safety benefits, with LT referring to light trucks and PC referring to passenger cars.

The range of camera costs assumes 130° camera with the display in the dashboard (lower cost) to the display in the mirror (higher cost; 75 Fed. Reg. 76,186, p. 76,238 note 91).

Table 2

Summary of benefits and costs passenger cars and light trucks (millions of 2010$) MY2018 and thereafter.

Data sourced from 79 Fed. Reg. 191,181 (2014). “The different estimates in this chart show some of the different potential technology options. The primary estimate is the lowest installation cost option (which assumes manufacturers will use a 130 ° camera and will utilize any existing display units already offered in their vehicles). The low estimate and high estimate provide the estimated minimum and maximum net impacts possible. The low estimate is the 180 ° camera and assumes that manufacturers will install a new display to meet the requirements of today’s rule. It represents the minimum overall benefit estimate as it has the largest negative net impact. Conversely, the high estimate is the 180 ° camera and assumes that manufacturers that currently offer vehicles with display units are able and choose to use those existing display units to meet the requirements of today’s rule. This represents the maximum overall benefit estimate because it has the smallest negative net impact.” 79 Fed. Reg. 191,181, n.13 (2014).

Table 1 does not include mirrors, but it does show that with different forms of mixing and matching, net costs can move in significantly different directions. Restricting cameras to light trucks, and exempting passenger cars, would have lower net costs (and also higher cost-effectiveness). The precise choice is less important than the general lessons. From this information, we might draw three such lessons: (i) if NHTSA seeks to comply with Executive Orders 13563, it should do nothing; (ii) if it must do something, it should be inclined to favor mirrors; and (iii) whatever it does, it ought to make distinctions among categories of vehicles in a way that maximizes net benefits or minimizes net costs.

But these conclusions raise their own complications. We should agree that the Act requires the NHTSA to do something, even if all options have negative net benefits. Inaction is not an option. Indeed, the statute appears to contemplate that the NHTSA must do something with respect to every class of vehicle – even if each class might be treated differently. There is also an argument that the Act does not permit NHTSA to choose an option that has a de minimis effect on the problem that motivated it.Footnote ⁶ If additional mirrors would achieve almost nothing in terms of the statutory goal, it might be taken to be inconsistent with the Act to mandate them, even if the more effective responses had significantly lower net costs. To that extent, engagement with the evidence may have the virtue of sharply narrowing the category of responses that NHTSA might select.

4.1 Quantification and its discontents

In terms of both costs and benefits, the most important change from the proposal stemmed from the fact that the automobile industry was moving rapidly in the direction of installing cameras on its own – which would decrease both costs and benefits substantially (79 Fed. Reg. 19,178 2014). According to the Department, about 73 % of covered vehicles would have rearview video systems by 2018, even without the regulation. With that assumption, the rule would cost $546 million to $620 million annually. This is a large amount in aggregate, but it is also useful to emphasize this finding: “We anticipate rear visibility systems will cost approximately $43-$45 for vehicles already equipped with a suitable visual display and between $132 and $142 for all other vehicles” (79 Fed. Reg. 19,178 2014, p. 19,181). (Most contemporary vehicles have a suitable display.)

In terms of social welfare, it is worthwhile asking whether those relatively small amounts might have modest adverse effects on consumers of new vehicles. A high but widely dispersed monetary cost might give a misleading picture of the welfare effects. If (say) a regulation requires 100 million people to pay $20 annually, the annual cost of $2 billion might overstate the welfare effect of the regulation (Dorman, Reference Dorman1996; Bronsteen et al., Reference Bronsteen, Buccafusco and Masur2013).

The rule would produce $265–$396 million in monetized benefits, including prevention of 13–15 annual deaths and 1125–1135 annual injuries. (“Thus, we believe that there will still be 13–15 fatalities and 1125–1332 injuries prevented annually that are a result of equipping the remaining 27 % of vehicles that we do not anticipate will have rear visibility systems by 2018” 79 Fed. Reg. 19,178 2014, p. 19,180.) The agency was aware that the proposed rule might itself have accounted for some of the growth of cameras and that without the proposal, adoption might be as low as 59 %, which would increase the costs to $827 million to $924 million, and increase the benefits from $398 million to $595 million.Footnote ⁷ Because of the growth of voluntary adoption of cameras, the agency expected to prevent only that small number of preventable deaths each year (79 Fed. Reg. 19,178 2014). In addition, the Department’s value of statistical life had changed, in the interim, to about $9 million (Trottenberg & Rivkin, Reference Trottenberg and Rivkin2013).

Table 2 summarizes various numbers, and these raise many further questions. Even with a degree of market penetration, mandatory cameras would cost hundreds of millions of dollars, without the kinds of safety benefits usually associated with rules having that expense. Return then to the options of sensors and mirrors; might either approach be preferable to cameras? The Department reiterated that for technical reasons, both would be inadequate. With respect to sensors, the agency found, on the basis of its own evidence, “that sensor-only systems have various technical limitations that lead to inconsistent object detection and that drivers with sensor-only systems generally either failed to respond to the sensor system’s audio warning, or paused only momentarily before resuming the backing maneuver” (79 Fed. Reg. 19,178 2014, p. 19,182). With respect to mirrors, the agency found “that drivers were unable to avoid targets behind the vehicle when assisted with additional rear-mounted mirrors such as rear convex ‘look-down’ or cross-view mirrors” (79 Fed. Reg. 19,178 2014, p. 19,182).

Perhaps surprisingly, and somewhat disturbingly, the agency did not offer actual benefits numbers for the two less expensive approaches, but it did say that “sensor-only and mirror-based rear visibility systems have demonstrated little to no success in inducing drivers to stop a backing maneuver to avoid a crash with a pedestrian behind the vehicle” (79 Fed. Reg. 19,178 2014, p. 19,183). Hence, “their lower cost is outweighed by the substantially reduced benefits that are likely to be achieved by these systems.” In these circumstances, cameras would be “not only the most effective systems at addressing the backover safety problem but also the most cost effective system” – and also the only way to fulfill the requirements of the Act, “as these other systems cannot be reasonably expected to address the backover crash problem” (79 Fed. Reg. 19,178 2014, p. 19,183). The agency added that ultrasonic sensor systems would be far more expensive than originally thought, costing between $79 and $138 per vehicle.

What about the fact that the quantifiable benefits were lower than the quantifiable costs? As it had in its proposal, NHTSA emphasized that “a simple quantitative analysis is not sufficient” (79 Fed. Reg. 19,178 2014, p. 19,235). It drew attention to equity. It noted that “victims of backover crashes are frequently the most vulnerable members of our society (such as young children, the elderly, or persons with disabilities)” (79 Fed. Reg. 19,178 2014, p. 19,180). It pointed to “strong reasons, grounded in unquantifiable considerations, to take action to prevent the deaths and injuries at issue here” (79 Fed. Reg. 19,178 2014, p. 19,236). It stated that “most people place a high value on the lives of children and that there is a general consensus regarding the need to protect children as they are unable to protect themselves” (79 Fed. Reg. 19,178 2014, p. 19,181). It said, “While the agency has used the Department’s standard monetary figure for the value of a statistical life, we acknowledge that various studies have placed the value of a statistical life at a higher value and the value of a statistical life of a child even higher” (79 Fed. Reg. 19,178 2014, p. 19,236).

It added: “In many cases, parents are responsible for the deaths of their own children. We continue to believe that avoiding that horrible outcome is a significant benefit which is not fully or adequately captured in the traditional measure of the value of a statistical life” (79 Fed. Reg. 19,178 2014, p. 19,236). It stated that “an exceptionally high emotional cost, not easily convertible to monetary equivalents, is often inflicted upon the families of backover crash victims” (79 Fed. Reg. 19,178 2014, p. 19,181). It emphasized: “Of course, any death of a young child is a tragedy, but we believe that this traditional measure also does not adequately account for the value of reducing the risk that parents will be responsible for the death of or serious injury to their own children” (79 Fed. Reg. 19,178 2014, p. 19,236). It referred to distributional impacts more broadly, including not only children but also people with disabilities and the elderly. “Especially in the context at issue, such people lack relevant control over the situation and are not in a good position to protect themselves. There are strong considerations, rooted in fairness and equity, to reduce these risks that they face” (79 Fed. Reg. 19,178 2014, p. 19,236).

It briefly noted as well that “[d]rivers will benefit in numerous ways from increases in rear visibility. For example, parking will be simplified, especially in congestion.” The increase in “ease and convenience” would provide “significant, but not yet quantifiable, benefits to drivers” (79 Fed. Reg. 19,178 2014, p. 19,236).

4.2 The absence of breakeven analysis

It is surprising and disappointing that the Department did not undertake a formal breakeven analysis. Such an analysis would have explored what the benefits would have had to be to justify the costs, and would have analyzed the assumptions that would support the conclusion that they did.Footnote ⁸ Perhaps it declined to do so on the ground that any such analysis would rest on highly speculative assumptions. But it would nonetheless have been possible. The Department did not offer a point estimate for the shortfall, but for purposes of analysis, let us stipulate that on reasonable assumptions, it was in the general vicinity of $200 million. Under breakeven analysis, the question was whether the unquantifiable values could make up the difference. Taken in the abstract, and without saying more, that question is difficult to answer. But the Department might have made a great deal of progress by saying a bit of about the relevant values.

Most obviously, the Department properly referred to the increased ease and simplification of driving. Suppose that the relevant improvement is valued, on average, at merely $25 annually, admittedly a somewhat arbitrary figure, but taken as a reasonable (and conservative) lower bound. Suppose too that the regulation would apply to eight million cars that would otherwise lack cameras. If so, it would produce $200 million in annual benefits. At that point, the monetized benefits are close to the monetized costs. We are essentially at the breakeven point.

The Department might have also noted that there is an emerging work on the valuation of children’s lives (Robinson et al., Reference Robinson, Raich, Hammitt and O’Keefe2019). At the time it finalized the rule, some preliminary work suggested that parents value a young child’s life at $18 million (Williams, Reference Williams2013) – a number that could add $90 million to its existing benefits figure (assuming that the rule would prevent five such deaths a year). At that point, the benefits exceed the costs. And indeed, that $18 million figure captures the parents’ valuation of children’s lives, not children’s valuation of their lives. It would have been an unusual step in view of the tentative nature of the existing research, but the Department might have undertaken a sensitivity analysis with values of $18 million and $27 million – both indicating that the benefits are comfortably in excess of costs. (Of course, it might be a form of illegitimate double-counting to attempt both to monetize children’s lives and to refer independently to “equity.”)

More recent and far more systematic work gives different numbers, generally suggesting higher values for children than for adults, in the range of a multiple of 1.5 (Robinson et al., Reference Robinson, Raich, Hammitt and O’Keefe2019). If we use that multiple for the $9 million VSL figure, the valuation of children would be $13.5 million, which would produce benefits of $67.5 million (again assuming prevention of five children’s deaths annually). As noted, the agency could plausibly say that the $200 million is a reasonable lower bound for the additional ease of driving and that even with that reasonable lower bound, the value of protecting young children would ensure that the benefits justify the costs. The discussion below of experience goods, with a relevant survey, strongly supports that conclusion.

Recall finally that we are speaking here of parents who would not only (only!) lose their children, but who would also be directly responsible for that loss. How much would it be worth to reduce the risk of that eventuality? Any reasonable lower-bound figure would fortify the conclusion that the costs would be justified. The numbers given here are of course speculative. But with an analysis of this admittedly tentative kind, the Department’s conclusion seems eminently sensible – not because of a laundry list of unquantifiable benefits, but because once we begin to speak of lower bounds and expected ranges, an apparently intractable puzzle begins to dissolve, or at least to look far more tractable.

5.1 Willingness-to-accept and willingness-to-pay

It is not easy to test the possibility that rearview cameras are experience goods. To obtain some information, I conducted a brief survey of 403 Americans in 2019, using Amazon’s Mechanical Turk:

The answers fell in the following categories:

• Less than $50 – 5.26 %

• $50–$99 – 7.77 %

• $100–$199 – 13.28 %

• $200–$299 – 16.79 %

• $300–$399 – 12.78 %

• $400–$499 – 8.77 %

• More than $500 – 35.34 %

The most remarkable finding here may that over one-third of respondents said that they would demand more than $500 to give up rearview cameras. It is also noteworthy that 94 % of respondents would demand, to give up cameras, more than the standard cost of cameras ($43–$45, for vehicles with modern displays) – and that a lower bound of 74 % would be willing to demand more than the high-end amount ($132–$142, for vehicles without such displays).

It might be expected that the driver of low numbers is low income, and that the driver of high numbers is high income. To test for that possibility, I divided participants into two groups, both with about 200 people: those who earned less $60,000 annually and those who earned more than that. Among those who earned less, 93 % would demand $50 or more. Of those who earned more, 97 % would demand $50 or more. More generally, the willingness to accept figures are closely comparable across the two groups.

We might hypothesize that driver experience contributes to the relatively high numbers; those who have driven cars with cameras do not want to drive cars without them. (On that hypothesis, see below.) A “pure” endowment effect might also be at work. Through the phrasing of the question, people are asked to assume that cars do have cameras, and the question is how much they would demand to give them up (rather than pay to get them in the first place). When an endowment effect is at work, willingness-to-accept will be higher than willingness-to-pay, and we might question whether willingness-to-accept captures the welfare effects of a good (Frederick et al., Reference Frederick, Novemsky, Wang, Dhar and Nowlis2009).

With these issues in mind, I did another survey with this question:

The results are consistent with the hypothesis of an endowment effect, accounting for a significant disparity between willingness to accept and willingness to pay. Notice the much lower numbers:

• Less than $50 – 12.90 %

• $50–$99 – 27.05 %

• $100–$199 – 29.03 %

• $200–$299 – 19.35 %

• $300–$399 – 5.71 %

• $400–$499 – 2.98 %

• More than $500 – 2.98 %

But, even with willingness to pay, more than 87 % would be willing to pay more than the standard cost of a camera. Here as well, I divided participants into two groups: those who earn more and less than $60,000. About 86 % of the lower-income group would be willing to pay at $50, compared to 88 % of the higher-income group. Here again, the distribution of answers is not radically different between the two groups (though 75 % of the lower-income group would pay less than $200, compared to 63 % of the higher-income group).

5.2 The effects of experience

For present purposes, the main issue lies elsewhere. In both versions of the survey, I also asked people whether they now own a car with a rearview camera. Hypothesizing that cameras are experience goods, I expected that drivers value such cars more as a result of experience; if they did not own one, their valuation would be lower. The hypothesis was supported. In the willingness to accept condition, the low figures were dominated by people who do not own such a car: of the 21 people who said they would demand less than $50, almost all (20) did not have one. The numbers are less dramatic but in the same direction in the willingness to pay condition: of the 52 people who said that they would be willing to pay less than $50, the vast majority (38) did not own such a car.

On the high end, the results were more ambiguous. Of the 141 who would demand more than $500, 72 owned such a car, as compared to 69 who did not. Of the 11 who would be willing to pay more than $500, nine owned such a car; because of the small number of people in this category, it would be a mistake to make much of the finding, but it is suggestive. (It is true, of course, that some people may have chosen a motor vehicle with a rearview mirror, and so those who place a high value on mirrors might simply have an ex ante preference for them. But because such vehicles really started to penetrate the market after the rule was proposed in 2012, that is not very likely – a point to which I will return.)

The most important findings here involve the low numbers, which are dominated by people with cars lacking rearview cameras. Whether we are speaking of willingness to accept or willingness to pay, those who have had experience with such cameras are overwhelmingly likely to believe that they are worth more than $50.Footnote ¹⁰ Those findings are consistent with the view that on net, consumers are benefiting from rearview cameras. It is worth pausing over the relevant numbers. Suppose that someone owns a car for 5 years. Is it even possible, for the vast majority of consumers, that having a rearview camera is worth less than $10 per year? We might well conclude that at a cost of $45 per vehicle, the rearview camera mandate is essentially a no-brainer, and the real question is whether the executive branch struggled so long and hard over it. But that question is meant to raise puzzles about economic analysis as it is standardly practiced, not about the competence of executive branch officials.

As I have noted, there is a possible confound in the surveys. Perhaps a significant number of respondents actually sought a vehicle with a rearview camera; perhaps the numbers reflect that fact. To get more clarity on this question, I repeated the survey, but with one additional question: I asked people if they had desired, and specifically sought out, vehicles with cameras. (There were about 160 participants in both willingness-to-pay and willingness-to-accept conditions.) About 1/3 of participants reported that they did so. The key question, of course, is whether those who had vehicles with rearview cameras, but did not desire and seek out that technology, would report low willingness to accept and willingness to pay figures.

The answer is that they did not. In the willingness to accept condition, 96 % of those who did not seek out rearview cameras would demand $50 or more; 92 % would demand $100 or more; and 60 % would demand $500 or more. In the willingness to pay condition, 80 % of those who did not seek out cameras would pay $50 or more and 65 % would pay $100 or more. Because of the relatively small sample size, it would be a mistake to give a great deal of weight to the particular numbers. The larger point is that the principal findings in the larger study are replicated here, in general and also among the class of people who did not specifically seek out motor vehicles with rearview mirrors.

5.3 Experience goods and regulatory policy

I have emphasized that the surveys should be regarded as preliminary, and the same is true for the analysis. To understand consumer behavior, a great deal more might be done, perhaps with nationally representative examples. To understand my broader claim, a great deal of conceptual as well as empirical work remains. In principle, government regulation might deliver experience goods that improve social welfare. It is fair to say that this is a distinct category of market failure, and also to say that it is extremely hard to engage in ex ante cost-benefit analysis that would show whether a welfare improvement will occur. There is more reason for confidence in the existence of the category than in the proposition that regulators should frequently or readily proceed in order to deliver experience goods. At least over time, markets generally do well on this count, and regulators might not.

It is important to emphasize that automobiles are among the largest investments that people make, and test drives are typically taken prior to purchase. The whole purpose of test drives is to offer a sense of how drivers will experience various product characteristics. If test drives are not adequate on that count (because we are dealing with durable goods for which learning may be slow), at least they provide something beyond what can be found through search alone. In addition, the market for new cars is heavily influenced by the expert assessments of consumer-focused reporters who do test drives of new models and report their reactions to new technologies and features. This process facilitates market diffusion of new technologies and features, even when experience goods are involved. It remains true, here as well, that the process may be inadequate.

These points suggest that the difference between search goods and experience goods might be best understood as one of degree; they might not be dichotomous. With some goods, search provides all the information that one needs to have. With other goods, experience is essential; search provides a small fragment of what is necessary. With still other goods, search provides a great deal of information, but consumers miss something that might be important. Cameras that provide rear visibility probably belong in the third category. Something similar might be said about other safety-related features of cars, such as the beeping sounds that many cars make when drivers leave their lane, or when they get close to hitting what is in back of them. (Should such features be mandated? Possibly, with the same analysis of costs and benefits, and of experience goods, offered here.)

I have also noted that when experience increases valuation, it is not always clear why. In the simplest cases, and it is my emphasis here, what people gain is information; they see what they have been missing, and they can do that only or best with experience. In other cases, people’s tastes shift, and not merely because of learning. In such cases, it would be true but inadequate to say that they have gained information. Extreme cases involve “transformative experiences,” rarely relevant to regulatory policy, but nicely illustrating situations in which new experiences produce new tastes and values (Paul, Reference Paul2016; Ullmann-Margalit, Reference Ullmann-Margalit2017). In some situations, the main effect is that a new experience makes an old one seem less pleasant or less valuable; people obtain less welfare from consuming or experiencing it, only because they have consumed or experienced something different or better. In such situations, the argument for regulatory intervention seems weak; regulators should not order adoption of New Technology B when and because those who use it will then dislike Old Technology A. In other situations, the new experience produces a kind of dependency, as when people become addicted to cigarettes; cameras that provide rear visibility do not entirely lack this feature. The argument for regulation is plausible or potentially strong only when experience produces some kind of welfare gain, and when consumers cannot anticipate that gain ex ante.

I have referred to the problem as one of limited foresight, and it helps illuminate a number of other actual or potential regulatory interventions; many such interventions might be increasing the provision of experience goods. One of the advantages of vehicles with high fuel economy is that drivers save time; they do not have to go to the gas station nearly as often. Do consumers sufficiently take account of this benefit when they purchase vehicles? We do not know, but perhaps not. The U.S. Department of Transportation has issued a series of rules designed to protect airline passengers, for example by forbidding certain delays on the tarmac. It is fair to ask: What is the market failure? Perhaps there is none; perhaps the rules are unjustified. But perhaps consumers have not been sufficiently attentive to the welfare costs of those delays, either because of insufficient experience or because they have been low-probability events. (It is true that in these cases, the real problem may be is that the good in question has numerous characteristics and consumers do not sufficiently focus on one, because it does not loom large, even though it can have a real effect on welfare.)

It is important to emphasize that markets can and do handle experience goods, as emphasized in the original treatment (Nelson, Reference Nelson1970; see also Israel, Reference Israel2005). Regulation is hardly justified merely because such goods are involved. People obtain experience all the time, and they make decisions accordingly. Word-of-mouth can do a great deal to communicate the value of experience goods; social learning is critical (Feldman et al., Reference Feldman, Papanastasiou and Segev2017). It is also true that producers can adopt innovative approaches to solve the problem – as, for example, through free or low-cost trial periods. What remains to be specified are the circumstances in which markets fail because (for example) social learning and innovative approaches on the part of producers turn out to fail. The only points here are that under imaginable assumptions, experience goods provide a plausible ground for regulation on social welfare grounds, and that rearview cameras appear to be a case in point. A great deal of work remains to be done on this topic.