of the inspection and, if the vehicle passes, the certificate needed for .... These geographic markets are: Apple Valley-Hesperia-Victorville ("AHV"), ..... From this I obtain the unconditional probability by integrating out over the distribution of :.
An Empirical Examination of Moral Hazard in the Vehicle Inspection Market
Thomas N. Hubbard
University of California, Los Angeles
I would like to thank participants of various seminars for their comments. Buzz Breedlove at the California Research Bureau, and Steve Gould, Janet Maglinte, and Tracy Shinmoto at the California Bureau of Automotive Repair have been extremely helpful. I would also like to thank Antonio Bernardo, Shane Greenstein, Wei-Yin Hu, Matthew Kahn, Peter Reiss, Andrea Shepard, and Scott Stern for offering advice and Roger Noll, Frank Wolak, and especially Tim Bresnahan for guidance. Finally, thanks to Rob Porter and two anonymous referees for their comments. All errors are mine. Financial support from the Lynde and Harry Bradley Foundation, and an Alfred P. Sloan Doctoral Dissertation Fellowship are gratefully acknowledged.
Abstract A moral hazard problem arises in “diagnosis-cure” markets such as auto repair and health care when sellers have an incentive to distort or misrepresent a buyer’s condition in order to increase demand for the treatments they supply. This paper examines one such market: California vehicle emission inspections. Using transaction-level data, I investigate whether the market provides incentives that lead inspectors to help vehicles pass. A novel feature of these data is that they contain inspections completed by state officials outside of the normal inspection market as well as inspections completed by individuals working at competitive private firms. Contrasts between these groups provide evidence regarding the strength of market incentives. I also examine how the behavior of inspectors varies with their firm’s competitive environment and organizational characteristics. I find that consumers are generally able to provide firms and inspectors strong incentives toward helping them pass, and I find cross-firm differences that are consistent with agency theory.
1. Introduction Buyers in auto repair, health care, and other “diagnosis-cure” markets generally are unable to determine their condition, and can neither perfectly observe nor costlessly verify sellers’ actions or recommendations. A moral hazard problem arises because sellers have an incentive to distort or misrepresent buyers’ condition to increase demand for the treatments they supply. In some cases, buyers and sellers can discourage post-contractual opportunism with contracts that are contingent on verifiable events — for example, lawyers sometimes base their fees on the outcome of the case. However, explicit contingent contracts have limited application in health care and auto repair markets, possibly because they create new adverse selection or moral hazard problems or because outcomes are not always verifiable. Whether buyers can give sellers incentives to act in their interest in the absence of contractual solutions is an open empirical issue. Anecdotal evidence suggests that at least some sellers in these markets mislead their customers — mechanics recommend unnecessary work, doctors prescribe excessive treatments — but whether these are representative is unknown. This paper studies one segment of the auto repair market — California vehicle emission inspections. In this market, inspectors can affect the probability vehicles fail in ways consumers can neither directly observe nor costlessly verify, and inspection suppliers profit when they repair failed vehicles. I examine two related research questions. First, to what extent do market incentives lead inspectors to help vehicles pass? Second, how does inspectors’ behavior vary with their firm’s competitive environment and organizational characteristics? I analyze transaction-level data supplied by the California Bureau of Automotive Repair. A novel feature of these data is that they contain inspections completed by state officials outside of the normal inspection market as well as inspections completed by individuals working at competitive private firms.
1
The empirical framework is simple. I specify whether individual vehicles fail inspections as a function of vehicle characteristics and whether the inspector is a state official or works at a private firm. When the inspector works at a private firm, I also include characteristics of the firm as independent variables. I attribute differences in the failure probability, conditional on vehicle characteristics, to differences in how inspectors conduct inspections. Contrasting the behavior of state officials and inspectors at private firms provides evidence regarding the strength of market incentives. Differences across private firms may reflect differences in their internal incentives. I therefore provide an incentive-theoretic interpretation of the cross-firm results. However, these differences could also be due to the process in which consumers select individual firms. Controlling for differences in vehicles’ characteristics and operating condition, inspection failure rates are more than twice as high in inspections conducted by state officials as in those completed at private firms. The main exception is when emission repairs are covered by a warranty at the firm inspecting the vehicle: for late model, low-mileage vehicles inspected at new car dealers. The incentives presented by consumers in this market are generally quite strong: when consumers incur the full repair costs, inspectors help vehicles pass. One explanation is that inspectors and their firms hope to establish or maintain reputations. I also find differences in inspection conduct across firms. Failure probabilities are slightly lower at firms with close geographic competitors. They also vary with firms’ organizational attributes. After controlling for differences in other firm characteristics such as station age, the number of inspectors, and whether inspectors and mechanics are paid piece rates, failure probabilities are higher at “chain stores” (such as Sears and Pep Boys) than at independent garages, service stations, new car dealers, and tune up shops. Owners face difficulties in providing managers
2
of auto repair shops effective performance incentives when managers are not residual claimants. Failure probabilities also increase with the number of inspectors employed by the firm. Free rider problems within firms may weaken individual inspectors’ incentives to help vehicles pass. Finally, failure probabilities are low at new car dealers for vehicles not under warranty. Klein and Leffler (1981) and Shapiro (1983) suggest that rents provide firms incentives to supply high quality services. New car dealers have higher prices, and probably also higher mark-ups, on inspections and repairs than other firms. The rest of the paper is organized as follows. Section 2 describes the inspection process and explains how inspectors can exercise discretion. I then describe inspectors’ objectives and show how their behavior may be influenced by consumers, regulators, and others in the firm. Section 3 describes the data used in the paper. Section 4 reports and interprets results from simple logits. Section 5 describes an empirical framework that allows unobserved vehicle characteristics to differ across station types. I then contrast estimates using this framework with the simple logits. Section 6 concludes. 2. The Market for California Vehicle Emission Inspections By state law, vehicles must receive emission inspections ("smog checks") in order to be registered in California. The state licenses individuals ("inspectors") to complete inspections and perform emission-related repairs. Inspectors are employed by private firms, which are also licensed by the state. Inspection prices are unregulated. They vary across firms for initial inspections. Most firms do not charge for reinspections of vehicles which initially failed an inspection at the firm, provided that the vehicle was not taken elsewhere for repairs between inspections. Inspections have two broad components: measurement of tailpipe emissions and a visual
3
check of emission-related parts. In the "emission component," inspectors place a device that measures hydrocarbon and carbon monoxide emissions in vehicles’ tailpipe. Vehicles pass this component if both hydrocarbon and carbon monoxide emissions are below model-year-specific standards when their engines are at idle speed and at 2500 RPM. In the "underhood component," inspectors examine the condition of emission-related parts, particularly those designed to limit evaporative and NOx emissions (which are not measured in the emission component). Vehicles pass this component when inspectors deem that all such parts are connected and functioning. The inspection procedure is guided by software routines embedded in machines specifically designed for smog checks. These machines prompt inspectors for vehicle-specific information, determine the relevant standards for the emission component, measure emissions, and ask inspectors whether the vehicle passes each part of the underhood component. They then print out the results of the inspection and, if the vehicle passes, the certificate needed for registration. This automation does not eliminate all discretion from the process. Inspectors can change the probability that vehicles fail inspections by adjusting their engine beforehand. They can also do so without making specific repairs. For example, warming up the engine promotes more efficient combustion and lower tailpipe emissions. This lowers the failure probability for the emission component. In the underhood component, inspectors can simply be more or less lenient in determining compliance. Vehicles fulfill inspection requirements when they pass inspections. Those which fail on the first try generally must be reinspected until they pass. 1 Between inspections, consumers may purchase repairs to lower emissions and increase the likelihood they pass. For some vehicles and
1
Consumers can obtain waivers of the inspection requirement when state officials verify that it would be prohibitively costly to bring their vehicles into compliance. 4
at some firms, emission-related repairs are covered by warranties. Federal law requires vehicle manufacturers to provide 5-year/50,000 mile warranties for emission-related repairs. These generally apply only at new car dealers. Firms’, Inspectors’, and Consumers’ Objectives and Decisions Firms choose their organizational characteristics, whether to supply inspections, and prices in order to maximize profits. Organizational characteristics include their other lines of business, hierarchical structure, and internal incentive schemes. For the purposes of this paper, entry, organizational characteristics, and prices will be taken as exogenously given. The analysis should be interpreted as conditional on these decisions. Inspectors conduct inspections to maximize their utility, which is a function of income and effort.
They are licensed to supply emission-related repairs, and are generally paid a function of
the work they complete. Inspectors have an incentive to help vehicles fail because it increases demand for emission-related repairs in the short run. However, inspectors are agents of as many as three sets of principals whose objectives differ. Two are regulators and consumers. Because inspectors’ objectives may not coincide with those of their firm, a third category of principals may be their firm’s owner or manager. Each of these interested parties may present incentives that influence how inspectors behave. Regulators prefer that the inspection of a vehicle measures its in-use emissions, and is not distorted by the inspector’s actions. If they can demonstrate that inspectors do not help high-emitters pass, regulators can claim that the inspection and maintenance program generates the emission reductions required by environmental legislation. Regulators monitor firms and inspectors and issue
5
citations if inspections are conducted improperly. 2 Examples of improper conduct are adjusting vehicles to make them more likely to fail, measuring the emissions of a different vehicle than the one receiving the certificate ("clean-piping"), and being overly strict or lenient on the underhood component. Regulators consider the latter the most pervasive. Because environmental objectives are important to regulators, and because false passes are considered more common than false failures, enforcement focuses on whether inspectors let high-emitting vehicles pass. During my sample period, regulators monitored inspectors by sending undercover staff to stations in vehicles that should have failed the inspection, and giving citations to firms and inspectors when they passed.3 In contrast, there was no special program to deter inspectors from failing low-emitting vehicles. Investigations of individual firms or inspectors were initiated by consumer complaints, and handled by the California Bureau of Automotive Repair in the same way as investigations of general auto repair fraud. Consumers choose a firm to maximize utility. While some consumers may have a non-costrelated component of utility,4 for most maximizing utility amounts to minimizing the expected cost of obtaining a passing inspection. This includes the price of the inspection, time and travel costs, and any costs associated with failing inspections. Inspection prices are known by consumers once
2
Firms and inspectors are fined $250 apiece for the first violation. The amount increases by $250 for each subsequent offense. Regulators may revoke the licenses of inspectors after their fourth offense, and must do so after their fifth. 3
During 1992, regulators audited each inspection station once on average. While they probably concentrated their efforts on firms suspected of misconduct, they did not target firms very well. For example, regulators did not use inspection data (such as that used in this paper) to target firms for investigation. They began to do so later. 4
For example, some may prefer the service of new car dealers, even for inspections. 6
they visit firms, and sometimes before. Consumers are unable to determine the true emissionrelated condition of their vehicles, and although they may choose among firms on the basis of reputations, they cannot perfectly anticipate how inspectors will exercise their discretion. Thus, repair costs, time and inconvenience costs, etc., are unknown even after a firm is chosen. They also can neither directly observe nor costlessly verify all inspectors’ actions. They are thus unable to determine with certainty ex post how much inspectors helped them pass or fail. One analytically convenient feature of this market is that one can characterize consumers’ preferences with respect to inspection outcomes irrespective of their vehicles’ underlying condition.5 When emission-related repairs are not covered by warranties, consumers prefer to pass. Passing relieves them of a regulatory requirement that is costly to fulfill. This is true even when inspections reveal that either the performance or the emission-related condition of their vehicle could be improved in ways that they privately value. Passing allows them to purchase repairs to the degree they wish. When repairs are covered by warranties which only apply when vehicles fail, consumers prefer failing inspections to passing them when the benefits they derive from the repairs outweigh any time costs they bear. This may be more likely when their vehicle is close to the end of its warranty period. Consumers may provide inspectors incentives explicitly through bribes. Alternatively, incentives may be implicit. Inspection outcomes may influence their future choice of firms. Inspectors may, in turn, believe that "passes" are rewarded with increases in demand in the future,
5
This is different than in health care markets, where sick patients usually wish to be diagnosed as sick. 7
both from consumers whose vehicles they inspect and possibly from others. 6 Inspectors’ behavior reflects the strength of these demand-side incentives. Consider cases where repairs are not covered by warranties. Finding that inspectors tend to take actions which increase the probability vehicles fail implies that these incentives are relatively weak. This is the situation that exists under some conditions in models of equilibrium fraud or inducement. 7 Finding instead that inspectors tend to help vehicles pass implies that demand-side incentives are relatively strong. They overcome both suppliers’ incentive to fail vehicles in general and regulators’ attempts to encourage inspectors to fail high-emitters. Firm Characteristics and Inspectors' Actions Some firms have only one inspector, who is also the owner. At these firms, inspectors’ behavior should reflect the incentives presented by only two sets of principals: regulators and consumers. For a given consumer and holding the level of regulatory scrutiny constant, inspectors' behavior might vary with their firms' competitive environment. Close geographic competition may lower inspectors’ expected gains from supplying failing inspections because it may decrease consumers’ cost of obtaining a second opinion.8 Conduct may also vary with the extent to which firms supply other automotive services. Firms may have a stronger incentive to pass consumers if
6
Although bribes are possible in this market, and those that are offered by undercover enforcement officials are sometimes accepted, they are considered by regulators to be a relatively uncommon event. Implicit rewards, on the other hand, may be more common. 7
Models of equilibrium fraud are in Darby and Karni (1973), Wolinsky (1993), and Taylor (1995). See Newhouse (1970) and Evans (1974) for early inducement models. There have been many empirical attempts to test for inducement in health care markets. Recent papers include Birch (1988), Grytten, Holst, and Laake (1990), and Gruber and Owings (1996). 8
This would be predicted from models with switching costs such as Klemperer (1987). 8
they also supply gasoline or a wide range of repairs. Most firms have multiple inspectors. At such firms, some inspections are completed by individuals who are not residual claimants. Although part of their total compensation is generally based on piece rates, these individuals receive only a share of the proceeds from the work they complete. Differences in conduct across firms may reflect differences in the extent to which their organizational characteristics align inspectors’ and firms’ incentives. For example, conduct may differ with the number of inspectors both because of diseconomies of scale in monitoring and because the incentives for individual inspectors to free ride may differ. Some firms have more complicated organizational forms. Chain stores, new car dealers, and tune up shops have an employee — either a head mechanic or a "service writer" — who serves as an intermediary between consumers and mechanics. This person receives service requests from consumers and allocates work to mechanics. This additional layer of hierarchy may hamper consumers’ ability to provide individual inspectors incentives toward helping them pass. Consumers may be less able to offer them incentives in the form of future business when they are assigned inspectors and mechanics rather then choose them. Absent incentives within the firm that would counteract this, inspectors may help consumers pass less. Another difference in organizational form is that chain stores and tune up shops are part of branded chains, but other firms are not. Differences in conduct between chain stores and tune up shops and firms which are not part of chains may reflect that internal incentives within these chains do not eliminate free riding at the outlet level. The chain stores and tune up shops in my sample differ in an important way: the tune up shops are franchised (and generally are managed by a residual claimant) but the chain stores are not (and always managed by employees). Internal incentive
9
schemes at chain stores may not fully resolve conflicts of interest between owners and non-owner managers. As a result, non-owner managers may monitor individual inspectors less intensively than owners, and inspection conduct may differ accordingly. 9 3. Data This paper uses two samples of inspection data. Both were collected by the California Bureau of Automotive Repair. The sample of "market" inspections includes almost all initial inspections completed between August 26 and November 11, 1992 at firms located in five geographic markets in central California. 10 This paper uses a one in five sample of the set of initial inspections. The geographic markets were chosen for two reasons. First, they are isolated from other geographic markets, so I am able to track vehicles through the inspection process and determine their initial inspection. Second, there is a single enforcement office of the Bureau of Automotive Repair located in each market. Thus, the regulatory environment is similar across markets. The other sample is from inspections completed by regulators at temporary inspection stations along state highways during 1993. State officials flag down vehicles11 and ask drivers if they
9
This idea is expounded in the franchising literature. See Caves and Murphy (1976), Klein and Saft (1985), Rubin (1978), and Brickley and Dark (1987). 10
These geographic markets are: Apple Valley-Hesperia-Victorville ("AHV"), Bakersfield, Fresno, Modesto, and Visalia-Hanford-Tulare ("VHT"). I begin with a sample of all initial inspections, but lose about 15% of the observations because firm characteristics are not available for some firms in these markets. This is because the firm characteristics data was collected in early 1994, over one year after the sample of inspections. By this time, some firms that had supplied inspections in late 1992 had exited the market. 11
For example, they pull over every fourth or fifth vehicle given that the inspection lane is
free. 10
are willing to have their vehicles inspected. If the driver assents, state officials then perform an inspection using the same equipment as private firms. Drivers are informed of the results, but are neither issued certificates if they pass nor ordered to obtain repairs if they fail. The state of California conducted roadside inspections annually between 1984 and 1993 to evaluate its vehicle inspection program. A broad measure of whether inspectors at private firms were completing inspections according to regulations compares the fraction of vehicles failing roadside inspections with the fraction failing inspections at private firms. By 1993, data from the roadside inspection program was providing little new information, and was thereby discontinued. In every year, a far higher fraction of vehicles failed roadside inspections than initial inspections at private firms.12 The second sample serves as a benchmark. My working assumption is that state officials take no actions that change vehicles’ operating condition, and that their judgement on the underhood component is in line with regulations. In conversations with state officials who completed some of these inspections, they told me that this was their intent.13 Relaxing this assumption to account for the idea that state officials may conduct inspections so that a higher proportion of vehicles fail than if they conducted them “by the books” does not fundamentally change things. As long as state officials would not punish inspectors at private firms for conducting inspections in the way that roadside inspectors do, differences in inspection conduct between roadside inspections and private
12
See California Air Resources Board (1991).
13
They related that there was little to be gained in distorting the inspection results. One did not need to distort the results to show that a much higher fraction of vehicles fail roadside inspections than at private firms. Furthermore, if roadside inspectors were intentionally lenient so that failure rates were comparable to those at private firms, the roadside inspection data would not have been considered credible. 11
firms reflect the strength of market incentives. Both samples contain many variables that are recorded at the time of the inspection.14 These fall into three general categories: identification, vehicle condition, and test results. — Identification. This includes identification of the vehicle and firm, and the date and time of the test. The vehicle identification contains many variables, including the model year, the odometer reading, the size of the engine, and whether it is carbureted or fuel injected. The model year and odometer reading allow me to construct a dummy variable (“warranty”) which equals one if the vehicle is covered by an emission warranty, and zero otherwise. — Vehicle Condition. The data contain the percentage of oxygen in the vehicle's exhaust. Oxygen levels indicate how “rich” or “lean” vehicles’ engines are running. Lean-running vehicles tend to have relatively low emissions because their fuel-air ratio is low; there is little unburned fuel in the exhaust. Oxygen levels may also proxy for vehicles’ general maintenance level. The variables in the market sample also include whether the inspection corresponded to a biennial registration renewal, a change in registration associated with a change of ownership, or the registration of a vehicle that was previously registered in another state. 15 This is a proxy for the condition of the emission equipment on the vehicle because it indicates the length of time since its previous inspection. — Inspection Results. This includes the result of the overall inspection, the emission
14
All but the results of the emission component are entered by the inspectors; emission component results are entered automatically. 15
These are three of the four conditions in which registration is required in California. In the fourth, the registration of a new vehicle never before registered, emission inspections are not required. 12
component, and the underhood component. .
These samples also include a dummy variable which indicates whether preinspection repairs
were completed. In the roadside inspections, this variable is always zero. In the market inspections, it equals one in about five percent of the observations; some actions taken by inspectors between the time the vehicle arrives at the firm and the time of the inspection are observed. This variable may not pick up all preinspection repairs. For example, inspectors may only record preinspection repairs for which they are compensated.16 Alternatively, they simply may not record some instances when preinspection repairs were completed, possibly because they were completed by other mechanics (and possibly at other firms). Some of the effects of these repairs may be captured in variables observed in the data such as oxygen level, but some might not be. Other ways in which inspectors can affect inspection outcomes, such as warming up the vehicle and being more or less lenient on the underhood component, are not recorded in the data. Sample means are in Table 1. The unconditional failure rate in roadside inspections is nearly twice that in market inspections. The difference is larger for the underhood than the emission component. These differences are caused by some combination of cross-sample differences in inspection conduct and vehicle mix. Contrasting the vehicle characteristic means, all but the odometer reading and warranty are statistically different from each other.17 Vehicles in roadside inspections are younger, but more intensively-driven. By the engine size means, they are also
16
The fact that consumers usually do not pay for reinspections makes their willingness to pay for preinspection repairs very low. It is usually a better strategy to see if the vehicle passes in its existing condition, then purchase the repairs if it does not. The only cost of the second inspection to the consumer would then be the value of the time it takes to conduct one. 17
Throughout this section, statistical tests are one-tailed t-tests of size 0.05. 13
smaller. The mean oxygen levels indicate differences in operating condition. Vehicles in roadside inspections tend to be running "richer" — with higher fuel-air ratios — than those in market inspections. For the sample of market inspections, I obtain characteristics of the firm conducting the inspection by incorporating data collected by the Bureau of Automotive Repair in station surveys, and by myself in telephone surveys. The station survey data includes "station type," "mode of ownership," station age, and number of inspectors. Station type indicates the nature of the firm's business: whether it is an independent garage, service station, smog check specialist, new car dealer, etc. Mode of ownership contains whether the firm is owned by an individual, corporation, or partnership. Station age denotes how many years the firm has been licensed by the Bureau of Automotive Repair to conduct general repairs. Number of inspectors is self-explanatory; it differs from the number of mechanics employed at the firm because not all mechanics are licensed for emission-related work. Using the firms' addresses, I construct a dummy variable ("Near") which equals one when a station shares a nine-digit zip code with one or more competitors, and zero otherwise.18 The telephone surveys collect two other variables. One is the price of the initial inspection.19 The other is, for chain stores and tune up shops, whether mechanics are paid as a
18
Nine-digit zip codes divide regions finely; hence, consumers purchasing inspections from a firm where "Near" equals one have an alternative practically next door or across the street. 19
I also asked whether the firms had a "pass or don’t pay" policy (one in which consumers pay only for passing inspections) and the price of reinspections. Almost none of the firms had "pass or don’t pay"; almost all of them did not charge for reinspections if any emission-related repairs were completed on site. 14
function of the work they complete (“commission”).20 Table 2 presents sample means by station type. Over half of the market inspections are supplied by independent garages. The next most common category is "smog check specialists." Although these firms supply some general auto repairs as well, most of their business is likely emissions-related. The number of inspections per firm is highest for this station type.21 Over threefourths of the inspections take place at either independent garages, service stations, or smog check specialists. In contrast, "chain stores," which include general merchandisers (e.g., Sears) and parts retailers (e.g., Pep Boys), complete a very small share of the inspections in my sample and have the smallest number of inspections/firm.22 Failure rates differ across station type.
Compared to independent garages, they are
significantly higher at chain stores, and significantly lower at new car dealers. There are also differences in vehicle mix. I note several here. The average age, odometer reading, and oxygen level of vehicles inspected at new car dealers are significantly lower than those inspected at independent garages. The average engine size of vehicles inspected at tune up shops is significantly lower than those inspected at independent garages; the average oxygen reading is significantly higher. Vehicles inspected at chain stores are more intensively driven and richer-running than those at independent
20
The survey was only over firms in these two station types because at all other station types, the practice is to pay mechanics an hourly rate plus piece rates. I asked each chain store and tune up shop in my sample how its mechanics were paid. In several cases, I was referred to a regional office. However, in all of these cases, in questioning other firms within the same chain, I was informed of the chain’s corporate policy toward payment of mechanics. 21
Because this is a one in five sample, the number of inspections/firm reported here understates the true number of inspections/firm by a factor of five for each station type. 22
"Other repairers" is a catch-all category which includes all other station types in the data, including brake shops, tire stores, foreign car repair specialists, etc. 15
garages, but the differences in the means are not significant. 4. Simple Logits This section reports results from simple logits in which the dependent variable is a dummy variable which equals one if the vehicle failed the overall inspection and zero otherwise. These examine whether, controlling for the vehicle characteristics observed in the data, there are correlations between the probability vehicles fail and firm characteristics. Assuming that unobserved vehicle characteristics are not correlated with the firm and vehicle characteristics included in the specification, the parameters on the firm characteristics indicate differences in inspector behavior. For expositional purposes, the rest of the section will discuss the results assuming that this holds. Later, I investigate whether the results change when allowing for differences in unobserved vehicle condition across station types and show that they generally do not. Table 3 presents results. There are three specifications. In each, I use a vehicle age polynomial, a prerepair dummy, oxygen level, a change of ownership dummy, and a warranty dummy as controls. Additional control variables such as odometer reading and idle speed were included in specifications not reported here (see Hubbard (1996a)). Omitting them does not affect the results. The first specification includes only the state and station type dummies as firm characteristics. The second includes a warranty*new car dealer interaction. The third includes all firm characteristics observed in the data. The results confirm that inspectors at private firms behave differently than state officials. They help vehicles pass. In each specification, the parameter on the state dummy is significantly larger than that on any of the station types except chain store, applying one-tailed t-tests of size 0.05. One can reject the null that the state and chain store coefficients are the same in the first two
16
specifications with one-tailed t-tests of size 0.10, and in the third with tests of size 0.05. Using the estimated parameters in the third specification and setting all firm characteristics to zero, I calculate the probability each vehicle in my sample would fail, were it inspected by a state official. Summing these probabilities across all observations, I derive a point estimate for the fraction of vehicles that would do so: 49.3%. From Table 2, only 21.6% actually failed inspections at private firms. Relative to that in roadside inspections, conduct at private firms cuts the fraction of vehicles which fail by more than half. The difference in the unconditional failure rates actually understates conduct differences, mainly because the vehicles in roadside inspections are newer. The Appendix reports results from specifications which use “emission component failure” and “underhood component failure” as the dependent variable. Inspectors at private firms tend to help vehicles pass the underhood as well as the emission component. This confirms that inspectors at private firms are helping vehicles pass in ways that regulators try to discourage. They are doing more than just warming vehicles up. Market incentives overcome both suppliers’ incentive to fail vehicles and regulators’ attempts to discourage inspectors from helping high-emitting vehicles pass. Turning to the station type coefficients, there are no systematic differences in conduct across the station types which perform most of the inspections in the market: independent garages, service stations, and smog specialists. Furthermore, the apparent difference between tune up shops and these station types from the first two specifications disappears in the third specification when one controls for differences in the number of inspectors. 23 The contrast between smog specialists and the other types suggests that the fraction of business composed of inspections does not affect inspection
23
One cannot reject the null that the coefficients on service station, smog specialist, and tune up shop are jointly zero. The Wald statistic is 5.44, which is significant only at the 0.15 level. 17
conduct. The lack of differences among all of these types suggests that the extent to which firms are integrated into supplying other repair-related services either does not affect inspectors’ incentives, or is counteracted by incentives provided within firms. From the contrast between tune-up shops (which are outlets within branded chains) and the other station types, there is little evidence of free riding at the outlet level at tune up shops. From the new car dealer coefficients in the second and third specifications, inspectors at new car dealers tend to help their non-warranty-covered consumers pass more than inspectors at independent garages. The probability derivative is negative 3-4 percentage points. This is much smaller than the difference in raw failure rates, but it implies that inspection conduct at new car dealers lowers the failure probability for the average vehicle in the sample by 15% relative to that at independent garages. From the vehicle age coefficients, this difference is about the same as the effect of reducing a vehicle’s age by one year.24 Inspection and repair prices (and probably markups) are far higher at new car dealers than at other station types. That inspectors help consumers pass more at such firms is consistent with the idea that rents provide firms strong incentives to supply high quality. Combined with the tune up shop coefficient, it is inconsistent with the idea that inspectors will help consumers pass less at firms where consumers generally do not deal with inspectors directly. The coefficient on the warranty*new car dealer interaction in the second and third specification is positive, but not statistically significant. The fact that there is only weak evidence that inspectors at new car dealers help their non-warranty-covered consumers pass more than their
24
For example, holding all other variables at their sample means, the predicted probability of failure for a five-year-old vehicle is 0.083. For a six-year-old vehicle it is 0.114. Differences are similar using other vintages except for the very oldest and youngest vehicles. 18
warranty-covered consumers is somewhat troubling: if consumers are indeed providing firms strong incentives, they should fail more when they do not bear the full cost of repairs. However, I find somewhat stronger evidence for this below. Inspectors at chain stores tend to help vehicles pass much less than inspectors at other firms. In the third specification, one can reject the null that the chain store coefficient is equal to any of the other station type coefficients using one-tailed t-tests at standard significance levels. The contrast between chain stores and tune up shops is particularly interesting, and implies that high failure rates at chain stores may be due to the weak incentives managers face to monitor and motivate their staff. This dovetails with the results of Shepard (1993), who shows that gasoline stations are less likely to be refiner-owned (and managed by salaried employees) when they have repair shops attached than when they do not, and attributes this to the difficulty of supplying managers who are not residual claimants incentives when their outlet supplies services that are hard to evaluate internally. From the third specification, inspectors help consumers pass less at firms with more licensed inspectors. The probability derivative implies that the failure probability is 1 percentage point, or about 5%, higher per inspector. This may be because internal monitoring is more difficult and inspectors have more of an incentive to free ride. From the coefficient on “Near,” conduct is slightly more consumer-friendly at firms with very close geographic competitors. One explanation is that it is less costly for consumers to switch firms and obtain second opinions in such circumstances. There is no evidence of relationships between inspector conduct and either the age of the station or how it is owned. Among chain stores and tune up shops, there is no evidence that paying inspectors piece rates affects how they conduct inspections. All else equal, vehicles inspected in the two least populous markets (AHV and VHT) failed more than those inspected in the other markets. One
19
possibility is that this reflects differences in vehicle condition. Smog check requirements only began to apply in AHV during 1991, so most of the vehicles in my sample from this market had never previously been inspected. Alternatively, this may reflect differences in inspector behavior arising from cross-market differences in the strength of competition. 5. Unobserved Vehicle Characteristics Framework If unobserved vehicle characteristics are not independent of the firm and vehicle characteristics included in the model, the estimates presented in the previous section are biased. This would be true, for example, if vehicles inspected by state officials tend to be “dirtier” than those inspected at private firms because they are surprise inspections for which consumers are unable to prepare their vehicles. Unobserved vehicle condition may differ systematically across firms as well. For example, new car dealers may tend to inspect vehicles that are well-maintained in ways the observed vehicle characteristics do not pick up. Although this form of selectivity can potentially arise with respect to any or all of the included station characteristics, it is of most concern with respect to the station type dummies. One reason for this concern is that these are the variables of central interest, another is that consumers are probably less likely to select firms on the basis of the other variables.25 I therefore assume that conditional on station type, other firm characteristics are independent of unobserved vehicle characteristics. I concentrate on possible differences across station types. The framework is adapted
25
Some of these they are unlikely to observe at all, such as the firm’s ownership structure and how it pays its inspectors. At larger firms, they are unlikely to know how many mechanics are licensed to complete inspections as well. 20
from Mroz and Guilkey (1992).26 Assume that the logit error term from the previous section is equal to the sum of two terms: , which may be correlated with the station type dummies, and
, which is assumed to be
independent of all right hand side variables and to be distributed logistically. Then: P(F|X, ) ' P(& X & '
eX
< )
%
1 % eX
(1) %
where P(F|X, ) is the probability that the vehicle failed the inspection conditional on X and , and X represents vehicle and firm characteristics observed in the data. From Table 2, there are systematic differences in the vehicles inspected at different station types. Therefore, I can construct a model which predicts station type as function of observables. Because
may be correlated with station type, it may have predictive power as well. Using a
multinomial logit framework, this implies that:
P(k*W,Z, ) '
e
Wk % Z
K
je
k
%
k
. Ws % Z
s
%
s
(2)
s'1
In this equation, k indexes station type (k {1,...,K}, where K is the number of station types).27 Wk is the average inspection price at station type k. I set Wstate=0; otherwise, Wk varies across
26
See also Heckman and Singer (1984), and Cameron and Taber (1994). See the Appendix for a short summary of work which examines the performance of this type of estimator. 27
Recall that the state and station type dummies divide the sample into mutually exclusive and exhaustive categories. In this section, “state” will be treated as if it were another station type. 21
geographic markets. Z is a vector containing vehicle-specific observables. 28 Assume that conditional on X, W, Z, and , the unobserved factors which affect the failure probability and where vehicles are inspected are independent. Then: P(k,o* X,W,Z, ) ' P(o* X, )P(k* W,Z, ).
where o is the outcome of the inspection, (o
(3)
O = {P,F}). (Recall that X includes station type
dummies.) From this I obtain the unconditional probability by integrating out over the distribution of : P(k,o*X,W,Z) '
m
P(k,o* X,W,Z, ) dF( ).
(4)
This integral is made tractable by assuming that F( ) is a step function; it becomes a weighted sum. If there are T points of support and pt, t=1,..,T, are the probabilities associated with these points, then: T
P(k,o* X,W,Z) ' j ptP(k,o* X,W,Z, ' t). t '1
(5)
The log-likelihood function is then obtained by substituting (5) into (6): J
K
log L' j j j d kolog P(k,o|X,W,Z) j '1 k '1 o 0O
(6)
where dko=1 if station type=k and outcome=o, and J is the number of observations. 29
28
This vector includes vehicle age, mileage, mileage/age (intensity of use), oxygen level, engine size, and dummies indicating whether the vehicle is a truck, whether it is on warranty, and whether it is changing ownership. 29
See the Appendix for the likelihood function. 22
The parameters of this model include those that determine the distribution of the discrete factor as well as those associated with the right hand side variables in each of the two stages. The discrete factor parameters are the distribution’s points of support ( t, t=1,...,T) and the density at each point of support (pt, t=1,...T). I normalize E( )=0, assume T=2, and let and
equal zero on one
of the multinomial logit branches to identify the model.30 To make estimation tractable, I reduce the number of station types to four: state, independent, dealer, and chain. This greatly reduces the number of parameters estimated in the stage explaining station type. Independent includes independent garages, service stations, smog specialists, used car dealers, and “other repairers”: small, independent businesses. Dealer includes new car dealers. Chain includes chain stores and tune up shops: outlets of branded auto repair chains. Admittedly, this classification is somewhat arbitrary. The intent is to group station types in a way that makes finding differences in unobserved vehicle condition likely. Consumers who particularly value having well-maintained vehicles in general may tend to choose new car dealers for inspections. Those who maintain them relatively infrequently may tend to choose chain stores and tune up shops. Those in between may tend to choose independent types. Although in principle all of the parameters of the model can be identified using maximum likelihood, in practice some of them — in particular the constants in the two equations, and
chain
state
,
dealer
,
— are difficult to identify separately without a further restriction. Below I report results
from specifications which restrict the distribution of
to be symmetric: p1=p2=1/2. The point
estimates and standard errors on the parameters of interest do not change when this restriction is
30
I normalize E( ) because one cannot separately identify the location of the distribution and the constant in the inspection result equation. The qualitative results do not change when increasing T. 23
relaxed, but near-multicollinearity makes the standard errors on the constants and the distributional parameters very large. Results Table 4 reports parameter estimates using the above procedure. In the first column,
t
is
restricted to equal zero. This restriction implies that unobserved vehicle condition does not differ across station types. The estimation results are the same as one would get estimating the two stages separately. The second relaxes this restriction. The log-likelihood improves considerably: one can reject the first specification in favor of the second using a likelihood ratio test. The most important difference in the point estimates is that the warranty*new car dealer coefficient becomes much larger. The point estimate is quite noisy: it is statistically significant at the 85% level but not at the 95% level. The fact that the point estimate on this interaction becomes much larger when allowing for differences in unobserved vehicle characteristics suggests that the unobserved condition of warranty-covered vehicles inspected at new car dealers is better than that of the rest of the sample. Specifications which did not account for this may understate differences in how warranty-covered and non-warranty-covered vehicles are inspected at new car dealers. This result lends more support to one of the main conclusions of this paper: even though consumers cannot verify everything inspectors do, they are able to provide inspectors strong incentives toward helping them obtain the inspection result they prefer at the firms they choose. 31 None of the other parameters of interest change much. This is also true in specifications not
31
The positive and significant coefficient on average inspection price reflects that there are omitted firm characteristics which are correlated with price. Examining a subset of the firms in this paper, Hubbard (1996b) shows that higher priced firms tend to have lower failure rates, conditional on station type. Here, price helps predict station type, but its coefficient does not have a single economic interpretation. 24
reported here which include all firm characteristics, and which allow the distribution of unobserved vehicle characteristics to be asymmetric. The results from this paper do not indicate that once one controls for, for example, vehicle age, there are important differences in unobserved heterogeneity across station type. One reason why not much changes may be that most variables that are correlated with vehicle condition are also correlated with vehicle age and are hence (indirectly) controlled for even in simple logits. The interpretations in section four do not change. 6. Conclusion Consumers are generally able to provide incentives that lead inspectors to act in their interest at the firms they choose. Inducement or consumer fraud is the exception rather than the rule. In a related paper (Hubbard (1996b)), I show that consumers’ choice of firms is strongly related to both a) the site and outcome of their previous inspection, and b) firms’ failure rates across all consumers. There, I claim this as evidence that multiperiod mechanisms provide firms and inspectors incentives to supply passing inspections. Whether these demand-side quality incentives are as strong in other “diagnosis-cure” markets is an open question. One important feature of the inspection market is that non-warranty-covered buyers generally know what treatment they prefer: none at all. This is only true under special circumstances in other diagnosis-cure markets, such as when individuals need to verify their health condition for insurance purposes. If buyers’ ability to bring competitive pressures to bear on firms depends on their knowing their preferred treatment, then the results from this market may not extend to many other markets. But there is reason to believe they do, because buyers can generally replicate the circumstances in the inspection market by purchasing diagnoses when they believe that “no treatment” is the correct one. That buyers generally do not do so in auto repair markets, in which
25
many firms offer free estimates, suggests that they believe that the value of this information is low. They may instead believe that they are already able to provide the sellers they choose incentives to act in their interest. Applying this paper’s results to health care markets, competitive pressures may encourage doctors to shade third-party-insured patients’ condition so that they receive more care. Insurers’ difficulties in counteracting these incentives when doctors are independent contractors may explain the rise of HMOs. Most inspections take place at small, independently-owned firms: garages, gas stations, tune up shops, and the like. The same is true for non-warranty-covered auto repairs, especially those which involve diagnoses. In this paper I show that inspectors act less in consumers’ interest at larger firms, especially those run by non-owner managers. Failure probabilities increase with the number inspectors. This suggests why most firms supplying non-warranty-covered repairs are small. Within large firms, it may be hard to provide internal incentives which lead mechanics to supply high quality diagnoses. Inspectors at chain stores help consumers pass much less than at other firms. The incentive structure within such firms may be designed to support the majority of the work they do — installing parts purchased within the store, tune ups, oil changes, etc. — but be ill-suited for supplying services involving diagnoses.32
32
In a celebrated incident, the California Bureau of Automotive Repair accused Sears Auto Centers of defrauding consumers during mid-1992. In the aftermath, Sears closed down many of its Auto Centers. At those which remained open, it decided not to supply most repairs that require diagnoses. This paper examines data from the period one to three months after this incident. The contrasts in this paper between chain stores and other firms are robust to a Sears firm effect. 26
Appendix Log-Likelihood Function
Log L ' J
K
j j (dk,fail log [p1 j '1
k '1
e K
je s'1
dk,pass log [p1
Wk % Z k %
e K
Ws % Z s %
Wk % Z k %
je
k 1
s 1
e
X %
1 % e
1
X %
1
% p2
e K
je s'1
k 1
Ws % Z s %
s 1
1 1 % e
X %
s'1
1
% p2
Wk % Z k %
e K
Ws % Z s %
Wk % Z k %
je
k 2
s 2
k 2
Ws % Z s %
s 2
e
X %
1 % e
2
] %
X %
2
X %
2
1 1 % e
] )
s'1
The specifications in the paper normalize i=0 on the “independent” branch, and restrict p 1=p 2=0.5 and
1
= - 2.
Properties of the Estimator Two studies shed light on the small sample performance of this type of estimator in different contexts. Mroz and Guilkey (1992) present Monte Carlo results for discrete factor models in twoequation discrete-continuous systems. Using 100 replications with N=1000, they find that discrete factor models generally perform well using the mean squared error of the parameter estimate of the coefficient on the endogenous dummy variable as a criterion, even when using only two or three points of support. They perform especially well relative to models which assume normality (such as maximum likelihood or two-step estimators) when the true disturbance is non-normal. Mroz and Guilkey also report that standard errors obtained through conventional methods (inverse Hessians) understate true standard errors by 10-50%, although White (1982) standard errors perform best . Cameron and Taber (1994) report Monte Carlo results for a ten period longitudinal discrete
27
choice model. Discrete factor models provide a means of correcting for heterogeneity bias in such contexts. Unlike Mroz and Guilkey, none of the right hand side variables are endogenous. Cameron and Taber’s estimator does not place assumptions on the number of points of support; they instead increase this number until the likelihood function does not further improve relative to some penalty function. The authors find that this semiparametric estimator performs extremely well in uncovering both the parameter estimates and the standard errors, even for sample sizes as low as 125. They also provide evidence that the estimates of the parameters of interest possess approximately %n normality.
28
References Birch, Stephen. “The Identification of Supplier-Inducement in a Fixed Price System of Health Care Provision: The Case of Dentistry in the United Kingdom.” Journal of Health Economics 7(June 1988):129-50. Brickley, James A., and Frederick H. Dark. “The Choice of Organizational Form: The Case of Franchising.” Journal of Financial Economics 18(June 1987):401-20. California Air Resources Board. Report and the ARB/BAR 1990 Random Roadside Inspection Survey. Sacramento: California Air Resources Board, 1991. Cameron, Stephen V. and Christopher R. Taber. “Evaluation and Identification of Semiparametric Maximum Likelihood Models of Dynamic Discrete Choice.” Manuscript, University of Chicago, November 1994. Caves, Richard E. and William F. Murphy. “Franchising: Firms, Markets, and Intangible Assets.” Southern Economics Journal 42(April 1976):572-86. Darby, Michael R. and Edi Karni. “Free Competition and the Optimal Amount of Fraud.” Journal of Law and Economics 16(April 1973):67-88. Evans, Robert G. “Supplier-Induced Demand: Some Empirical Evidence and Implications.” In The Economics of Health and Medical Care, edited by Mark Perlman. New York: Wiley, 1974. Gruber, Jonathan and Maria Owings. “Physician Financial Incentives and the Diffusion of Cesarean Section Delivery.” Rand Journal of Economics 27(Spring 1996):99-123. Grytten, Josstein, Dorthe Holst, and Petter Laake. “Supplier Inducement: Its Effect on Dental Services in Norway.” Journal of Health Economics 9(1990):483-91. Heckman, James J. and Burton Singer. “Econometric Duration Analysis.” Journal of Econometrics 24(January/February 1984):63-132. Hubbard, Thomas N. “Agency Relationships in the Vehicle Inspection Market: Empirical Analysis and Public Policy Implications.” Doctoral Dissertation, Stanford University, 1996a. Hubbard, Thomas N. “Consumer Beliefs and Buyer and Seller Behavior in the Vehicle Inspection Market.” Manuscript, University of California, Los Angeles, 1996b. Klein, Benjamin and Keith B. Leffler. “The Role of Market Forces in Assuring Contractual Performance.” Journal of Political Economy 89(August 1981):615-41. Klein, Benjamin and Lester F. Saft. “The Law and Economics of Franchise Tying Contracts.” 29
Journal of Law and Economics 28(1985):345-61. Klemperer, Paul. “The Competitiveness of Markets with Switching Costs.” Rand Journal of Economics 18(Spring 1987):138-150. Mroz, Thomas A. and David K. Guilkey. “Discrete Factor Approximations for Use in Simultaneous Equation Models with Both Continuous and Discrete Endogenous Variables.” Manuscript, University of North Carolina, Department of Economics, 1992. Newhouse, Joseph P. “A Model of Physician Pricing.” Southern Economic Journal 37(October 1970):174-183. Rubin, Paul H. “The Theory of the Firm and the Structure of the Franchise Contract.” Journal of Law and Economics 21(April 1978):223-33. Shapiro, Carl. “Premiums for High Quality Products as Returns to Reputations.” Quarterly Journal of Economics 98(November 1983):659-79. Shepard, Andrea. “Contractual Form, Retail Price, and Asset Characteristics in Gasoline Retailing.” Rand Journal of Economics 24(Spring 1993):58-77. Taylor, Curtis R. “The Economics of Breakdowns, Checkups, and Cures.” Journal of Political Economy 103(February 1995):53-74. Wolinsky, Asher. “Competition in a Market for Informed Experts’ Services.” Rand Journal of Economics 24(Autumn 1993):380-398. White, Halbert. “Maximum Likelihood Estimation of Misspecified Models.” Econometrica 50(1982):1-25.
30
Table 1 Sample Means Variable
Roadside Market Inspections Inspections
Overall Failure Emission Component Failure Underhood Component Failure
36.8% 23.4% 26.0%
21.6% 16.9% 9.7%
Vehicle Age Odometer (thousands) Odometer/Age Engine Size (liters) % Oxygen Truck Change of Ownership Warranty Prerepairs
8.32 71.33 11.96 3.07 1.85 0.23 N/A 0.20 N/A
9.84 70.79 9.28 3.55 3.21 0.37 0.25 0.18 0.05
419
14774
Sample Size
Table 2 Sample Means Market Inspections
Ind. Garages
Service Stations
Number of Inspections Number of Firms Inspections/Firm
14774 409 36.1
7613 204 37.3
1670 68 24.6
1925 17 113.2
Overall Failure Emission Component Failure Underhood Component Failure
21.6% 16.9% 9.7%
23.5% 18.4% 10.5%
19.6% 15.5% 8.1%
Vehicle Age Odometer (thousands) Odometer/Age Engine Size (liters) % Oxygen Truck Change of Ownership Warranty Prerepairs
9.84 70.79 9.28 3.55 3.21 0.37 0.25 0.18 0.05
10.58 75.27 9.08 3.64 3.18 0.40 0.25 0.14 0.05
Inspectors Station Age Corporation Partnership Commission Near
2.99 5.18 0.24 0.25 0.02 0.13
2.36 4.02 0.12 0.29 N/A 0.13
Variable
Smog New Car Specialists Dealers
Used Car Dealers
Chain Stores
Tune Up Shops
Other Repairers
1201 55 21.8
304 8 38.0
108 12 9.0
1449 20 72.5
923 25 36.9
21.7% 17.9% 9.4%
8.3% 7.6% 2.3%
20.7% 18.8% 3.3%
38.0% 29.6% 16.7%
25.3% 16.8% 15.3%
26.8% 18.0% 16.3%
9.34 69.26 9.34 3.55 2.89 0.35 0.16 0.18 0.03
10.30 71.90 9.16 3.54 3.93 0.37 0.22 0.14 0.06
5.41 49.77 11.01 3.34 1.75 0.31 0.60 0.52 0.02
11.66 62.28 7.54 3.88 4.33 0.39 0.36 0.10 0.07
9.54 74.87 10.21 3.23 1.97 0.28 0.03 0.14 0.04
8.97 68.77 9.68 3.16 4.03 0.32 0.07 0.19 0.04
9.60 67.37 9.78 3.36 2.34 0.32 0.14 0.15 0.07
2.65 9.20 0.22 0.19 N/A 0.03
2.87 2.22 0.09 0.41 N/A 0.16
3.67 13.71 0.94 0.01 N/A 0.14
1.91 3.74 0.14 0.30 N/A 0.30
1.57 8.87 0.95 0.05 0.84 0.25
6.84 4.05 0.54 0.06 0.18 0.15
1.26 2.04 0.07 0.16 N/A 0.06
Table 3 Parameter Estimates — Simple Logits Dependent Variable: Mean of Dependent Variable: Number of Observations:
Overall Inspection Failure 0.22 15193
Log of Likelihood Function
-6961.6 Parameter Estimate
Std. Error
Constant
-4.706
0.206
State Service Station Smog Specialist New Car Dealer Used Car Dealer Chain Store Tune Up Shop Other Repairer
1.303 -0.116 -0.045 -0.271 -0.294 0.840 0.325 -0.294
0.166 0.071 0.065 0.117 0.150 0.219 0.072 0.123
-6960.7 Probability Derivative
0.253 -0.015 -0.006 0.033 -0.036 0.147 0.049 -0.036
Warranty*New Car Dealer
Parameter Estimate
Std. Error
-4.712
0.206
1.305 -0.116 -0.045 -0.321 -0.294 0.841 0.326 -0.295
0.167 0.071 0.065 0.124 0.150 0.219 0.072 0.123
0.464
0.344
-6928.7 Probability Derivative
Parameter Estimate
Std. Error
-4.891
0.214
0.252 -0.015 -0.006 0.038 -0.036 0.146 0.049 -0.036
1.580 -0.134 -0.105 -0.321 -0.206 0.931 0.025 -0.414
0.180 0.076 0.068 0.141 0.153 0.274 0.098 0.129
0.323 -0.017 -0.014 0.039 -0.026 0.168 0.003 -0.049
0.058
0.491
0.344
0.063
0.078 -0.005 -0.110 -0.067 -0.138 0.045 0.190 -0.006 -0.032 0.245
0.013 0.005 0.070 0.053 0.069 0.174 0.078 0.062 0.069 0.083
0.010 -0.001 -0.015 -0.009 -0.018 0.006 0.027 -0.001 -0.004 0.035
0.666 -0.032 4.9E-04 -0.854 -0.013 -0.209 -0.839
0.046 0.003 6.5E-05 0.110 0.005 0.053 0.153
0.090 -0.004 6.6E-05 -0.089 -0.002 -0.027 -0.095
Inspectors Station Age Corporation Partnership Near Commission AHV Fresno Modesto VHT Age Age**2 Age**3 Prerepair % Oxygen Change of Ownership Warranty
0.658 -0.032 4.8E-04 -0.845 -0.011 -0.225 -0.775
0.046 0.003 6.5E-05 0.109 0.005 0.052 0.143
0.089 -0.004 6.5E-05 -0.088 -0.001 -0.029 -0.089
Note: Omitted dummies are "independent garages," "individual ownership," and "Bakersfield."
0.660 -0.032 4.8E-04 -0.845 -0.011 0.228 -0.833
0.046 0.003 6.5E-05 0.109 0.005 0.052 0.152
0.090 -0.004 6.6E-05 -0.088 -0.001 -0.030 -0.095
Probability Derivative
Table 4 Parameter Estimates Dependent Variables: Overall Inspection Failure, Station Type Number of Observations: 15193 Log of Likelihood Function
-17337.1
-17277.3
No
Yes
Control for Differences In Unobserved Characteristics? Parameter Estimate
Std. Error
Parameter Estimate
Std. Error
-4.819
0.216
-4.810
0.217
State New Car Dealer Chain Store/Tune Up Shop
1.053 -0.279 0.406
0.113 0.098 0.067
1.044 -0.313 0.355
0.113 0.099 0.067
Warranty*New Car Dealer
0.516
0.329
1.924
1.190
0.671 -0.032 4.9E-04 -0.868 -0.011 -0.226 -0.827
0.050 0.003 7.2E-05 0.113 0.005 0.052 0.157
0.671 -0.032 4.9E-04 -0.869 -0.011 -0.224 -0.796
0.050 0.003 7.2E-05 0.113 0.005 0.053 0.155
Inspection Results Equation Constant
Age Age**2 Age**3 Prerepair % Oxygen Change of Ownership Warranty Site of Inspection Equation Avg. Price State Branch: Constant Age Odometer Odometer/Age Oxygen Truck Engine Size Warranty New Car Dealer Branch: Constant Age Odometer Odometer/Age Oxygen Truck Engine Size Warranty Change of Ownership Chain/Tune Up Branch: Constant Age Odometer Odometer/Age Oxygen Truck Engine Size Warranty Change of Ownership Distributional Parameters Theta1 Rho(State) Rho(New Car Dealer) Rho(Chain/Tune Up)
0.098
0.003
0.118
0.003
-0.581 -0.024 -0.001 0.011 -0.062 -0.484 0.048 -0.048 -3.980 -0.069 -0.006 0.022 -0.030 -0.285 0.126 0.834 1.478 -0.957 -0.021 -0.001 0.002 0.039 -0.197 -0.128 0.115 -1.361
0.128 0.010 0.001 0.006 0.010 0.089 0.024 0.095 0.158 0.011 0.002 0.006 0.006 0.071 0.022 0.101 0.063 0.115 0.007 0.001 0.006 0.006 0.063 0.019 0.087 0.105
-0.253 -0.027 -0.002 0.011 -0.055 -0.484 -0.050 0.069 -5.208 -0.079 0.008 0.028 -0.036 -0.398 0.152 1.066 1.687 -6.295 -0.027 -0.002 0.008 0.041 -0.294 -0.136 0.410 -1.265
0.118 0.009 0.001 0.006 0.010 0.082 0.022 0.088 0.226 0.013 0.002 0.009 0.008 0.092 0.029 0.128 0.086 2.453 0.009 0.001 0.009 0.007 0.075 0.023 0.113 0.118
-0.042 15.77 45.59 155.85
0.003 1.208 2.940 47.003
Notes: Omitted station type includes independent garages, service stations, smog specialists, used car dealers, and "other repairers." Engine Size in liters, Odometer in tens of thousands of miles. Bold indicates statistical significance of two-tailed t-test of size 0.05. Standard errors are White (1982) standard errors.
Table A1 Emission and Underhood Component Logits Number of Observations:
15193
Dependent Variable: Mean of Dependent Variable:
Emission Component Failure 0.17
Underhood Component Failure 0.10
-6221.1
-4122.1
Log of Likelihood Function Parameter Estimate
Std. Error
Probability Derivative
Parameter Estimate
Std. Error
Probability Derivative
Constant
-4.693
0.228
-0.515
-8.126
0.391
-0.369
State Service Station Smog Specialist New Car Dealer Used Car Dealer Chain Store Tune Up Shop Other Repairer
1.216 -0.126 -0.043 -0.220 0.022 0.732 -0.073 -0.502
0.197 0.082 0.073 0.147 0.157 0.292 0.107 0.143
0.203 -0.013 -0.005 -0.023 0.002 0.106 -0.008 -0.046
2.107 -0.157 -0.117 -0.355 -1.342 0.825 0.018 -0.325
0.234 0.108 0.095 0.229 0.330 0.355 0.131 0.182
0.253 -0.007 -0.005 -0.015 -0.037 0.058 0.001 -0.013
Warranty*New Car Dealer
0.445
0.365
0.050
-0.437
1.047
-0.012
Inspectors Station Age Corporation Partnership Near Commission AHV Fresno Modesto VHT
0.040 -2.8E-04 -0.009 -0.006 -0.103 -0.037 0.378 0.046 0.002 0.302
0.014 0.005 0.075 0.057 0.073 0.199 0.082 0.067 0.076 0.088
0.004 -3.0E-05 -0.001 -0.001 -0.011 -0.004 0.045 0.005 0.000 0.035
0.169 -0.016 -0.431 -0.303 -0.156 0.414 -0.043 -0.048 -0.146 0.195
0.017 0.007 0.103 0.075 0.100 0.213 0.110 0.085 0.097 0.115
0.008 -0.001 -0.019 -0.014 -0.007 0.023 -0.002 -0.002 -0.006 0.010
Age Age**2 Age**3 Prerepair % Oxygen Change of Ownership Warranty
0.589 -0.030 4.9E-04 -0.643 -0.017 -0.144 -1.057
0.050 0.003 7.1E-05 0.116 0.005 0.056 0.175
0.065 -0.003 5.4E-05 -0.057 -0.002 -0.015 -0.091
1.005 -0.046 6.2E-04 -1.287 -0.005 -0.177 -0.036
0.082 0.005 1.1E-04 0.187 0.006 0.074 0.026
0.046 -0.002 2.8E-05 -0.036 -0.000 -0.008 -0.002
Omitted dummies are "independent garages," "individual ownership," and "Bakersfield."
