# Hedonic analysis of reliability and safety for new automobiles.

The purpose of this study is to test whether prices of new
automobiles reflect their varying degrees of safety and reliability,
controlling for other characteristics. Hedonic price analysis (Rosen
1974) is employed and produces estimates of the prices (or the
contributions toward the total price) for each characteristic.

The characteristics likely to command price differences are many and varied and include such factors as performance, size, and type of vehicle (e.g., sports car versus luxury car), accessories or options included, financing and trade-in information, and geographic region. The present study extends the analysis to determine whether or not differences in safety and reliability are also captured in the prices of new automobiles.

Unlike some of the more obvious observable characteristics, such as type of car, accessories, and even fuel efficiency (which by law must be posted on the car), safety and reliability are not visible to the prospective buyer at the time of purchase. While there are alternative sources for obtaining such information, whether by word of mouth, stories in the media, or consumer group studies, it is not as easily obtained or widely available.

From a public policy standpoint, an issue of consumer protection may exist if prices do not reflect variations in safety and reliability.(1) In such a case, automobile manufacturers might not have the incentive to produce safer or more reliable cars because they will not be adequately compensated for the costs involved in providing these added characteristics. In this way, the problem resembles that studied by Akerlof (1970). If, on the other hand, automobile prices do reflect variations in safety and reliability, then the argument for consumer protection may be lessened.

Other hedonic analyses of automobile prices have been conducted, though no study has included both safety and reliability. Multicollinearity has plagued earlier hedonic studies, a problem dealt with here through the use of principal component analysis. The present study also employs transaction prices rather than list prices. This study finds that as the levels of safety and reliability in new cars increase, consumers pay more for cars, all else held constant. On this basis one would reject the hypothesis that the market fails to convey consumer valuations of the safety and reliability characteristics. As such, producers should have an incentive to produce safer and more reliable automobiles.

PREVIOUS RESULTS

One of the earliest uses of hedonic analysis with automobiles was to attempt to adjust index numbers for quality changes. Triplett (1969), Cowling and Cubbin (1971), and Griliches (1971) used hedonic analysis for this purpose. These studies all used physical characteristics such as weight, horsepower, length, number of gears, and options as quality measures. Because these variables are highly correlated, the results are subject to a great deal of multicollinearity. These studies, as is true with almost all later studies, used list prices rather than transaction prices. This could result in incorrect estimates if there is differential discounting by different dealers or by makes of cars.

Several studies aimed at analyzing legislation calling for increased fuel efficiency utilized hedonic analysis; among them are Cato, Rodekohr, and Sweeney (1976), Cassella and Rabe (1978), Falvey et al. (not dated), Goodman (1983), and Ohta and Griliches (1986). Each of these studies suffered from multicollinearity, which some authors corrected by dropping some of the explanatory variables. These studies also used list rather than transaction prices.

Finally, Agarwal and Ratchford (1980) used hedonic analysis to estimate demand equations for various characteristics of cars. They carried their hedonic analysis a step further than most in that they did not stop at estimating the implicit prices for each characteristic, but used these prices along with other variables to estimate the underlying supply and demand equations for each characteristic.

In summary, hedonic analysis has been used extensively, but none of the studies answered the question posed here. In order to overcome the problem of multicollinearity one cannot simply eliminate variables as others have done as this may lead to omitted variable bias. This study utilizes principal component analysis to deal with that problem. Also, if discounts from list price vary by model, using list price results in incorrect estimates. Transaction prices for new automobile sales are used. In addition, the main contribution of this study is the inclusion of safety and reliability, two characteristics hypothesized to be of importance to consumers in buying a new car. Other studies have included only one or the other or dropped them because of multicollinearity.

THE HEDONIC MODEL

The hedonic model is based on the work of Rosen (1974). Each car is comprised of a set of characteristics such as fuel efficiency, reliability, safety, and roominess. When consumers purchase cars, what they purchase is a bundle of characteristics. Consumers are assumed to have preferences over those attributes. There are maximum amounts they are willing to spend (known as the consumer's bid) for any given combination they purchase. This bid depends upon the consumer's income and preferences as well as the bundle of characteristics.

The producers of cars are assumed to face two decisions: what bundle of characteristics to produce and how many cars to produce. Producers choose the bundle that maximizes their profits subject to the prices of inputs, their production function, number of units produced, and market price for each alternative bundle of characteristics. Thus the producers' "offer" function indicates the minimum unit price the producer will accept for the car produced.

Market equilibrium is the tangency of these bid and offer functions. The prices seen in the market are these tangency points; the price consumers "bid" or are willing to pay for each characteristic is equal to the amount producers "offer" or are willing to accept for each characteristics. These tangencies or implicit prices are what this study attempts to estimate through the use of a hedonic equation. The hedonic equation is a reduced-form equation reflecting both supply and demand influences.

The price used in the hedonic equation is the actual price paid for the car, or what is referred to as the transaction price. The characteristics of the car should be those which directly enter the consumers' utility function, not just physical characteristics. The characteristics include reliability, safety, power or performance, fuel economy, styling, size, and options. This list obviously does not include all possible options or characteristics. This does not bias the estimates for reliability and safety, the variables of primary interest, as long as the missing variables are uncorrelated with them. A complete list of variables is included in Table 1.

In addition to these attributes, a set of dummy variables denoting the region of the country is used. Prices may be higher in certain parts of the country due to transportation costs, limited supplies, or, in the case of California, due to stricter emission standards. The value of any car traded in is also included to test whether the amount allowed for a trade-in affects the price paid for the new car. It is often thought that dealers who allow consumers more for their old car simply raise the price of the new car (i.e., reduce the amount of the discount). If this is true, then the amount of the trade-in must be controlled for in the analysis. The finance charge and the percentage of the purchase price actually financed are also included for those who financed their cars. The reasoning here is similar to that for trade-ins: individuals who receive a lower finance rate may pay a higher price than those with a higher finance rate so that the net price is the same. Dummy variables are used to indicate whether the car is a sports or a luxury car. Sports and luxury cars contain features that affect the prices of cars, but are not captured in the variables listed. For example, those cars may have better quality upholstery, leather seats, or roomier seats. A list of the cars designated as sports and TABULAR DATA OMITTED luxury cars is included in Appendix Table 1.

The equation to be estimated can be written in the following form:

lnP = a + |B.sub.j~|X.sub.j~ + |C.sub.k~|W.sub.k~ + |Z.sub.m~|Q.sub.m~ + e (1)

where P = the price at which the car actually sold

X = a vector of characteristics of the car

W = a vector of dummy variables indicating car options

Q = a vector of the market characteristics

e = error term.

The equation is estimated in semi-logarithmic functional form. This is the most commonly used form in previous studies of the automobile market and thus allows direct comparison. In addition, this form allows for the fact that as the amount of a characteristic increases, all else held constant, the marginal valuation of the car ought to decline. It is also important to note that the choice of this particular functional form may affect the statistical results.

Once the hedonic equation has been estimated, the coefficients on each variable can be used to derive the price elasticities for each characteristic. If the coefficients differ significantly from zero and if the price of the car increases as the amount of the characteristic increases, the market works, as consumers will not pay more if they do not know that the characteristics are present. However, if the coefficients are not statistically significant or if the price of the car does not increase with increased amounts of the characteristic, the null hypothesis that the market does not work in handling the characteristics, reliability and safety, cannot be rejected.

DATA

Most of the data are from the 1983 Customer Satisfaction with Dealer Service survey conducted by J. D. Power and Associates. The data set consists of 7,109 observations on individuals who purchased new cars in the spring of 1982, of which 2,637 were used in this study. The survey was sent to a sample of new car buyers drawn randomly from among R. L. Polk new car registrations recorded during March 1982. The measure of reliability used in this study is the trouble index (otherwise known as the frequency of repair index) from Consumer Reports.(2) This overall trouble index rates cars in terms of the number of systems experiencing problems relative to the number in all cars of the same age. The trouble index has been adjusted to take into account the number of miles different consumers drive in a year. The index takes on values of 1 through 5, where 5 indicates that the car has a much better than average rating, 3 an average rating, and 1 a much worse than average rating.(3)

The value of the trouble index used is for the latest year available in the 1982 Consumer Reports (see Appendix Table 2). For example, if the latest year a trouble index was available in the 1982 report was 1979, then that is the rating used. As consumers would have made their judgments on the most recent available information this seems to justify this methodology rather than a methodology which simply dropped all cars not having a 1981 rating. Other automobile studies have used the Consumer Reports' frequency of repair ratings (Goodman 1983; Irvine 1979; Ohta and Griliches 1976).(4) . The measure of safety used in this study is an index created by the Highway Loss Data Institute (HLDI), part of the Insurance Institute for Highway Safety.(5) Each year the Institute publishes Injury and Collission Loss Experience which summarizes recent insurance injury and collision loss experiences of passenger cars. This study uses the 1983 edition. Three different indices are presented. Two of the indices cover injury losses, which are claims filed under Personal Injury Protection (PIP) coverage. An overall injury index is the frequency of all medical claims filed under PIP, while a severe injury index represents the frequency of claims over $500. The third measure, a collision index, presents losses in terms of average loss payment per insured vehicle year. The indices are presented in relative terms, with 100 representing the result for all cars in each category. Thus an index of 96 is four percent better than average and 122 is 22 percent worse than average. The results presented in the HLDI's publication have been adjusted to eliminate distortions due to operator age group. The overall injury index is used as a measure of safety in this study. The severe injury index was ruled out because of the small number of cars for which it was available. Overall injury is used rather than the collision index as the collision index is easily influenced by the cost of the car. That is, a more expensive car may receive the same amount of physical damage as a less expensive car, but it costs more to have it repaired or replace.(6)

Power or performance is measured by the ratio of horsepower to weight: the more horsepower the greater the acceleration, but the heavier the car the slower the acceleration. This information is available for most cars from Consumer Reports. For those cars not in Consumer Reports, the data are from Ward's Automotive Report. Measures of styling are luxury, sports car, size (e.g., compact, sub-compact, 4-door), and options included. The list of possible options is measured simply by the use of dummy variables with the variable taking on the value one if the option is present.

Fuel efficiency is measured by the miles per gallon given in the 1982 Gas Mileage Guide published by the Department of Energy. The interior volume index and the trunk capacity index, which are also reported in the Gas Mileage Guide, are used as measures of roominess.

The J. D. Power and Associates data set does not provide information on the amounts individuals received for trade-ins, but does indicate whether or not they traded in a car and its make, model, type, and year. The March 1982 edition of the Used Car Trade-In Guide (The National Automobile Dealers Used Car Guide Company) was used to establish prices for the trade-ins. These booklets only provide information on cars not more than five years old. For older cars The Gold Book Official Used Car Value Guide (Craft 1982) was used. The 1982 edition was used as it was the edition in effect at the time the individuals in the sample traded in their cars.

ESTIMATION

The number of observations used in the final analysis, 2,637, is substantially less than the 7,109 in the original data set. Some observations were eliminated due to nonresponse or missing data. Consumer Reports does not provide frequency of repair ratings on every car manufactured, nor does the Highway Loss Data Institute provide safety information on all cars. Cars not in these two sources were eliminated.

Table 2 (column 1) presents the results of estimating equation (1) when all performance variables and available options are included. As the equation is estimated in semi-logarithmic functional form, each coefficient can be interpreted as an approximation of the percentage change in the dependent variable for a one unit change in the independent variable, everything else held constant. The t-statistic for each coefficient is presented beneath the coefficient in Table 2. Among the performance variables, miles per gallon (MPG) and safety are the only ones with coefficients significant at the five percent level. Significance levels of the coefficients on options, finance charge, trade-in value, and region are indicated on the table.

As indicated previously, inclusion of all the option and performance TABULAR DATA OMITTED variables leads to a high degree of multicollinearity. Fuel efficiency, as measured by miles per gallon (MPG) and miles per gallon squared (MP|G.sup.2~), and in particular the safety measure suffer from a high degree of collinearity with the other variables. In order to reduce some of the multicollinearity without having to eliminate the highly correlated variables, which leads to omitted variable bias, principal component analysis is used. Principal component analysis seeks to estimate a linear combination of variables which are highly correlated so that the linear combination captures as much of the variation in these variables as possible while remaining uncorrelated with the rest of the variables in the equation. This principal component is then entered into the equation in place of the variables of which it is comprised.

The variables used to form the principal component for this study are variables which are not of primary interest in this study, but are highly correlated with fuel efficiency, safety, reliability, and performance. These are engine size, number of cylinders, transmission, type of car, and passenger and trunk compartment room. The regression equations are then re-estimated using the principal component. Thus the problem of multicollinearity is reduced, yet the effect of the otherwise omitted variables is still captured by the principal component. It is not possible to interpret the coefficient on the principal component used, as the variables used to construct it are not all measured in the same terms. However, the primary variables of interest are not included in the principal component, so this is not a significant concern.

Equation (1) is re-estimated using the principal component. First reliability is entered linearly and with a squared term. Next, reliability is entered as a set of dummy variables, with average as the reference category. Finally, reliability is again entered as a set of dummy variables, but the other performance variables are entered only with linear terms. Again, the equations are estimated in semi-logarithmic functional form, thus each coefficient can be interpreted as an approximation of the percentage change in price, the dependent variable, for a one unit change in the independent variable, all else held constant. Each of these equations is briefly discussed.

Before presenting the elasticities for fuel efficiency, safety, reliability, and performance, it is of interest to note the changes due to the use of the principal component. The primary difference is the increased significance of the reliability measures (RELIABILITY) and performance (HORSEPOWER/WT and |(HORSEPOWER/WT).sup.2~) variables. Because the principal component reduces the extent of multicollinearity, the variance of the estimated coefficients is reduced and significance levels increase. No substantial change is found in the significance level of any of the other variables.

Returning to the elasticities for fuel efficiency, safety, reliability, and performance, the results are shown in Table 3. Note that these are not the price elasticities of demand. These present the percentage change in price for a one percent increase in the relevant variable. These elasticities are calculated at the mean for each of the variables.

A one percent reduction in safety, which is reflected by an increase in the safety index, results in a decrease in the price of a new car. This result has the proper sign and is consistent with the hypothesis; the less safe a car, the less consumers are willing to pay for the car, all else held constant. The results for reliability also support the hypothesis. As reliability increases (e.g., as it moves from much worse than average to worse than average or from worse than average to average), the price of a car increases, holding all else constant. Thus, consumers pay more for a car that is more reliable. A one percent increase in fuel efficiency, measured in miles per gallon, results in a decrease in the price paid for the car. This result to be discussed further is unexpected. Increases in performance, HORSEPOWER/WT, result in a decrease in the price of the car.

In order to test the robustness of the results for reliability, equation (1) is estimated entering reliability with a linear term and a nonlinear term, while all other variables are entered as previously stated. Table 2 column 2 presents the results of this estimation. Both terms are significant, with a positive coefficient on the linear term and a negative coefficient on the nonlinear term. Thus the results indicate that the percentage change in price increases with increases in reliability at a decreasing rate. Next, equation (1) is estimated with dummy variables used in place of the linear and squared terms for reliability. The results are presented in Table 2 column 3. There are several reasons for estimating the equation using the dummy variables. First, it allows another test of the robustness of the results. In addition, the measure used for reliability is a discrete and not a continuous measure. The previous equations do not take account of this.

The results using the dummy variables indicate that, holding all else constant, cars which are worse and much worse than average cost less than cars which are average in terms of frequency of repair. As hypothesized, much worse than average has a coefficient greater in absolute value than that of worse than average. The coefficient on better than average is insignificant and, therefore, not different from average. The coefficient on much better than average is significant and negative. The unexpected sign may possibly be due to a correlation between reliability and other variables. Finally, equation (1) is estimated again with dummy variables for the reliability measure, but utilizing only linear terms for the performance variables, miles per gallon, safety, and horsepower/weight. This is done in order to see the effect of the nonlinear terms on the estimation. Results are presented in Table 2 column 4. The only major change is that safety is now significant. The insignificant coefficients on the safety variables in the other equations may be due to the high correlation between the quadratic and linear terms.(7)

During the course of this study several other specifications were tried. Some were an attempt to reduce the multicollinearity problem, some used a cost index rather than a trouble index for reliability, and others simply omitted variables. The results for the safety measure were not affected by the specification. The results for reliability were not as robust, with the magnitude of the change sensitive to which variables were included or excluded.

As noted, for fuel efficiency, MPG, the sign is unexpected. As miles per gallon increases, consumers should pay more for the car, everything else held constant. A possible explanation for the negative coefficient on miles per gallon (MPG) may be that if fuel economy is increased, it results in more than offsetting decreases in other characteristics, thus making the overall impact on price negative. Alternatively, if all models in a particular year provide the fuel efficiency consumers want, then for a model to achieve even greater fuel economy other characteristics, such as style, may have to be sacrificed and therefore consumers rate increased fuel economy as a negative factor. Other studies have yielded similar findings. Cowling and Cubbin (1971) reported a negative coefficient for fuel consumption which they attributed to collinearity among the attributes they included. Hogarty (1975) found that the coefficients on the measures of fuel economy were either insignificant or had a negative sign. Falvey et al. (not dated) could not obtain a hedonic equation in which fuel economy had a significant positive coefficient; a result which they also attributed to the high degree of collinearity with other included variables. Cassella and Rabe (1978) estimated their equations by manufacturer and obtained positive and significant results only for General Motors. It may be that the fuel efficiency measures from the Department of Transportation are not an appropriate measure of fuel efficiency. While use of principal components reduces the problem of multicollinearity, it does not eliminate all of the collinearity.(8) In particular, the tests indicate that MPG is still correlated with several included variables, which may be the reason for the results found here.

The results for some of the other variables included in the regression equations can also be examined. Turning to the finance charges, those individuals who finance 25 percent or less of the purchase price pay a higher price, which increases with the interest rate, all else held constant. Such individuals may have higher search costs so they spend less time searching for the best deal, thus they end up paying a higher price for the car chosen. Alternatively, it could simply be that there is a correlation between the amount financed and the type of car purchased. The impact of the trade-in is statistically significant, with a $1,000 trade-in resulting in roughly a $40 increase in the price of the new car. The hedonic prices for each of the characteristics which comprise the principal component are shown in Appendix Table 3, along with the calculations which yielded the prices.

The results for safety and reliability can be compared to other studies which included these variables. The safety measure is either insignificant or only significant at the ten percent level as in earlier studies which used different measures of safety (Hogarty 1975; Irvine 1979). Winston and Mannering (1984) used safety data from the Highway Loss Data Institute, though they did not use hedonic analysis. They found that consumers were willing to pay for increased safety, which is the finding in this paper. Turning to the results for frequency of repair or reliability, Goodman (1983) and Irvine (1979) used the Consumer Reports' frequency of repair ratings. Goodman found that consumers paid less for cars whether they were worse or better than average, while Irvine found demand to be sensitive to the frequency of repair ratings. While the reliability measure used in this study is not robust in magnitude, the direction and significance do seem to be robust. This may be attributed to several factors. First, the problem of multicollinearity has been greatly reduced in this study. Also, transaction prices rather than list prices are used, which may provide a more accurate measure of prices.

CONCLUSION

This study uses actual transaction prices of new cars to determine whether or not the market functions in pricing safety and reliability of new automobiles. This study finds that as both safety and reliability increase, the price of the car increases, all else held constant. The findings show that the market does work in pricing these characteristics. As the market is observed to work, the need to rely on consumer protection legislation to guarantee safety and reliability is lessened. This study shows that not only do consumers value these characteristics but producers have an incentive to produce safer and more reliable cars. The recent increase in the inclusion of air bags and the advertising spent on making consumers aware of them is evidence of both consumers' value of safety and the responsiveness of the industry.

The major contributions of this study are (1) use of transaction prices rather than list prices, (2) use of principal component analysis to reduce multicollinearity, and (3) inclusion of safety and reliability. While this study concludes that the market does work and thus producers do have an incentive to offer reliability and safety, it says nothing about the degree to which consumer information, product reputation, warranties and service contracts, or the threat of government intervention are responsible. Hopefully further research can be carried out to determine the role each of these has in the market for new automobiles.

APPENDIX

Cheryl Carleton Asher is Assistant Professor, Department of Economics, College of Commerce and Finance, Villanova University, Villanova, PA.

The author would like to thank Martin Asher, Cliff Huang, Pualine Ippolito, Jim Lacko, participants in workshops at the Federal Trade Commission, and two anonymous reviewers for helpful comments.

1 For an overview of FTC actions in this area see Calfee and Ford (1984).

2 How much each car owner (from a random sample representing all types of cars) spent on repairs (not maintenance), where it was done (at home, dealer), as well as some measure of time, trouble, and inconvenience would be ideal. Some parts may be easy to repair and may infrequently break while others may be less expensive to replace but require more time in the shop and/or break down more often. With this data an average cost of repair could be calculated for each make and model of car. Such a measure, however, is not available.

3 Consumer Reports provides two possible alternative measures. In this annual auto issue Consumer Reports presents both a cost index and a trouble index, along with trouble spots. The cost index, one possible measure of reliability, rates cars in terms of average maintenance and repair costs relative to all other cars for which data are available in that year. The cost index was not used for two reasons. First, it includes both maintenance and repair costs. A car could conceivably receive a cost rating which is much worse on average simply because it has high maintenance costs (for example Mercedes-Benz or Volvo). Second, the year of interest, 1982, did not have enough responses to calculate the cost index.

4 This measure is far from perfect, however it is the best measure currently available. The measure is based on the number of systems experiencing problems. Using such a measure assumes that a car which has problems in one system, regardless of the number of such problems, is more "reliable" than a car which has problems in several systems, again regardless of total number of problems.

Another problem with the measure is that it is not continuous; it can only take on discrete values. The difference between much worse than average and worse than average is assumed to be the same as between worse than average and average or between any two of the ratings. This may or may not be a valid assumption. Another potential problem is the sample involved. The ratings come from information that Consumer Reports' readers submit in response to a yearly survey. It is possible that there is a bias in who responds. For example, those who are strongly dissatisfied or who take better care of their cars may be overrepresented. To the extent that this occurs, it may affect the amounts being spent for repairs or the numbers of repairs reported. However, unless there is a difference in who responds across cars, this should not affect the relative ratings. That is, if only people who are dissatisfied respond to the questionnaire, this will not affect the relative rankings as long as only dissatisfied people respond regardless of which type of car they own. To the extent that this is not true, there may be some bias in the relative rankings as well.

5 Probably the best known measure of safety is the crash tests carried out by the National Highway Traffic Safety Administration (NHTSA). These tests are designed to see how well cars protect belted front seat occupants in a 35 mile per hour head-on crash into a rigid barrier. The main limitation to using this measure is that not all cars are tested.

6 The overall injury index is certainly not the most ideal measure of safety. First it only covers cars which are insured. If there is some bias as to which cars are insured there may be some bias in the resulting index. However, most people are apt not to insure older rather than newer cars. Bias is not likely to occur in the present study because it analyzes prices of new cars, all of which are included in the injury index. A second possible problem is that the indices are aggregated for three model years, 1980 through 1982, making it impossible to separate out the results for any one model year.

7 I thank an anonymous referee for making this point.

8 To test for any remaining collinearity, each of the independent variables is individually regressed on all of the other independent variables. The |R.sup.2~ for the primary variables of interest is as follows:

Thus some collinearity still exists between MPG, HLDI, and the other independent variables. In particular, for MPG, the t-statistics indicate a highly significant relationship between MPG and HORSEPOWER/WT, HLDI, the principal component, and reliability. For HLDI, the t-statistics indicate a highly significant relationship between HLDI and HORSEPOWER/WT, MPG, the principal component, and reliability.

REFERENCES

Agarwal, Manoj K. and Brian T. Ratchford (1980), "Estimating Demand Functions for Product Characteristics: The Case of Automobiles," Journal of Consumer Research, 7 (December): 249-262.

Akerlof, George A. (1970), "The Market for 'Lemons': Quality Uncertainty and the Market Mechanism," Quarterly Journal of Economics, 84(August): 488-500.

Brown, James N. and Harvey S. Rosen (1982), "On the Estimation of Structural Hedonic Price Models," Econometrica, 50(May): 765-768.

Calfee, John E. and Gary T. Ford (1984), "The FTC's Product Defects Program and Consumer Perceptions of Product Quality," in Perceived Quality: How Consumers View Stores and Merchandise, Jacoby and Olson (eds.), Lexington, MA: Lexington Books, D.C. Heath and Company: 175-191.

Cassalla, M. A. and F. T. Rabe (1978), The Relationship of Automobile to List Prices and Profit Margins--A Preliminary Analysis, Newton, MA: EIC Corporation (August).

Cato, Derriel, Mark Rodekohr, and James Sweeney (1976), "The Capital Stock Adjustment Process and the Demand for Gasoline: A Market-Share Approach," in Econometric Dimensions of Energy Demand and Supply, A. Bradley Askin and John Kraft (eds.), Lexington, MA: Lexington Books, D.C. Heath and Company: 29-52.

Consumer Reports (various issues), published by Consumers Union, Mount Vernon, NY.

Cowling, Keith and John Cubbin (1971), "Price, Quality and Advertising Competition: An Econometric Investigation of the United Kingdom Car Market," Econometrica, 38 (November): 378-394.

Craft, Quentin (1982), The Gold Book Official Used Car Value Guide, vol. 4, no. 2, El Paso, TX.

Falvey, Rodney E., Jeff Frank, Harold O. Fried, and Mark Babunovic (not dated), "Fuel Economy Standards and Automobile Prices," mimeo, Tulane University, New Orleans, LA.

Goodman, Allen C. (1983), "Willingness-to-pay for Car Efficiency," Journal of Transport Economics and Policy, 18 (September): 247-266.

Griliches, Zvi (ed.) (1971), Price Indexes and Quality Change, Cambridge, MA: Harvard University Press.

Highway Loss Data Institute (1983), HLDI Injury and Collision Loss Experience, Washington, DC (September).

Hogarty, Thomas F. (1975), "Price-Quality Relations for Automobiles: A New Approach," Applied Economics, 7: 41-51.

Irvine, F. Owen, Jr. (1979), "Estimated Demand Equations for Individual Automobile Models with Implications for Regulatory Issues and Gasoline Conservation," mimeo, Wesleyan University, Middletown, CT (July).

The National Automobile Dealers Used Car Guide Company (1982), Used Car Trade-In Guide, McLean, VA (March).

Ohta, Makoto and Zvi Griliches (1976), "Automobile Prices Revisited: Extensions of the Hedonic Hypothesis," in Household Production and Consumption, N. Terleckyj (ed.), New York: Columbia University Press: 325-390.

Ohta, Makoto and Zvi Griliches (1986), "Automobile Prices and Quality: Did the Gasoline Price Increase Change Consumer Tastes in the U.S.?" Journal of Business and Economic Statistics, 4(April): 187-198.

Rosen, Sherwin (1974), "Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition," Journal of Political Economy, 82(January): 34-55.

Triplett, Jack E. (1969), "Automobiles and Hedonic Quality Measurement," Journal of Political Economy, 77: 408-417.

United States Department of Energy (1982), 1982 Gas Mileage Guide, Washington, DC.

Ward's Automotive Report (various issues), published by Ward's Communications, Inc., Detroit, MI.

Winston, Clifford and Fred Mannering (1984), "Consumer Demand for Automobile Safety," American Economic Review, 74(May): 316-319.

The characteristics likely to command price differences are many and varied and include such factors as performance, size, and type of vehicle (e.g., sports car versus luxury car), accessories or options included, financing and trade-in information, and geographic region. The present study extends the analysis to determine whether or not differences in safety and reliability are also captured in the prices of new automobiles.

Unlike some of the more obvious observable characteristics, such as type of car, accessories, and even fuel efficiency (which by law must be posted on the car), safety and reliability are not visible to the prospective buyer at the time of purchase. While there are alternative sources for obtaining such information, whether by word of mouth, stories in the media, or consumer group studies, it is not as easily obtained or widely available.

From a public policy standpoint, an issue of consumer protection may exist if prices do not reflect variations in safety and reliability.(1) In such a case, automobile manufacturers might not have the incentive to produce safer or more reliable cars because they will not be adequately compensated for the costs involved in providing these added characteristics. In this way, the problem resembles that studied by Akerlof (1970). If, on the other hand, automobile prices do reflect variations in safety and reliability, then the argument for consumer protection may be lessened.

Other hedonic analyses of automobile prices have been conducted, though no study has included both safety and reliability. Multicollinearity has plagued earlier hedonic studies, a problem dealt with here through the use of principal component analysis. The present study also employs transaction prices rather than list prices. This study finds that as the levels of safety and reliability in new cars increase, consumers pay more for cars, all else held constant. On this basis one would reject the hypothesis that the market fails to convey consumer valuations of the safety and reliability characteristics. As such, producers should have an incentive to produce safer and more reliable automobiles.

PREVIOUS RESULTS

One of the earliest uses of hedonic analysis with automobiles was to attempt to adjust index numbers for quality changes. Triplett (1969), Cowling and Cubbin (1971), and Griliches (1971) used hedonic analysis for this purpose. These studies all used physical characteristics such as weight, horsepower, length, number of gears, and options as quality measures. Because these variables are highly correlated, the results are subject to a great deal of multicollinearity. These studies, as is true with almost all later studies, used list prices rather than transaction prices. This could result in incorrect estimates if there is differential discounting by different dealers or by makes of cars.

Several studies aimed at analyzing legislation calling for increased fuel efficiency utilized hedonic analysis; among them are Cato, Rodekohr, and Sweeney (1976), Cassella and Rabe (1978), Falvey et al. (not dated), Goodman (1983), and Ohta and Griliches (1986). Each of these studies suffered from multicollinearity, which some authors corrected by dropping some of the explanatory variables. These studies also used list rather than transaction prices.

Finally, Agarwal and Ratchford (1980) used hedonic analysis to estimate demand equations for various characteristics of cars. They carried their hedonic analysis a step further than most in that they did not stop at estimating the implicit prices for each characteristic, but used these prices along with other variables to estimate the underlying supply and demand equations for each characteristic.

In summary, hedonic analysis has been used extensively, but none of the studies answered the question posed here. In order to overcome the problem of multicollinearity one cannot simply eliminate variables as others have done as this may lead to omitted variable bias. This study utilizes principal component analysis to deal with that problem. Also, if discounts from list price vary by model, using list price results in incorrect estimates. Transaction prices for new automobile sales are used. In addition, the main contribution of this study is the inclusion of safety and reliability, two characteristics hypothesized to be of importance to consumers in buying a new car. Other studies have included only one or the other or dropped them because of multicollinearity.

THE HEDONIC MODEL

The hedonic model is based on the work of Rosen (1974). Each car is comprised of a set of characteristics such as fuel efficiency, reliability, safety, and roominess. When consumers purchase cars, what they purchase is a bundle of characteristics. Consumers are assumed to have preferences over those attributes. There are maximum amounts they are willing to spend (known as the consumer's bid) for any given combination they purchase. This bid depends upon the consumer's income and preferences as well as the bundle of characteristics.

The producers of cars are assumed to face two decisions: what bundle of characteristics to produce and how many cars to produce. Producers choose the bundle that maximizes their profits subject to the prices of inputs, their production function, number of units produced, and market price for each alternative bundle of characteristics. Thus the producers' "offer" function indicates the minimum unit price the producer will accept for the car produced.

Market equilibrium is the tangency of these bid and offer functions. The prices seen in the market are these tangency points; the price consumers "bid" or are willing to pay for each characteristic is equal to the amount producers "offer" or are willing to accept for each characteristics. These tangencies or implicit prices are what this study attempts to estimate through the use of a hedonic equation. The hedonic equation is a reduced-form equation reflecting both supply and demand influences.

The price used in the hedonic equation is the actual price paid for the car, or what is referred to as the transaction price. The characteristics of the car should be those which directly enter the consumers' utility function, not just physical characteristics. The characteristics include reliability, safety, power or performance, fuel economy, styling, size, and options. This list obviously does not include all possible options or characteristics. This does not bias the estimates for reliability and safety, the variables of primary interest, as long as the missing variables are uncorrelated with them. A complete list of variables is included in Table 1.

In addition to these attributes, a set of dummy variables denoting the region of the country is used. Prices may be higher in certain parts of the country due to transportation costs, limited supplies, or, in the case of California, due to stricter emission standards. The value of any car traded in is also included to test whether the amount allowed for a trade-in affects the price paid for the new car. It is often thought that dealers who allow consumers more for their old car simply raise the price of the new car (i.e., reduce the amount of the discount). If this is true, then the amount of the trade-in must be controlled for in the analysis. The finance charge and the percentage of the purchase price actually financed are also included for those who financed their cars. The reasoning here is similar to that for trade-ins: individuals who receive a lower finance rate may pay a higher price than those with a higher finance rate so that the net price is the same. Dummy variables are used to indicate whether the car is a sports or a luxury car. Sports and luxury cars contain features that affect the prices of cars, but are not captured in the variables listed. For example, those cars may have better quality upholstery, leather seats, or roomier seats. A list of the cars designated as sports and TABULAR DATA OMITTED luxury cars is included in Appendix Table 1.

The equation to be estimated can be written in the following form:

lnP = a + |B.sub.j~|X.sub.j~ + |C.sub.k~|W.sub.k~ + |Z.sub.m~|Q.sub.m~ + e (1)

where P = the price at which the car actually sold

X = a vector of characteristics of the car

W = a vector of dummy variables indicating car options

Q = a vector of the market characteristics

e = error term.

The equation is estimated in semi-logarithmic functional form. This is the most commonly used form in previous studies of the automobile market and thus allows direct comparison. In addition, this form allows for the fact that as the amount of a characteristic increases, all else held constant, the marginal valuation of the car ought to decline. It is also important to note that the choice of this particular functional form may affect the statistical results.

Once the hedonic equation has been estimated, the coefficients on each variable can be used to derive the price elasticities for each characteristic. If the coefficients differ significantly from zero and if the price of the car increases as the amount of the characteristic increases, the market works, as consumers will not pay more if they do not know that the characteristics are present. However, if the coefficients are not statistically significant or if the price of the car does not increase with increased amounts of the characteristic, the null hypothesis that the market does not work in handling the characteristics, reliability and safety, cannot be rejected.

DATA

Most of the data are from the 1983 Customer Satisfaction with Dealer Service survey conducted by J. D. Power and Associates. The data set consists of 7,109 observations on individuals who purchased new cars in the spring of 1982, of which 2,637 were used in this study. The survey was sent to a sample of new car buyers drawn randomly from among R. L. Polk new car registrations recorded during March 1982. The measure of reliability used in this study is the trouble index (otherwise known as the frequency of repair index) from Consumer Reports.(2) This overall trouble index rates cars in terms of the number of systems experiencing problems relative to the number in all cars of the same age. The trouble index has been adjusted to take into account the number of miles different consumers drive in a year. The index takes on values of 1 through 5, where 5 indicates that the car has a much better than average rating, 3 an average rating, and 1 a much worse than average rating.(3)

The value of the trouble index used is for the latest year available in the 1982 Consumer Reports (see Appendix Table 2). For example, if the latest year a trouble index was available in the 1982 report was 1979, then that is the rating used. As consumers would have made their judgments on the most recent available information this seems to justify this methodology rather than a methodology which simply dropped all cars not having a 1981 rating. Other automobile studies have used the Consumer Reports' frequency of repair ratings (Goodman 1983; Irvine 1979; Ohta and Griliches 1976).(4) . The measure of safety used in this study is an index created by the Highway Loss Data Institute (HLDI), part of the Insurance Institute for Highway Safety.(5) Each year the Institute publishes Injury and Collission Loss Experience which summarizes recent insurance injury and collision loss experiences of passenger cars. This study uses the 1983 edition. Three different indices are presented. Two of the indices cover injury losses, which are claims filed under Personal Injury Protection (PIP) coverage. An overall injury index is the frequency of all medical claims filed under PIP, while a severe injury index represents the frequency of claims over $500. The third measure, a collision index, presents losses in terms of average loss payment per insured vehicle year. The indices are presented in relative terms, with 100 representing the result for all cars in each category. Thus an index of 96 is four percent better than average and 122 is 22 percent worse than average. The results presented in the HLDI's publication have been adjusted to eliminate distortions due to operator age group. The overall injury index is used as a measure of safety in this study. The severe injury index was ruled out because of the small number of cars for which it was available. Overall injury is used rather than the collision index as the collision index is easily influenced by the cost of the car. That is, a more expensive car may receive the same amount of physical damage as a less expensive car, but it costs more to have it repaired or replace.(6)

Power or performance is measured by the ratio of horsepower to weight: the more horsepower the greater the acceleration, but the heavier the car the slower the acceleration. This information is available for most cars from Consumer Reports. For those cars not in Consumer Reports, the data are from Ward's Automotive Report. Measures of styling are luxury, sports car, size (e.g., compact, sub-compact, 4-door), and options included. The list of possible options is measured simply by the use of dummy variables with the variable taking on the value one if the option is present.

Fuel efficiency is measured by the miles per gallon given in the 1982 Gas Mileage Guide published by the Department of Energy. The interior volume index and the trunk capacity index, which are also reported in the Gas Mileage Guide, are used as measures of roominess.

The J. D. Power and Associates data set does not provide information on the amounts individuals received for trade-ins, but does indicate whether or not they traded in a car and its make, model, type, and year. The March 1982 edition of the Used Car Trade-In Guide (The National Automobile Dealers Used Car Guide Company) was used to establish prices for the trade-ins. These booklets only provide information on cars not more than five years old. For older cars The Gold Book Official Used Car Value Guide (Craft 1982) was used. The 1982 edition was used as it was the edition in effect at the time the individuals in the sample traded in their cars.

ESTIMATION

The number of observations used in the final analysis, 2,637, is substantially less than the 7,109 in the original data set. Some observations were eliminated due to nonresponse or missing data. Consumer Reports does not provide frequency of repair ratings on every car manufactured, nor does the Highway Loss Data Institute provide safety information on all cars. Cars not in these two sources were eliminated.

Table 2 (column 1) presents the results of estimating equation (1) when all performance variables and available options are included. As the equation is estimated in semi-logarithmic functional form, each coefficient can be interpreted as an approximation of the percentage change in the dependent variable for a one unit change in the independent variable, everything else held constant. The t-statistic for each coefficient is presented beneath the coefficient in Table 2. Among the performance variables, miles per gallon (MPG) and safety are the only ones with coefficients significant at the five percent level. Significance levels of the coefficients on options, finance charge, trade-in value, and region are indicated on the table.

As indicated previously, inclusion of all the option and performance TABULAR DATA OMITTED variables leads to a high degree of multicollinearity. Fuel efficiency, as measured by miles per gallon (MPG) and miles per gallon squared (MP|G.sup.2~), and in particular the safety measure suffer from a high degree of collinearity with the other variables. In order to reduce some of the multicollinearity without having to eliminate the highly correlated variables, which leads to omitted variable bias, principal component analysis is used. Principal component analysis seeks to estimate a linear combination of variables which are highly correlated so that the linear combination captures as much of the variation in these variables as possible while remaining uncorrelated with the rest of the variables in the equation. This principal component is then entered into the equation in place of the variables of which it is comprised.

The variables used to form the principal component for this study are variables which are not of primary interest in this study, but are highly correlated with fuel efficiency, safety, reliability, and performance. These are engine size, number of cylinders, transmission, type of car, and passenger and trunk compartment room. The regression equations are then re-estimated using the principal component. Thus the problem of multicollinearity is reduced, yet the effect of the otherwise omitted variables is still captured by the principal component. It is not possible to interpret the coefficient on the principal component used, as the variables used to construct it are not all measured in the same terms. However, the primary variables of interest are not included in the principal component, so this is not a significant concern.

Equation (1) is re-estimated using the principal component. First reliability is entered linearly and with a squared term. Next, reliability is entered as a set of dummy variables, with average as the reference category. Finally, reliability is again entered as a set of dummy variables, but the other performance variables are entered only with linear terms. Again, the equations are estimated in semi-logarithmic functional form, thus each coefficient can be interpreted as an approximation of the percentage change in price, the dependent variable, for a one unit change in the independent variable, all else held constant. Each of these equations is briefly discussed.

Before presenting the elasticities for fuel efficiency, safety, reliability, and performance, it is of interest to note the changes due to the use of the principal component. The primary difference is the increased significance of the reliability measures (RELIABILITY) and performance (HORSEPOWER/WT and |(HORSEPOWER/WT).sup.2~) variables. Because the principal component reduces the extent of multicollinearity, the variance of the estimated coefficients is reduced and significance levels increase. No substantial change is found in the significance level of any of the other variables.

Returning to the elasticities for fuel efficiency, safety, reliability, and performance, the results are shown in Table 3. Note that these are not the price elasticities of demand. These present the percentage change in price for a one percent increase in the relevant variable. These elasticities are calculated at the mean for each of the variables.

TABLE 3 Elasticities (Results from Table 2) Equation Two Four Means SAFETY -.347 -.507 101.380 RELIABILITY .041 3.088 MPG -.328 -.187 26.725 HORSEPOWER/WT -.003 -.069 3.300 Note: The elasticities were calculated at the mean for each variable listed. The formula used for Equation Two was |elasticity.sub.i~ = |B.sub.i~|X.sub.i~ + 2|B.sub.i~|X.sub.i~. The formula used for Equation Four was |elasticity.sub.i~ = |B.sub.i~|X.sub.i~.

A one percent reduction in safety, which is reflected by an increase in the safety index, results in a decrease in the price of a new car. This result has the proper sign and is consistent with the hypothesis; the less safe a car, the less consumers are willing to pay for the car, all else held constant. The results for reliability also support the hypothesis. As reliability increases (e.g., as it moves from much worse than average to worse than average or from worse than average to average), the price of a car increases, holding all else constant. Thus, consumers pay more for a car that is more reliable. A one percent increase in fuel efficiency, measured in miles per gallon, results in a decrease in the price paid for the car. This result to be discussed further is unexpected. Increases in performance, HORSEPOWER/WT, result in a decrease in the price of the car.

In order to test the robustness of the results for reliability, equation (1) is estimated entering reliability with a linear term and a nonlinear term, while all other variables are entered as previously stated. Table 2 column 2 presents the results of this estimation. Both terms are significant, with a positive coefficient on the linear term and a negative coefficient on the nonlinear term. Thus the results indicate that the percentage change in price increases with increases in reliability at a decreasing rate. Next, equation (1) is estimated with dummy variables used in place of the linear and squared terms for reliability. The results are presented in Table 2 column 3. There are several reasons for estimating the equation using the dummy variables. First, it allows another test of the robustness of the results. In addition, the measure used for reliability is a discrete and not a continuous measure. The previous equations do not take account of this.

The results using the dummy variables indicate that, holding all else constant, cars which are worse and much worse than average cost less than cars which are average in terms of frequency of repair. As hypothesized, much worse than average has a coefficient greater in absolute value than that of worse than average. The coefficient on better than average is insignificant and, therefore, not different from average. The coefficient on much better than average is significant and negative. The unexpected sign may possibly be due to a correlation between reliability and other variables. Finally, equation (1) is estimated again with dummy variables for the reliability measure, but utilizing only linear terms for the performance variables, miles per gallon, safety, and horsepower/weight. This is done in order to see the effect of the nonlinear terms on the estimation. Results are presented in Table 2 column 4. The only major change is that safety is now significant. The insignificant coefficients on the safety variables in the other equations may be due to the high correlation between the quadratic and linear terms.(7)

During the course of this study several other specifications were tried. Some were an attempt to reduce the multicollinearity problem, some used a cost index rather than a trouble index for reliability, and others simply omitted variables. The results for the safety measure were not affected by the specification. The results for reliability were not as robust, with the magnitude of the change sensitive to which variables were included or excluded.

As noted, for fuel efficiency, MPG, the sign is unexpected. As miles per gallon increases, consumers should pay more for the car, everything else held constant. A possible explanation for the negative coefficient on miles per gallon (MPG) may be that if fuel economy is increased, it results in more than offsetting decreases in other characteristics, thus making the overall impact on price negative. Alternatively, if all models in a particular year provide the fuel efficiency consumers want, then for a model to achieve even greater fuel economy other characteristics, such as style, may have to be sacrificed and therefore consumers rate increased fuel economy as a negative factor. Other studies have yielded similar findings. Cowling and Cubbin (1971) reported a negative coefficient for fuel consumption which they attributed to collinearity among the attributes they included. Hogarty (1975) found that the coefficients on the measures of fuel economy were either insignificant or had a negative sign. Falvey et al. (not dated) could not obtain a hedonic equation in which fuel economy had a significant positive coefficient; a result which they also attributed to the high degree of collinearity with other included variables. Cassella and Rabe (1978) estimated their equations by manufacturer and obtained positive and significant results only for General Motors. It may be that the fuel efficiency measures from the Department of Transportation are not an appropriate measure of fuel efficiency. While use of principal components reduces the problem of multicollinearity, it does not eliminate all of the collinearity.(8) In particular, the tests indicate that MPG is still correlated with several included variables, which may be the reason for the results found here.

The results for some of the other variables included in the regression equations can also be examined. Turning to the finance charges, those individuals who finance 25 percent or less of the purchase price pay a higher price, which increases with the interest rate, all else held constant. Such individuals may have higher search costs so they spend less time searching for the best deal, thus they end up paying a higher price for the car chosen. Alternatively, it could simply be that there is a correlation between the amount financed and the type of car purchased. The impact of the trade-in is statistically significant, with a $1,000 trade-in resulting in roughly a $40 increase in the price of the new car. The hedonic prices for each of the characteristics which comprise the principal component are shown in Appendix Table 3, along with the calculations which yielded the prices.

The results for safety and reliability can be compared to other studies which included these variables. The safety measure is either insignificant or only significant at the ten percent level as in earlier studies which used different measures of safety (Hogarty 1975; Irvine 1979). Winston and Mannering (1984) used safety data from the Highway Loss Data Institute, though they did not use hedonic analysis. They found that consumers were willing to pay for increased safety, which is the finding in this paper. Turning to the results for frequency of repair or reliability, Goodman (1983) and Irvine (1979) used the Consumer Reports' frequency of repair ratings. Goodman found that consumers paid less for cars whether they were worse or better than average, while Irvine found demand to be sensitive to the frequency of repair ratings. While the reliability measure used in this study is not robust in magnitude, the direction and significance do seem to be robust. This may be attributed to several factors. First, the problem of multicollinearity has been greatly reduced in this study. Also, transaction prices rather than list prices are used, which may provide a more accurate measure of prices.

CONCLUSION

This study uses actual transaction prices of new cars to determine whether or not the market functions in pricing safety and reliability of new automobiles. This study finds that as both safety and reliability increase, the price of the car increases, all else held constant. The findings show that the market does work in pricing these characteristics. As the market is observed to work, the need to rely on consumer protection legislation to guarantee safety and reliability is lessened. This study shows that not only do consumers value these characteristics but producers have an incentive to produce safer and more reliable cars. The recent increase in the inclusion of air bags and the advertising spent on making consumers aware of them is evidence of both consumers' value of safety and the responsiveness of the industry.

The major contributions of this study are (1) use of transaction prices rather than list prices, (2) use of principal component analysis to reduce multicollinearity, and (3) inclusion of safety and reliability. While this study concludes that the market does work and thus producers do have an incentive to offer reliability and safety, it says nothing about the degree to which consumer information, product reputation, warranties and service contracts, or the threat of government intervention are responsible. Hopefully further research can be carried out to determine the role each of these has in the market for new automobiles.

APPENDIX

TABLE 1 Sports and Luxury Cars Sports Cars Luxury Cars Pontiac Firebird Buick Electra Ford Mustang Buick Riviera Mercury Cougar XR-7 Oldsmobile 98 Mercury Capri Oldsmobile Toronado Honda Prelude Cadillac De Ville Nissan 200SX Cadillac Eldorado Toyota Celica BMW 320i VW Scirocco Audi 5000 Lincoln Continental Mark VI Chrysler New Yorker Nissan Maxima/810 Cadillac Seville TABLE 2 Consumer Reports Trouble Index Trouble Index Number of Observations Much Worse Than Average (1) 386 Worse Than Average (2) 384 Average (3) 1,093 Better Than Average (4) 625 Much Better Than Average (5) 149 TABLE 3 Hedonic Prices of the Characteristics Included in the Principal Component Variable Price Gasoline -.013 Turbo -.009 Diesel .023 Four cylinders .580 Six cylinders -.234 Eight cylinders -.346 Automatic -.556 Four speed .180 Five speed .376 2D sedan -.081 2D hatchback .231 4D sedan -.295 4D hatchback .060 Wagon .076 Convertible .011 Interior room -18.320 Trunk room -4.563 These prices are derived from the following set of equations: Price of |characteristic.sub.i~ = -.002|V.sub.i~P Where -.002 is the coefficient on the principal component variable from Table 2, Equation Four; |V.sub.i~ is a vector of the coefficients of each of the variables in the principal component, and P is the average price of a new car ($9,452.1736). These equations are derived by first differentiating the hedonic equation with respect to the principal component, which yields the coefficient on the principal component and then differentiating the principal component equation with respect to each of the variables listed.

Cheryl Carleton Asher is Assistant Professor, Department of Economics, College of Commerce and Finance, Villanova University, Villanova, PA.

The author would like to thank Martin Asher, Cliff Huang, Pualine Ippolito, Jim Lacko, participants in workshops at the Federal Trade Commission, and two anonymous reviewers for helpful comments.

1 For an overview of FTC actions in this area see Calfee and Ford (1984).

2 How much each car owner (from a random sample representing all types of cars) spent on repairs (not maintenance), where it was done (at home, dealer), as well as some measure of time, trouble, and inconvenience would be ideal. Some parts may be easy to repair and may infrequently break while others may be less expensive to replace but require more time in the shop and/or break down more often. With this data an average cost of repair could be calculated for each make and model of car. Such a measure, however, is not available.

3 Consumer Reports provides two possible alternative measures. In this annual auto issue Consumer Reports presents both a cost index and a trouble index, along with trouble spots. The cost index, one possible measure of reliability, rates cars in terms of average maintenance and repair costs relative to all other cars for which data are available in that year. The cost index was not used for two reasons. First, it includes both maintenance and repair costs. A car could conceivably receive a cost rating which is much worse on average simply because it has high maintenance costs (for example Mercedes-Benz or Volvo). Second, the year of interest, 1982, did not have enough responses to calculate the cost index.

4 This measure is far from perfect, however it is the best measure currently available. The measure is based on the number of systems experiencing problems. Using such a measure assumes that a car which has problems in one system, regardless of the number of such problems, is more "reliable" than a car which has problems in several systems, again regardless of total number of problems.

Another problem with the measure is that it is not continuous; it can only take on discrete values. The difference between much worse than average and worse than average is assumed to be the same as between worse than average and average or between any two of the ratings. This may or may not be a valid assumption. Another potential problem is the sample involved. The ratings come from information that Consumer Reports' readers submit in response to a yearly survey. It is possible that there is a bias in who responds. For example, those who are strongly dissatisfied or who take better care of their cars may be overrepresented. To the extent that this occurs, it may affect the amounts being spent for repairs or the numbers of repairs reported. However, unless there is a difference in who responds across cars, this should not affect the relative ratings. That is, if only people who are dissatisfied respond to the questionnaire, this will not affect the relative rankings as long as only dissatisfied people respond regardless of which type of car they own. To the extent that this is not true, there may be some bias in the relative rankings as well.

5 Probably the best known measure of safety is the crash tests carried out by the National Highway Traffic Safety Administration (NHTSA). These tests are designed to see how well cars protect belted front seat occupants in a 35 mile per hour head-on crash into a rigid barrier. The main limitation to using this measure is that not all cars are tested.

6 The overall injury index is certainly not the most ideal measure of safety. First it only covers cars which are insured. If there is some bias as to which cars are insured there may be some bias in the resulting index. However, most people are apt not to insure older rather than newer cars. Bias is not likely to occur in the present study because it analyzes prices of new cars, all of which are included in the injury index. A second possible problem is that the indices are aggregated for three model years, 1980 through 1982, making it impossible to separate out the results for any one model year.

7 I thank an anonymous referee for making this point.

8 To test for any remaining collinearity, each of the independent variables is individually regressed on all of the other independent variables. The |R.sup.2~ for the primary variables of interest is as follows:

Variable |R.sup.2~ MPG .693 SAFETY .649 HORSEPOWER/WT .255 MUCH WORSE THAN AVERAGE .433 WORSE THAN AVERAGE .311 BETTER THAN AVERAGE .375 MUCH BETTER THAN AVERAGE .201

Thus some collinearity still exists between MPG, HLDI, and the other independent variables. In particular, for MPG, the t-statistics indicate a highly significant relationship between MPG and HORSEPOWER/WT, HLDI, the principal component, and reliability. For HLDI, the t-statistics indicate a highly significant relationship between HLDI and HORSEPOWER/WT, MPG, the principal component, and reliability.

REFERENCES

Agarwal, Manoj K. and Brian T. Ratchford (1980), "Estimating Demand Functions for Product Characteristics: The Case of Automobiles," Journal of Consumer Research, 7 (December): 249-262.

Akerlof, George A. (1970), "The Market for 'Lemons': Quality Uncertainty and the Market Mechanism," Quarterly Journal of Economics, 84(August): 488-500.

Brown, James N. and Harvey S. Rosen (1982), "On the Estimation of Structural Hedonic Price Models," Econometrica, 50(May): 765-768.

Calfee, John E. and Gary T. Ford (1984), "The FTC's Product Defects Program and Consumer Perceptions of Product Quality," in Perceived Quality: How Consumers View Stores and Merchandise, Jacoby and Olson (eds.), Lexington, MA: Lexington Books, D.C. Heath and Company: 175-191.

Cassalla, M. A. and F. T. Rabe (1978), The Relationship of Automobile to List Prices and Profit Margins--A Preliminary Analysis, Newton, MA: EIC Corporation (August).

Cato, Derriel, Mark Rodekohr, and James Sweeney (1976), "The Capital Stock Adjustment Process and the Demand for Gasoline: A Market-Share Approach," in Econometric Dimensions of Energy Demand and Supply, A. Bradley Askin and John Kraft (eds.), Lexington, MA: Lexington Books, D.C. Heath and Company: 29-52.

Consumer Reports (various issues), published by Consumers Union, Mount Vernon, NY.

Cowling, Keith and John Cubbin (1971), "Price, Quality and Advertising Competition: An Econometric Investigation of the United Kingdom Car Market," Econometrica, 38 (November): 378-394.

Craft, Quentin (1982), The Gold Book Official Used Car Value Guide, vol. 4, no. 2, El Paso, TX.

Falvey, Rodney E., Jeff Frank, Harold O. Fried, and Mark Babunovic (not dated), "Fuel Economy Standards and Automobile Prices," mimeo, Tulane University, New Orleans, LA.

Goodman, Allen C. (1983), "Willingness-to-pay for Car Efficiency," Journal of Transport Economics and Policy, 18 (September): 247-266.

Griliches, Zvi (ed.) (1971), Price Indexes and Quality Change, Cambridge, MA: Harvard University Press.

Highway Loss Data Institute (1983), HLDI Injury and Collision Loss Experience, Washington, DC (September).

Hogarty, Thomas F. (1975), "Price-Quality Relations for Automobiles: A New Approach," Applied Economics, 7: 41-51.

Irvine, F. Owen, Jr. (1979), "Estimated Demand Equations for Individual Automobile Models with Implications for Regulatory Issues and Gasoline Conservation," mimeo, Wesleyan University, Middletown, CT (July).

The National Automobile Dealers Used Car Guide Company (1982), Used Car Trade-In Guide, McLean, VA (March).

Ohta, Makoto and Zvi Griliches (1976), "Automobile Prices Revisited: Extensions of the Hedonic Hypothesis," in Household Production and Consumption, N. Terleckyj (ed.), New York: Columbia University Press: 325-390.

Ohta, Makoto and Zvi Griliches (1986), "Automobile Prices and Quality: Did the Gasoline Price Increase Change Consumer Tastes in the U.S.?" Journal of Business and Economic Statistics, 4(April): 187-198.

Rosen, Sherwin (1974), "Hedonic Prices and Implicit Markets: Product Differentiation in Pure Competition," Journal of Political Economy, 82(January): 34-55.

Triplett, Jack E. (1969), "Automobiles and Hedonic Quality Measurement," Journal of Political Economy, 77: 408-417.

United States Department of Energy (1982), 1982 Gas Mileage Guide, Washington, DC.

Ward's Automotive Report (various issues), published by Ward's Communications, Inc., Detroit, MI.

Winston, Clifford and Fred Mannering (1984), "Consumer Demand for Automobile Safety," American Economic Review, 74(May): 316-319.

Printer friendly Cite/link Email Feedback | |

Author: | Asher, Cheryl Carleton |
---|---|

Publication: | Journal of Consumer Affairs |

Date: | Dec 22, 1992 |

Words: | 6177 |

Previous Article: | Investigation of the effects of disclosure statements in rental car advertisements. |

Next Article: | Personal factors related to consumer product disposal tendencies. |

Topics: |