An objective measure of search versus experience goods.

Laband, David N.

I. INTRODUCTION

"Search" goods and "experience" goods have been part of the lexicon of economics for twenty years now. During this time a substantial body of research has been influenced by Nelson's 1970 and 1974 papers, which enlarged upon Stigler's [1961] seminal paper. Despite the appeal of the search versus experience dichotomy to members of the economics profession, empirical development and application of the concepts involved has not been impressive. (1)

Stigler [1961] argued that consumers' efforts to acquire information about the quality or performance of products and vendors are a function of expected benefits and costs. Nelson's [1970; 1974] work attempted to measure the expected benefits and, to a lesser extent, expected costs to consumers of acquiring product and vendor information. Unfortunately, although this search versus experience dichotomy now is part of economists' conventional wisdom, it has led to little empirical analysis beyond Nelson's investigations. The root problem seems to be a lack of agreement over the definition of a search good versus an experience good. To Nelson [1970], the latter is one whose qualitative characteristics are not ascertainable by buyer inspection prior to purchase. Thus, information about the product's quality or performance can be obtained only through buying and using the item. We can, however, ascertain performance information about search goods prior to purchase.

Porter's [1976] analysis of the acquisition of product information by consumers classified items as convenience and non-convenience goods. The former are characterized by low unit price and high frequency of purchase; the opposite holds for non-convenience goods. The low unit price of convenience goods implies that relevant performance information will be acquired via sampling (experience), even though one might be able to acquire the information from other sources, but at higher cost. An experience good is not such by design, rather by virtue of consumer choice in the face of varying informational costs. Thus, Porter's measure of the incentive to acquire product/vendor information is product price, as opposed to Nelson's ad hoc search/experience dichotomy. Using Porter's classification, Meisel [1979] argues that the degree to which consumers' efforts to acquire information take the form of search rather than sampling/experience depends on the expected costs and benefits for any given product. Similarly, Laband [1986] argues that consumer demand for product and vendor information in advance of purchase is affected by both price and frequency of purchase.

Nelson's classification scheme is at odds with his definitions. For example, he regards food as an experience item, although much relevant information about it is available prior to purchase. (2) Product classification is admittedly ambiguous (e.g., jewelry and silverware). In fact, the search vs. experience classification more logically represents the two extremes of product classification, with most items falling somewhere in-between. By subdividing experience goods into durable and nondurable categories, Nelson [1974] validates this assertion.

Given this ambiguity, it is understandable that empirical research in this nexus of producer/consumer interaction has been limited. To use the search-experience classification for empirical analysis, one must either borrow Nelson's existing scheme, which is limited in product coverage, or create new product classifications which are of dubious merit. On the other hand, the Porter/Meisel analyses, which do provide the micro-foundations of consumer behavior in the face of uncertain product/vendor quality, have never supplanted or integrated Nelson's work into their broader framework.

The purpose of this paper is to bridge these two strands of the advertising as information literature. I argue that the discussion of search versus experience goods and, more importantly, the behavior of buyers and sellers, is driven by the cost to the buyer of a disappointing purchase. As the cost of making a disappointing purchase increases, the would-be buyer rationally seeks to acquire additional information prior to purchase regarding product quality and performance. At the other extreme, for some items the cost of making a disappointing purchase is relatively small. Information about product quality for these items may be obtained cheaply through sampling and experience.

As the purchase price of an item rises, so does the cost of making a disappointing purchase and, accordingly, so do the benefits from pre-purchase efforts to acquire information, ceteris paribus. Using this definition, the awkward dichotomy of search versus experience can be replaced by a continuous variable--price. My model generates results similar to those obtained by Nelson but is not limited in its application to the specific items he identified. Moreover, product price is an objective, continuous measure of the benefits of information search; this suggests replication of past, present, and future results by other economists, as well as the potential for investigating the shape of the response function.

The paper is organized as follows. In section III briefly discuss the concept of a disappointing purchase. A more focused analysis of the role played by product price in determining consumers' efforts to acquire product- and vendor-specific information follows. I then discuss the relationship between product price and the supply of product performance information by sellers. A simple test of my model is devised and conducted in section III. Producers supply information about their products via advertising. As the price of an item rises, the cost of making a disappointing purchase rises and potential buyers' demand for ex ante information about the performance of the product increases, generating profit opportunities for sellers who provide such information. My hypothesis that producer supply of informational signals is related positively to product price is supported by data gathered from nearly 14,000 product ads placed in the Baltimore Sun daily newspaper during 1986. Implications of these findings are discussed in section IV.

II. THEORETICAL DISCUSSION

Disappointing Purchases

A consumer, i, expects to obtain a certain level of utility from the purchase and consumption of commodities. This expected utility is valued at V[E([U.sub.ij])], for item j. After purchasing j, (s)he obtains some actual level of utility from consumption of it, which is valued at V[[U.sub.ij]]. A disappointing purchase occurs whenever a buyer's pre-purchase expectations of the performance/quality of a product are higher than the actual performance of that product (in terms of the consumer's realized utility) in the post-purchase period:

(1) V[E([U.sub.ij])] > V[[U.sub.ij]].

Product Price, Disappointing Purchases and Demand for Ex Ante Information

Consumers demand product and vendor information for two reasons: (a) they are uncertain in the pre-purchase period about the ability of the product to satisfy them vis-a-vis their reasons for purchasing the item, and (b) there is a positive incidence of defective products in the world (i.e., variation around the mean performance characteristics of specific items). The latter factor, of course, influences the former.

As Spence [1977, 562] notes, a disappointing purchase is indisputably costly to (at a minimum) the buyer of the item. Three cases merit examination: (a) V[[U.sub.ij]] = 0 < V[E([U.sub.ij])], (b) V[[U.sub.ij]] < 0 < V[E([U.sub.ij])], and (c) 0 < V[[U.sub.ij]] < [P.sub.j] < V[E([U.sub.ij])], where [P.sub.j] is the purchase price of j.

When V[[U.sub.ij]] = 0, the purchaser makes no use of the product purchased, and his nominal loss is equal to the purchase price paid. The real loss is the value of his expected utility from consumption of that item minus the realized level of utility, in this case zero. The real loss will almost always exceed the nominal loss, since a necessary condition for purchase to occur is that V[E([U.sub.ij])] > [P.sub.j], and [P.sub.j] thus represents the lower bound on the real cost of making a disappointing purchase when V[[U.sub.ij]] = 0.

When consumption of j is harmful to the buyer, V[[U.sub.ij]] < 0. The consumer of cyanide-laced capsules receives negative utility from the product purchased, even though the expected utility was positive. The consumer suffers a loss greater than that experienced when the product is simply worthless to him. That loss is greater by the value of the harm incurred. Again, in the limit the minimum cost to the buyer of making a disappointing purchase is [P.sub.j].

Finally, when 0 < V[[U.sub.ij]] < [P.sub.j] < V[E([U.sub.ij])], the buyer gains positive utility from consumption of j. However, actual utility received is lower than expected utility. The nominal cost of making the disappointing purchase in this case is equal to [P.sub.j], since V[[U.sub.ij]] < [P.sub.j] defines a purchase that would not have been made under conditions of perfect information. The real loss to the consumer is the foregone purchase of some other commodity k, where V[[U.sub.ij]] < [P.sub.j] < V[E([U.sub.ik])]. This loss equals V[E([U.sub.ik])] -V[[U.sub.ij]]. As V[[U.sub.ij]] approaches [P.sub.j] and V[E([U.sub.ik])] approaches [P.sub.j] = [P.sub.k], that loss approaches zero as a minimum. (3)

Stigler [1961] and others have argued that consumer search for product- and vendor-specific information prior to purchase of an item will occur as long as the benefits of additional search exceed the costs. As the nominal purchase price of a product rises, all other things held constant, the expected cost of making a disappointing purchase also rises, and the value of additional search must rise. Therefore, holding constant the unit cost of information acquisition, we should expect more search behavior for higher-priced products/services, irrespective of the frequency of purchase. (4) This holds true even though nominal product price is an imperfect proxy for the costliness of making a disappointing purchase. (5)

A Simple Model of Consumer Demand for Pre-Purchase Information

An individual, i, is assumed to maximize the value of his expected utility from the purchase and consumption of one unit of good j:

(2) max V[E([U.sub.ij])] = V[[U.sub.ij]](1-[phi]) - [P.sub.j],

where [phi] is the probability that j is defective and [P.sub.j] is the nominal purchase price of j. (6) The individual's utility in consumption is influenced by product price and the probability of product defect. Alternatively, we may view the consumer as attempting to minimize his expected loss from the purchase of item j:

(2') min E(L) = -V[[U.sub.ij]] (1-[phi]) + [P.sub.j],

where E(L) = {-V[E([U.sub.ij])]}. There are two relevant cases to consider: (a) the probability of defect is identical across products, and (b) the probability of defect differs across products.

Case (a): When the probability of defect (in the broad sense) is constant across products, the consumer's expected gain/loss from purchasing item j is influenced purely by [P.sub.j]. For a given item, the greater the purchase price, [P.sub.j], the less likely is expected utility going to net out positive. From one product to another, the size of the loss incurred by a defective unit is also purely a function of the purchase price. A buyer incurs a greater loss from a defective $10,000 automobile than from a defective $1 cantaloupe.

Case (b): When the probability of defect is variable across products, the consumer's expected gain/loss from purchasing j depends upon [P.sub.j] and [phi]. Within a product class, an observed price dispersion may imply that the probability of product defect varies from one seller to another. Price itself is one means whereby sellers convey information to buyers with respect to defect probability. Specifically, [partial derivative][phi]/[partial derivative] [P.sub.j] < 0. (7) In this case, it is unclear whether [P.sub.j] provides information with respect to the ex post losses due to product defect or the ex ante probability of defect. However, the existence of a relationship between price and the probability of defect within a product class has no necessary bearing on the importance of product price to the incentive to search across products. Across products there is no reason to suspect the [partial derivative][phi]/[partial derivative][P.sub.j] < 0 relationship to hold. Indeed, I maintain that the relationship across products is reversed, [partial derivative][phi]/[partial derivative][P.sub.j] > 0.

Typically, expensive items include performance dimensions that are not included for inexpensive items. Assuming a producer's cost curve for detecting and/or correcting defective units slopes upward, additional dimensions imply a greater incidence of product defect. Thus, buyers will expect more expensive items to have greater rates of defect than less expensive items ([partial derivative][phi]/[partial derivative] [P.sub.j] > 0). All other things being equal, this provides a motivation for consumers to search more intensively for ex ante information about product/seller performance.

The consumer's ex ante demand for information about product/seller performance depends upon his expected loss:

(3) D(I) = f [E(L)],

where f' > 0, and f" need not be signed. Since E(L) = g[[P.sub.j], [phi] ([P.sub.j])], substituting into (3) yields:

(3') D(I) = g[[P.sub.j],[phi]([P.sub.j])].

Given f' > [partial derivative], E(L)/[partial derivative][P.sub.j] > 0

and [partial derivative]E(L)/[partial derivative][phi] x [partial derivative] [phi]/[partial derivative][P.sub.j], [partial derivative]D(I)/ [partial derivative][P.sub.j]

must be positive by transitivity.

Product Price and Producer Supply of Ex Ante Information

Given normal assumptions about the shape of the cost functions faced by buyers and sellers for producing information about product quality ex ante, sellers will produce at least some ex ante product performance information for buyers, with the quantity produced being related in a positive fashion to product price. Indeed, supply-side considerations have heretofore given meaning to search and experience goods. Suppose [C.sub.b] and [C.sub.s] represent the cost functions to buyers and sellers for producing information. Holding quality of information fixed, the equilibrium condition implied by efficient production of information is

[partial derivative][Q.sub.b]/[partial derivative][C.sub.b] = [partial derivative][Q.sub.s]/[partial derivative][C.sub.s]

where Q refers to quantity of performance information produced. As the price of a product rises, demand for ex ante performance information rises. Assuming the information acquisition efforts of buyers are subject to diminishing returns, efficient production of information implies that sellers of high-priced products will supply more performance information than will sellers of low-priced products. (8)

III. TESTING THE DISAPPOINTING PURCHASE MODEL

While Nelson's search/experience dichotomy may exhibit predictive power in a model of search behavior, a continuous product price variable, if found to be important, would permit the scope of empirical inquiry to be broadened considerably, since one would not then be constrained to examine only those items falling within Nelson's search/experience classification. To place the search/experience dichotomy in the broader context of product price, the empirical investigation must be able to:

(1) replicate Nelson's results for search/experience goods with the data contemplated for testing the hypothesized relationship between search and product price (to ensure my results do not result from peculiar data);

(2) demonstrate that the hypothesized relationship between product price and measures of seller provision of product/vendor information obtains, even in the presence of a search/experience variable; and

(3) demonstrate that the relationship between product price and search behavior exists more generally in a large, random sample of all products.

My analysis suggests that consumers will search for additional information on performance prior to product purchase as the nominal purchase price of the product rises. Ceteris paribus, the location of the demand curve for product/vendor information is dependent upon the price of the product: the higher the price of the product, the further out the demand curve. That is, at a given unit price of product information, demand for that information rises with the price of the product. In a general equilibrium framework, as shown by Ehrlich and Fisher [1982], the demand for pre-purchase information will incite a supply response, from both consumers and sellers of products.

Although most investigators in the advertising-as-information tradition have analyzed gross quantity of advertising, I am concerned with the content of advertising. The empirical tests reported below are based on data collected from all product/service ads (excepting commercial real estate ads and classified ads) placed in the Baltimore Sun morning newspaper during 1986. Data were collected from the Nation, Sports, Business, and Entertainment sections of the newspaper, every day of the week. Information was collected for each ad regarding product price, product type, selection offered by the seller, seller's longevity in business, and whether or not the seller indicated explicitly: (a) that a guarantee or warranty was offered on the product, (b) that store financing was available to potential buyers, (c) address/ phone number information, (d) days of the week/hours per day of operation, and (e) that third party credit cards were an acceptable form of payment. The data set consisted of 30,940 ads.

Price as a Measure of Disappointing Purchases: Evidence

I began by deleting from the collection of ads all those for items not classified by Nelson [1970, 319]. Ads for which no specific product price was available were also deleted. The sample of ads satisfying both of these criteria numbered 6,972. Equation (4) details the model estimated in an attempt to verify Nelson's results on my data collected from the Sun:

(4) SELLER PROVIDED INFORMATION =

[[alpha].sub.0] + [[alpha].sub.1]ADSIZE + [[alpha].sub.2]EXPERIENCE + [[alpha].sub.3]LOG OF PRICE + [[epsilon].sub.i],

where

SELLER PROVIDED INFORMATION =

a dummy variable, assigned a value of one if the seller indicated in the ad (a) that the product(s) being sold was guaranteed or warranted in some respect, (b) the number of years the seller had been in business, or (c) that the seller offered financing on the item being purchased;

ADSIZE =

number of square inches of the advertisement;

EXPERIENCE GOOD =

a zero-one dummy variable, which assumes a value of one for products classified by Nelson [1970, 319, Table 2] as experience Items and a value of zero for Nelson-classified search goods;

LOG OF PRICE =

the log of the nominal price of the product being advertised; where multiple products were advertised, price is defined as the average of the spread, i.e., the highest plus the lowest price, divided by two; and

[epsilon] =

a binomially-distributed error term.

At issue is the extent to which sellers provide specific information in their ads. Once the size of the ad is fixed, the seller's marginal cost of including any specific information equals zero, since he can include as few or as many signals in a given ad space as desired, at no extra cost (the cost is for the space, not the content). (9) The predicted relationship between experience goods and the dependent variable, SELLER PROVIDED INFORMATION, is straightforward. Since qualitative characteristics of experience goods can be ascertained only after purchase, whereas inspection prior to purchase may reveal performance information for search goods, guarantees and warranties will be particularly valuable to buyers of experience goods, ceteris paribus. Similarly, the fact that the seller signals longevity in business and whether or not s(he) offers financing on purchases is of particular value to buyers of experience goods. (10) Conversely, this information should not be of much value to buyers of search goods. In a general equilibrium framework, this information is less likely to be included in the advertising provided by sellers of search goods. (11)

I employed the LOGIT procedure for estimation of a regression equation with a dichotomous dependent variable. Estimation results for three alternative specifications of equation (4) on the restricted sample of Nelson-defined items are reported in Table I and discussed below. The regression results reported in column (a) essentially replicate Nelson's earlier findings. Seller provision of at least one of the three specific types of information discussed above was predicted to occur more often for experience goods as compared to search goods. The prediction is borne out by the data; the coefficient on EXPERIENCE goods is positive, sizable and statistically significant at the 0.01 level.

TABLE I
LOGIT Regression Results for Equation (4), Restricted Sample

Variable (a) (b) (c)

Intercept -1.0887 *** -1.0839 *** -1.3887 ***
 (0.0747) (0.0998) (0.1048)

ADSIZE 0.0073 *** 0.0076 *** 0.0076 ***
 (0.0004) (0.0004) (0.0004)

EXPERIENCE GOODS 1.2119 *** 1.0658 ***
 (0.0711) (0.0794)

LOG OF PRICE 0.1324 *** 0.0538 ***
 (0.0116) (0.0131)

N 6972 6972 6972

# with S.P.I. = 1 4419 4419 4419

Model Chi-square 730.53 561.99 747.40

-2 Log Likelihood 8429.16 8597.69 8412.28

*** Chi-square statistic significant at 0.01 level.

Standard errors are in parentheses.

In order to investigate the comparative explanatory power of product price on

seller supply of informational signals, I re-estimated the model, replacing the dichotomous EXPERIENCE variable with a continuous LOG OF PRICE variable; the results are reported in column (b). Since the value of, and thus demand for, product and vendor information embodied in explicit guarantees and warranties and implicit guarantees (seller experience and financing) increases as the purchase price of the product rises, I expected [[alpha].sub.2] to sign positively; producers respond to this demand by advertising guarantees/warranties. Seller longevity and seller financing should also be a positive function of product price.

The results reported in column (b) demonstrate that product price is a statistically significant predictor of explicit and implicit guarantee and warranty provision by sellers, results that support my claim that product price may serve as a useful proxy for the cost to consumers of making a disappointing purchase. In column (c) the EXPERIENCE and LOG OF PRICE variables are both included in the estimated equation. Both variables demonstrate independent and statistically significant explanatory power.

This suggests that aspects of the Nelson classification are not addressed by product price, and vice versa. Within the confines of this restricted sample, the predictive ability of the price model is 23 percent lower than that of the pure search/experience model. However, investigation of models using price as a measure of the cost of disappointing purchases need not be limited to Nelson-classified items. One advantage of the price model is its potential applicability to all items, services as well as products.

Accordingly, I re-estimated equation (4) across the unrestricted sample of all ads placed in the Baltimore Sun during 1986, that conveyed price information with respect to specific items (N=13964). LOGIT estimation results for the general sample are reported in Table II and discussed below.

TABLE II
LOGIT Regression Results for the Unrestricted Sample

Variable (a) (b)

Intercept 2.2401 *** 2.2809 ***
 (0.0523) (0.0535)

ADSIZE 0.0047 *** 0.0048 ***
 (0.0002) (0.0002)

LOG OF PRICE 0.2839 *** 0.2872 ***
 (0.0071) (0.0072)

SERVICES 0.4795 ***
 (0.1153)

DURABLES 0.1373 **
 (0.0601)

N 13964 13964

# with S.P.I. = 1 6502 6502

Model Chi-square 2306.63 2323.37

-2 Log Likelihood 16985.53 16968.79

Variable (c) (d)

Intercept 2.1891 *** 2.2208 ***
 (0.0566) (0.0573)

ADSIZE 0.0047 *** 0.0049 ***
 (0.0002) (0.0002)

LOG OF PRICE 0.2644 *** 0.2631 ***
 (0.0111) (0.0111)

SERVICES 0.5228 ***
 (0.1161)

DURABLES 0.1719 ***
 (0.0606)

N 13964 13964

# with S.P.I. = 1 6502 6502

Model Chi-square 2311.85 2331.40

-2 Log Likelihood 16980.32 16960.76

*** Chi-square statistic significant at the 0.01 level.

** Chi-square statistic significant at the 0.05 level.

Standard errors are in parentheses.

The results for the general sample mirror those for the restricted sample. Product price, by itself, is a statistically significant predictor, in the expected positive direction, of informational signalling through the advertisement of warranties and experience by would-be sellers. The strong showing of LOG OF PRICE in both the restricted and unrestricted samples is all the more impressive because, to the extent product price understates the true cost of making a disappointing purchase, it is less likely that consumer demand for information, and thus producer supply of information, will be revealed by the data.

I note that over the more general sample the price model is a much better predictor of quality assurance signalling by sellers than is the search/experience model over its restricted sample. Compare regression (a) in Table I (the search/experience model) against regression (a) in Table II (the price model): the model Chi-square of the latter is more than triple that of the former.

Why the relatively poor showing of LOG OF PRICE in Table I? I suspect this reflects a shortcoming of restricting the sample of items analyzed to those classified by Nelson. Of the 6972 items recognized in the restricted sample, 4419 were associated with quality assurance signalling by sellers (63.4 percent), whereas in the unrestricted sample only 6502 ads out of 13964 signalled quality guarantees (46.6 percent). The restricted sample thus apparently over-represents the distribution of quality assurance signals as compared to the true distribution of those signals; it ought to yield a strong coefficient on the EXPERIENCE GOOD coefficient.

The regression estimates reported in column (b) and (c) include separate zero-one dummy variables for services, an inclusion heretofore impossible due to the restricted nature of Nelson's sample, and durables. Supply of information containing explicit and implicit quality assurances is shown to be significantly higher for services than for products, a result that dovetails nicely with the theory of advertising as information. Qualitative aspects of services are generally more difficult to ascertain, in both the pre- and post-purchase periods, than for products. I expect demand for guarantees to be greater for services than for products and for sellers to react appropriately.

Durables are costly, not only in terms of initial price, but also on account of maintenance expenditures. Arguably, consumers have plenty of incentive to acquire ex ante performance information about durables; the supply-side response by sellers is predictable. Without diminishing the impact or importance of LOG OF PRICE on SELLER PROVIDED INFORMATION, the facts support my expectations about durables. Producer supply of quality assurance signals is significantly greater for durables than non-durables. This result squares nicely with those of Gerstner [1985] and Tellis and Wernerfelt [1987].

Recent work by Milgrom and Roberts [1986], Kihlstrom and Riordan [1984], Gerstner [1985], Tellis and Wernerfelt [1987], Bagwell and Riordan [1988], and others, suggests that product price may serve to inform consumers about the ex ante probability of product defect, in addition to measuring the ex post losses to consumers from experiencing a defective product. (12) Empirical support for a positive price-quality relationship within product categories is weak and highly product-specific, although it seems to hold more strongly for durables than non-durables. (13) Since my data consist of advertisements for numerous products, placed by multiple sellers within each product category, there is a possibility that my empirical results derive from positive within-product price-quality relationships that have more to do with ex ante probabilities of defect than with ex post estimates of the losses suffered by purchasers of defective products. To explore the seriousness of this objection to my results, I aggregated my data by product category and generated price means for each. In addition, I generated a coefficient of variation for price, as recommended by Tellis and Wernerfelt [1987, 246]. I then examined the relationship between the fraction of all sellers of each product who provided at least one of the three quality cues (guarantee/warranty, seller experience, seller financing) and mean product price and coefficient of price variation, using Ordinary Least Squares (OLS) regression. The results are reported in Table III.

TABLE III
Product Price and Seller Supply of Quality Signals

Variable Coefficient Std. Error t-statistic

Constant 0.3587262 .2850200 12.586 ***
Mean Price 0.0000039 .0000010 3.749 ***
Price Variation 0.0323412 .0222187 1.456

N = 176
[R.sup.2] = .0900
[F.sub.reg] = 8.501 ***

*** Significant at .01 level.

Seller supply of quality signals is related positively to mean product price. By construction, this finding cannot result from ex ante signalling by sellers of the probability of defect. (14) This finding supports my contention that the incentive for consumers to search for pre-purchase information is a function of the price of the item being considered for purchase.

IV. IMPLICATIONS AND CONCLUDING THOUGHTS

Product price governs the pre-purchase demand for performance information, since price is what the buyer is at risk for. A buyer risks more when he spends $10,000 on an item than when he spends $10 on an item. The expected loss is therefore the money spent, and people spend more to protect items worth $10,000 from theft than for items worth less. Product price, as a measure of the cost to consumers of making a disappointing purchase, and hence the benefits of engaging in search, has attractive properties: (a) it is a readily available attribute of products to the empirical researcher, (b) it is an objective, continuous measure that stands to reduce squabbling about the classification of specific items, and (c) a much broader spectrum of products than heretofore agreed upon may be subject to scrutiny. Moreover, previous theoretical and empirical results can be replicated by this model; they also can be extended to encompass a broad spectrum of products and services.

The unit cost of acquiring information has been assumed to remain constant. A complete model of actual search behavior must, of course, take both the costs and benefits of search into account. Although the cost side of the model has been addressed informally by Laband [1986], no commonly accepted proxy variable(s) for the cost of search exists at this time. My results may be interpreted as shedding light on a related subject, namely, the division of labor between buyers and sellers in the production of pre-purchase information about product quality. This is a topic whose veil has yet to be lifted by economists. (15) Guarantees and warranties imply that firms are not the least-cost detectors of defective units, consumers are. Sellers pay for the detection, but consumers do the detecting. Across items, as price rises the number of dimensions on which quality may be judged undoubtedly rises (with admitted exceptions, such as diamonds), making detection of defective units that much more costly to firms with normally-shaped cost functions. Thus, deciding this division of labor, the degree to which producers of different products turn over the detection task to consumers will be related positively to product price. This is exactly my finding.

Implications for policy and further research abound. To the extent the regulatory authorities insist on requiring informational disclosure by product manufacturers and retailers, my model suggests that higher-priced products be regulated before lower-priced products, ceteris paribus. My model provides an objective basis for regulating false or deceptive advertising, with stiffness of penalty being a function of product price and number of buyers. The model suggests that private, independent suppliers of product information, such as Consumer Reports, will make more money by concentrating their focus on higher-priced items rather than low-priced items, ceteris paribus. With respect to future research, the model helps open analysis of information search across product lines, and within product lines. Again, the development of measures of the cost of search is important. A comprehensive listing of products and the benefits of search, ranging from high-benefit (search) goods to low-benefit (experience) goods is now possible. Research into the content of advertising, i.e., the types of information being signalled from sellers to buyers, as a function of product price would also appear to be a fruitful area of inquiry.

REFERENCES

Akerlof, G. A. "The Market for 'Lemons': Qualitative Uncertainty and the Market Mechanism." Quarterly Journal of Economics, August 1970, 488-500.

Anderson, J. E. and F. M. Gollop. "The Effect of Warranty Provisions on Used Car Prices," in Consumer Protection Economics, edited by E M. Ippolito and D. T. Scheffman. Washington, D.C.: Federal Trade Commission, 1984, 67-102.

Bagwell, K. and M. Riordan. "High and Declining Prices Signal Product Quality." Northwestern Discussion Paper, October 1988.

Barzel, Y. "Measurement Cost and the Organization of Markets." Journal of Law and Economics, April 1982, 27-48.

Chan, Y. S. and H. Leland. "Prices and Qualities in Markets with Costly Information." Review of Economic Studies, October 1982, 499-516.

Clarkson, K. W. and R. L. Miller. Industrial Organization. New York: McGraw-Hill Book Company, 1982.

Cooper, R. and T. Ross. "Prices, Product Qualities and Asymmetric Information: The Competitive Case." Review of Economic Studies, April 1984, 197-207.

Curry, D. and P. Riesz. "Prices and Price/Quality Relationships: A Longitudinal Analysis." Journal of Marketing, January 1988, 36-51.

Dardis, R. and N. Gieser. "Price and Quality of Durable Goods: Are They More Closely Related in the Seventies Than in the Sixties?" Journal of Consumer Policy, 4(3), 1980, 238-48.

Ehrlich, I. and L. Fisher. "The Derived Demand for Advertising." American Economic Review, June 1982, 366-88.

Farrell, J. "Prices as Signals of Quality." Ph.D. dissertation, Brasenose College, Oxford, 1980.

Friedman, M. P. "Quality and Price Considerations in Rational Consumer Decision Making." Journal of Consumer Affairs, Summer 1967, 13-23.

Geistfeld, L. V. "The Price-Quality Relationship Revisited." Journal of Consumer Affairs, Winter 1982, 334-46.

Gerstner, E. "Do Higher Prices Signal High Quality?" Journal of Marketing Research, May 1985, 209-15.

Greer, D. F. Industrial Organization and Public Policy. New York: Macmillan, 1980.

Grossman, S. "The Informational Role of Warranties and Private Disclosure about Product Quality." Journal of Law and Economics, December 1981, 461-84.

Judd, K. L. and M. H. Riordan. "Price and Quality in a New Product Monopoly." Boston University Working Paper, January 1989.

Kihlstrom, R. E. and M. H. Riordan. "Advertising as a Signal." Journal of Political Economy, June 1984, 427-50.

Klein, B. and K. B. Leffler. "The Role of Market Forces in Assuring Contractual Performance." Journal of Political Economy, August 1981, 615-41.

Laband, D. N. "Advertising as Information: An Empirical Note." Review of Economics and Statistics, August 1986, 517-21.

Laband, D. N. and M. T. Maloney. "Seller Financing, Quality Assurance, and Bilateral Opportunism." Unpublished manuscript, Clemson University, 1989.

Martin, S. Industrial Economics. New York: Macmillan, 1988.

Meisel, J. B. "Demand and Supply Determinants of Advertising Intensity Among Convenience Goods." Southern Economic Journal, July 1979, 233-43.

Milgrom, P. and J. Roberts. "Price and Advertising Signals of Product Quality." Journal of Political Economy, August 1986, 796-821.

Nelson, P. "Information and Consumer Behavior." Journal of Political Economy, March/April 1970, 311-29.

--."Advertising as Information." Journal of Political Economy, July/August 1974, 729-54.

Olson, J. C. "Price as an Informational Cue," in Consumer and Industrial Buying Behavior, edited by A. G. Woodside, J. N. Sheth and P. D. Bennett. New York: Elsevier, North Holland, 1977.

Palfrey, T. R. and T. Romer. "An Experimental Study of Warranty Coverage and Dispute Resolution in Competitive Markets," in Consumer Protection Economics, edited by P. M. Ippolito and D. T. Scheffman. Washington, D.C.: Federal Trade Commission, 1984, 307-72.

Porter, M. E. Interbrand Choice, Strategy and Bilateral Market Power, Cambridge, MA: Harvard University Press, 1976.

Riesz, P. C. "Price Versus Quality in the Marketplace, 1961-1975." Journal of Retailing, Winter 1978, 15-28.

--."Price Quality Correlations for Packaged Food Products." Journal of Consumer Affairs, Winter 1979, 236-47.

Spence, A. M. "Consumer Misperceptions, Product Failure, and Producer Liability," Review of Economic Studies, October 1977, 561-72.

Sproles, G. B. "New Evidence on Price and Product Quality." Journal of Consumer Affairs, Summer 1977, 63-77.

Stigler, G. J. "The Economics of Information." Journal of Political Economy, June 1961, 213-25.

Tellis, G. and B. Wernerfelt. "Competitive Price and Quality Under Asymmetric Information." Marketing Science, 6, Summer 1987, 240-53.

Telser, L. "Supply and Demand for Advertising Messages." American Economic Review, Papers and Proceedings, May 1966, 457-66.

Wolinsky, A. "Prices as Signals of Product Quality." Review of Economic Studies, October 1983, 647-58.

(1.) The appeal of the search vs. experience dichotomy is revealed by its inclusion in undergraduate Industrial Organization textbooks (e.g., Martin [1988]; Greer [1980]; Clarkson and Miller [1982]).

(2.) See Nelson [1970, 325]. Although taste cannot be determined a priori, ingredient content, which may bear a strong relationship to taste, caloric information, price, preparation instructions and other information may be secured prior to purchase. Grocers do not normally kick shoppers out of their stores for sampling grapes. Indeed, some food manufacturers provide tasting of their product prior to checkout in grocery stores. On the other hand, Baskin Robbins outlets provide prepurchase tasting of the ice cream they sell, whereas most restaurants do not offer such tasting. I have no ready explanation of producer behavior in this domain.

(3.) It is, of course, possible for one to make a purchase that is disappointing, but not costly, although it seems rather unlikely. IF the actual utility received by consumer i from purchase of commodity j is lower than the expected utility, V[[U.sub.ij]] < V[E([U.sub.ij])], than he has made a disappointing purchase. However, if V[[U.sub.ij]] > [P.sub.j] and, more importantly, if V[[U.sub.ij]] > V[E([U.sub.ik])] that is no opportunity loss or nominal purchase price loss associated with the disappointing purchase.

(4.) A model of risk-averse consumers also generates the result that consumer demand for product information is a positive function of product price.

(5.) The statement that as nominal purchase price increases so does the expected cost of making a disappointing purchase is undeniably correct for cases (a) and (b), where price understates the true cost of making a disappointing purchase. Even for (c), in the particular circumstance that price overstates the true cost of making a disappointing purchase, the statement is also correct, since as [P.sub.j] = [P.sub.k] rises, the opportunity cost of not having purchased commodity k also rises and the opportunity loss V[E([U.sub.ik])] - V[[U.sub.ij]] increases. Thus, as product price rises, the probability that price overstates the true cost of the disappointing purchase falls.

(6.) To avoid a bundling/unbundling problem with respect to price, [P.sub.j] refers to the price paid for an item, even if several consumption units of the item must be purchased as a bundle. For example, you can purchase potatoes individually or by the bag. I assume it is the purchase price of the bundle, not the price of the consumption unit, that the consumer places at risk. For simplicity, I assume that the individual is not harmed by purchase of a defective product, i.e., [partial derivative]V[[U.sub.ij]]/ [partial derivative][phi] = 0. [partial derivative]V[[U.sub.ij]]/ [partial derivative][phi] < 0 complicates the equations, but does not alter the basic relationship between V[E([U.sub.ij])] and [P.sub.j].

(7.) There is a substantial literature on the static and dynamic within-product relationship between price and product quality. For an excellent, summary of this literature, see Bagwell and Riordan [1988].

(8.) Note that not only will sellers of high-priced items supply more information about product performance than sellers of low-priced items, buyers of high-priced products will self-produce more ex ante information about product performance than will buyers of low-priced products. More pre-purchase information is demanded and supplied, in total, for higher-priced items than lower-priced items, ceteris paribus.

(9.) of course, ad size itself may convey information from sellers to buyers. However, the information conveyed has meaning only for comparisons between sellers of the same product and not between products. Since the latter is investigated here, it seems unlikely that the results are tainted by inclusion of the ad size variable.

(10.) Years of experience signals high (average) product quality and satisfied customers (else there would be no such longevity), while seller financing serves as an implicit satisfaction guarantee. This aspect of seller financing is argued and tested for empirically by Laband and Maloney [1989]. For explicit discussions of the role of guarantees and warranties in the nexus of producer/consumer interaction, consult Anderson and Gollop [1984], Spence [1977], Akerlof [1970], Grossman [1981], and Palfrey and Romer [1984].

(11.) I explicitly assume that the ubiquitous search for profits motivates sellers to respond appropriately to consumer demand for product/vendor information, especially since they may be the least cost producers of such information. I further assume that sellers (qua advertisers) attempt to convey relevant information to consumers where such information has the most value. In other words, the decision by the firm to convey specific information to potential consumers via advertising is a filter that refines the data for the purpose of testing the theory. One of the eeteris paribus conditions of the theory is the degree of knowledge possessed by the buyer. By examining advertisements, we let the seller tell us what the buyer is ignorant about.

(12.) See, for example, Chan and Leland [1982], Cooper and Ross [1984], Curry and Riesz [1988], Dardis and Gieser [1980], Farrell [1980], Friedman [1967], Geistfeld [1982], Olson [1977], Riesz [1979], Sproles [1977], and Wolinsky [1983].

(13.) For example, Gerstner [1985] finds significant (at the .05 level) Kendall correlation coefficients between price and quality for only twenty-four of eighty-nine infrequently purchased items (28 percent) and seven of fifty-nine frequently purchased items (12 percent). With respect to my data, I was able to investigate the strength of the within-product price-quality relationship for ninety items. For these products, I regressed the probability that the seller advertised at least one of the three quality signals referred to earlier against advertised price and ad size. Sixty percent (54/90) of the estimated coefficients on the price variable were negative, and only 22 percent (20/90) of all coefficients were statistically significant at the .10 level. Only 17 percent (15/90) were significant at the .05 level. However, only eight of the twenty significant coefficients signed positive; the rest signed negative. At best, my results suggest no general price-quality relationship within products; any such relationship is apparently highly product specific. Indeed, as argued by Bagwell and Riordan [1988], such a relationship may exist only temporarily, for new producers.

(14.) I also re-estimated regression (a) from Table II with zero-one variables for each product category, except the control category (drapes). The coefficient estimate for LOG OF PRICE is .293567, which is significant at the .0001 level.

(15.) Barzel's [1982] analysis of measurement cost and the organization of markets is an insightful prolegomena on the problem.

DAVID N. LABAND, Department of Economics and Center for Policy Studies, Clemson University. Helpful comments received from Cotton M. Lindsay, Robert E. McCormick, Raymond D. Sauer, Mason Gerety, T. Bruce Yandle, Roger Meiners, John P. Sophocleus and an anonymous reviewer are gratefully acknowledged. The usual caveat applies.