question Nelson's conjecture that more heavily advertised brands are likely to be better buys. It is a curious feature of attribute advertising that it informs some buyers not to shop among certain brands. Thus the incentives for advertising are clearly related to expected size of the market. On the one hand those varieties that would be less viable when information is less imperfect obviously have less incentives to advertise. In this sense the incentive to advertise depends on the seller's expectations of how well his brands cater to market preferences. Hence more heavily advertised brands are likely to be better buys. On the other hand, customers with minority preferences suffer in the presence of imperfect information because it is more costly for sellers to seek them out and tailor products to their needs and preferences. Specialized minority varieties may still appear in the market without much advertising if minority preferences are sufficiently intense relative to high volume alternative goods and their private costs of search are not too large. Clearly more heavily advertised brands are more likely to cater to "mass preferences" in the market and are unlikely best buys for those with more esoteric tastes.
This is an analysis of the optimum search strategies of consumers in markets of the type described above, pp. 165-179, in the presence of imperfect information and costs of transacting. George Stigler and J. J. McCall examined search behavior in markets for identical and homogeneous goods that sustain price dispersion by transaction costs.27 Here I attempt to generalize their results to markets where price dispersion is also sustained by transactions costs but where goods are heterogeneous. Again the heterogeneity is indexed by the attributes embodied by goods.
The model with full information in the second section of this paper showed that knowledge of the price-attributes function p(z) contained all relevant data the consumer needed to make a correct decision. In the case of incomplete information, the immediate generalization of p(z) is a joint probability distribution of product prices and characteristics. Denote the joint density defined over (p,z) by ϕ(p,z) dpdz. The conditional distribution of price given z represents observed price dispersion among varieties in the market with identical attributes z. If all goods are homogeneous and equivalent in characteristics, the conditional density completely describes market observations, just as in Stigler and McCall. Thus one can view the present problem as a "distribution of____________________