Influential forecasts occur when the forecast itself determines whether the forecast is tested. New product sales forecasts are often influential because a low forecast may cause a firm not to launch a new product so that actual sales are never observed. This paper considers a dilemma we face as influential forecasters. Our client requests an unbiased forecast but pressures sometimes exist to provide a bias forecast. From theoretical and empirical perspectives, we discuss the impact of these pressures on the quality of forecasts. We find that: • Noninfluential forecasts, generally, create pressure for statistically biased forecasts. • As influence increases, the pressures increase. • When our forecasts eliminate alternatives, (e.g., product designs, advertising campaigns), not all forecasts are tested. • Not validating all forecasts causes two effects: Survivor's Curse and Prophet's Fear. • Survivor's Curse makes statistically unbiased forecasts appear optimistic (i.e., overestimate actual sales) because, often, only optimistic forecasts are tested. • Forecasts appearing statistically unbiased or pessimistic might cause concern. Perhaps, some failures are justified. • Prophet's Fear encourages pessimistic forecasts because these forecasts cause hidden opportunity losses while optimistic forecasts cause observable actual losses. • Tested forecasts may appear completely unbiased despite a pessimistic pre-launch bias. • Although no perfect solution exists, clients may lessen bias with experimentation and by seeking more accurate forecasts. Forecasters may lessen bias with forecasts conditioned on launching and by seeking more accurate forecasts.forecasting, new product research, channel relationships, bias, brand management

This paper reports the development and testing of a normative model for determining how firms should select the countries to be used in the information search for foreign direct investment. After a subset of countries is selected in the first stage of the decision process, a final selection process chooses the country with the best score within the subset: A "percentile method of subset selection" for singling out clusters of countries for information search and for identifying the best of the subset is presented that performs better than maximum country rankings ("top means") and maximum uncertainty ("top variance") techniques of subset selection. As an illustration, the percentile selection method is applied empirically.The research reported in this paper focuses on the development and testing of a normative model on how firms should specify the countries for which to carry out the information search for foreign direct investment (F.D.I.) The term "foreign direct investment" is used herein as the investment in a manufacturing facility in a host country to produce a given * Chaim Ehrman, receivedhis Ph.D. in Marketing from the Wharton School, University of Pennsylvania, 1984. He is an AssistantProfessorat LoyolaUniversity, and this paper was written while he was an AssistantProfessorat the University of Illinois at Chicago. His currentresearchinterestsareInternational Marketing, Consumer Behavior, and subsetSelectionProceduresfor Business Applications. ** Morris Hamburg is Professorof Statistics and Operations Researchat the Wharton School. His publications include research monographs,books, and numerousarticles in professionaljournals. He has been the directorof researchstudiesandhas servedas a consultant to corporations and as an expert witnessto law firmsandto government agencies. This articlestemmedfrom work begun by ChaimEhrmanin a doctoraldissertation supervisedby Morris Hamburg. t The authors acknowledge The Center for InternationalBusinessStudies at TheWhartonSchool of the University of Pennsylvania for financial support. The first author acknowledges ProfessorAbba Krieger,University of Pennsylvania andDr. Harold R. Shirefor guidance,encouragement, andfor constructivecomments.

No abstract

The field of direct mail advertising is becoming increasingly important. Many selection decisions must be made by direct marketers, such as those concerning package testing and list and segment within list selection. These decisions can be quite complex, especially when sample sizes and average order size per package and list are not equal. In this article, Bayesian and non‐Bayesian statistics are applied to these problems to generate optimal decision rules for package testing and list evaluation and selection. An example is given using real data from test results.