Memory Seeding: Representations Underlying Quantitative Estimations Nadezhda N. LaVoie, Lyle E. Bourne Jr., and Alice F. Healy University of Colorado at Boulder N. R. Brown and R. S. Siegler (1996) found that training participants on a subset of country populations improved estimations for novel transfer country populations, an effect called seeding that remained intact over time. They attributed this effect to the abstraction by participants of a general metric framework for estimating populations not dependent on specific country anchors. In a series of 3 follow-up experiments, the authors found that training on seed populations produces both general metric information and durable specific country information. Moreover, minimal amounts of general (mean or range of populations) or specific (1 or 3 countries) information made available for inspection while estimating produced a significant seeding effect. Retention over long intervals was facilitated by both presenting 3 seed countries as opposed to 1 and providing names for the seed countries. Making quantitative estimations is an important part of our daily lives, from figuring out how long it will take to get to work in the morning to deciding what time to put dinner in the oven. Although we are fairly accurate when it comes to estimating some routine quantities like these, we seem to have a poor ability to estimate or to remember other specific numeric information, such as country populations or the distances between cities (Brown & Siegler, 1993; Paulos, 1988). This poor performance raises the question of what type of representations can support accurate quantitative estimations, and what kinds of training may lead to the develop- ment of such representations. Forming representations of quantities is likely to be affected by many factors, including existing domain knowledge, the amount of new information available, and whether estimation heuristics are applied. Brown and Siegler (1993) proposed a framework with two major processes as potential explanations of the representation and use of country population data: mapping and abstract metrics. Mapping entails knowledge of which countries are large and small so as to map a new, novel, or unknown country to an appropriate population estimate. Metrics, on the other hand, refers to the statistical properties of the populations, such as their distribution, central tendency, and variability. Participants in Brown and Siegler’s (1993) studies seemed to rely both on heuristics based on familiarity and on domain-specific knowledge (e.g., the fact that industrialized countries may be more populated than undeveloped countries) as tools to map countries to populations. Estimates often reflected familiarity with the coun- tries’ names so that population estimates of countries that had been in the news recently, such as the Soviet Union or Iraq, were higher, although not more accurate, than population estimates for more obscure countries such as Burkina Faso. The participants’ rating of familiarity of each country was the best predictor of the size of their estimate. The use of familiarity by participants is reminiscent of Tversky and Kahneman’s (1973) availability heuristic for de- cision making and suggests that participants were using a sense of how easily a country came to mind as an index of how populated that country was. The biggest estimation errors were caused by assuming that many of the small, but familiar, European countries, such as Switzerland, had large populations. Provided with correct examples of countries and their populations, participants were able to improve their ability to map the countries appropriately, which was demonstrated by improved rank-order correlations between estimated and actual country populations. Participants in this study clearly relied on metric information, in addition to mapping, in some situations. In fact, providing participants with distributional parameters such as the mean population and the range of popula- tions decreased the familiarity bias. Brown and Siegler (1993, 1996) demonstrated that estimates of countries’ populations in general can be dramatically improved by training on a subset of populations. They called this effect “seeding the knowledge base.” The process begins by presenting partici- pants with the correct information for some subset of countries and then asking them to estimate the populations of other countries for which no prior information has been provided. Improvement in performance on the untrained countries indicates the extent to which participants are able to transfer their knowledge of the trained subset of populations to the remaining populations. Using this procedure, Brown and Siegler were able to demonstrate im- proved estimations for a number of different quantities such as the distance between cities (Brown & Siegler, 2001) and latitudes and longitudes for geographic locations (Friedman & Brown, 2000) in addition to country populations. The seeding procedure is extremely effective in improving individuals’ estimates in quantitative domains in which the details are relatively unfamiliar before seeding. How do these improve- ments take place? As noted by Brown and Siegler (1996), partic- ipants may use a strategy of attempting to store each country’s population and then using stored individual facts as anchors for future comparisons when estimating new country populations. This proposal is similar to the context theory of category learning Nadezhda N. LaVoie, Lyle E. Bourne Jr., and Alice F. Healy, Depart- ment of Psychology, University of Colorado at Boulder. This research was supported in part by Army Research Institute Con- tracts DASW01-96-K-0010 and DASW01-99-K-0002 awarded to the Uni- versity of Colorado. We are indebted to Norman Brown and two anony- mous reviewers for their thoughtful comments on an earlier version of this article. Correspondence concerning this article should be addressed to Na- dezhda N. LaVoie, Department of Psychology, University of Colorado at Boulder, Muenzinger Psychology Building, Campus Box 345, Boulder, Colorado 80309-0345. E-mail: nole@psych.colorado.edu Journal of Experimental Psychology: Copyright 2002 by the American Psychological Association, Inc. Learning, Memory, and Cognition 2002, Vol. 28, No. 6, 1137–1153 0278-7393/02/$5.00 DOI: 10.1037//0278-7393.28.6.1137 1137