Judgments of Category Size: Now You Have Them, Now You Don't

Estimating the frequency of events from unnatural categories

“…Many of the presentation frequencies in the present study were substantially higher than those used in the studies by Barsalou and Ross (1986) and Freund and Hasher (1989); presentation frequency ranged from 0 to 4 in the former study and from 0 to 8 in the latter study. The reason for the high presentation frequencies in the present study was to create the conditions for participants to use nonnumerical estimation strategies, which have been observed primarily when frequencies are high (e.g., Brown, 1995Brown, , 1997Conrad et al, 1998).…”

“…In particular, when the true frequency was zero, these participants reported having studied, on average, 2.75 items with the test properties; in contrast, the explicit property and the taxonomic groups reported studying only 1.21 and 0.34 instances of the same properties, respectively. In conjunction with high-end underestimation, the implicit property group's low-end overestimation essentially reproduces the flat frequency function observed by Barsalou and Ross (1986) and Freund and Hasher (1989). This contrasts with moderate, across the board underestimation by the explicit property and taxonomic groups and is reflected by the lower rank order correlation between the actual and the estimated frequency for the implicit property group than for the other two groups.…”

“…Because the implicit property group relied predominantly on simple enumeration, we would expect them to underestimate, relative to normative values (Underwood & Richardson, 1956), across all levels of frequency. However, Barsalou and Ross (1986) and Freund and Hasher (1989) observed pronounced regression effects-low-end overestimation and high-end underestimation-among participants whose task resembled that of our implicit property group. 4 The average size of the estimates for the 10 levels of frequency, relative to actual frequency, did in fact differ between the three groups [interaction of group 3 frequency, F(18,189) = 3.66, MS e = 7.671, p < .001].…”

“…And presumably, underreporting is more prevalent for high-frequency events: Forgetting could be large, since there are more legitimate events to forget and this could swamp intrusions. This pattern of error-low-end overestimation and high-end underestimation-could produce the flat estimation functions observed by Barsalou and Ross (1986) and Freund and Hasher (1989).…”

“…For example, is it possible to form a qualitative impression of frequency, such as that happened a lot, for a category that one is not thinking of when potential instances occur? The estimation of property frequency provides an excellent test bed for this question, because people do not seem to encode events in terms of properties unless the properties are made particularly salient (Barsalou & Ross, 1986, Experiment 3;Freund & Hasher, 1989, Experiment 2).…”