Examining the Agreement of Alternative Estimates of Category Exemplar Typicality
By Taylor Curley1, Nichol Castro1,2, Christopher Hertzog1
1School of Psychology, Georgia Institute of Technology
2Department of Communicative Disorders and Sciences, University of Buffalo
Jump to:
ABSTRACT:
Semantic categories (e.g., ANIMAL) are composed of exemplars (e.g., dog). The typicality of a given exemplar indicates how prototypical or representative that item is of its superordinate category (e.g., dog is a more typical animal than whale) which can influence item processing and subsequent memory. Typicality estimates in published category norms (e.g, Battig & Montague, 1969; Castro, Curley, & Hertzog, in press) are the mean frequency of a word being generated when cued by the category name. It may underestimate typicality at the low end of the scale. Thus, we assessed alternative methods of obtaining exemplar typicality with a cross-sectional (18-96) sample of 447 adults who completed two typicality estimation tasks: 1) rank ordering category exemplars from most to least typical, and 2) rating provided exemplar typicality using a 1-10 (least to most) Likert scale. Typicality patterns were relatively consistent between free-response and ranking methods with two exceptions: abstract categories (e.g., “things that are green”) and low-typicality exemplars. We interpret these new scalings of typicality with respect to existing category norms and argue they are better suited for item-selection in memory and language studies.
Loading...
Reload document 
Open in new tab Cross-Method Comparisons (Rank-Order Correlations)
Free Response vs. Rank-Ordered Typicality Estimates:
Table Key:
Category: The normative category label.Num_Exemplars: Free-response and rank-based methods use different numbers of exemplars. We can only compare the estimates for the exemplars that both methods employ.Max_Exemplars: The number of exemplars in the method with the larger number of recorded exemplars (i.e. free-response).Num_Discrepancies: How many exemplars are excluded from the correlational comparison?All_Corr_All: Across all participants, calculate the correlation between the average ranking and average typicality. (Spearman’s rank-order correlation.)- NOTE: These correlations are multiplied by -1 in order to reflect the differences in rank (increases in typicality: high –> low) and free-response (increases in typicality: low –> high) scoring.
All_Corr_First: Across all participants, calculate the correlation between the percentage of times an item was ranked or produced first.YA_Corr_All: Across young adults, calculate the correlation between the average ranking and average typicality. (Spearman’s rank-order correlation.)- NOTE: These correlations are multiplied by -1 in order to reflect the differences in rank (increases in typicality: high –> low) and free-response (increases in typicality: low –> high) scoring.
YA_Corr_First: Across young adults, calculate the correlation between the percentage of times an item was ranked or produced first.MA_Corr_All: Across middle-aged adults, calculate the correlation between the average ranking and average typicality. (Spearman’s rank-order correlation.)- NOTE: These correlations are multiplied by -1 in order to reflect the differences in rank (increases in typicality: high –> low) and free-response (increases in typicality: low –> high) scoring.
MA_Corr_First: Across middle-aged adults, calculate the correlation between the percentage of times an item was ranked or produced first.OA_Corr_All: Across older adults, calculate the correlation between the average ranking and average typicality. (Spearman’s rank-order correlation.)- NOTE: These correlations are multiplied by -1 in order to reflect the differences in rank (increases in typicality: high –> low) and free-response (increases in typicality: low –> high) scoring.
OA_Corr_First: Across older adults, calculate the correlation between the percentage of times an item was ranked or produced first.
Free Response vs. Likert Typicality Estimates:
Table Key:
Category: The normative category label.Num_Exemplars: Free-response and rank-based methods use different numbers of exemplars. We can only compare the estimates for the exemplars that both methods employ.Max_Exemplars: The number of exemplars in the method with the larger number of recorded exemplars (i.e. free-response).Num_Discrepancies: How many exemplars are excluded from the correlational comparison?All_Corr_All: Across all participants, calculate the correlation between the average Likert response and average typicality. (Spearman’s rank-order correlation.)All_Corr_First: Across all participants, calculate the correlation between the percentage of times an item was rated as or produced first.YA_Corr_All: Across young adults, calculate the correlation between the average Likert response and average typicality. (Spearman’s rank-order correlation.)YA_Corr_First: Across young adults, calculate the correlation between the percentage of times an item was rated as or produced first.MA_Corr_All: Across middle-aged adults, calculate the correlation between the average Likert response and average typicality. (Spearman’s rank-order correlation.)MA_Corr_First: Across middle-aged adults, calculate the correlation between the percentage of times an item was rated as or produced first.OA_Corr_All: Across older adults, calculate the correlation between the average Likert response and average typicality. (Spearman’s rank-order correlation.)OA_Corr_First: Across older adults, calculate the correlation between the percentage of times an item was rated as or produced first.
Items highlighted in yellow have correlations between 0.5 and 0.7. Items highlighted in red have correlations below 0.5. Un-highlighted items have correlations greater than 0.7.
Rank-Ordered vs. Likert Typicality Estimates:
Table Key:
Category: The normative category label.Num_Exemplars: The number of exemplars used in the correlation estimate.Max_Exemplars: It’s unlikely that the rank-order and Likert methods have different numbers of exemplars; if they do, however, this gives the number of exemplars in the method with the larger number of exemplars.Num_Discrepancies: How many exemplars are excluded from the correlational comparison? (Most likely 0.)All_Corr_All: Across all participants, calculate the correlation between the average ranking and average Likert rating. (Spearman’s rank-order correlation.)- NOTE: These correlations are multiplied by -1 in order to reflect the differences in rank (increases in typicality: high –> low) and Likert response (increases in typicality: low –> high) scoring.
All_Corr_First: Across all participants, calculate the correlation between the percentage of times an item was ranked or rated as first.YA_Corr_All: Across young adults, calculate the correlation between the average ranking and average Likert rating. (Spearman’s rank-order correlation.)- NOTE: These correlations are multiplied by -1 in order to reflect the differences in rank (increases in typicality: high –> low) and Likert response (increases in typicality: low –> high) scoring.
YA_Corr_First: Across young adults, calculate the correlation between the percentage of times an item was ranked or rated as first.MA_Corr_All: Across middle-aged adults, calculate the correlation between the average ranking and average Likert rating. (Spearman’s rank-order correlation.)- NOTE: These correlations are multiplied by -1 in order to reflect the differences in rank (increases in typicality: high –> low) and Likert response (increases in typicality: low –> scoring.
MA_Corr_First: Across middle-aged adults, calculate the correlation between the percentage of times an item was ranked or rated as first.OA_Corr_All: Across older adults, calculate the correlation between the average ranking and average Likert rating. (Spearman’s rank-order correlation.)- NOTE: These correlations are multiplied by -1 in order to reflect the differences in rank (increases in typicality: high –> low) and Likert response (increases in typicality: low –> high) scoring.
OA_Corr_First: Across older adults, calculate the correlation between the percentage of times an item was ranked or rated as first.
Items highlighted in yellow have correlations between 0.5 and 0.7. Items highlighted in red have correlations below 0.5. Un-highlighted items have correlations greater than 0.7.
Cross-Method Comparisons (Rank-Order Correlations) – Split Half Analyses
Free Response vs. Rank-Ordered Typicality Estimates:
Free Response vs. Likert Typicality Estimates:
