The Variability Puzzle in Human Memorymemory.psych.upenn.edu/files/pubs/KahaAgga17.poster.pdfThe...
1
The Variability Puzzle in Human Memory While much is known about between-subject variability in performance on cognitive tasks, relatively little is known about within-subject variability. Researchers typically attribute such within-subject variability to well-known factors such as item or list difficulty, effects of practice, or fatigue. To characterize variability in delayed free recall across both subjects and lists, we developed a statistical model to remove influences of known variables. Introduction Methods Penn Electrophysiology of Encoding and Retrieval Study (PEERS): Each of 50 subjects contributed data on 576 lists of 24 common nouns across 24 sessions. Statistical models: • Across Session variability: • Within Session variability: Multiple Regressions Contact: [email protected] – (610) 750-4961 Strength of Variables Michael J. Kahana, Eash V. Aggarwal Across Session Variability Fig. 1: Observed Variability in Memory Performance. For each subject, dots represent range of performance from average lowest lists to average highest lists across all sessions. Histogram represents ranges of performance. Within Session Variability Fig. 1: Removed Variability from Predictive Model. Figure represents the reduction in range of memory performance as accounted for by the regression models, where for each subject, dots represent the subject’s average performance added to the model’s predictive error for that list, and averaged from lowest list to highest list. Within Session Model Residuals P(rec ) = b 0 + b 1 ( session ) + b 2 ( sleep) + b 3 ( alertness ) + b 4 (time) + b 5 ( day ) P (rec ) = b 0 + b 1 ( session ) + b 2 ( block ) + b 3 (list ) + b 4 ( frequency ) + b 5 (emotionality ) + b 6 (concreteness ) Multiple Regressions Strength of Variables Fig. 1: Observed Variability in Memory Performance. For each subject, dots represent range of performance from lowest session to highest session. Histogram represents ranges of performance. Fig. 1: Removed Variability from Predictive Model. Figure represents the reduction in range of memory performance as accounted for by the regression models, where for each subject, dots represent the subject’s average performance added to the model’s predictive error for that session. Across Session Model Residuals Note: Shaded bars and dots represent significant models. (p < 0.05) Note: Shaded bars and dots represent significant models. (p < 0.05) Rabbit Steak Ball 3+5+7 3+5+7 3+5+7 Steak Country Ball Encoding Distractor Free Recall … …
Transcript of The Variability Puzzle in Human Memorymemory.psych.upenn.edu/files/pubs/KahaAgga17.poster.pdfThe...
The Variability Puzzle in Human Memory
While much is known about between-subject variability in performance on cognitive tasks, relatively little is known about within-subject variability.
Researchers typically attribute such within-subject variability to well-known factors such as item or list difficulty, effects of practice, or fatigue.
To characterize variability in delayed free recall across both subjects and lists, we developed a statistical model to remove influences of known variables.
Introduction
Methods
Penn Electrophysiology of Encoding and Retrieval Study (PEERS):
Each of 50 subjects contributed data on 576 lists of 24 common
nouns across 24 sessions.
Statistical models:
• Across Session variability:
• Within Session variability:
Multiple Regressions
Contact: [email protected] – (610) 750-4961
Strength of Variables
Michael J. Kahana, Eash V. Aggarwal
Across Session Variability
Fig. 1: Observed Variability in Memory Performance. For each subject, dots represent range of performance from average lowest lists to average highest lists across all sessions. Histogram represents ranges of performance.
Within Session Variability
Fig. 1: Removed Variability from Predictive Model. Figure represents the reduction in range of memory performance as accounted for by the regression models, where for each subject, dots represent the subject’s average performance added to the model’s predictive error for that list, and averaged from lowest list to highest list.
Within Session Model Residuals
P(rec) = b0 + b1(session)+ b2(sleep)
+ b3(alertness)+ b4(time)+ b5(day)
P(rec) = b0 + b1(session)+ b2(block)+ b3(list)
+ b4( frequency)+ b5(emotionality)
+ b6(concreteness)
Multiple Regressions Strength of Variables
Fig. 1: Observed Variability in Memory Performance. For each subject, dots represent range of performance from lowest session to highest session. Histogram represents ranges of performance.
Fig. 1: Removed Variability from Predictive Model. Figure represents the reduction in range of memory performance as accounted for by the regression models, where for each subject, dots represent the subject’s average performance added to the model’s predictive error for that session.
Across Session Model Residuals
Note: Shaded bars and dots represent significant models. (p < 0.05)
Note: Shaded bars and dots represent significant models. (p < 0.05)
Rabbit
Steak
Ball
3+5+7
3+5+7
3+5+7
Steak
Country
Ball
Encoding Distractor Free Recall… …
