Memory accuracy in old age: Cognitive, metacognitive, and neurocognitive determinants

Abstract

Age differences in memory accuracy were examined within a conceptual framework specifying the mediating role of metacognitive monitoring and control processes (Koriat & Goldsmith, 1996b). Replicating previous results, older adults showed poorer memory quantity and accuracy performance compared to young adults. Even when memory quantity performance was equated, by dividing the young adults’ attention during encoding, the difference in memory accuracy was not eliminated. Examination of the underlying metacognitive processes revealed that the age-related reduction in memory accuracy stemmed partly from less effective memory monitoring, apparently the result of poorer encoding, and also from differences in two aspects of metacognitive control: (1) a more liberal report criterion—greater tendency to volunteer incorrect (and correct) answers, and (2) reduced control sensitivity—less reliance on subjective monitoring as a basis for responding. This latter control reduction was associated with lower neuropsychological measures of executive functioning, suggesting a decline in frontal-lobe efficiency.

Keywords:

• Memory accuracy
• Metamemory
• Ageing
• Executive processes
• Memory control

Acknowledgements

This research was supported by the Charles E. Smith Extended Grant on the Subject of Memory, awarded by the National Institute for Psychobiology in Israel to AK, MG, and AP.

Notes

¹We did not administer these neuropsychological tests to the young adults because, based on previous research (see Rhodes, 2004), we did not expect to find much meaningful variance for this population.

²In light of the findings of two studies that were published after the current study was conducted (Kelley & Sahakyan, 2003; Rhodes & Kelley, 2005), perhaps we were overcautious regarding these considerations.

³Because of the small sample sizes, these path analyses were conducted using least-squares multiple regression rather than structural equation modelling (Kline, 1998). Power analyses based on Green (1991), with power set at .80, indicated adequate sample size to detect partial correlations ≥.43 in Models A and B and ≥.57 in Model C. Thus, the nonsignificant path coefficients in these models should be interpreted with caution, as they may be concealing low-to-moderate sized effects. (Note that the path coefficients are standardised beta coefficients, which are generally lower than the corresponding partial correlations.)