Age effects on explicit and implicit memory

Abstract

It is well-documented that explicit memory (e.g., recognition) declines with age. In contrast, many argue that implicit memory (e.g., priming) is preserved in healthy aging. For example, priming on tasks such as perceptual identification is often not statistically different in groups of young and older adults. Such observations are commonly taken as evidence for distinct explicit and implicit learning/memory systems. In this article we discuss several lines of evidence that challenge this view. We describe how patterns of differential age-related decline may arise from differences in the ways in which the two forms of memory are commonly measured, and review recent research suggesting that under improved measurement methods, implicit memory is not age-invariant. Formal computational models are of considerable utility in revealing the nature of underlying systems. We report the results of applying single and multiple-systems models to data on age effects in implicit and explicit memory. Model comparison clearly favors the single-system view. Implications for the memory systems debate are discussed.

Keywords: aging; implicit memory; models of memory; priming; recognition.

Figure 1

Recognition performance for immediate and delayed items in Ward et al. (2013), Experiment 1. Left panel: discriminability performance (d′) [significant main effect of age group, F_{(1, 38)} = 4.26, p = 0.04, η²_p = 0.10, and delay, F_{(1, 38)} = 5.76, p = 0.02, η²_p = 0.13, and no significant interaction, F_{(1, 38)} = 0.16, p = 0.69]. Right panel: proportion of hit and false-alarm responses. Bars indicate experimental data (error bars indicate SE of the mean), and symbols indicate the mean expected result from each model when fit to each individual's data. I, immediate; D, delayed; SS, single-system model; MS1, multiple-systems-1 model; MS2, multiple-systems-2 model.

Figure 2

Continuous identification (CID) task performance in Ward et al. (2013), Experiment 1. Left panel: priming effects [no significant main effect of age group, F_{(1, 38)} = 1.78, p = 0.19, or delay, F_{(1, 38)} = 0.02, p = 0.89, and no significant interaction, F_{(1, 38)} = 0.04, p = 0.85]. The proportional priming effect was calculated as the difference in the median identification RT to new and old items divided by the median identification RT for new items. Bars indicate the mean of the median priming across participants (error bars indicate SE of the mean), and symbols indicate the mean expected proportional priming (relative to the expected identification RT for new items, see Table 3) from each model when fit to each individual's data. Right panel: mean identification RTs (ms) of immediate, delayed, and new items. Bars indicate experimental data (error bars indicate SE of the mean), and symbols indicate the mean expected result from each model when fit to each individual's data. Imm, immediate; Delay, delayed; SS, single-system model; MS1, multiple-systems-1 model; MS2, multiple-systems-2 model.

Figure 3

Model selection results. Each bar represents the percentage of participants best fit by each model according to the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). SS, single-system model; MS1, multiple-systems-1 model; MS2, multiple-systems-2 model.

Figure 4

Predictions of the SS, MS1, and MS2 models (Berry et al., 2012) in the older group (panel A), younger group (panel B), and collapsed across groups (panel C). “Priming: Overall - judged new” refers to the difference in the overall priming effect and the priming effect for items judged new. Prediction 1 concerns whether fluency effects occur within new and old items (i.e., CR − FA and M − H). Prediction 2 concerns whether the magnitude of the priming effect overall (across all items) is greater than the priming effect for items judged new. Bars indicate 95% confidence intervals. I, immediate; D, delayed; CR, correct rejection, FA, false alarm; M, miss; H, hit; SS, single-system model; MS1, multiple-systems-1 model; MS2, multiple-systems-2 model.