The effects of print exposure on sentence processing and memory in older adults: Evidence for efficiency and reserve

ABSTRACT

The present study was an examination of how exposure to print affects sentence processing and memory in older readers. A sample of older adults (N = 139; Mean age = 72) completed a battery of cognitive and linguistic tests and read a series of sentences for recall. Word-by-word reading times were recorded and generalized linear mixed effects models were used to estimate components representing attentional allocation to word-level and textbase-level processes. Older adults with higher levels of print exposure showed greater efficiency in word-level processing and in the immediate instantiation of new concepts, but allocated more time to semantic integration at clause boundaries. While lower levels of working memory were associated with smaller wrap-up effects, individuals with higher levels of print exposure showed a reduced effect of working memory on sentence wrap-up. Importantly, print exposure was not only positively associated with sentence memory, but was also found to buffer the effects of working memory on sentence recall. These findings suggest that the increased efficiency of component reading processes that come with life-long habits of literacy buffer the effects of working memory decline on comprehension and contribute to maintaining skilled reading among older adults.

Keywords:

• Cognitive aging
• Reading
• Print exposure
• Compensation
• Sentence processing
• Text memory

Acknowledgments

We are grateful for support from the National Institute on Aging (Grants R01 AG029475 and R01 AG013935). We also wish to thank Pat Hill and Joshua Jackson for comments on an earlier draft of this article. Portions of this article were presented at the 63rd annual meeting of the Gerontological Society of America.

Notes

¹The use of mixed-effects modeling has been prevalent in social science, education, and behavioral research for some time (Singer, 1998; Snijders & Bosker, 1999). Recently, these modeling techniques have begun to gain ground in psycholinguistic and cognitive psychology research as well (Baayen, Davidson & Bates, 2008; Jaeger, 2008; Locker, Hoffman, & Boviard, 2007; Quene & van den Bergh, 2004, 2008). In conventional psycholinguistic experiments, the use of GLMM incurs several benefits, such as allowing the researcher to analyze fixed and random effects across subjects and items simultaneously, which avoids the need for separate F1 (by-subjects) and F2 (by-items) analyses. Additionally, GLMM are capable of (1) modeling predictors of subject and item level variability simultaneously, (2) modeling both discrete and continuous variables simultaneously, (3) modeling unbalanced designs, and (4) explicitly modeling variances and covariances, allowing for violations of sphericity and homogeneity of error variance (Snijders & Bosker, 1999).

²It is important to note that, while these methods for decomposing continuous interactions were originally designed in the context of ordinary least squares regression, these methods are also appropriate and equivalent in the context of GLMM, since we are only probing a parameter from the fixed portion of the model.

³Although our analyses focused on verbal working memory as the major indicator of individual differences in processing capacity, similar effects were found for the reading time models when speed of processing was used in place of vWM as a proxy of fluid cognition. That is, re-fitting the reading time models while replacing speed with vWM revealed that processing speed was a significant predictor of reading time (p < .001) and interacted with sentence and clause wrap-up (SB × Speed, p < .001; IntSB × Speed, p < .001). Importantly, the effects of processing speed on sentence wrap-up were buffered by print exposure (ART × SB × Speed, p < .001), much in the same way that the effects of vWM were buffered by greater print exposure. Thus, it appears that individual differences in processing speed acted in a similar fashion to vWM as an indicator of cognitive capacity among older adults.

⁴Unlike the reading time models (see footnote 3), using processing speed in place of vWM revealed that speed did not uniquely predict sentence recall (p = .20) nor was its effect on recall moderated by print exposure (p = .21). The robust effects of processing speed on the reading time models may reflect the fact that psychomotor speed is a stronger predictor of reaction time variables than those based on memory processes. Given that speed did not predict recall, while vWM did (and was influenced by print exposure), this suggests that working memory has broader predictive power for measures of language comprehension, perhaps because complex span measures tap multiple abilities (e.g., executive attention, processing capacity, and resistance to proactive interference; Lustig, May, & Hasher, 2001; Whitney, Arnett, Driver, & Budd, 2001; Engle, 2002) that underlie language comprehension.

⁵Verbal working memory, like most measures of fluid cognition, shows monotonic declines across the entire lifespan. However, within this sample, the correlation between age and vWM was only marginally significant. This is due largely to the restricted age range in the current study (64–92). Nevertheless, there was a great deal of variance in vWM among the older adults, suggesting that our older sample showed substantial individual differences in cognitive ability.