Text Comprehension Products and Processes in Young, Young-Old, and Old-Old Adults

Abstract

Aging may be associated with an increase in generalized text processing, particularly in adults older than 75 years. The current study examined text comprehension in young, young-old, and old-old adults. Experiment 1 included a comprehension measure (product) and Experiment 2 examined inferences generated during reading (process). Comprehension scores were lowest in old-old adults. Generalized and elaborative inference scores were highest in old-old adults. Participants over 65 years with the lowest scores on cognitive integrity variables also had significantly lower comprehension scores, but there was no effect of cognitive integrity on inference scores. This dissociation suggests that inferential processes may be maintained even when cognitive integrity and comprehension declines are present. Relevance to cognitive aging theories addressing text processing and self-regulatory processes is addressed.

EVEN when their texts of choice are newspapers, magazines, and mail, most older adults read daily, taking pride in their ability to comprehend well (Meyer & Pollard, 2006; Smith, 2000). Until recently, the bulk of the language and aging literature suggested that text comprehension is maintained in community-dwelling older adults (Burke, MacKay, & James, 2000; Kemper, 2006; Meyer & Pollard, 2006; Wingfield & Stine-Morrow, 2000). As is typical of cross-sectional studies in this literature, adults with a mean age around 70 years (young-old adults) were compared with young adults. Augmenting this literature are studies that included a sample of adults with a mean age around 80 years (McGinnis & Zelinski, 2000, 2003). In these studies, we reported sizable differences between young-old and old-old on measures of comprehension; results that are consistent with large-scale studies that consistently demarcate 74 years as an age when verbal processes start to decline, with declines observed across methodologies (e.g., cross-sectional, longitudinal) and across instruments (Lindenberger & Baltes, 1997; Schaie, 1996). Given these differences between young-old and old-old adults, accurately characterizing the cognitive decline in late life may depend on including a sample with a mean age around 80 years.

Stine-Morrow, Miller, and Hertzog (2006) proposed that older adults may be more likely than young adults to lower their threshold for the acceptability of an interpretation to “gist” levels. The mechanisms proposed in their Self-Regulated Language-Processing Model (SRLP) characterize progressive age-related declines in the specificity of cognitive processes and products as influenced by age-related changes in processing-resource integrity, resource self-regulation, and socioemotional processes. The shift to generalized processing may accelerate in old-old adults as they experience and compensate for accelerating declines in processing-resource integrity.

In our previous studies, a measure of unfamiliar word interpretation was used to examine the product of text comprehension (McGinnis & Zelinski, 2000, 2003). Participants were presented with paragraph-length stories that included one unfamiliar word and content supporting interpretation. In these studies, old-old adults were more likely than young and young-old to focus on overgeneralized interpretations (passage summary). Hence, the unfamiliar word methodology enabled the identification of a tendency to generate overgeneralized or global interpretations for text in old-old adults.

Interpretations that are generalized or inaccurate may be more likely when readers generate inferences that are semantically inappropriate in some manner. Think-aloud protocols for adults over 75 years included greater quantities of generalized inferences, suggesting that declines in the specificity of inferential processes could be one aspect of age-related changes in text processing (McGinnis & Zelinski, 2003). The results of our comprehension and inferential processing studies are consistent with the age-related specificity declines described in the SRLP model.

To examine the relationship among inference and interpretation variables, interpretation and inference tasks were commingled in our earlier study (McGinnis & Zelinski, 2003). Because comprehending words central to a narrative's meaning may garner more attention, commingling comprehension and think-aloud tasks could result in the allocation of fewer resources to the think-aloud task. It is also possible that comprehension may take precedence over other text processing tasks. Hence, commingling thinking-aloud and interpretation tasks could have had a deleterious impact on interpretation and/or inference variables, possibly affecting old-old adults disproportionally.

Specificity problems observed in community-dwelling old-old adults may be associated with declines in cognitive integrity. Even though disease-related impairments are unlikely to be present in most community-dwelling adults, these impairments do affect some, with the proportion increasing with age (Sliwinski, Lipton, Buschke, & Stewart, 1996). As researchers sample more adults older than 75 years, participants experiencing the earliest manifestations of neurodegenerative disorders may be inadvertently included (Albert, Moss, Tanzi, & Jones, 2001; Sliwinski et al., 1996). In addition, specificity declines such as those observed in our comprehension research qualitatively resemble semantic impairments observed in prodromal and early phases of Alzheimer's disease (Aronoff et al., 2006; Chapman et al., 2002; Kempler, 1995; Linn et al., 1995; Noble, Glosser, & Grossman, 2000). Hence, the ability to process text at an appropriate level of specificity may be influenced by cognitive impairments due to subclinical diseases in some participants.

THE PRESENT STUDY

Similar to our earlier studies, participants were asked to interpret words presented in supportive contexts in Experiment 1. By excluding thinking aloud, Experiment 1 was designed to emphasize the product of comprehension. Based on previous research, it was hypothesized that old-old adults would favor less accurate interpretations. Experiment 2 was designed to examine an aspect of the process of comprehension, inferential specificity. To focus more precisely on process, participants were asked to think aloud while reading passages that did not include an unfamiliar word. It was hypothesized that the think-aloud protocols for older adults would include more overgeneralized inferences.

Both experiments included measures considered sensitive to cognitive integrity (Jacobs et al., 1995; Masur et al., 1989; O’Connell & Tuokko, 2002; Sliwinski, Lipton, Buschke, & Wasylyshyn, 2003). If impairment processes associated with diseases are related to overgeneralizations, then scores on cognitive integrity variables should be related to interpretation and inference scores. It was hypothesized that participants who score highest on cognitive integrity measures will have higher interpretation scores compared with participants scoring lower on cognitive integrity measures (Experiment 1). Similarly, participants with high scores on cognitive integrity measures should generate fewer generalized inferences compared with participants with low scores (Experiment 2). Cluster analysis was used to group participants into cognitive integrity subgroups enabling these comparisons. Because categorizing community-dwelling participants as impaired with any precision is impossible, the correlational associations examined herein are exploratory.

To summarize, the primary goal of the current study was to examine the increase in generalized text processing in older adults by examining interpretation and inference processes in separate experiments. A secondary goal was to examine the relationship among comprehension and cognitive integrity variables.

Experiment 1

Methods

Subjects.—

A total of 137 adults participated: 43 young adults (12 males, 31 females); 48 young-old (19 males, 29 females); and 46 old-old adults (12 males, 34 females). The means for age were young: 21.84 (5.37); young-old: 67.73 (3.86); and old-old: 80.96, (4.78) and for years of education: 13.07 (1.24); 14.27 (2.47); and 13.76 (2.12), respectively. Education differences were significant, F(2, 136) = 3.97, p = .021; however, only the mean difference for young and young-old was significant at p ≤ .05. Older adult subjects were recruited through flyers and newspaper announcements. Recruitment protocol included telephone screening for serious medical conditions, dementia, low education (<11th grade), and vision and hearing problems. After participating, subjects were mailed $10.00 gift cards. The young adult sample consisted of Oakland University undergraduates who participated for course credit. Two subjects who scored below 10 on the Wechsler Adult Intelligence Scale-Revised (WAIS-R) vocabulary subtest were excluded from this sample.

Measures.—

Background questionnaire.

Along with demographic questions, two questions about reading frequency were included in the background questionnaire: (1) “How many times a week do you sit down to read for 30 minutes or more?” and (2) “How many days of the week do you sit down to read at least one newspaper or one magazine article?” Responses ranged from 0 to 20 for (1) and from 0 to 7 for (2).

WAIS-R Vocabulary.

Subjects were asked to provide written definitions for the last 20 words of the 40-item vocabulary subtest of the WAIS-R (Wechsler, 1981). Written responses were scored according to WAIS guidelines. Scores ranged from 1 to 40.

Selective Reminding Test.

Subjects were presented with 12 words at 2-s intervals on Trial 1 (Buschke, 1973). On Trials 2–6, the experimenter stated only words not recalled on the immediately preceding trial. Subjects were asked to restate all 12 words on each trial. To assess delayed recall, subjects were asked to restate all 12 words after a 5-min delay. Selective reminding (SR) scores correspond to the average across all six trials (SR-average) and to delayed recall (SR-delayed).

Category Fluency.

Subjects were given 60 s to state exemplars of the following categories: colors, fruits, animals, and towns. CF scores consist of the average across the four category trials.

Mini-Mental State Exam.

The Mini-Mental State Exam (MMSE) (Folstein, Folstein, & McHugh, 1975) was administered, yielding scores reflecting overall cognitive integrity.

Unfamiliar Word Definition Task.

Subjects read eight narratives. Each narrative consisted of approximately 150 words. Passages were modifications of those developed by Daneman and Green (1986) used in an earlier study (McGinnis & Zelinski, 2003). As in the earlier study, these narratives were presented in Times New Roman 14. Each narrative described a problem that was resolved as the narrative progressed, and each contained an unfamiliar word, as well as sufficient contextual information to determine the unfamiliar word's meaning. Each subject read these passages silently. After reading each passage, subjects were presented with four definitions. The passage and the definition options were presented on a single sheet of paper with the narrative at the top, followed by the definitions. During all phases of this task, subjects had complete access to the passage. The four definition options corresponded to: (1) precise: the precise definition; (2) less precise: a generalization of the precise definition; (3) summary: a summary of the narrative; and (4) irrelevant: passage information irrelevant to the word's meaning. Subjects were asked to use a 7-point scale (0–6, poor to excellent) to rate each definition, yielding four scores. As expected, most subjects rated precise and less precise options higher than summary and irrelevant. The eight passages were presented in four orders, but no effect of passage order on scores was observed statistically. A sample passage and the corresponding definitions are provided in the Appendix.

For each word, precise and less precise ratings were added to obtain a good definition sum. Summary and irrelevant ratings were added to obtain a poor definition sum. The mathematical difference between the good and the poor sums were averaged for each subject to compute a definition discrimination score for each participant. Cronbach's alpha reliabilities were .84 for definition discrimination, .59 for precise, .64 for less precise, .88 for summary, and .82 for irrelevant. These statistics resemble those reported in McGinnis and Zelinski (2003).

Procedure.—

After completing the background questionnaire, subjects were presented with the following tasks: SR, Category Fluency, and SR-delayed recall. Next, subjects were given the unfamiliar word definition task, followed by vocabulary and MMSE. Subjects completed these tasks at their own pace, with all completing the protocol in less than 90 min.

Results and Discussion

Univariate analysis of variance (ANOVA) was used to examine age differences. Because of its robustness when variances are unequal, Dunnett's T3 procedure was used to examine the significance of pairwise mean differences. Group means and ANOVA statistics are reported in Table 1. Vocabulary differed across age groups, but post hoc analyses revealed that none of the pairwise differences was significant. There was a main effect of age group on all the measures reflecting cognitive integrity (MMSE, SR-average, SR-delayed, and Category Fluency), with old-old having the lowest means across all four variables. There was also a main effect of reading frequency and daily reading, with old-old adults reporting reading the most frequently. For the comprehension variables, old-old adults did not discriminate among definition choices as well as adults younger than 75 years. The largest effect size was obtained for summary ratings, with old-old adults rating summary definitions higher on average compared with the other age groups. Post hoc analyses revealed that the means for old-old adults differed from the means for young adults in every case. For summary and irrelevant ratings, the pairwise differences were all significant. The results of Experiment 1 suggest that interpretation scores when participants are not thinking aloud are similar to those observed when participants are thinking aloud. Participants may view unfamiliar word comprehension as critical for narrative comprehension, necessitating attention to that task even when it is commingled with a think-aloud task.

Table 1.

Background and Interpretation Means by Age Group (Experiment 1)

	Young		Young-Old		Old-Old
	N = 43		N = 48		N = 46		ANOVA
	M	SD	M	SD	M	SD	F(2, 134)	η2
Background
Vocabulary	19.58	6.62	23.19	7.70	20.54	7.20	3.10*	.046
MMSE	29.30ab	0.86	28.23a	1.42	27.59b	1.48	19.65***	.227
SR-average	8.81a	1.00	7.34a	1.79	6.24a	1.53	32.92***	.329
SR-delayed	8.95ab	2.25	6.06a	2.43	5.20b	2.35	30.96***	.316
CF	18.73a	3.37	17.15b	4.14	14.65ab	3.05	14.53***	.178
Rdg Frq	5.87ab	3.28	8.40a	4.66	9.83b	4.70	9.63***	.126
Daily Frq	3.02ab	1.88	6.19a	1.65	6.07b	1.97	42.28***	.387
Interpretation
Precise	4.86	0.73	4.93	0.83	5.14	0.88	1.41	.021
Less Precise	3.86	0.88	4.02	1.17	4.34	1.10	2.36	.035
Summary	0.53a	0.65	1.17a	1.35	2.18a	1.71	17.62***	.208
Irrelevant	0.27a	0.36	0.55b	0.88	1.24ab	1.36	11.89***	.151
Discrimination	3.95a	0.79	3.62	1.38	3.03a	1.51	6.17*	.084

Notes: MMSE = Mini-Mental State Exam. SR = selective reminding; CF = Category Fluency; Rdg Freq = how many times a week read for 30 min or more; Daily Frq = how many days a week a newspaper or magazine is read. Matching subscripts across rows indicate significant mean differences.

*p ≤ .05; **p < .01; ***p < .001.

EXPERIMENT 2

Experiment 2 enabled the examination of age differences in the specificity of inferences generated without a concurrent comprehension task. An inference consists of information used to facilitate passage comprehension, information that is implied but not explicitly stated. Generating an inference during text comprehension necessitates the integration of world knowledge with explicitly stated information. Skilled comprehension depends on a range of inferential processes (Graesser, Millis, & Zwaan, 1997; Graesser, Swamer, Baggett & Sell, 1996; Kintsch, 1974).

Inferences can be categorized as necessary or elaborative. Necessary inferences are critical for developing an accurate mental representation of passage information and events. A story about getting fired (see Appendix) may elicit an inference pertaining to the reason the protagonist was fired (e.g., “whipping children”), a necessary inference. Inferences that embellish an interpretation without being critical are elaborative. In that same story, readers could generate inferences about the consequences of getting fired (e.g., having to look for another job), and these inferences are elaborative.

In our previous study, the think-aloud protocols of adults aged 75 years and older were more likely to include overgeneralized inferences (e.g., “problem” instead of whipping children). This increase in generalized inferences may reflect a tendency to utilize or emphasize gist-level or global information when processing and interpreting text (Adams, 1991; Adams, Smith, Nyquist, & Perlmutter, 1997; Dixon & Backman, 1993; Stine-Morrow et al., 2006). Similarly, the age-related increase in generalized interpretations may reflect a late-life shift to elaborations if older adults are generating story interpretations that tend to be more personally relevant than young adults (Blanchard-Fields, Horhota, & Mienaltowski, 2008; Carstensen, 1995; Carstensen & Mikels, 2005).

Because methodologically commingling tasks may result in less attention to thinking aloud, possibly diminishing or altering inference generation, Experiment 2 was conducted to examine inferential processes in the absence of a concurrent comprehension task. The passages for Experiment 2 were the same as those in Experiment 1, but each unfamiliar word was replaced with its definition. It was hypothesized that the think-aloud protocols for old-old adults would include more overgeneralized and more elaborative inferences compared with the protocols for young and young-old adults. A secondary goal was to examine the relationship among inference and cognitive integrity variables. To accomplish this, statistical analyses enabling cognitive integrity subgroup identification as well as analyses addressing subgroup differences on interpretation and inference variables were conducted.

Methods

Subjects.—

A total of 95 adults participated: 33 young adults (10 males, 23 females); 31 young-old (10 males, 21 females); and 31 old-old adults (6 males, 25 females). The means for age were young: 20.79 (2.83); young-old: 67.68 (4.41); and old-old: 81.19 (5.64), and for years of education: 13.24 (1.39); 13.68 (2.48); and 12.94 (1.90), respectively. The main effect of age group on years of education was not significant. The recruitment protocol was identical to that used in Experiment 1, and there was no overlap in subjects across the two experiments. Two participants were excluded from the sample after vocabulary screening.

Measures.—

Participants were asked to complete the background questionnaire and all the following: WAIS-R Vocabulary, SR, Category Fluency, and MMSE. Refer to Experiment 1 for descriptions.

Think-aloud passages and inference scores.

Subjects read the same eight narratives used in Experiment 1 with the unfamiliar words replaced with a phrase corresponding to the precise meaning for each word. As before, these narratives were presented in Times New Roman 14. Half of the passages were preceded by instructions to “read for entertainment”; and the other half by instructions to “read for a test.” The instructional manipulation yielded nonsignificant main effects and interactions. Because of this, the data reported herein are collapsed across this variable. Methodological differences (e.g., age of samples, narrative versus expository) may explain the differences between the current study and an earlier study (van den Broek, Lorch, Linderholm, & Gustafson, 2001).

Word counts.

The number of words in each think-aloud protocol was obtained using the word count algorithm in Microsoft Word. Word count scores reflect the average word count across all eight passages.

Procedure.—

After completion of the background questionnaire, subjects were presented with the think-aloud task (the first four passages), SR, Category Fluency, and SR-delayed. Subjects continued with the second half of the think-aloud passages, WAIS-R Vocabulary, and MMSE. Most subjects finished in less than 90 min.

Think-aloud instructions encouraged participants to report every thought (Ericsson & Simon, 1980; Suh & Trabasso, 1993). Subjects paused to think aloud at five designated breaks, delineated by heavy dashed lines. Previously read portions of the passage remained uncovered, minimizing memory demands. Think-aloud sessions were tape-recorded and transcribed verbatim.

Content analysis of the think-aloud protocols.

During all phases of content analysis, coders viewed only the think-aloud protocol, ensuring that coders were blind to all other variables. During the first phase, each inference was circled by two coders independently and all discrepancies resolved through discussion. During the second phase, inference master sheets with the circled inferences were coded by two coders independently. Each inference was coded on two dimensions: necessary or elaborative and specific or generalized. A list of the most common necessary inferences was used, facilitating consistency and interrater reliability. If an inference was not necessary for basic comprehension but embellished the story in some manner, it was coded as elaborative. Similarly, a strict criterion was used for generalized or specific coding. An inference was coded as generalized if it reflected a broad category without any mention of a subcategory. For example, inferences such as “problem” or “things,” which could have been more specific given the story content, were coded as generalized. Inferences that were more specific were coded as specific (e.g., whipping children). If a broad inference included adjectival modification, it was coded as specific. Coding training sessions transpired until 10 protocols were reliable 90% of the time. After training, each protocol was distributed to two of the six coders, counterbalanced so that all possible coding pairs were included. A third coder examined and resolved discrepancies. Prior to discrepancy analysis, the average interrater reliability was 88%.

Content analysis resulted in four inference scores: specific necessary, generalized necessary, specific elaborative, and generalized elaborative. Cronbach's alpha reliabilities were as follows: specific necessary, .86; specific elaborative, .91; generalized elaborative, .77; and generalized necessary, .68. The reliability for all specific inferences was .85 and for all generalized .77. The reliability for all elaborative inferences was .89 and for all necessary inferences .84. The lower reliability for generalized necessary inferences is due to the low scores observed for this variable (often 0), whereas the other inferences were produced more reliably across participants. Before proceeding with other data analyses, the effect of passage order on inference counts was examined, with no effect of passage order observed.

Results and Discussion

Age differences in inference variables.—

A 3 (age group: young, young-old, old-old) by 2 (inference type: necessary, elaborative) by 2 (inference specificity: specific, generalized) repeated-measures ANOVA, with inference scores as the dependent variable, was used to examine the main effects and interactions. Table 2 provides the means, standard deviations, and ANOVA results for cognitive integrity, inference, and word count variables. Overall, think-aloud protocols included more elaborative than necessary inferences, F(1, 92) = 299.62, p < .001, η² = 0.765, and more specific than generalized inferences, F(1, 92) = 402.96, p < .001, η² = .814. The type by specificity interaction was also significant, F(1, 92) = 144.24, p < .001, η² = .611, reflecting large specific elaborative scores.

Table 2.

Background and Inference Means by Age Group (Experiment 2)

	Young		Young-Old		Old-Old
	N = 33		N = 31		N = 31		ANOVA
	M	SD	M	SD	M	SD	F(2, 92)	η2
Background
Vocabulary	17.61	4.85	20.77	7.75	20.52	7.89	2.08	.043
MMSE	29.24ab	0.75	28.00a	1.50	27.79b	1.62	10.96***	.192
SR-average	9.03a	1.31	7.61a	1.46	5.69ab	2.23	30.77***	.401
SR-delayed	8.82ab	2.47	6.45a	2.60	4.58b	3.42	17.73***	.278
Category Fluency	18.48a	3.89	17.86b	3.58	14.88ab	3.04	9.38***	.169
Rdg Frq	5.18bc	4.50	10.03a	6.68	10.77a	6.93	8.00***	.148
Daily Frq	2.97bc	2.02	6.42a	1.37	5.52a	2.39	26.48***	.365
Inference types by specificity
Necessary	17.94	12.19	13.65	8.46	17.84	12.54	ns	.031
Specific	13.97	10.79	10.29	6.94	13.00	9.38	ns	.029
Generalized	3.97	3.29	3.35	2.85	4.84	4.80	ns	.026
Elaborative	61.88	24.18	53.42a	24.42	74.74a	26.95	5.64**	.109
Specific	47.91	17.67	39.42a	20.84	55.58a	20.66	5.20**	.102
Generalized	13.97	8.86	14.00	6.53	19.16	10.79	3.54*	.071
Specific total	61.88	20.04	49.71a	22.80	68.58a	23.78	5.76**	.111
Generalized total	17.94	9.46	17.35a	7.92	24.00a	13.45	3.82*	.077
Inference total	79.82	26.52	67.06a	27.23	92.58a	32.26	6.11**	.117
Word counts	130.22	68.41	111.16	72.52	138.30	71.22	ns	.025

Notes: MMSE = Mini-Mental State Exam. SR = selective reminding. Rdg Freq = how many times a week read for 30 min or more. Daily Frq = how many days a week a newspaper or magazine is read. Matching subscripts across rows indicate significant mean differences.

*p < .05; **p < .01; ***p < .001.

The main effect of age group on total inference counts was significant: old-old had the highest inference total scores, followed by young, with young-old generating the fewest: F(2, 92) = 6.11, p = .003, η² = .117. The age group by inference specificity interaction was significant, F(2, 92) = 3.87, p = .024, η² = .078, reflecting the differences observed in generalized inference counts for old-old adults compared with young and young-old, as well as the difference observed in specific inference counts for young-old and old-old adults. Old-old adults also generated a disproportionate quantity of elaborative inferences compared with young and young-old, yielding a significant age group by inference type interaction, F(2, 92) = 3.57, p = .032, η² = .072. The post hoc pairwise comparisons reported in Table 2 reveal that the significant differences across age group were due to differences between young-old and old-old adults. Because it was possible that age differences could be explained by age-related increases in verbosity, age differences in word counts were analyzed, yielding a nonsignificant outcome.

The age-related increase in generalized inference counts replicates the results reported in McGinnis and Zelinski (2003). However, the age differences in elaborative inference counts were not significant in our earlier study. Examining inferential processes in the absence of a comprehension task may have resulted in the liberal generation of elaborative inferences, with the possibility that this type of inferential processing is typical during reading. In addition, the age-related increase in elaborative inferences suggests that old-old adults may generate greater quantities of elaborative inferences during reading compared with young-old adults. The results of Experiments 1 and 2 suggest that comprehension may take precedence over other tasks. Alternatively, think-aloud tasks may be affected by the demand characteristics of concurrent processing tasks, particularly if one of the tasks is related to generating an interpretation.

Experiments 1 and 2: cognitive integrity analyses.—

Cluster analysis was used to identify subgroups presenting different configurations across cognitive integrity variables (MMSE, SR-average, SR-delayed, and Category Fluency). Because the interest herein is the maintenance of cognitive integrity in older adults, only the data from the young-old and old-old samples were included. Combining the data across experiments ensured that the categorization process was consistent, maintaining the integrity of the grouping process. K-means cluster analysis produces groupings that maximize the distances among cluster centers, resulting in statistically homogeneous groups (Hair & Black, 2000). Overall, this analysis yielded three cognitive integrity groups: Low, Medium, and High. The first group (N = 57) reflects individuals with high standardized cluster center scores across all four variables. The cluster centers were .91 for MMSE, .80 for SR-average, .75 for SR-delayed, and .78 for Category Fluency). The second group (N = 47) had low cluster centers across all four variables (−.33; −1.14; −1.04; and −.81, respectively). The third group (N = 52) had scores in between the lowest and highest groups, with the exception of a very low MMSE score (−.69; .15; .12; −.12, respectively).

Discriminant analysis was used to examine the integrity of the cluster groupings. Discriminant analysis resulted in two significant functions. The first accounted for 83% of the variance with a canonical correlation of .86. The second function accounted for 17% of the variance with a canonical correlation of .606. Both discriminant functions yielded significant Wilks’ lambda statistics: Δ₁ = .164; Δ₂ = .632, p < .001. The first function was associated with the largest correlation coefficients for SR-average, SR-delayed, and Category Fluency: .773, .636, and .510, respectively, and a lower nonsignificant coefficient for MMSE, .448. These results suggest that the High-Integrity participants may include individuals who have maintained high levels of cognitive integrity into late adulthood, whereas the Low-Integrity participants may include individuals experiencing changes related to age and/or disease processes. The second significant function is associated with a greater MMSE coefficient, .851, and lower nonsignificant coefficients for SR-average, SR-delayed, and Category Fluency: −.345, −.248, and .077, respectively, mirroring the final cluster center pattern observed for the Medium subgroup. Hence, participants in the Medium subgroup may be more likely to include participants who have experienced age-related cognitive decline, possibly resembling “cognitively impaired, no-dementia” (CIND) individuals (Chertkow et al., 2007).

Classification integrity can be ascertained by examining the correspondence between cluster membership and membership predicted using the discriminant functions. The two discriminant functions yielded 96.8% accurate prediction. Five possible misclassifications were observed: Two who were classified as Medium instead of Low, two as Medium instead of High, and one as Low instead of Medium. These five participants were excluded prior to examining cluster group differences.

Table 3 provides the means for background, cognitive integrity, interpretation, and inference variables across High,- Medium-, and Low-Integrity groups. The interpretation performance in the Low-Integrity group differed statistically from the interpretation performance for the High-Integrity group, yielding sizable effect sizes. The Low-Integrity group generated more inferences in every cell (specificity by type) than the High and Medium groups, but this trend did not yield statistical significance (all the Fs were below 1.5). The effect sizes for main effects and interactions were small (.034–.046), with the latter reflecting the effect of cluster group on overall inference generation.

Table 3.

Means by Integrity Subgroups

	Low		Medium		High
	M	SD	M	SD	M	SD	ANOVA
Background	N = 46		N = 48		N = 57		F(2, 150)	η2
Age	78.98ab	8.11	73.88a	9.07	71.89b	5.98	11.08***	.130
Education	13.13a	1.96	13.38b	1.98	14.58ab	2.61	6.73**	.079
Vocabulary	18.22a	6.46	19.19b	5.70	25.77ab	8.02	18.90***	.203
MMSE	27.43a	1.45	26.78a	0.93	29.28a	0.77	77.85***	.512
SR-average	4.55a	1.31	7.10a	0.94	8.26a	1.13	139.77***	.654
SR-delayed	2.74a	1.83	6.02a	1.70	7.65a	2.04	89.01***	.546
Category Fluency	12.67a	2.40	15.16a	2.98	18.73a	3.23	56.17***	.431
Interpretation	N = 25		N = 32		N = 33		F(2, 89)	η2
Precise	5.04	0.77	4.93	0.81	5.18	0.82	0.72	.016
Less precise	4.32	1.12	4.30	1.07	3.98	1.24	0.85	.019
Summary	2.36a	1.88	1.85b	1.50	0.74ab	0.99	9.49***	.169
Irrelevant	1.43a	1.44	0.89	1.15	0.36a	0.62	6.92**	.137
Discrimination	2.79a	1.41	3.24b	1.42	4.03ab	1.15	6.62**	.132
Inference type	N = 21		N = 16		N = 24		F(2, 58)	η2
Necessary	17.81	11.49	12.63	7.80	15.83	12.02	1.03	.034
Elaborative	71.24	30.12	59.00	28.32	62.33	24.89	1.01	.034
Inference specificity
Specific	65.24	26.19	52.94	27.41	58.67	22.26	1.11	.037
Generalized	23.81	15.00	18.69	9.16	19.50	9.10	1.14	.038
Total	89.05	35.15	71.63	32.97	78.17	28.81	1.41	.046

Notes: MMSE = Mini-Mental State Exam. SR = selective reminding. Older participants (aged 60+ years) from Experiments 1 and 2 (N = 151). Matching subscripts across rows indicate significant mean differences.

*p < .05; **p < .01; ***p < .001.

GENERAL DISCUSSION

Age Differences in Products and Processes

In Experiment 1, old-old adults did not discriminate among definition choices as well as young and young-old adults. More specifically, old-old adults were more likely to assign high ratings to summary and irrelevant definition options even when two of the four definition options were more specific: results that mirror those obtained in our previous experiments (McGinnis and Zelinski, 2000, 2003). As such, age-related declines in the product of comprehension processes have been replicated across studies, with these declines reflecting a tendency to generate interpretations that are overgeneralized or irrelevant.

In Experiment 2, the think-aloud protocols of old-old adults included more generalized and more elaborative inferences, yielding significant age by inference interactions. The age-related increase in generalized and elaborative processes may reflect the emergence of a narrative propensity for interpretations that are global and personally relevant, a propensity that could be an artifact of the richness of narrative experiences possible across the life span (Adams, 1991; Adams et al., 1997; Dixon & Backman, 1993). In addition, the tendency to process text in a global or elaborative manner may reflect socioemotional selectivity predispositions associated with increasing positive and personally relevant experiences (Blanchard-Fields et al., 2008; Carstensen, 1995; Carstensen & Mikels, 2005).

Self-regulated Language Processing

Stine-Morrow et al. (2006) proposed that age-related language-processing problems may be due to declines in the effective regulation of processing resources. These scientists hypothesize that older adults may lower their threshold for acceptable interpretations to “good enough,” with gist-level interpretations sufficing in many comprehension scenarios. According to Craik and his colleagues, comprehension processes in late life could be affected by an age-related increase in reliance on habits, overlearned mechanisms, and schemas (Craik & Bialystok, 2007; Naveh-Benjamin, Craik, Guez, & Kreuger, 2005). Using stereotypical schemas or habitual approaches may contribute to encoding that is insufficiently specific, compromising specificity in the meanings generated. Consistent with these frameworks, overgeneralized interpretations may reflect compensatory adjustments of comprehension goals when older adults find tasks challenging. The interpretation and inferential results reported herein are consistent with these theories, thereby providing empirical support for a late-life shift to overgeneralized and/or schema-based information processing.

Effects of Commingling Comprehension Tasks

Compared with our earlier think-aloud study, Experiment 2 of the current study demonstrated that participants generated an abundance of elaborative inferences when narratives did not include unfamiliar words. Hence, commingling tasks may affect inference generation. The age differences in unfamiliar word ratings observed in the present study resemble those previously observed, suggesting that older adults focus on unfamiliar word interpretation regardless of task commingling. Taken together, these findings suggest that it may be prudent to separate some text processing tasks when testing older adults.

Cross-Sectional Designs With Young-Old and Old-Old Samples

Because age-related changes in verbal processes have been observed across methodologies and across measures, the differences between young-old and old-old adults are now reasonably substantiated (MacDonald, Hultsch & Dixon, 2003; Meyer & Pollard, 2006; Zelinski & Burnight, 1997; Zelinski & Stewart, 1998). Including sufficient older adults around 80 years of age, in addition to the more typical sample of older adults around 70, facilitates the statistical exploration of cognitive change across a broader range of age in late life. The reluctance to recruit two samples or more is understandable, reflecting practical concerns about time and expense. On the other hand, generalizability and reliability would be enhanced if two-sample approaches were more common (also see Bäckman, Small, Wahlin, & Larsson, 2000; Baltes & Smith, 2003). Empirical inconsistencies across cross-sectional studies in the cognitive aging literature could be due to variations in the proportion of adults older than 75 years.

Cognitive Integrity and Comprehension in Older Adults

Variables sensitive to cognitive integrity were included in both experiments, enabling a correlational analysis of the relationship among cognitive integrity variables and comprehension processes. Cognitive integrity in presumed healthy older adults could be influenced by disease-related neurodegeneration as well as by “normal” age-related changes. The prevalence of Alzheimer's disease and vascular dementia increases substantially with age (Evans et al., 1989; Katzman & Kawas, 1994). Because the disease can be present up to a decade before diagnosis, the assumption that 10%–30% of presumed healthy older adult participants around 80 years are in a transitional phase is reasonable (Bäckman et al., 2000; Bondi, Salmon, & Kaszniak, 1996; La Rue & Jarvik, 1987; Linn et al., 1995; Masur, Sliwinski, Lipton, Blau, & Crystal, 1994; Sliwinski, Hofer, & Hall, 2003). Because a sample of older individuals may include people who are unimpaired as well as people who are experiencing cognitive deficits, it may be appropriate to view presumed healthy samples of older adults as reflecting differing cognitive configurations: adults who are minimally affected by age-related cognitive deficits, adults who are affected by deficits related to aging but minimally affected by disease-related changes (possibly CIND), and adults who are prodromal for some type of dementia. Neuropsychological indicators of cognitive integrity that had demonstrated sensitivity to preclinical Alzheimer's disease in previous studies were included in the current experiment (Jacobs et al., 1995; Masur et al., 1989; O’Connell & Tuokko, 2002; Sliwinski, Lipton, Buschke, & Wasylyshyn, 2003; Sliwinski, Hofer, & Hall, 2003).

The unfamiliar word interpretation results of the current study suggest that overgeneralization tendencies may characterize participants in Low- and Medium-Integrity groups. It is noteworthy, however, that the interpretation scores for the High-Integrity group resemble the scores reported for young adults in Experiment 1. It is also noteworthy that inference means did not differ across cognitive integrity levels. Hence, these data suggest that text processing could be reasonably well maintained even when participants may be experiencing diminished cognitive integrity. Because our earlier studies yielded correlations among comprehension and working memory measures for older adults, declines in text processing in healthy young-old adults may be due primarily to age-related declines in processing resources. In contrast, language-processing declines in healthy adults around 80 years may reflect multiple explanations, including explanations associated with early neurodegeneration. Hence, explanations for cognitive changes may be more diverse and complex in old-old adults compared with young-old adults.

Methodologically, these data suggest that using cluster and/or discriminant analysis to identify cognitive integrity groups could be informative. Significant group differences for chronological age, vocabulary, and unfamiliar word interpretation provided convergent validity. Because cognitive integrity declines may affect particular cognitive domains differently (e.g., producing interpretations versus inferential processing), the empirical use of cognitive integrity classification has potential to distinguish realms of cognition vulnerable to disease-related changes from those less vulnerable.

Limitations

Even though it is reasonable to assume that community samples include a small proportion of prodromal individuals, cross-sectional data, like the data reported here, preclude future diagnoses. At present, only a few longitudinal studies include second- or third-wave clinical diagnoses, permitting the eventual identification of prodromal individuals (Sliwinski, Hofer, & Hall, 2003). Even though the results of the current study and relevant implications are speculative, they may stimulate future studies of semantic deficits in old-old adults, with an emphasis on explanations for age-related changes in comprehension and other cognitive processes.

CONCLUSIONS

The results of these two experiments suggest that community-dwelling old-old adults may have problems generating appropriately specific interpretations of narrative text and may focus on information that is generalized and elaborative during text processing, replicating results reported in our earlier studies (McGinnis & Zelinski, 2000, 2003). In addition, factors associated with the age-related declines in interpretation specificity (product) may pertain to an individual's overall cognitive integrity, whereas inferential processing (process) may be less susceptible to declines in cognitive integrity, reflecting divergent age-related influences on text processes.

APPENDIX

Sample narrative and definitions

Johnson had held his present job since 1991, but in the past year, there had been a number of complaints. After his superiors investigated, Johnson was fired for “dippoldism.” When his superiors found out what Johnson had been doing at the school where he taught, they were quite disturbed. In some cases, the welts had taken a long time to heal. Many of the victims had broken rules and disciplining them was within Johnson's duties, but his method had injured a number of them. This practice, although once fairly common, was no longer considered acceptable. Even though most of them had been too scared to complain about someone they were supposed to obey, eventually the story came out. Johnson's replacement put more emphasis on learning than on discipline, and there was general satisfaction among parents with the decision.

Please rate how closely each definition corresponds to your idea of a good definition for the word dippoldism.

Whipping of school children
0	1	2	3	4	5	6
Poor						Excellent
Cruel treatment of other people
0	1	2	3	4	5	6
Poor						Excellent
Losing a job for misconduct
0	1	2	3	4	5	6
Poor						Excellent
A practice that was common
0	1	2	3	4	5	6
Poor						Excellent

Key. Precise = whipping of school children. Less Precise = cruel treatment of other people. Summary = losing a job for misconduct. Irrelevant = a practice that was common.