Skip to main content
PMC home page
The Journals of Gerontology Series B: Psychological Sciences and Social Sciences logoLink to The Journals of Gerontology Series B: Psychological Sciences and Social Sciences
. 2010 May 24;65B(4):449–458. doi: 10.1093/geronb/gbq034

List and Text Recall Differ in Their Predictors: Replication Over Samples and Time

Kayan L Lewis 1, Elizabeth M Zelinski 1,✉
PMCID: PMC2949299  PMID: 20498454

Abstract

This study tested the hypothesis that latent list and text recall invoke somewhat different processes. A bivariate outcome path model of latent list and text recall evaluated the effects of age, latent speed, working memory, and vocabulary as their predictors. Independent of age, working memory reliably predicted both recall variables, whereas speed reliably predicted list recall only. The relationship between vocabulary and recall was mediated by age, working memory, and speed. The generalizability of this model, based on data from the 1994 testing of the Long Beach Longitudinal Study, was evaluated across samples by testing its invariance on baseline data from an additional panel and for eventual attrition at baseline and at a subsequent testing of retested participants and dropouts. Results showed that the model was invariant over all groups, supporting a replicable distinction between list and text recall.

Keywords: Cognition, Measurement, Working memory

ALTHOUGH episodic recall is generally thought to decline with advancing age, regardless of the nature of the materials to be remembered (e.g., Zacks, Hasher, & Li, 2000), recent work suggests that different mechanisms may be associated with particular recall tasks (Siedlecki, 2007). Studies training memory in older adults have failed to show transfer: Deployment of specific resources and abilities may be required depending on the task (see, e.g., Rebok, Carlson, & Langbaum, 2007). Models of broad memory factors may therefore not adequately distinguish sources of age and individual differences in specific remembering activities.

Although list and text recall have been considered to be part of a latent verbal memory construct (e.g., van der Linden et al., 1999) and may be treated as markers of a higher order memory factor (e.g., Hertzog, Dixon, Hultsch, & MacDonald, 2003), including list and text recall in a single latent variable permits only the common memory variance from these tasks to be predicted. However, list recall shows larger average age declines than text recall when scores are identically calibrated with Rasch scaling (Zelinski & Kennison, 2007). List recall may decline more than text recall because it is more affected by age effects on fluid-like abilities or resources, such as speed or working memory, whereas text recall may be less affected by aging because fluid-like deficits are balanced by stability in crystallized-like abilities, such as vocabulary (see, e.g., Stine-Morrow, Miller, Gagne, & Hertzog, 2008, but see Johnson, 2003).

Another reason why list and text recall should be examined separately is that recall of unrelated words and of discourse very likely involves different encoding and retrieval processes (e.g., Jefferies, Lambon Ralph, & Baddeley, 2004). Word list recall entails encoding and retrieving contextual and semantic item information. Light (1992) suggested that age deficits in list recall occur because of difficulty in remembering the nonsemantic context rather than the semantic information. Discourse memory, in contrast, involves semantic and linguistic processing: analysis of syntactic and semantic relations between linguistic units, construction of inferences, the development of a text base of the discourse, and a situation model of the gist of the text base (e.g., Kintsch & van Dijk, 1978; see Stine-Morrow et al., 2008, for a detailed review). Text recall may therefore be less impaired with age because it relies more heavily on semantic and other linguistic processes than list recall.

The role of processing resources associated with age differences in recall may thus vary across tasks. Processing speed has been associated with word list recall in a number of studies (e.g., Verhaeghen & Salthouse, 1997). Salthouse’s (1996) model suggests that age-related slowing limits the amount of information that can be processed in immediate memory; recall of unrelated words in a list may invoke rapid initial phonological processes as well as slower associative ones (e.g., Jefferies et al., 2004), which could contribute to reduced episodic recall. In contrast, text recall is supported by automatic processes in semantic or long-term memory (Jefferies et al.), so speed may not predict discourse recall in older adults (see Stine-Morrow et al., 2008, for findings of differential prediction of abilities for text vs. sentence recall).

Unlike speed with its differential influences, working memory is likely to be a predictor of age and of individual differences in both list and text recall, as it is associated with list recall in older adults (e.g., Zacks et al., 2000), and comprehension of and memory for discourse (see, e.g., Daneman & Merikle, 1996). Additionally, vocabulary ability is an important independent predictor of text memory and of list recall in people older than 55 years (Hedden, Lautenschlager, & Park, 2005), suggesting that it is another covariate of both types of recall in older adults.

Zelinski and Stewart (1998) analyzed 16-year data from 106 individuals in the Long Beach Longitudinal Study (LBLS) and found that baseline age and change in reasoning, which is highly correlated with working memory (e.g., Kyllonen & Christal, 1990), predicted changes in text recall, whereas age and change in the Schaie-Thurstone Adult Mental Abilities (Schaie, 1985) Recognition Vocabulary test, which has a strong speed component (see Hertzog, 1989), predicted changes in list recall.

In the present study, we evaluated a more sophisticated bivariate outcome model of latent list and text recall as predicted by age, and latent speed, working memory, and vocabulary to determine whether there are differences in their predictors, using data from the LBLS. The model tested is shown in Figure 1A. Working memory and speed were treated as processing resources that covaried but directly predicted list and text recall, which also covaried. Treating working memory as a covariate rather than a mediator of speed is different from many of the individual differences studies published earlier (e.g., Hedden et al., 2005); the rationale is that current findings do not confirm that changes in a single resource such as speed underlie age declines in other abilities (see also McArdle, Hamagami, Meredith, & Bradway, 2000). Increasing evidence additionally suggests that age does not fully account for changes in the covariation between speed and memory (e.g., Ferrer, Salthouse, McArdle, Stewart, & Schwartz, 2005; McArdle, Ferrer-Caja, Hamagami, & Woodcock, 2002; Sliwinski & Buschke, 1999; see also Verhaeghen & Salthouse, 1997) and that a general change factor does not account for most of the variance in ability decline (Hertzog et al., 2003). By allowing a covariance between the two latent abilities, the model is silent as to a chain of causal influence. We also wanted to test whether each of these latent variables, with their covariance accounted for, independently predicted latent list and text recall.

Figure 1.

Figure 1.

Open in a new tab

(A) Structural model of the relationships between age, speed, working memory, vocabulary (Voc), list and text recall. (B) Observed model with nonsignificant paths removed.

Vocabulary was predicted by speed and working memory because changes in fluid-like abilities precede verbal ability changes in older adults (e.g., Ghisletta & de Ribaupierre, 2005; Ghisletta & Lindenberger, 2003). Direct effects of vocabulary on list and text recall and the direct and indirect role of age on the outcome variables were also directly assessed (see Anstey, Hofer, & Luszcz, 2003a). It was expected that, independent of age, latent list recall would only be predicted by speed, but that both list and text recall would be predicted by working memory and by vocabulary.

A second question in the present study is that of the replicability of the model in different samples and subgroups of participants. Few studies have assessed whether models are invariant across samples, despite the importance of replication for scientific generalization (e.g., Horn & McArdle, 1992). A related question is that of the effects of potential attrition on findings. Attrition is often not considered problematic with respect to generalization in cross-sectional designs. However, cross-sectional samples include participants who would drop out if the study was extended. By not differentiating those who would drop out from those who would continue, cross-sectional results might not accurately represent the relationships between variables for either the retested or the dropout subgroups. This could affect conclusions of models of “normal” cognitive aging studies, which are typically cross-sectional (cf., Zelinski, Kennison, Hall, & Lewis, 2009); there may be important subgroup differences in the measurement of constructs studied as well as their interrelationships (e.g., Sternberg & Berg, 1987). For example, eventual dropouts experiencing health or cognitive difficulties may show larger between-variable correlations at baseline than healthier individuals (e.g., Sliwinski, Hofer, & Hall, 2003; Wilson et al., 2002). In a sample with a substantial proportion of potential dropouts, this could lead to average cross-sectional relationships indicating that the variables studied have strong predictive effects on both list and text recall, leading to findings suggesting that their predictors are identical, and supporting combining list and text recall in models of memory aging. Alternatively, if a sample had few potential dropouts, average relationships would be smaller and possibly not different from zero. This could, in turn, suggest that different predictors are associated with list and with text recall, supporting distinct latent recall variables. Thus, models of memory aging could be contaminated by effects that arise due to attrition.

One of the features of LBLS is that multiple cross-sectional panels are recruited at different times, and data are also available across testings, enabling evaluation of the generality of the observed model across samples and over time with tests of invariance. Metric invariance refers to absolute measurement equivalence over individuals, samples, and occasions to ensure that the same constructs and their relationships have been assessed. Configural invariance, on the other hand, suggests that relative measurement equivalence exists because the constructs and relationships are similar but not precisely identical. Thus, the observed model results may not be fully replicable. However, if configural invariance is not present, the generalizability of model findings is severely limited (see Horn & McArdle, 1992).

Two major aspects of invariance were tested. First, invariance of the best-fitting path model was evaluated in a second panel of LBLS participants, and second, its invariance was tested at baseline for subgroups of those retested and of dropouts at a first follow-up, 3 years later, and within the retested group, at baseline and the first follow-up for those who continued in the study, and those who dropped out at a second follow-up 6 years after baseline. This allowed for the examination of the replicability of the model between different groups of participants, between attrition groups, and over time.

METHODS

Participants

The sample was from the LBLS. The study began in 1978 with Panel 1 consisting of members of a Southern California-based health maintenance organization (HMO), Family Health Plan (see Zelinski, Gilewski, & Schaie, 1993). A retest occurred in 1981. In 1994–1995, 106 individuals were retested from that original panel (see Zelinski & Burnight, 1997) with a wider set of measures than for the first two testings, and 42 of them were retested 3 years later on those measures. Panel 2 was drawn from the 1992 membership of the HMO for the 1994–1995 testing and was retested at 3-year intervals. This panel consisted of individuals aged 30–97 years and is described elsewhere (Zelinski & Lewis, 2003). Panels 1 and 2 were collapsed for analyses using the 1994–1995 measures because the factor solution across panels was invariant, and there were relatively few participants from Panel 1 at the 1994–1995 wave (Zelinski & Lewis); in this article, the panels are referred to as Panels 1–2.

Panel 3 was either resampled from the 1992 HMO database or recruited with age-targeted direct mail addressing and advertisements in senior newspapers. Participants were aged 30–98 years and were first tested in 2000–2001. They were retested once approximately 3 years after their first testing. Table 1 provides demographic and health information on the participants in each panel. Participants rated their health relative to their age on a 6-point scale ranging from “very poor” to “ very good” from the Seattle Health Behaviors Questionnaire (Schaie, 2005, p. 242ff).

Table 1.

Mean (SDs) of Demographic and Performance Variables for Participants With Full Information at Each Wave by Panel

Panels 1–2
 Panel 3
Testing Time 1 Time 2 Time 3 Time 4 Time 1 Time 2
Sample size 613 322 162 127 822 486
Number of women 324 170 81 67 454 276
Age
    M 67.61 71.16 71.93 73.58 68.76 71.74
    SD 13.87 11.91 10.97 9.53 13.71 12.45
Self-reported health
    N providing information 563 315 148 124 772 478
        M 2.55 2.06 2.13 2.03 2.10 2.16
        SD 1.36 1.01 1.03 0.86 1.07 1.05
Years of education
    N providing information 565 311 138 125 762 479
        M 13.95 14.15 14.4 14.03 14.86 15.23
        SD 2.98 2.70 2.06 2.57 2.60 2.61
Open in a new tab

Notes: Health rating on a scale of 1–6, with 1 being excellent. All performance scores were scaled as z scores across panels and times of testing.

Screening for Dementia

LBLS testers administered the Mini-Mental State Examination (MMSE; Folstein, Folstein, & McHugh, 1975) to those participants having difficulty following instructions in completing the first two tests of the first session. Those who scored less than 25 on the MMSE were permitted to complete the first session, but their data were not scored and they were excused from further testing. The screening was not repeated for any indication of impairment at follow-up; virtually no participants who returned for testing showed impairment.

Measures

The model comprises five latent factors with three measures per factor except for list recall. The measures and their reliabilities are Vocabulary 1 (1.0), Vocabulary 2 (0.88), Recognition Vocabulary (0.66), Number Comparison (1.0) Letter Comparison (0.96), Pattern Comparison (0.56), Sentence Memory Hard (0.77), Sentence Memory Easy (0.88), Sentence Span (1.0), Text Recall 1 (0.85), Text Recall 2 (0.95), Text Recall 3 (1.0), List Recall 1 (0.92), and List Recall 2 (1.0). Reliabilities reflect the factor loadings squared in the measurement model from Zelinski and Lewis (2003). Reliabilities of 1.0 indicate the measure used to standardize the latent factor. The latent factors from which the measures were derived were vocabulary, speed, working memory, text recall, and list recall, respectively, as described elsewhere (Zelinski & Lewis). All scores were standardized to z scores across occasions and panels for the analyses. A description of all measures in the present study is in the Appendix.

Analyses and Treatment of Missing Data

The measurement model for the five factors was derived using maximum likelihood extraction with Equamax rotation (see Zelinski & Lewis, 2003). Structural equations were fitted using the Mx program (Neale, 1997) using individuals with full information on each test. The indices used to examine model fit were χ2; Akaike’s (1987) information criterion (AIC), with lower values indicating a better fit; the root mean error of approximation (RMSEA; Browne & Cudeck, 1993), with values at or less than .08 a good fit and those between .08 and .10 a reasonable fit; and the comparative fit index (CFI; Bentler, 1990), with values more than .90 indicating a good fit. The level of significance for nested χ2 tests of fit was p < .001 because of the large number of tests conducted.

Invariance testing compared a highly constrained model with models that systematically relaxed parameters between groups. This approach allowed us to test initial null hypotheses that the groups did not differ on any parameters and was an indication of whether each less stringent model reduced misfit compared with its more constrained predecessor, as suggested by Horn and McArdle (1992). Means were modeled in all analyses. Effects of age on the structural model were modeled in the invariance analyses, and results were identical regardless of its inclusion, so only tests of parameters associated with the latent variables are reported. The relations of age with each latent variable are reported in Table 2.

Table 2.

Model Parameters for the Time 1 Data of all Panels 1–2 and Panel 3 Participants and for Configurally Invariant Models of the Time 1–Time 2 Data From the Time 3 Retested and Dropout Participants

Participant group
Panels 1–2
 Panel 3
 Panels 1–2 by status at Time 3
Measurement occasion
Baseline
 Baseline
 Retested
 Dropout
Parameter Time 1 Time 1 Time 1 Time 2 Time 1 Time 2
Age to latent factors
    Age → speed −0.65 −0.53 −0.75 −0.67 −0.55 −0.60
    Age → working memory −0.46 −0.40 −0.41 −0.41 −0.48 −0.47
    Age → vocabulary 0.37 0.44 0.45 0.22 0.36 0.28
    Age → text recall −0.06 −0.07 −0.07 −0.00 −0.05 −0.21
    Age → list recall −0.10 −0.15 −0.30 −0.14 −0.10 −0.20
Latent factor to latent factor
    Speed → vocabulary 0.28 =0.28 0.11 0.23 0.37 0.44
    Working memory → vocabulary 0.93a =0.93 1.00 0.52** 0.75 0.67
    Speed → text recall −0.01 =−0.01 −0.07 0.08 0.16 0.03
    Speed → list recall 0.21 =0.21 0.23 0.42 0.27 0.19
    Working memory → text recall 0.81 =0.81 0.88 0.60 0.70 0.64
    Working memory → list recall 0.54 =0.54 0.18 0.56** 0.64** 0.74**
    Vocabulary → text recall 0.06 =0.06 0.07 0.06 −0.01 0.05
    Vocabulary → list recall 0.01 =0.01 0.12 −0.04 −0.07 −0.01
Latent factor correlations
    Speed ↔ working memory 0.49 0.49 0.43 0.51 0.46 0.56
    Text recall ↔ list recall 0.21 0.16 −0.01 0.42** 0.42** 0.57**
Latent factor SDs
    Speed 0.71 0.69 0.63 0.67 0.76 0.76
    Working memory 0.66 0.68 0.68 0.59 0.63 0.59
    Vocabulary 0.57 0.53 0.53 0.49 0.46 0.52
    Text recall 0.43 0.53** 0.46 0.35 0.51 0.34
    List recall 0.56 0.56 0.56 0.68 0.60 0.61
Open in a new tab

Notes: Parameters are from the model identified in the Results section as best fitting for the analysis (baseline structural model for Panels 1–2; Model 3 for Panel 3, Model 4 for Panels 1–2 by status at Time 3). Bold parameters for the baseline Panel 1 model are significantly different from zero at p < .01. Equal signs before the parameter indicate that the parameters remained fixed to those of the comparison group.

a

The path coefficient of .93 from working memory to vocabulary is larger than reported elsewhere and most likely due to cooperative suppression from age. Cooperative suppression is observed when there is a positive relationship between predictors like working memory and age with an outcome like vocabulary, but the predictors themselves are negatively related, like working memory and age. The negative relationship between predictors involves some variance that is unrelated to the outcome, so that when the predictors are partialled from each other, the relationship with the outcome becomes larger than the zero-order correlation (Cohen & Cohen, 1975, p. 90). An analysis in which the path from age to vocabulary was removed reduced the working memory to vocabulary path substantially.

**p < .01 difference from comparison group.

RESULTS

Bivariate Model of List and Text Recall

The bivariate path model shown in Figure 1 was evaluated with the baseline data of Panels 1–2 to determine whether the paths differed significantly from zero. This constituted a test of the hypothesis that age and the latent variables predicted list and text recall.

The model fit was reasonable, with χ2 = 467, df = 70, AIC = 327, RMSEA = .096, and CFI = .943. Figure 1B shows the paths that differed significantly from zero, and the first column of Table 2 shows the parameters obtained. For comparison with previous work (e.g., Hedden et al., 2005), a path from speed to working memory instead of a correlation was tested, but results were essentially identical to those in Table 2. Other alternative models removed the direct effects of age on list and text recall and the direct effect of speed on text recall, but no changes in fit were observed.

Age had direct effects, as shown in Table 2, with negative paths to speed, working memory, and to list recall, and with a positive path to vocabulary. Its path to text recall did not differ from zero. Speed and working memory had direct and positive paths to vocabulary. Speed was directly positively associated with list but not with text recall. Working memory had direct positive paths to both list and text recall. Vocabulary did not directly predict either recall factor. These results support a distinction between latent list and text recall.

Replication of the Bivariate Model Across Panels

The parameters of the baseline structural model for Panels 1–2 were tested for invariance against the baseline data of Panel 3. Results are presented in Table 3. Model 1, the most stringent model with identical (fixed) parameters for the factor paths, factor standard deviations (SDs), and correlations (Horn & McArdle, 1992), indicated a reasonable fit, as shown in Table 3. Model 2 relaxed the factor correlations to configural invariance, that is, it allowed factor correlations to vary by panel but kept the factor SDs and latent factor relationships metrically invariant. There was no improvement in fit for this model. Model 3 relaxed the factor SDs and factor correlations but kept the latent factor relationships identical. The fit of Model 3 was significantly better than that of Model 2, as shown in Table 3. The freed and fixed parameters for the Panel 3 participants at baseline from Model 3 are shown in the second column of Table 2. Tests of 99% confidence intervals identified the parameters of Panel 3 that differed from those of Panels 1–2 and revealed that the SD for text recall in Panel 3 was significantly larger than that for text recall in Panels 1–2. Model 4 was a configurally invariant model of factor SDs, factor correlations, and latent factor relationships so that all these parameters were free to vary across samples. Model 4 did not further improve fit, as shown in Table 3, so the parameters of Model 3 were accepted as the best fitting. This indicates that the latent variables were consistently measured, and all but one parameter were identical across panels at baseline, supporting the generality of the path model findings in Panels 1–2 in a second sample.

Table 3.

Fit Indices for Time 1 Models Testing Different Degrees of Measurement Invariance of the Structural Model Between Panels 1–2 and Panel 3 at Time 1

Model χ2 df AIC RMSEA CFI Δχ2a Δdfa
Equal factor paths, SDs, and correlations 959 170 619 .081 .948
Free correlations only 956 166 624 .082 .948 3 4
Free factor SDs and correlations 922 156 610 .084 .950 34*** 10
Configural invariance (all parameters free) 907 140 627 .088 .950 15 16
Open in a new tab
a

Nested comparison of the current (more relaxed) model fit with the prior (more constrained) one.

***p < .001.

Attrition Studies Within Panels

In order to test the hypothesis that the baseline parameters of the path model differed between those who dropped out of the study and those who were retested, invariance of the structural model was evaluated within each panel and across testings.

Baseline (Time 1) comparison of 3-year Time 2 status: retested vs. dropouts.—

Parameters from the baseline data of those retested at Time 2, 3 years after baseline, were compared with those who dropped out at Time 2 for each of the panels. In both cases, Model 1, the strictest invariance model, held, with no improvement in fit for the less stringent models, as shown in the top two sections of Table 4. This indicates that the parameters of the path model at baseline were metrically invariant for both the retested participants and Time 2 dropouts within the two panels.

Table 4.

Fit Indices for Models Testing Different Degrees of Measurement Invariance of the Structural Model for Retested Participants and Dropouts Within Panels

Model χ2 df AIC RMSEA CFI Δχ2a Δdfa
Time 1: retested vs. dropout at Time 2
 Panels 1–2
    Equal factor paths, SDs and correlations 529 170 189 .083 .944
    Free correlations only 529 166 197 .084 .943 0 4
    Free factor SDs and correlations 514 156 202 .087 .944 15 10
    Configural invariance (all parameters free)
 510
 140
 230
 .093
 .942
 4
 4
Time 1: retested vs. dropout at Time 2
 Panel 3
    Equal factor paths, SDs and correlations 563 170 223 .075 .951
    Free correlations only 553 166 221 .075 .952 10 4
    Free factor SDs and correlations 535 156 223 .071 .953 18 10
    Configural invariance (all parameters free)
 512
 140
 232
 .081
 .954
 23
 16
Time 1 and 2: retested vs. dropout at Time 3
 Panels 1–2
    Equal factor paths, SDs and correlations 764 340 84 .092 .929
    Free correlations only 732 332 68 .091 .933 32*** 8
    Free factor SDs and correlations 692 312 68 .092 .937 40 20
    Configural invariance (all parameters free) 592 280 32 .089 .948 100*** 32
Open in a new tab
a

Nested comparison of the current (more relaxed) model fit with the prior (more constrained) one.

***p < .001.

Differences over time and between Time 3 retested and dropout subgroups.—

This analysis tested the hypothesis of differences in the path model parameters at each testing occasion in the Panels 1–2 participants retested at Time 2. They were subdivided into those who dropped out at the third testing, 6 years after baseline, and those who continued in the study. Even though the model of Time 1 parameters was metrically invariant for the 3-year dropouts and retested participants, only 26% of the baseline sample participated in the third testing, so a retrospective analysis of returnees and dropouts identified by their status at the second retest was conducted to determine whether longer term attrition affected the model at baseline. In addition, parameters for the first follow-up in both groups were tested for invariance within individuals over time. A simultaneous four-group model based on Time 3 status was evaluated: retested participants’ Time 1 data, their Time 2 data, the dropouts’ Time 1 data, and their Time 2 data (Table 2, columns 3 through 6). Model 1 fixed the parameters of the groups to those of the retested participants at baseline. Results shown in the bottom panel of Table 4 indicate that Model 1, the test of strictest invariance, showed greater misfit than the analyses with more relaxed parameters. Model 2 had an improved fit, but Model 4, with all parameters freed, had the best fit. As shown in the third column of Table 2, the continuing participants’ Time 1 correlation between list and text recall was −0.01. It did not differ significantly from zero. This correlation was significantly smaller than that of the same participants at Time 2 (0.42) (Table 2, column 4), and for that at both Times 1(0.42) and 2 (0.57) of those who left the study 6 years after baseline (Table 2, columns 5 and 6). In addition, the continuing participants had a significantly smaller baseline factor path of 0.18 from working memory to list recall than they did at Time 2 (0.56) and than the 6-year dropouts did at both testings (0.64 at Time 1 and 0.74 at Time 2). The 6-year dropouts’ parameters are presented in columns 5 and 6 in Table 2. Finally, the continuing participants had a significantly smaller path from working memory to vocabulary at Time 2 than at Time 1, suggesting a reduced relationship with practice.

These results indicate that those who dropped out at the third testing, 6 years after baseline, had significantly larger factor relationships at baseline and the first follow-up compared with the baseline of those who remained in the study, confirming that dropouts differ from retested participants in the relative strength of parameters (Sliwinski et al., 2003; Wilson et al., 2002). This implies that the Time 3 dropouts’ data affected one of the eight factor relations at Time 1 for Panels 1–2. However, the configurations of abilities that significantly predicted list and text recall did not vary with respect to subgroups, suggesting that the same relative relationships were preserved in the cross-sectional Time 1 model and at Time 2 despite substantial attrition. Thus, the model shown on the lower section of Figure 1 was essentially replicated for different groups, with attrition, and over time.

DISCUSSION

The structural model tested hypotheses about whether list recall and text recall have different predictors with respect to age, speed, working memory, and vocabulary. Speed only directly affected list recall; working memory had direct effects on both list and text recall. Vocabulary had no significant association with either type of latent recall after removing effects of speed and working memory. As shown on the lower half of Figure 1, age had both indirect and direct effects on list recall and indirect effects on text recall. Thus, there were some differences in predictors for the two latent recall variables. These findings were metrically invariant for two samples and the subgroups of 3-year dropouts and retested participants, and configurally invariant for 6-year continuing participants and dropouts over two testings.

Age Effects

The findings indicate that the effect of age on list and text recall is multifaceted. Age had a significant and positive relationship with vocabulary, and negative relationships with speed (e.g., Hedden et al., 2005), working memory, and list recall (see, e.g., Park et al., 2002). Age also directly affected list recall in the cross-sectional analyses, which is contrary to some (see, e.g., Salthouse, 1996), but not all studies (e.g., Siedlecki, 2007). Zelinski and Stewart (1998) found that change in a timed vocabulary task, which was assumed to reflect speed rather than verbal ability changes, as well as age, was independently associated with list recall change; the structural analysis here confirms previous findings with manifest variables. One difference between LBLS and other studies examining structural models of age and ability predictors on memory is that the initial sample sizes are two to three times larger, and therefore, power to detect the age effect on list recall, independent of abilities, was greater (see, e.g., Salthouse, 2004).

Direct effects of age may serve as a proxy for both sociodemographic variables and health in predicting baseline list recall (e.g., Anstey et al., 2003a). For example, cohort differences (Zelinski & Kennison, 2007) and other demographic variables such as education (e.g., Zelinski & Gilewski, 2003) may contribute to some of the cross-sectional age effects. Health-related phenomena that may be associated with list recall include presence of specific medical conditions, health ratings, and depression (e.g., Zelinski & Gilewski), as well as multiple comorbidities (e.g., Wilson et al., 2002). Finally, age may be a surrogate for neuropsychological indicators of frontal or executive function deficits (see, e.g., Bugg, Zook, DeLosh, Davalos, & Davis, 2006). In contrast, age had no direct effects on text recall, perhaps because of weaker effects of variables such as cohort on text compared with list recall (Zelinski & Kennison).

Processing Resources

As expected, latent speed was both directly and indirectly associated with latent list recall (e.g., Salthouse, 1996; Verhaeghen & Salthouse, 1997). Recent work suggests that variance in memory remains after partialling speed (e.g., Salthouse, 2004), which was confirmed here, as working memory, also mediated by age, directly predicted list recall (see Hedden et al., 2005; Park et al., 2002). Finally, though vocabulary had no effects on list and text recall in the structural model, a significant covariance of vocabulary with both latent recall variables in the measurement model suggested classical suppression due to the partialling of effects of age and processing resources, as found elsewhere (e.g., Hedden et al.).

The effect of speed on text recall was not reliable, supporting recent findings that speed does not underlie declines in all cognitive abilities (e.g., Ferrer et al., 2005; Salthouse, 2004), and in text recall, in particular, contrary to other suggestions (e.g., Johnson, 2003). Older adults may be able to compensate for declining working memory because they rely on the linguistic and conceptual information available in discourse to self-regulate its processing (see Stine-Morrow et al., 2008) and memory. Thus, speed may not predict text recall as it does list recall. The nonsignificant findings with vocabulary also suggest that vocabulary knowledge does not appear to play a direct compensatory role in text recall in LBLS samples.

That working memory predicted text recall confirms a substantial literature, indicating that it is critical for discourse comprehension and recall (e.g., Daneman & Merikle, 1996) and that its effects are direct and independent of age (e.g., Kemper & Liu, 2007). Working memory and reasoning are strongly associated (e.g., Kyllonen & Christal, 1990), thus the longitudinal findings of Zelinski and Stewart (1998), showing change in reasoning associated with text recall change, were confirmed with latent variables, in a new sample, and in subgroups of the sample over time. As a predictor of both list and text recall, working memory may be a common processing resource of both types of episodic memory (e.g., Verhaeghen & Salthouse, 1997). This is not surprising, given that the working memory tasks in the present study all required recall, though performance on working memory span tasks is not completely determined by episodic memory (see van der Linden et al., 1999).

Differences in predictors of list and text recall in bivariate analysis are contradictory to the suggestion that list and text recall reflect only a common memory process, as assumed in studies using both list and text recall measures that combine them into a composite (e.g., van der Linden et al., 1999; but see Hertzog et al., 2003). In the present study, the correlation between the two types of latent recall was 0.58 in the measurement model, which suggested less than complete overlap in variance. In the structural model, shown in Table 2, the correlation was reduced to .21, suggesting that removal of variance due to the effects of the other predictors suppressed that relationship. Coupled with the finding that age, speed, and working memory affected list recall, but only working memory affected text recall, the present findings suggest that list and text recall reflect somewhat different processing activities. This suggests that collapsing list and text recall into a memory factor may produce misleading results.

This may also be true of other memory tasks. Park and colleagues (2002) found that visuospatial and verbal recall tasks required separate factors in latent variable analyses. Within verbal tasks, it was also necessary to identify separate factors based on the environmental support required, so that cued recall, for example, was differentiated from free recall (Hedden et al., 2005; Siedlecki, 2007). Siedlecki confirmed that this was not the case with visuospatial tasks. Verbal memory may therefore be more multifaceted than visuospatial memory with respect to latent constructs. However, the domain of visuospatial tasks studied thus far is limited and fewer of its component abilities have been identified.

Generality of the Path Model

This is the first study we are aware of to evaluate invariance of a path model of memory with a second sample of participants as well as of invariance within subsamples of retested participants and dropouts. Analyses of a second sample and of 3-year dropouts and retested individuals at baseline confirmed metric structural invariance of the model of list and text recall.

The more extensive retrospective evaluation of attrition effects for those remaining in the study 6 years after baseline, at Time 3 compared with Time 3 dropouts, however, showed some parameter differences, with several smaller parameters for continuing participants at both baseline and Time 2 than the third-testing dropouts. This supports the suggestion of less variability and weaker relations in a retest group at the first testing, as well as greater selection associated with a subsequent testing, as proposed by Ferrer and colleagues (2005) and Wilson and colleagues (2006). This may imply the existence of subgroups who do not manifest the identical relationships of abilities to recall outcomes at baseline or first follow-up. Unfortunately, additional comparisons of the structural model could not be made for those who left the study after the Time 3 retest because of sample size considerations, or for Panel 3, for which a third retest is currently in progress, so the hypothesis of parameter differences in subgroups identified from attrition beyond a second retest and in a second sample could not be confirmed at this time. On balance, it has been suggested that relatively few individuals in large sample studies experience changes that affect multiple cognitive functions in a systematic way (e.g., Anstey, Hofer, & Luszcz, 2003b; Wilson et al., 2002; Zelinski & Lewis, 2003).

CONCLUSIONS

This study found differential prediction of latent list and text recall in cross-sectional samples spanning the range from age 30 to 98 years. Though both list and text recall obviously involve memory, their predictors vary somewhat; list recall is associated with speed of processing and working memory, whereas text recall is associated with working memory alone. These differences have implications for models of memory decline with age that focus on predictors of recall without regard to the demands of specific recall tasks. On an applied level, they suggest that in order to improve older adults’ memory for different types of materials, training should specifically address the task. For example, to improve memory for lists, activities that improve speed of processing and expansion of working memory should be practiced. To improve discourse recall, speed may be less relevant. Interestingly, a trial of a computer program emphasizing speeded auditory processing of sound sweeps, phonemes, and syllables, but with much less emphasis on discourse, produced improvement in list and span measures but not recall of stories (Smith et al., 2009). Greater emphasis on discrimination in working memory for complex contextual material might have improved discourse recall.

The study also suggests the observed model held across samples and that attrition did not markedly change results in both samples and over more than one testing. This implies that the patterns of interindividual differences across adulthood in list and text memory are generalizable.

FUNDING

The Long Beach Longitudinal Study is an ongoing study conducted at the Andrus Gerontology Center, University of Southern California, funded by R01 AG10569 and T32 AG00156.

Acknowledgments

K.L.L. and E.M.Z. contributed equally to this work.

Appendix

Measures Test Description Source Scoring
Vocabulary Advanced Vocabulary 1 Each test has the participant identify the definition of a word from a list of five possible definitions. Each test is 4 min. Ekstrom, French, Harmon, and Dermen (1976) An overall score was calculated from the total correct. The range is 0–18 items correct.
Advanced Vocabulary II
List recall Immediate Recall 1 Participant study a list of 20 concrete high-frequency nouns for 3.5 min. At the end of the studying time, participants immediately recall as many of the words as possible and have unlimited time to recall them. The Long Beach Longitudinal Study has two immediate recall lists—similar in composition to each other—administered at different testing sessions. Zelinski and Lewis (2003) The proportion of words correctly recalled serves as the score for each task.
Immediate Recall 2
Text recall Text Recall 1 Text recall comprises three separate essays requiring the participant to hear the essay out loud and then immediately recall as much of the essay as possible in writing. The participant follows along reading the essay while it is being read to them. Text Recall 1 was a 227-word essay with 104 idea units used in 1978. Zelinski, Light, and Gilewski (1984) Text recall for each of the three essays was computed as the proportion of idea units recalled.
Text Recall 2 Text Recall 2 was a 204-word essay with the same idea units as the 1978 text. Two versions were counterbalanced by participant and testings.
Text Recall 3 Text Recall 3 was a story with 209 words with 93 idea units. Two versions were counterbalanced by participant and testings.
Working memory Syntax Memory Easy These tasks measure memory for sentences that have identical words but are parsed in such a way that they are easy- or hard-to process syntactically. Each of the 16–19 word sentences has one of two syntactic manipulations. In one type of manipulation, a verb and particle (e.g. “pick up”) are either placed together in a sentence (easy) or are separated in the sentence (hard). The easy version occurs more frequently in the English language. The other syntactical manipulation involves the connection or separation between a prepositional phrase and a verb where the connection is easy to remember and separation of the two by other words is harder to remember. Syntax Memory Easy and Hard comprised eight counterbalanced sentences with varying syntactical manipulation. Participants rate the grammatical soundness of the sentence (i.e., grammatically correct or incorrect) and then are asked to write the sentence that they just heard. See Zelinski and Lewis (2003) for more details Each sentence has six to eight propositions that are scored for gist. The proportion of correctly recalled proportions is the total score for each task.
Syntax Memory Hard
Sentence Span Participants listened to sentences in groups of two or three sentences each. Each sentences is in the form of subject–verb–object. Some sentences are semantically correct and others are not correct (i.e., “The sleep ate the doughnut”). They were asked to hold the subject and object in mind. After each set of two to three sentences (for a total of four sets), participants rated whether the sentence was semantically correct. Finally, they were asked to recall either the subject or the object of the sentences in the order presented. Baddeley, Logie, Nimmo-Smith, and Brereton (1985) Regardless of whether participants were asked to remember the subject of the sentences in the order given or the object of each sentence, a point was given for the correct recall in the order given. The maximum score was 20 points.
Perceptual speed Letter Comparison Participants compare two strings of letters and determine whether they are the same or different. They have 30 s to compare as many as they can. The strings are 3-, 6-, and 9- letters long. Salthouse and Babcock (1991) The final score is the total number correct across all three tasks (i.e., 3-, 6-, and 9- item)
Number Comparison The same procedure as letter comparison but with strings of numbers. The time limit is 30 s.
Pattern Comparison The same procedure as letter comparison but with strings of patters. The time limit is 30 s.
Open in a new tab

References

  1. Akaike H. Factor analysis and AIC. Psychometrika. 1987;52:317–332. [Google Scholar]
  2. Anstey KJ, Hofer SM, Luszcz MA. Cross-sectional and longitudinal patterns of dedifferentiation in late-life cognitive and sensory function: The effects of age, ability, attrition, and occasion of measurement. Journal of Experimental Psychology: General. 2003a;132:470–487. doi: 10.1037/0096-3445.132.3.470. [DOI] [PubMed] [Google Scholar]
  3. Anstey KJ, Hofer SM, Luszcz MA. A latent growth curve analysis of late-life sensory and cognitive function over 8 years: Evidence for specific and common factors underlying change. Psychology and Aging. 2003b;18:714–726. doi: 10.1037/0882-7974.18.4.714. [DOI] [PubMed] [Google Scholar]
  4. Baddeley A, Logie R, Nimmo-Smith I, Brereton N. Components of fluent reading. Journal of Memory and Language. 1985;24:119–131. [Google Scholar]
  5. Bentler PM. Comparative fit indices in structural models. Psychological Bulletin. 1990;107:238–246. doi: 10.1037/0033-2909.107.2.238. [DOI] [PubMed] [Google Scholar]
  6. Browne MW, Cudeck R. Alternative ways of assessing model fit. In: Bollen KA, Long JS, editors. Testing structural equation models. Newbury Park, CA: Sage; 1993. pp. 136–162. [Google Scholar]
  7. Bugg JM, Zook NA, DeLosh EL, Davalos DB, Davis HP. Age differences in fluid intelligence: Contributions of general slowing and frontal decline. Brain and Cognition. 2006;62:9–16. doi: 10.1016/j.bandc.2006.02.006. [DOI] [PubMed] [Google Scholar]
  8. Cohen J, Cohen C. Applied multiple regression/correlation analysis for the behavioral sciences. Hillsdale, NJ: Erlbaum; 1975. [Google Scholar]
  9. Daneman M, Merikle P. Working memory and language comprehension: A meta-analysis. Psychonomic Bulletin & Review. 1996;3:422–433. doi: 10.3758/BF03214546. [DOI] [PubMed] [Google Scholar]
  10. Ekstrom R, French J, Harman H, Dermen D. Manual for kit of factor-referenced cognitive tests. Lawrenceville, NJ: Educational Testing Service; 1976. [Google Scholar]
  11. Ferrer E, Salthouse TA, McArdle JJ, Stewart WF, Schwartz BS. Multivariate modeling of age and retest in longitudinal studies of cognitive abilities. Psychology and Aging. 2005;20:412–422. doi: 10.1037/0882-7974.20.3.412. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Folstein MF, Folstein SE, McHugh PR. “Mini-mental state”: A practical method for grading the cognitive state of patients for the clinician. Journal of Psychiatric Research. 1975;12:189–198. doi: 10.1016/0022-3956(75)90026-6. [DOI] [PubMed] [Google Scholar]
  13. Ghisletta P, de Ribaupierre A. A dynamic investigation of cognitive dedifferentiation with control for retest: Evidence from the Swiss Interdisciplinary Longitudinal Study on the Oldest-old. Psychology and Aging. 2005;20:671–682. doi: 10.1037/0882-7974.20.4.671. [DOI] [PubMed] [Google Scholar]
  14. Ghisletta P, Lindenberger U. Age-based structural dynamics between perceptual speed and knowledge in the Berlin Aging Study: Direct evidence for ability dedifferentiation in old age. Psychology and Aging. 2003;18:696–713. doi: 10.1037/0882-7974.18.4.696. [DOI] [PubMed] [Google Scholar]
  15. Hedden T, Lautenschlager G, Park DC. Contributions of processing ability and knowledge to verbal memory tasks across the adult life-span. Quarterly Journal of Experimental Psychology. 2005;58A:169–190. doi: 10.1080/02724980443000179. [DOI] [PubMed] [Google Scholar]
  16. Hertzog C. Influences of cognitive slowing on age differences in intelligence. Developmental Psychology. 1989;25:636–651. [Google Scholar]
  17. Hertzog C, Dixon RA, Hultsch DF, MacDonald SWS. Latent change models of adult cognition: Are changes in processing speed and working memory associated with changes in episodic memory? Psychology and Aging. 2003;18:755–769. doi: 10.1037/0882-7974.18.4.755. [DOI] [PubMed] [Google Scholar]
  18. Horn JL, McArdle JJ. A practical and theoretical guide to measurement invariance in aging research. Experimental Aging Research. 1992;18:117–144. doi: 10.1080/03610739208253916. [DOI] [PubMed] [Google Scholar]
  19. Jefferies E, Lambon Ralph MA, Baddeley AD. Automatic and controlled processing in sentence recall: The role of long-term and working memory. Journal of Memory and Language. 2004;51:623–643. [Google Scholar]
  20. Johnson RE. Aging and the remembering of text. Developmental Review. 2003;23:261–346. [Google Scholar]
  21. Kemper S, Liu C. Eye movements of young and older adults during reading. Psychology and Aging. 2007;22:84–93. doi: 10.1037/0882-7974.22.1.84. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Kintsch W, van Dijk TA. Toward a model of text comprehension and production. Psychological Review. 1978;85:363–394. [Google Scholar]
  23. Kyllonen P, Christal R. Reasoning ability is (little more than) working memory capacity?! Intelligence. 1990;14:389–433. [Google Scholar]
  24. Light LL. The organization of memory in old age. In: Craik FIM, Salthouse TA, editors. Handbook of cognition and aging. Hillsdale, NJ: Erlbaum; 1992. pp. 111–166. [Google Scholar]
  25. McArdle JJ, Ferrer-Caja E, Hamagami F, Woodcock RW. Comparative longitudinal structural analyses of the growth and decline of multiple intellectual abilities over the life span. Developmental Psychology. 2002;38:115–142. [PubMed] [Google Scholar]
  26. McArdle JJ, Hamagami F, Meredith W, Bradway KP. Modeling the dynamic hypotheses of Gf-Gc theory using longitudinal life-span data. Learning and Individual Differences. 2000;12:53–79. [Google Scholar]
  27. Neale MC. Mx: Statistical modeling. 4th ed. Richmond, VA: Department of Psychiatry; 1997. [Google Scholar]
  28. Park DC, Lautenschlager G, Hedden T, Davidson NS, Smith AD, Smith PK. Models of visuospatial and verbal memory across the adult life span. Psychology and Aging. 2002;17:299–320. [PubMed] [Google Scholar]
  29. Rebok GW, Carlson MC, Langbaum JBS. Training and maintaining memory abilities in healthy older adults: Traditional and novel approaches. Journal of Gerontology. 2007;62B(Special Issue I):53–61. doi: 10.1093/geronb/62.special_issue_1.53. [DOI] [PubMed] [Google Scholar]
  30. Salthouse TA. The processing-speed theory of adult age differences in cognition. Psychological Review. 1996;103:403–428. doi: 10.1037/0033-295x.103.3.403. [DOI] [PubMed] [Google Scholar]
  31. Salthouse TA. Localizing age-related individual differences in a hierarchical structure. Intelligence. 2004;32:541–561. doi: 10.1016/j.intell.2004.07.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Salthouse TA, Babcock RL. Decomposing adult age differences in working memory. Developmental Psychology. 1991;27:763–776. [Google Scholar]
  33. Schaie KW. Schaie-Thurstone adult mental abilities test. Palo Alto, CA: Consulting Psychologists Press; 1985. [Google Scholar]
  34. Schaie KW. Developmental influences on adult intelligence. Oxford, England: Oxford University Press; 2005. [Google Scholar]
  35. Siedlecki KL. Investigating the structure and age invariance of episodic memory across the adult lifespan. Psychology and Aging. 2007;22:251–268. doi: 10.1037/0882-7974.22.2.251. [DOI] [PubMed] [Google Scholar]
  36. Sliwinski M, Buschke H. Cross-sectional and longitudinal relationships among age, cognition, and processing speed. Psychology and Aging. 1999;14:18–33. doi: 10.1037//0882-7974.14.1.18. [DOI] [PubMed] [Google Scholar]
  37. Sliwinski MJ, Hofer SM, Hall C. Correlated and coupled cognitive change in older adults with and without clinical dementia. Psychology and Aging. 2003;18:672–683. doi: 10.1037/0882-7974.18.4.672. [DOI] [PubMed] [Google Scholar]
  38. Smith GE, Housen P, Yaffe K, Ruff R, Kennison RF, Mahncke HW, Zelinski EM. A cognitive training program based on principles of brain plasticity: Results from the Improvement in Memory with Plasticity-based Adaptive Cognitive Training (IMPACT) Study. Journal of the American Geriatrics Society. 2009;57:594–603. doi: 10.1111/j.1532-5415.2008.02167.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Sternberg RJ, Berg CA. What are theories of adult intellectual development theories of? In: Schooler S, Schaie KW, editors. Cognitive functioning and social structure over the life course. Norwood, NJ: Ablex Publishing Corporation; 1987. pp. 3–23. [Google Scholar]
  40. Stine-Morrow EAL, Miller LMS, Gagne DD, Hertzog C. Self-regulated reading in adulthood. Psychology and Aging. 2008;23:131–153. doi: 10.1037/0882-7974.23.1.131. [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. van der Linden M, Hupet M, Feyereisen P, Schelstraete M.-A., Bestgen Y, Bruyer R, Lories G, Al Ahmadi A, Seron X. Cognitive mediators of age-related differences in language comprehension and verbal memory performance. Aging, Neuropsychology, and Cognition. 1999;6:32–55. [Google Scholar]
  42. Verhaeghen P, Salthouse TA. Meta-analyses of age-cognition relations in adulthood: Estimates of linear and nonlinear age effects and structural models. Psychological Bulletin. 1997;122:231–249. doi: 10.1037/0033-2909.122.3.231. [DOI] [PubMed] [Google Scholar]
  43. Wilson RA, Beckett LA, Barnes LL, Schneider JA, Bach J, Evans DA, Bennett DA. Individual differences in rates of change in cognitive abilities of older persons. Psychology and Aging. 2002;17:179–193. [PubMed] [Google Scholar]
  44. Wilson RS, Li Y, Bienias JL, Bennett DA. Cognitive decline in old age: Separating retest effects from the effects of growing older. Psychology and Aging. 2006;21:774–789. doi: 10.1037/0882-7974.21.4.774. [DOI] [PubMed] [Google Scholar]
  45. Zacks RT, Hasher L, Li KZH. Human memory. In: Craik FIM, Salthouse TA, editors. Handbook of cognition and aging. 2nd ed. Mahwah, NJ: Erlbaum; 2000. pp. 293–358. [Google Scholar]
  46. Zelinski EM, Burnight KP. Sixteen-year longitudinal and time-lag changes in memory and cognition in older adults. Psychology and Aging. 1997;12:503–523. doi: 10.1037//0882-7974.12.3.503. [DOI] [PubMed] [Google Scholar]
  47. Zelinski EM, Gilewski MJ. Effects of demographic and health variables on Rasch scaled cognitive scores. Journal of Aging and Health. 2003;15:435–464. doi: 10.1177/0898264303253499. [DOI] [PubMed] [Google Scholar]
  48. Zelinski EM, Gilewski MJ, Schaie KW. Individual differences in cross-sectional and three-year longitudinal memory performance across the adult lifespan. Psychology and Aging. 1993;8:176–186. doi: 10.1037//0882-7974.8.2.176. [DOI] [PubMed] [Google Scholar]
  49. Zelinski EM, Kennison RF. Not your father’s test scores: Cohort reduces psychometric aging effects. Psychology and Aging. 2007;22:546–557. doi: 10.1037/0882-7974.22.3.546. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Zelinski EM, Kennison RF, Hall AW, Lewis KL. Convergence between cross-sectional and longitudinal studies: Cohort matters. In: Hertzog C, Bosworth H, editors. Cognition in aging methodologies and applications: Festschrift for K. Warner Schaie. Washington, DC: American Psychological Association; 2009. pp. 101–118. [Google Scholar]
  51. Zelinski EM, Lewis KL. Adult age differences in multiple cognitive functions: Differentiation, dedifferentiation, or process-specific change? Psychology and Aging. 2003;18:727–745. doi: 10.1037/0882-7974.18.4.727. [DOI] [PubMed] [Google Scholar]
  52. Zelinski EM, Light LL, Gilewski MJ. Adult age differences in memory for prose: The question of sensitivity to passage structure. Developmental Psychology. 1984;20:1181–1192. [Google Scholar]
  53. Zelinski EM, Stewart S. Individual differences in 16-year memory changes. Psychology and Aging. 1998;13:622–630. doi: 10.1037//0882-7974.13.4.622. [DOI] [PubMed] [Google Scholar]

Articles from The Journals of Gerontology Series B: Psychological Sciences and Social Sciences are provided here courtesy of Oxford University Press

RESOURCES