2 Empirical evaluation of three research questions
2.2 Individual differences in verbal learning in old age
2.2.1 An empirical analysis of verbal learning in old age
The data used in the sequel come from the Zurich Longitudinal Study on Cognitive Aging (ZULU), an ongoing longitudinal study on cognitive and learning abilities of elderly persons in Switzerland (Zimprich et al., in revision). At first measurement occasion (T1: 2005), the sample of the Zurich Longitudinal Study on Cognitive Aging (ZULU) comprised 364 participants who where between 65 to 80 years of age (Mean age: 73 years, SD = 4.4 years; 46% women). The majority of the sample was married and resided with others. On average, participants reported about 13 years of formal education. For the sample, there were no signs of cognitive impairments or pronounced depressive affect. The majority of participants judged their health as ”good” and, in addition, no participant reported any severe hearing or vision difficulties. Part of the cognitive testing protocol in ZULU were three measures of processing speed (Number Comparison, Identical Pictures, Letter Digit Substitution), a verbal learning measure that comprised five trials of a word list recall task, and three measures of memory (Paired Associates, Story Recall, Picture Memory).
Number Comparison (Ekstrom, French, Harman, & Dermen, 1976) required participants to compare as rapidly as possible whether two numbers presented on the computer screen were identical or not. Scored was the number of correct answers, which could range between 0 and 60. After two practice items during the instruction phase, the time to work on the task was 90 seconds. Identical Pictures (Ekstrom et al., 1976) required participants to choose one out of five objects that was identical to a reference object as rapidly as possible. Scored was the number of correctly answered items, which could range from 0 to 60. After two practice items during the instruction phase, the time to work on the task was 90 seconds. Eventually, Letter Digit Substitution consisted of 75 items. For each item, a table that assigned five different letters to the numbers one to five was displayed on the top of the computer screen. Below the table, a single letter was presented together with a question mark. Participants were supposed to press the number that belonged to the single letter according to the presented coding table. For each item, there was a different coding table. Scored was the number of correctly answered items, which could range from 0 to 75. After two practice items, participants had 90 seconds to work on the task.
Verbal Learning was assessed by five consecutive trials of a word list recall task. The task comprised 27 meaningful, but unrelated words that were taken from the German Version of Rey Auditory Verbal Learning Test (Helmstädter, Lendt, & Lux, 2001). The 27 words
appeared on a computer screen at a rate of two seconds each and participants were required to read them aloud. After the presentation of all 27 words, participants were asked to recall as many words as possible in any order. This procedure was repeated five times, with the order of words being different for each trial. At each trial, the number of correctly recalled words was scored, ranging between 0 and 27.
Paired Associates comprised 12 semantically unrelated word pairs taken from the German version of the Wechsler Memory Scale-Revised (WMS-R: Härting et al., 2000) and from the Munich Verbal Memory Test (MVGT: Ilmberger, 1988). After two examples during instruction, word pairs were presented for four seconds each and participants had to read them aloud. Following a pause of one second, the next word pair was displayed. After presentation of all 12 word pairs, only the first word of a pair appeared on the screen as a cue and the second one was replaced by a question mark (e.g. salad - ?), using a different order than during encoding. Scored was the number of correctly recalled target words, which could range from 0 to 12. Story Recall consisted of story A of the Logical Memory subtest of the German version of the Wechsler Memory Scale-Revised (Härting et al., 2000). The 66-word story was read by the experimenter during 60 seconds. Participants were asked to listen closely and, when the story was finished, to immediately recall as many details as possible. Scored was the number of correctly recalled propositions, which could range from 0 to 25. Finally, Picture Memory encompassed 12 pictures taken from the Nuremberg Age Inventory (Nürnberger-Alters-Inventar: Oswald & Fleischmann, 1999). For each item, a picture of a simple object for 2.75 seconds and participants were required to name the shown object aloud (e.g., “apple”). Followed by a pause of one second, the next picture was displayed. Immediately after presentation of all 12 pictures, participants were asked to verbally recall as many of the seen objects as possible. Scored was the number of correctly recalled objects, which could range between 0 and 12.
All analyses reported below were conducted using Mx (Neale, Boker, Xie, & Maes, 2003). Nonlinear learning models were specified as structured growth models (Browne, 1993; Browne & Du Toit, 1991). The absolute goodness-of-fit of models was evaluated using the χ2
-test and two additional criteria, the Comparative Fit Index (CFI) and the Root Mean Square Error of Approximation (RMSEA). Values of the CFI above .95 are considered to be adequate, whereas for the RMSEA values less than .06 indicate an acceptable model fit (cf. Hu & Bentler, 1999). In comparing the relative fit of nested models, we used the χ2-difference
test where appropriate. Due to its dependency on sample size and due to the fact that we also wanted to compare non-nested models, we mainly relied on calculating 90% RMSEA confidence intervals for the models estimated (MacCallum et al., 1996). Because the RMSEA is virtually independent of sample size, the comparison of RMSEA confidence intervals, that is, whether they do or do not overlap, provides an effective, alternative method of assessing relative model fit of nested and non-nested models. Throughout, we refer to a significance level of p < .05 if a parameter estimate is denoted as statistically significant.