Presented at the Cognitive Aging Conference, Atlanta, Georgia, April 1994

There is increased attention in the field to the relationships among activity and age-related changes in cognition (Clarkson-Smith & Hartley, 1990; DeCarlo, 1974; Hultsch, Hammer & Small, 1993). However, the types of activities and frequency of engagement in these activities by different age groups have yet to be established (Salthouse, 1991). A questionnaire pertaining to different types of everyday activities was administered to 199 subjects (101 older adults 61 to 87 years and 98 young adults 18 to 40). The questionnaire was designed to survey a variety of everyday activities that may relate to cognitive performance. Eight factors were extracted that explained the relationships among the activity variables for the entire sample. The eight factors are: physical activity, verbal activity, social activity, cultural activity, mental fitness activity, news seeking activity, travel activity, and entertainment activity.

In order for group differences on any construct to be meaningful, it must first be established that constructs are reliably measured within each group (Horn & McArdle, 1992). A multiple group factorial invariance test was performed across the two age groups. The eight activity factors extracted for the entire sample failed to demonstrate invariance suggesting that separate factor solutions are needed for each age group. This analysis suggests that old and young subjects organize their activities differently. The possibility that different scales are needed to assess constructs in old and young age groups is discussed. The relationship between cognitive abilities and activity levels is discussed for the oldest subjects.

Recently there has been an increase in reports that include multiple factors that may relate to age-related change in cognitive performance. For example, we know that physical fitness (Clarkson-Smith & Hartley, 1990; DeCarlo, 1974), as well as everyday activities relate to performance on some cognitive tasks. Hultsch, Hammer and Small (1993) have recently reported on some relationships among health, activity life style and cognitive performance in a group of adults aged 55-86 years. They measured cognitive performance with eight different indicators (such as working memory and word recall) and found that health and activity were significant contributors to variance in performance in the cognitive tasks. Their results suggest, for example, that keeping oneself "mentally fit" is one possible key to successful cognitive aging. Similarly, DeCarlo (1974) suggested that the physically fit tend toward better performance (see also Perlmutter & Nyquist, 1990). Yet rarely are these factors considered in the bulk of the cognitive literature. Furthermore, even when such factors are considered, it is often in studies considering only a small sets of tasks or tasks that tap a single domain.

Thus, in the 'perfect world' we would want to consider all factors that relate to cognitive performance, and we would want to measure each and every cognitive activity possible. Furthermore, every one of our measures would be perfectly reliable and valid. We are far from such a utopian goal, although in some ways we are working toward this. For example, Clarkson-Smith and Hartley (1990), by examining structural relations, report a relationship between physical activity and cognitive ability. What differs in their report is the causal nature of their investigation. The use of structural modeling (e.g., Jö reskog & Sö rbom, 1979) has the advantage of allowing a comparison among hypothesized and competing, alternative models. Structural modeling also has the advantage of comparing changes in patterns of relationships among different samples or groups (Horn and McArdle, 1992). This is a particularly useful tool for studying changes among age groups, allowing us to uncover any differences in the underlying measurement of variables in different samples (i.e. construct validity).

Although the relationship between everyday activities (such as physical activity) and cognition are under exploration; the types and frequencies of different activities performed by older and younger adults have yet to be established (Arbuckle, Gold & Andres, 1986; Salthouse, 1991, p.124). Since frequency of engagement in different types of activities across ages is not known at this time, this report has an immediate goal of providing some information to this regard.

A second, and in some ways more primary goal of this research, is to assure valid measurement as we assess activity life styles across the adult life span. Specifically, we question the use of a single measurement instrument for both young and old adults for understanding relationships among such things as activity engagement and cognitive change in adulthood.

Subjects

Ninety-eight young adults ranging in age from 18 to 40 (mean=22.3, SD=4.9) and 101 older adults ranging in age from 61 to 87 (mean=73.8, SD=6.4) completed the activities questionnaire in group settings with no time constraints. The young adults, consisting of 29 males and 69 females, were volunteers from undergraduate psychology classes at the University of Southern California, the University of Calgary and from the greater Los Angeles community. The older adults, consisting of 29 males and 72 females, were community-dwelling volunteers in the greater Los Angeles community. The older subjects were given a battery of cognitive tests in addition to the activities questionnaire.

Measures

The activities questionnaire, consisting of 34 items, was initially created with an a priori ten factor structure in attempt to assess a wide variety of life-style activities in which old as well as young persons engage. Subjects were encouraged to place a check mark at any point on a continuum to best indicate the frequency with which they presently engage in a specified activity. Items were coded from 1 (lowest frequency) to 9 (highest frequency) including checks in between anchors. An example of a typical response pattern is as follows:

never yearly monthly weekly daily

1 2 3 4 5 6 7 8 9

The older subjects were also given a cognitive battery consisting of tests from the Gf-Gc Sampler (Horn, 1985) which includes measures of Fluid ability (Gf), Crystallized ability (Gc), short-term memory (Gsm), and speed (Gs).

Whenever it is important to ask whether or not two groups differ on any construct, it is also important to determine whether or not the same construct is being measured in each of the groups. If, for example, a mean difference across age groups is hypothesized, it should first be established that the construct being measured is the same for each age group. If a similar factor structure can be found for groups being compared, then it can be assumed that there is invariance of measurement across those groups. Mean differences on the construct are then, and only then, meaningful (Horn & McArdle, 1992). Because many studies in aging are interested in comparing the activity levels of both old and young subjects, a factorial invariance test was performed to ensure construct validity across age groups. A factor solution was first established for all subjects analyzed together, then this factor solution was imposed on both old and young subjects separately in a two group LISREL model.

Factor Solution for the Entire Sample

An inspection of the scree plot of eigenvalues (Cattell, 1966) suggested an eight factor solution. Because correlations between the activity constructs were not expected to be zero, an oblique factor rotation method (PROMAX) was used to achieve maximum simple structure (Thurstone, 1947). It was necessary to combine some items into two-item parcels due to high item intercorrelations creating two new composite variables; SPORT (two items having to do with participation in sports), and COOK (two items having to do with creative cooking). Combining highly correlated items ensures nonlinearity and independence of items (Cattell, 1952). The eight factors extracted were: Physical Activity, Verbal Activity, Social Activity, Cultural Activity, Mental versus Physical Fitness, Current Events, Travel, and Entertainment Activity. Items, factor loadings, factor intercorrelations and alpha reliability coefficients are reported in Table 1 for this analysis on the entire sample.

Invariance Test

First, all items were standardized across the entire sample. Then, covariance matrices for the young group and the older group were extracted separately from these standardized items to establish data for LISREL input. This procedure makes it possible for means and variances to vary from one group to another and is essential for assessing meaningful multiple group comparisons (Cudeck, 1989; Cunningham, 1991). A multiple group LISREL model was then proposed where the factor configurations were required to be invariant across the two groups (i.e., salient loadings were required to be salient, hyperplane loadings were set to zero for each factor). Factor covariances and item uniquenesses were allowed to be free. This is a test of "configural" invariance which is the least stringent of invariance tests (Horn & McArdle, 1992; Horn, McArdle & Mason, 1980). The proposed model (chi-square = 1697, df=810, GFI=.678) was rejected suggesting that the factor configurations cannot be considered to be the same across age groups. Because the invariance test failed, separate factor solutions were explored for each age group independently.

Factor Solution for the Young Subjects

An inspection of the scree plot of the eigenvalues suggested a seven factor solution for the young subjects. Again, oblique rotation (PROMAX) was used to achieve maximum simple structure for the solution. The seven factors extracted include: Physical Activity, Verbal Activity, Social Activity, Cultural Activity, Community Involvement, Current Events, and Mental Activity. Three items were left out of the final solution due to multiple loadings (loading on three or more factors at once) and/or skewness within the age group. Items, factor loadings, factor intercorrelations, and alpha reliability coefficients are reported in Table 2.

Factor Solution for the Older Subjects

An inspection of the scree plot of the eigenvalues suggested a seven factor solution for this group as well. Oblique rotation (PROMAX) yielded the following seven factor solution with maximum simple structure: Physical Activity, Verbal Activity, Cultural Activity, Community Involvement, Social Activity, Domestic Activity, and Mental Activity. Four items were left out of the final solution due to multiple loadings and/or skewness within the age group. Items, factor loadings, factor intercorrelations, and alpha reliability coefficients are reported in Table 3.

Correlations of Activities and Cognitive Performance

The extent to which the cognitive abilities correlated with these seven factors is reported in Table 4. The Verbal Activities factor correlated significantly with the common word analogies test (a measure of Fluid ability) and the picture number test (a measure of short-term memory). The Cultural Activities factor correlated significantly with the compare letters test (a measure of speed). The Social Activities factor correlated significantly with the cross out test (a measure of speed). The Mental Activities factor correlated with two separate vocabulary tests (both measures of Crystallized ability), the esoteric analogies test (also a measure of Crystallized ability), and the common word analogies test (a measure of Fluid ability). Males scored higher on the Verbal Activities factor than did females, and females scored higher on the Domestic Activities factor than did males. Within this older sample there were no significant age correlations with any of the activity factors.

The model which forced factor configurations to be the same across age groups did not fit the data well. This suggests that different factor configurations are needed to adequately describe the various activities in which young and old subjects engage. This finding brings to light some important methodological considerations in both the design and use of questionnaires that attempt to assess similar constructs in samples of different ages.

The importance of considering issues of factorial invariance at the design stages of research cannot be underestimated. Questionnaires should be written with the idea of obtaining maximum variance for each question --- both old and young subjects should be able to respond to all items. Even if such care is taken in questionnaire construction, it is possible that groups of interest will organize, and even interpret, a priori constructs differently. Such is the case in the present study.

It is clear that both young and old subjects engage in physical activities, but older adults think of "aerobic exercise" and "attending exercise classes" differently than do younger subjects. Older adults don't view these types of physical activities in the same manner as they do an "active daily routine" or "engaging in sports" where younger subjects seem to make little distinction between aerobic exercise/exercise classes and an active daily routine/engaging in sports. Thus, the Physical Activities factor is slightly different for old and young subjects.

Similarly, both age groups engage in activities which involve verbal expression such as "telling stories", "giving speeches", and "writing papers", but older adults consider "writing in journals" an aspect of this verbal expression while younger adults view "writing in journals" as an aesthetic or culture building activity. Although the verbal expression construct is similar, it differs slightly across age groups.

For the youngest subjects, going out to see "live music" or engaging in the "performing arts" are both viewed as primarily social activities. For older adults, social activities include "discussing current events", "communicating thoughts and feeling to others", and "playing cards." The Social Activities factor takes on quite a different flavor depending on which group is being studied.

Perhaps the most interesting group difference revolves around what are interpreted as "Mental Activities." Older adults simply rate themselves as being "mentally active" and report doing "crossword puzzles." Younger adults, on the other hand, include "card playing", watching "TV news", and "watching many hours of TV per week" as mental activities. "TV watching", for older subjects, is seen as an activity that keeps one from engaging in other worthwhile endeavors such as cultural activities and in no way resembles a "mental activity" (note that "hours per week of TV watching" loads -.52 on the Cultural Activities factor in the older age group). Mental Activities are surprisingly different constructs for old and young subjects.

Another important distinction between these two age groups is the appearance of a "domestic activities" factor only in the older group and the appearance of a "current events" factor only in the young group. These unique factors indicate that the importance of certain leisure activities varies with age significantly (see also, Nussbaum, Thompson & Robinson, 1989).

It is not our attempt to suggest that we have arrived at a reliable, factorially invariant activities questionnaire. Much work needs to be done before a replicable, reliable factor structure can be realized that will demonstrate invariance across age groups. However, it IS our attempt to bring to the attention of developmental researchers the problems that may occur when constructs are not reliably measured across groups of interest. Group mean comparisons CANNOT be made if constructs are not invariant across groups.

We find here that certain types of activities are related to performance in some cognitive tasks in the older group. For example, consistent with the findings of Hultsch et al (1993), mental activity is highly related to measures of both fluid and crystallized abilities. We suspect that those individuals who engage actively in mental or physical activities will be more likely to maintain optimal cognitive performance over time. The causal nature of this relationship, however, can not be determined at this time. Regardless of the direction of causality, maintenance of higher activity levels in a variety of domains appear to relate to maintained cognitive performance with age (see also Arbuckle, Gold & Andres, 1986; Craik , Byrd & Swanson, 1987). We suggest that a carefully designed scale be included in studies investigating age-related changes in cognition. Including such a measure will allow us to consider the proportion of variance seen in different cognitive tasks that can be accounted for by different types of everyday activities.