Age and Ageing 1996:25:72-80
Activity Levels and Cognitive Functioning in
an Elderly Community Sample
HELEN CHRISTENSEN, AILSA KORTEN, A. F. JORM,
A. S. HENDERSON, RUTH SCOTT, A. J. MACKINNON
Summary
The influence of self-reported and informant-reported activity levels on Crystallized Intelligence, Fluid
Intelligence, Memory and the Mini-Mental State Examination was investigated in a sample of 858 communitydwelling elderly subjects. Both self-reported and informant-reported activity levels explained variance beyond
that accounted for by sex, sensory functioning, activities of daily living, medical conditions, current health
problems and education. Age accounted for additional variance once activity and the other contextual variables
were entered. Interaction effects indicated that inactivity was associated with poorer performance on fluid
intelligence in older rather than younger elderly subjects and that inactivity was predictive of poor crystallized
intelligence at younger ages. Higher informant-rated activity levels moderated the effects of education, so that
higher activity offset effects associated with low education on memory tasks. The amount of variance explained
by activity levels was modest.
Introduction
There is a widespread belief, particularly among elderly
people, that activity prevents cognitive decline in old
age and protects against the onset of dementia. This
view is reflected in the philosophy of many organizations, such as the University of the Third Age [1],
which claim that mental activity is crucial in preventing
cognitive impairment. The importance of an active
lifestyle has also received support within the scientific
community [2-6]. A recent editorial in the British
Medical Journal [7] stated that 'stimulating mental
activity is worthwhile' (p. 952) in protecting against
dementia and called for further experimental investigation of the issue.
The present study investigated the role of activity on
cognitive performance in a large community sample of
elderly subjects. The primary aim was to determine
whether differences in cognitive performance could be
accounted for by activity, particularly when the
influence of other variables, such as sex, sensory
functioning [8], health [9], education [7] and disability
were taken into account. These variables have been
found previously to be predictive of cognitive performance. A secondary aim was to examine whether age
continued to be predictive of cognitive performance
once account had been taken of activity, health,
education, sex, disability and sensory functioning.
This allowed investigation of the hypothesis that
activity and other variables account for age-associated
changes in cognitive functioning. This hypothesis
predicts that once the effects of activity are accounted
for, no further age effects should be present. A third
aim was to examine the hypothesis that activity was
more influential in older than younger subjects.
Christensen and Mackinnon [10] and Hultsch et al.
[6] noted that maintaining high levels of activity may be
of greater importance in very elderly subjects than in
younger elderly people. Furthermore, because earlier
findings suggested that activity levels were not independent of education, and that activity may offset to
some degree education disadvantage [10, 11], the
interactions between activity and education were also
examined.
Methods
The sample: The subjects were a sample of elderly persons
living in the community in the Australian city of Canberra
and adjacent town of Queanbeyan". Elderly people were
sampled from the electoral roll which is a register of eligible
voters. Registration on the electoral roll is compulsory for all
Australian citizens aged 18 or more. The sample was drawn at
random within three age strata to give an equal achieved
sample of men and women. For each sex, there were three age
strata (70-74 years, 75-79 and more than or equal to 80
years), with each stratum sampled to give an achieved sample
which was proportional to the number of individuals in the
population at that age group.
The persons sampled were sent a letter inviting participation in the survey and then approached at home by trained
interviewers. The purposes of the study and what was asked
of the participant were explained orally. If a person was
unable to give informed consent owing to disability, the
principal caring relative was asked. If an interview was
granted, the subject was asked at the end to nominate a close
ACTIVITY LEVELS AND COGNITIVE FUNCTIONING
relative or friend who knew him/her well and would be able to
describe his/her present circumstances. An interview was
then sought from the informant nominated. Thirty-one per
cent of those approached refused to participate. This refusal
rate is similar to those recently obtained in other community
samples. Sixty-five per cent of refusers were women, but the
age range of the refusers did not differ from that of the
participants. The sample used in the present paper consisted
of 858 subjects aged 70-89 years of age who completed data
on the self-activity scales, the Mini-Mental State Examination (MMSE) [12] and sensory functioning. Of these, 664
subjects had informant data available. Subjects were divided
into four age groups, 70-74, 75-79, 80-84 and 85-89 years.
General procedure: Subjects were interviewed in their own
homes by trained interviewers. Information necessary for the
diagnosis of dementia and depression was collected, as well as
information on social background, personality, social support,
self-reported cognitive function and the use of services.
Prevalence rates for depression and dementia have been
reported elsewhere, as have data on health, neuroticism and
cognitive functioning [13-18].
Measures of inactivity: Both informants and subjects were
asked about how often 'these days' the subject read a
newspaper, engaged in physical activity (such as sport,
walking, heavy gardening, cleaning), was involved in interests
and hobbies, spent time sitting around without doing very
much (e.g. resting, dozing, watching the TV without caring
what the programme was), spent time in planned activities,
and had a daily nap. Each of these activities was rated on a
scale with from 2 to 4 points. For example, subjects were
asked 'How often do you read a newspaper, book or magazine
these days?' (0 = every day, 1 = m o s t days, 2 = once a week,
3 = less than once a week). The six items were combined to
form a scale. Five items had 4 points and one item had 2
points, making 16 the maximum score for the scale. Scores
were pro-rated if up to three of the six items were completed.
If fewer than three items were completed, the scale was listed
as missing. Informants' reports were summed in the same
manner to give scores from 0 to 16, with higher scores
indicating greater inactivity. The correlation between informant and subject total scores on the activity measures was 0.74
(n = 664), showing strong agreement. Cronbach alpha for the
self-report scale was 0.64 (n = 841) and for the informant scale
was 0.69 (n = 603).
Because the items on the scale measured a variety of
physical, social and mental activities, the scales were
subjected to principal components analysis to confirm that a
general activity factor was measured. Two factors with
eigenvalues greater than 1 were revealed on the self-report
scale. All six items loaded on the first factor (eigenvalue of
2.27) with 38% of variance accounted for. Unrotated loadings
for the items were as follows: read newspaper 0.44; physical
activities 0.64; interests and hobbies 0.62; time sitting around
0.78, planned activities 0.77; and have a nap 0.28. Two items
loaded on the second factor (eigenvalue 1.02) accounting for a
further 17% of variance. These were 'time sitting around' and
'have a nap'. The informant scale produced a similar pattern
of factors with factor 1 accounting for 42% of the variance.
These factor analyses confirmed that the scales measured a
general activity factor and that self-reported and informantreported scales had a very similar factor structure.
Measures of health: Two measures of health were used.
These were summary measures found useful in an earlier
study of this population [14]. In the present study, these
measures controlled for individuals having multiple health
problems and current multiple symptoms. More detailed
73
analysis of effects of specific health problems on cognitive
performance are reported elsewhere [14]. The first health
measure asked subjects to note whether they had ever had any
of 28 medical conditions such as stroke, heart attack, severe
psychiatric disorder (such as schizophrenia), diabetes, cataract, thyroid dysfunction, and so on. The number of medical
conditions reported by the subjects was aggregated to form
the medical condition scale. For the second measure, subjects
noted whether they suffered currently from any of 21
symptoms, such as cramps, breathing difficulties, indigestion,
and headaches. The number of complaints was summed to
obtain the second health score which was labelled current
health problems [14].
Measures of sensory functioning: Subjects were asked about
their eyesight and hearing on ten items. Two of the items
required overall ratings of corrected vision and hearing
['Would you say your eyesight (with glasses) is generally
good (score 4), fair (3), poor (2) or blind (1)?'; 'Would you say
your hearing (with a hearing aid) is generally good (3), fair (2)
or poor (1)?']. Other items referred to specific behaviours.
The questions were: 'Under the best conditions, can you see
to read newspapers?; letters and mail?; labels on medicines?;
street signs or bus numbers?'. 'Are you able to hear what
somebody is saying to you when there is background noise
from radio and television?; when others are talking?; out in the
street?; travelling?'. These items were scored 1 for 'no', 2 for
'yes, with difficulty' and 3 for 'yes'. In addition, the
interviewer rated the subject's hearing and vision at the end
of the interview. 'How well did the subject seem to hear you?'
(4 = very well, 3 = quite well, 2 = somewhat impaired,
4 = profoundly deaf)- 'How well was the subject able to see
the papers given to him/her?' (4 = very well, 3 =quite well,
2 = somewhat impaired (could not see photographs),
1 = interviewer had to read aloud or omit questions requiring
sight). All 12 items were summed to provide a score ranging
from 1 to 26. There were 194 (22%) subjects who achieved the
maximum score. A further 28% scored 24 or 25 points. Higher
scores indicated better sensory functioning.
Measures of Activities of Daily Living (ADL):
Each
informant and subject was asked about the subject's capacity
to get about when travelling, to walk about 400 metres
without pain or difficulty, to get in and out of bed, to get
dressed, bathed and showered, to get in and out of an
armchair, to dress completely without difficulty, to take care
of feet and toe nails and to get to the toilet. The eight items
were combined to form a scale. Each item was rated on a scale
with from 3 to 5 points. The items were summed separately
for subjects and informants to give scores ranging from 0 to
22, with high scores indicating disability. The correlation
between the informant and subject for the eight items was
0.88 indicating excellent agreement. The scale has face and
content validity. Moreover, scores on the ADL scale have
been found to be related to informant ratings of ability to live
independently, and, if not able, whether this was due to
physical disability or, to a lesser extent, behavioural disability
[17].
Education: The total number of years of schooling and
post-school training was used to form the education measure.
Measures of cognitive functioning: The cognitive battery
consisted of tests of memory, crystallized and fluid intelligence, and speed, as well as the MMSE. Details of the tests
are reported elsewhere [13]. Briefly, the cognitive battery
consisted of 14 tasks including standard tests. They were: (1)
the National Adult Reading Test (NART) [19] which relies
on the reading of words that are not pronounced phonetically;
(2) The Symbol Letter Modalities Test (SLMT) which is a
74
H. CHRISTENSEN ET AL.
modified version of the (oral) Symbol Digit Modalities Test
[20]; (3) Choice and simple reaction time, each measured over
20 trials, with the choice reaction time task requiring subjects
to respond to one of two lights; (4) Vocabulary, which
consisted of three items from the Wechsler Adult Intelligence
Scale—Revised (WAIS-R) [21]; (5) Similarities, which
consisted of three items from the WAIS-R; (6) Information,
which consisted of four items identifying historicalfigures;(7)
Cube drawing, where subjects were required to produce a
copy of a cube from a drawn two-dimensional figure; (8)
Memory for a figure, which consisted of Item 1 (flags) from
the Visual Reproduction subtest of the Wechsler Memory
Scale [22]; (9) Recall of words, a task requiring the recall of
three words learned to criterion; (10) Address recall, which
required subjects to recall a name and address learned to
criterion 1 to 2 minutes earlier; (11) Face recognition; which
was an intentional face recognition task, similar to that used in
the Rivermead Behavioural Memory Test [23]; (12) Word
recognition, which was an incidental memory task in which
subjects were asked to answer 13 questions (including 1 buffer
item), each of which contained key words, and which were
later recognized from 24 items (12 targets and 12 distractors);
(13) Verbal Fluency, where subjects were asked to say aloud
as many animals as they could within 30 seconds; and [14]
MMSE, a brief screening test for dementia. Because of the
high correlations among outcome measures, data from the
cognitive tests (excluding the MMSE) were subjected to a
principal components analysis with varimax rotation. Three
factors which emerged with eigenvalues greater than 1.00
were retained in the factor solution. From their factor
loadings, these factors were labelled Crystallized Intelligence,
Fluid Intelligence and Memory. These factor scores were
used in the analyses thereby reducing the number of outcome
measures. Because not all subjects completed all tests in the
cognitive battery, only 703 subjects contributed to the factor
scores. There were 858 subjects who provided data from the
MMSE.
Results
Differences in education, activity, health, ADL and
cognitive functioning as a function of age: Table I
shows means and standard deviations in age groups
for sensory functioning, medical conditions, current
health problems, education, self- and informantreported measures of activities of ,daily living. The
statistical significance of mean differences of these
variables was examined using ANOVA. Sensory
functioning declined significantly across age groups,
while education and health did not. Both subject- and
informant-reported measures of A D L were higher in
the older age groups, indicating that disability, as
measured by the A D L scales, was greater in older age
groups.
The data for the self-reported inactivity and informant-reported inactivity scales were examined separately using multivariate analysis of variance (see Table
I). Overall, both self-reported and informant-reported
inactivity scores were higher in older age groups
indicating lower activity levels in the older age groups
[Wilks' lambda F(18, 2502) =2.98 for self-reported
activity, and Wilks' lambda F(18, 1788) = 2.12 for
informant-reported activity]. With the exception of
'read newspaper', all items on the self-report inactivity
scale indicated lower activity levels in the older age
groups. For the informant scale, all items except 'little
involvement in interests and hobbies' and 'daily nap
taking' indicated lower activity levels in older groups.
Cognitive test scores are also shown in Table I.
Differences among age groups for crystallized intelligence, fluid intelligence and memory factors were
examined using multivariate analysis of variance.
With the exception of crystallized intelligence, older
groups had lower factor scores than did younger
groups. ANOVA indicated that older age groups had
lower MMSE scores than did younger age groups.
Correlations among activity, age and cognitive
performance: Correlations between activity measures
and cognitive performance are shown in Table II. Both
self- and informant-reported activity measures correlated significantly, albeit weakly, with the three factors
and the MMSE. The correlations ranged from —0.32 to
—0.10. The strength of the correlations between the
cognitive factors and activity levels was assessed using
the T2 statistic described by Steiger [24]. These tests
examined whether activity was associated more
strongly with specific cognitive factors. Correlations
may have been expected to be weaker between activity
and crystallized intelligence since crystallized intelligence shows the least age-associated decline. Thus,
tests were undertaken to compare the magnitude of the
correlation between activity and crystallized intelligence with the correlation between activity and each of
the cognitive test scores. Taking the subject reports of
activity, the correlation with crystallized intelligence
was not significantly different from the correlation with
fiuid intelligence scores [T 2 (704)= 1.75, p>0.05] or
the correlation with Memory [T 2 (704) = 0.19,p>0.05],
but it was significantly weaker than the correlation of
activity with MMSE [T2(855) = 2.71, p < 0.01]. For the
informant reports of activity, the correlation with
crystallized intelligence was again not significantly
different from the correlation with memory
[T 2 (559) = 0.68, p>0.05] but was significantly weaker
than both the correlation of activity with fluid
intelligence [T 2 (561) = 3.87 = p <0.01] and with the
M M S E [T 2 (661) = 2.82, p<0.01].
Hierarchical regression analysis: Hierarchical regression analysis was employed to test three issues: (i)
whether activity made a significant contribution to
variance once account had been taken of sex, sensory
functioning, health and education, (ii) whether age
accounted for further variance after controlling for the
effects of activity and the other contextual variables;
and (iii) whether activity influenced cognitive performance to a greater extent in older elderly groups.
Separate regressions were undertaken for self-reported
and informant reported activity measures. Four dependent variables (crystallized intelligence, fluid intelligence, memory and the MMSE) were examined for
each of these two data sets. Thus, to take account of the
probable influence of contextual variables, the following variables were entered on steps 1 to 6 of the
ACTIVITY LEVELS AND COGNITIVE FUNCTIONING
75
Table I. Demographic, health, ADL, activity and cognitive variables as a function of age
Age (years)
Variable
Demographic, health, ADL,
Percentage of men
Sensory function (n = 858)*
Medical conditions (n = 858)
Current health problems (n = 858)
Education (years) (n = 855)
Mean ADL (n = 855)*
Mean Informant ADL (n = 639)»
Self-reported activity scale
Read newspaper (n = 841)
Physical activity (n = 841)*
Hobbies and interests (n = 841)#
Time sitting around (n = 841)
Planned activities (n = 841)*
Have nap (n = 841)#
Total score (n = 858)*f
Informant-reported activity scale
Read newspaper (n = 603)*
Physical activity (n = 603)*
Hobbies and interests (n = 603)*
Time sitting around (n = 603)*
Planned activities (n = 603)*
Have nap (n = 603)
Total score (n = 664)#f
Cognitive measures
Crystallized intelligence (n = 707)
Fluid intelligence (n = 707)»
Memory (n=707)*
MMSE (n = 858)*
70-74
(n = 371)
Mean (SD)
75-79
(n = 277)
Mean (SD)
80-84
(n=155)
Mean (SD)
85-89
(n = 55)
Mean (SD)
55
47
50
49
23.31 (3.22)
3.20(2.17)
3.24(2.92)
11.62(2.61)
1.25(2.34)
1.45(2.74)
21.91 (4.31)
3.15(2.00)
3.40(2.67)
11.20(2.54)
1.70(1.98)
2.12(2.54)
20.96 (4.40)
3.39(2.01)
3.65 (2.82)
11.14(2.45)
2.49 (2.42)
2.64 (2.36)
19.33 (5.92)
2.95(1.69)
4.05 (2.64)
10.95 (2.78)
3.82(3.41)
4.14(4.10)
1.22(0.73)
1.95(1.28)
2.47(1.43)
1.67(0.73)
2.59(1.09)
1.30(0.46)
4.06 (2.62)
1.31 (0.90)
2.24(1.46)
2.67(1.50)
1.85(0.82)
2.84(1.13)
1.39(0.49)
4.90 (3.05)
1.37(0.96)
2.44(1.58)
2.80(1.51)
1.86(0.89)
2.95(1.07)
1.41 (0.49)
5.28 (2.98)
1.39(1.02)
3.00(1.71)
3.16(1.52)
2.16(0.95)
3.22(1.17)
1.45(0.50)
6.44(3.38)
1.22(0.72)
2.12(1.33)
2.48(1.42)
1.69(0.75)
2.60(1.09)
1.47(0.50)
4.37 (2.92)
1.27(0.84)
2.39(1.56)
2.48(1.47)
1.82(0.81)
2.68(1.15)
1.51 (0.50)
4.75 (3.05)
1.28(0.81)
2.44(1.56)
2.84(1.51)
1.89(0.86)
2.89(1.19)
1.52(0.50)
5.30 (2.97)
1.70(1.40)
3.18(1.69)
2.78(1.58)
2.23 (1.10)
3.23 (1.17)
1.60(0.49)
6.49 (4.01)
0.03 (0.96)
0.28 (0.84)
0.19(0.86)
27.97(1.93)
-0.04(1.03)
-0.07 (0.98)
0.05 (0.99)
27.33 (2.63)
0.09(1.03)
-0.38(1.04)
-0.26(1.19)
26.72 (2.76)
-0.31 (1.05)
-0.53(1.41)
-0.35(1.21)
25.73 (4.22)
•p<0.05.
f Higher numbers contributed to total score because subjects with pro-rated scores were included.
analyses: sex, sensory functioning, ADL, past medical
conditions, current health problems and education.
Activity was entered on step 7, followed by age (step 8)
and then by the interaction variables of activity-byeducation (step 9), age-by-activity (step 10) and ageby-education (step 11). The interaction terms were
created using the products of the relevant variables. In
this analysis, product terms moderate the effects of
variables, and thus, for example, if the interaction of a
variable and age adds significantly to the regression
equation this implies that the effect of the predictor
variable varies as a function of age [25]. The results of
the regression analysis using self-reported activity
measures are shown in Table III.
Self-reported and informant-reported activity measures: The findings indicated that activity contributed
significantly, albeit modestly, to the amount of variance
accounted for in the four cognitive measures beyond
that contributed by sensory functioning, sex, ADL,
Table II. Correlation coefficients (r) between activity and
perfonnance on cognitive factors and the MMSE and age
Activity measure
Self report
Crystallized
intelligence
Fluid
intelligence
Memory
MMSE
Age
Informant report
r
No.f
r
No.t
-0.17*
707
-0.10*
564
-0.26*
-0.16*
-0.27*
0.21*
707
707
858
858
-0.32*
-0.14*
-0.22*
0.17#
564
562
664
664
•p<0.05.
t Number in group.
>p<0.05.
1 Sex
2 Sensory functioning
3 ADL
4 Medical conditions
5 Health problems
6 Education
7 Activity
8 Age
9 Activity x education
10 Age x activity
11 Age x education
0.05
0.05
0.00
0.00
0.08
0.82
-1.32
0.02
-0.03
1.26
-0.41
0.02
0.09
0.09
0.10
0.12
0.43
0.44
0.44
0.44
0.45
0.45
0.00
0.01*
0.00
0.00
0.00
0.17*
0.01»
0.00
0.00
0.01*
0.00
-0.13
-0.02
-0.11
-0.03
-0.01
0.56
0.88
0.06
0.13
-1.19
-0.46
0.17
0.23
0.31
0.31
0.31
0.36
0.38
0.42
0.42
0.42
0.42
R
0.03*
0.02*
0.05*
0.00
0.00
0.03*
0.02*
0.03*
0.00
0.01*
0.00
R change
Beta
R change
Beta
R
Fluid intelligence (n = 7O3)
Crystallized intelligence
(n = 7O3;).
0.14
0.04
0.06
-0.05
0.00
0.17
-0.59
-0.12
0.34
0.13
-0.21
Beta
0.12
0.18
0.19
0.19
0.19
0.21
0.24
0.29
0.30
0.30
0.30
R
Memory (n = 703)
0.02*
0.02*
0.00
0.00
0.00
0.01*
0.01*
0.03*
0.00
0.00
0.00
R change
0.07
0.09
-0.02
-0.02
0.07
-0.37
0.57
-0.19
0.20
-0.95
0.53
Beta
0.03
0.22
0.25
0.25
0.26
0.35
0.38
0.40
0.40
0.41
0.41
R
MMSE (n = 852)
0.00
0.05*
0.01*
0.00
0.01*
0.06*
0.02*
0.02*
0.00
0.00
0.00
R change
Table III. Summary of hierarchical regressions predicting crystallized intelligence, fluid intelligence, memory, and MMSE using self-reported activity scores
kj
z
w
Z
m
so
a>
X
a
X
ACTIVITY LEVELS AND COGNITIVE FUNCTIONING
health and education. Age continued to be associated
with performance levels on fluid intelligence, memory
and MMSE performance beyond that contributed by
activity and the other contextual variables. This finding
indicates that age-associated differences in cognitive
performance cannot be accounted for entirely by these
contextual variables.
Age-by-activity interaction effects were present for
crystallized and fluid intelligence but not for memory or
for the MMSE. For crystallized intelligence, the
interaction effect indicated that low levels of activity
were associated with lower levels of crystallized
intelligence at younger rather than older ages. For
fluid intelligence, the interaction was in the opposite
direction, where low activity was associated with lower
levels of fluid intelligence but only in older elderly
subjects.
There were several other findings of interest. Across
all three factors and the MMSE, education, sex and
sensory functioning were found to be strongly and
fairly consistently associated with cognitive performance. ADL was found to be predictive of fluid
intelligence scores but not of the other test scores,
probably because tests loading on fluid intelligence
required speed and motor co-ordination (for example,
S L M T , choice and simple reaction time).
The findings from the regression analysis using the
informant-reported activity scales were very similar to
those for the self-reported scale. There was, in addition,
a significant education-by-activity effect for the
memory factor. This indicated that high activity was
associated with higher performance levels in less well
educated subjects. Activity made little difference to
performance in highly educated subjects.
Exclusion of low Mini-Mental State scorers, DSMIII-R Dementia and Depression: A small number of
low-scoring subjects may have been responsible for
many of the significant relationships between activity
and cognitive performance. This possibility was examined in a further analysis by excluding all individuals
scoring below 24 on the MMSE, a level regarded as
predictive of probable dementia. Data on the MMSE
were available for 673 subjects. Overall, the results
were very similar to those found for the total sample
using both self and informant scales. The 23/24 cut-off
on the MMSE may identify moderate to severe cases
but is less likely to identify mild, early onset or possible
dementia cases. Further analysis of subjects with
MMSE scores of 27 and above was conducted. This
analysis produced results that were very similar to those
for the subsample with M M S E scores above the 23/24
cut-off. A final analysis was undertaken excluding
subjects with DSM-III-R Depression and Dementia
as diagnosed using the Canberra Interview for the
Elderly [15, 16]. Similar findings were found, supporting the view that the results could not be accounted for
by subjects with dementia or depression.
Activity as a 'proxy' for cognitive competence: One of
the major problems with cross-sectional analyses of this
kind is that activity may simply serve as a proxy
77
measure of cognitive competence. In this respect, it
would be reassuring to find that activity continued to
influence cognitive performance once the effects of
'cognitive competence' had been accounted for. Chronological age might reflect changes in the nervous system
intrinsically linked to 'biological ageing' and therefore
emerge as an indicator of cognitive competence. This
issue can be investigated by introducing the effect of age
in the first step of a regression analysis followed by
examination of the independent contribution of other
variables. These sets of analyses with age entered first,
for self- and informant-reported activity, were performed. For self-reported activity measures, variables
in addition to those of age which accounted for
significant variance were identical to those found in
the original set of analyses. Most importantly, activity
continued to contribute independently to variability in
cognitive performance for both self-reported and
informant-reported analyses.
Discussion
This study investigated the relationship between
activity and performance on a cognitive test battery in
a sample of elderly people living in the community. The
primary aim was to determine whether age differences
in cognitive performance could be accounted for by
activity, particularly when the influence of other
variables, such as sex, sensory functioning, health,
education and disability were taken into account. We
expected that levels of inactivity would increase with
age while levels of cognitive performance would
decrease, that low levels of activity would be associated
with poor fluid and memory performance, and that
accounting for the effects of activity would reduce age
differences. In addition, we examined whether activity
was of more importance in reducing age differences for
older age groups than for younger groups or those with
poor education compared with those with more extensive education.
Our study differs in important respects from earlier
ones: (i) Reports of activity levels were taken from both
the subject and an informant who knew the subject
well. No earlier studies have attempted to substantiate
self-reported activity measures, (ii) We used a representative sample. The most relevant earlier studies [e.g.
4, 6] examined volunteers, (iii) The range of outcome
measures was greater than in previous studies. We
included measures of fluid intelligence, crystallized
intelligence and memory. Earlier studies used mainly
verbal memory tasks. By including a wide range of
measures it was possible to examine whether activity
correlated more strongly with certain types of cognitive
functioning. It might be expected that the association
between activity level and cognitive ageing would be
more pronounced for fluid and memory tasks than for
crystallized, since the former tasks are known to have
the greatest age-associated decline [6]. (iv) Finally, we
attempted to control for sensory functioning and
physical disability, variables that have previously not
H. CHRISTENSEN ET AL.
been examined in studies of this kind but which may
not be independent of activity levels.
The effect of activity on cognitive test performance: We
found that both self- and informant-rated activity levels
declined significantly with increasing age and that
activity levels influenced the level of cognitive performance independently of sensory functioning, activities
of daily living, education, and health. Importantly, age
continued to be associated with variance of the
cognitive measures once the influence of activity (and
the other contextual variables) had been taken into
account. This latter finding refutes the hypothesis that
cognitive changes in ageing can be accounted for
entirely by the contextual variables activity, health,
sensory functioning and ADL. When control for
mental competence was attempted, by entering age
first and by excluding low M M S E scorers, the influence
of activity was still present consistently. The fact that
activity accounted for additional variance, over and
above that of sensory functioning and ADL, suggests
that activity influenced cognitive performance independently of disability and sensory dysfunction. This
then lowers the probability that better performance on
speeded tasks (such as those used in the measurement
of fluid intelligence) was due to improved psychomotor performance rather than to enhanced cognitive
competence.
Correlational analysis established that the contribution of self-reported activity to cognitive test performance was no greater for fluid intelligence or memory
than for crystallized intelligence, although it was greater
for the MMSE. Once account was taken of contextual
variables in the regression analysis, however, it appeared
that activity contributed somewhat greater variance to
fluid intelligence scores than to the MMSE (see Table
III). On the basis of the present data it would be
premature to conclude that activity influenced particular
cognitive test scores more than others.
The present study found some support for the view
that activity levels were of greater importance in old
than in young subjects. For fluid intelligence, significant interactions were found between age and activity.
This interaction suggested that activity 'offsets' the
effects of age for fluid intelligence measures. For
crystallized intelligence, activity was of greater importance as a predictor of lower scores in younger than in
older elderly subjects. There are various possible
explanations for the effect. Low activity in younger
subjects may be a sensitive indicator of dementia, which
is known to affect crystallized intelligence. Low activity
at older ages may be a less sensitive indicator of
dementia because factors other than dementia, such as
health and motivation, may influence activity levels at
more advanced ages. Alternatively, activity may be
associated with improved cognitive functioning but
only above a certain level of participation. Since activity
levels decline with age, activity levels at older ages may
be below the threshold above which they influence
cognitive processes.
The findings for fluid intelligence and activity were
consistent with earlier work. Christensen and Mackinnon [10] found significant interactions between age
and physical activity on fluid intelligence tasks,
suggesting that activity level 'compensated' for loss of
cognitive abilities on fluid intelligence tasks. However,
our findings for crystallized intelligence were not
consistent with an earlier study on activity levels
using largely verbal tasks [6]. However, differences in
measurement and sample composition were likely to be
responsible for these differences.
Despite the significant relationships we found
between activity and cognitive performance, one of
the clearest findings from the study was that the
influence of activity on cognitive performance was
modest. Across all factors and for both self and
informant report, the variance accounted for by activity
ranged from 0 to 3%. This was lower than that for sex
(0—4%), sensory functioning (range 0—7%), ADL (range
0-6%), education (range 1-18%) and for age (range 0—
4%), but not for health which failed to contribute
significantly to any of the factors. The magnitude of
variance explained by activity is very similar to that
noted by Hutsch et al. [6]. Nevertheless, despite the
modest amount of variance explained, the fact that both
self-rated and informant-rated activity consistently
predicted significant variance for all factors points to its
importance as a correlate of cognitive performance in
old age.
Importance of other contextual variables in predicting
performance: Recent reports have emphasized the
importance of sex, sensory functioning, education and
health factors in predicting cognitive performance of
elderly people. In particular, visual and auditory acuity
have been reported to become more important as age
advances and account for a large proportion of variance.
It has been proposed that age differences in intelligence, including speeded tasks, may be mediated by
vision and hearing [8]. While the present study found a
consistent relationship between sensory measures and
outcome, the relationship was much weaker than that
reported by Lindenberger and Baltes [8]. We found
simple correlations between sensory functioning and
crystallized intelligence to be 0.09, fluid intelligence to
be 0.14, Memory to be 0.14 and MMSE to be 0.15.
Lindenberger and Baltes [8] used objective measures of
hearing (uncorrected) and vision, had a smaller sample
of volunteers, used a slightly greater age range (70—103
years), had a different sampling distribution, and used a
different cognitive test battery. These differences may
account for discrepancies between studies, although the
strength of correlations found in Lindenberger and
Baltes' study needs confirmation from other work.
Threats to validity of findings: Two potential problems with the activity scale were that it was limited in
the range of activities covered and, like other selfreports of activity, it may be subject to bias [26]. In an
epidemiological study, the number of questions asked is
necessarily restricted. T o overcome this limitation,
each question was framed so as to cover a broad range of
activities, without presuming which specific 'organized'
ACTIVITY LEVELS AND COGNITIVE FUNCTIONING
activities were engaged in, and to include activities that
were almost universal (e.g. reading newspaper). Methods of assessing activity used in other studies, such as
summing items on an activity list, may themselves be
biased because they allow those who engage in a variety
of organized activities to score highly. However, despite
their limitations, there were a number of indications
that our activity scales had validity, (i) The results from
the self-report measure were substantiated by an
informant. Although, this does not entirely counter
the criticism that such shared perceptions are nevertheless biased [see 27], it has been established that
informants can be reasonably competent when assessing aspects of behaviour, such as cognitive decline, in
their elderly relatives [18]. (ii) The factor analysis
identified the presence of one major factor and the
factor structure for the informant and self report were
the same. It is significant that the items, although
measuring seemingly different characteristics, did form
a scale, suggesting there may be an underlying single
construct.
Potential problems with the health scales were that
they aggregated equally weighted disparate medical
conditions and current symptoms. More comprehensive scales measuring health factors known to influence
cognitive functioning may be seen as more useful to
control for health factors in the present analysis.
However, the present scales were designed to control
for individuals with multiple health complaints. Moreover, previous analysis of the same population using a
variety of health measures failed to find relationships
with cognitive functioning [14]. The failure to find
strong relationships between cognitive functioning and
health has been reported for other elderly populations
[9]. Consequently, the use of these scales in this
population sample to control for multiple health
problems seems justified.
Because the study is cross-sectional, the activity
relationships may be due to cohort effects. Although
this explanation cannot be ruled out entirely, there is no
evidence or explanation for a more sedentary life-style
being present in the older cohorts. Also, because the
data are cross-sectional, it is difficult to draw conclusions about the direction or nature of the relationship
between cognitive performance and activity. As noted
by others, for example Salthouse et al. [9], high levels of
cognitive function may be a prerequisite for certain
sorts of experiences. In these cases, rather than activity
improving meptal functioning, mental functioning may
influence the type of activity participated in and thus
reflect cognitive competence. The direction of causation between activity levels and cognitive performance
can better be clarified in longitudinal studies, although
the direction of the relationship, even there, can be
difficult to determine [29].
The data from the present study are from the first
wave of a longitudinal study and the relationship of
activity with cognitive decline will be examined after
wave 2 data are collected. Nevertheless, the present
study confirms that it is 'prudent to recommend to
79
elderly people that stimulating mental activity is
worthwhile' [7]. However, the study findings might
suggest the following caveats: (i) activity levels will only
make a small amount of difference; (ii) other factors
such as sensory and motor functioning are also
important; and (iii) despite the contribution of all
these factors, cognitive performance will tend to get
worse with advancing age.
References
1. Swindell R. A model for successful ageing. Ageing Soc
1993;13:245-66.
2. Gribbin K, Schaie WK, Parham IA. Complexity of life
style and maintenance of intellectual abilities. J Soc Issues
1980;36:47-61.
3. Paulson JS, Weisston CC, Heaton RK. The neuropsychology of aging. Curt Opin Psych 1994;7:347-53.
4. Arbuckle TY, Gold DP, Andres D. Cognitive functioning of older people in relation to social and personality
variables. Psychol Aging 1986;l:55-62.
5. Arbuckle TY, Gold DP, Andres D, Schwartzman A,
Chaikelson J. The role of psychosocial context, age, and
intelligence in memory performance of older men.
Psychol Aging 1992;7:25-36.
6. Hultsch DF, Hammer M, Small BJ. Age differences in
cognitive performance in later life: Relationships to selfreported health and activity life style. J Gerontol Psychol
Scil993;48:l-ll.
7. Orrell M, Sahakian B. Education and dementia: research
evidence supports the concept of 'use it or lose it', BrMed
J 1995;310:951-2.
8. Lindenberger U, Bakes P. Sensory functioning and
intelligence in old age: a strong connection. Psychol
Aging 1994;9:339-55.
9. Salthouse TA, Kausler DH, Saults JS. Age, self-assessed
health status and cognition. J Gerontol Psychol Sci
1990;45:156-60.
10. Christensen H, Mackinnon A. The association between
mental, social and physical activity and cognitive
performance in young and old subjects. Age Ageing
1993;22:175-82.
11. Huppert FA. Age-related changes in memory, learning
and remembering information. In: Boiler F, Grafman, J,
eds, Handbook of neuropsychology. Amsterdam: Elsevier,
1988.
12. Folstein MF, Folstein SE, McHugh PR. 'Mini-Mental
State': a practical method for grading the cognitive state of
patients for the clinician. JPsychiatr Res 1975; 12:179-98.
13. Christensen H, Mackinnon AJ, Jorm AF, Henderson AS,
Scott LR, Korten AE. Age differences and interindividual cognition in community dwelling elderly.
Psychol Aging 1994;9:381-90.
14. Christensen H, Jorm AF, Henderson AS, Mackinnon AJ,
Korten AE, Scott LR. The relationship between health
and cognitive functioning in a sample of elderly people in
the community. Age Ageing 1994;23:204-12.
15. Henderson AS, Jorm AF, Mackinnon A et al. The
prevalence of depressive disorders and the distribution of
depressive symptoms in later life: a survey using Draft
ICD-10 and DSM-III-R. Psychol Med 1993;23:719-29.
16. Henderson AS, Jorm AF, Mackinnon A et al. A survey of
dementia in the Canberra population: experience with
ICD-10 and DSM-III-R criteria. Psychol Med
1994;24:473-82.
17. Jorm AF, Henderson S, Scott R, Mackinnon AJ, Korten
8o
18.
19.
20.
21.
22.
23.
24.
25.
26.
H. CHRISTENSEN ET Ah.
AE, Christensen H. The disabled elderly living in the
community: care received from family and formal
services. Med J Aust 1993;158:383-7.
Jorm AF, Christensen H, Henderson AS, Koten AE,
Mackinnon AJ, Scott LR. Complaints of cognitive
decline in the elderly: a comparison of reports by subjects
and informants in a community survey. Psychol Med
1994;24:365-74.
Nelson HE. National Adult Reading Test (NART).
Berkshire, England: NFER-Nelson, 1982.
Smith A. The Symbol Digit Modalities Test. Los Angeles:
Western Psychological Services, 1973.
Wechsler D. Wechsler Adult Intelligence Scale—Revised.
New York: Psychological Corporation, 1981.
Wechsler D. A standardized memory scale for clinical
use. J Psychol 1945;19:87-95.
Wilson B, Cockburn J, Baddeley A. The Rivermead
Behavioural Memory Test. Reading, Berks.: Thames
Valley Test Company, 1985.
Steiger JH. Test for comparing elements of a correlation
matrix. Psychol Bull 1980;87:245-51.
Fisher GA. Problems in the use and interpretation of
product variables. In: Long JS, ed. Common problems/
proper solutions. Beverly Hills: Sage, 1988.
Bunce DJ, Warr PB, Cochrane T. Blocks in choice
responding as a function of age and physical fitness.
Psychol Aging 1993;8:26-33.
27. Ross M. Relation of implicit theories to the construction
of personal histories. Psychol Rev 1989;96:341-57.
28. Jorm AF, Mackinnon AJ, Christensen H, Henderson S,
Scott R, Korten A. Cognitive functioning and neuroticism in an elderly community sample. Pers Individ Differ
1993;15:721-3.
29. Farrell AD. Structural equation modeling with longitudinal data: Strategies for examining group differences
and reciprocal relationships. J Consult Clin Psychol
1994;62:477-87.
Authors' addresses
H. Christensen, A. Korten, A. F. Jorm,
A. S. Henderson, R. Scott
National Health and Medical Research Council,
Social Psychiatry Research Unit,
The Australian National University,
Canberra 0200, Australia
A. J. Mackinnon
Mental Health Research Institute,
Parkville, Victoria, Australia
Received in revised form 22 June 1995