• Aging, Neuropsychology, and Cognition 1382-5585/00/0702-069$15.00

Aging, Neuropsychology, and Cognition 1382-5585/00/0702-069$15.00
2000, Vol. 7, No. 2, pp. 69-85 © Swets & Zeitlinger
Cognitive Performance and the Role of Control Beliefs
in Midlife*
Lisa M. Soederberg Miller and Margie E. Lachman
Brandeis University
ABSTRACT
Midlife has been touted as being a time of peak performance in many different areas of functioning. In the
present study, we investigated whether this was true for cognitive functioning on tasks assessing speed,
reasoning, short-term memory, and vocabulary. We also explored the extent to which levels of cognitive
functioning could be attributed to individual differences in general control beliefs. Middle-aged adults
showed little or no cognitive declines on speed, reasoning, and short-term memory measures relative to the
young and outperformed the young on vocabulary. Relative to the elderly, middle-aged adults scored
higher on all tasks except vocabulary, for which there were no differences. Adults in midlife, on the other
hand, had lower scores on measures of general control beliefs compared to younger adults. Thus, although
midlife is a time of high cognitive functioning, it is also a time of lower beliefs about control. To investi-
gate the relationship between control beliefs and cognitive performance, we used structural equation mod-
eling. The models showed that for adults in midlife, control beliefs were predictive of performance but only
for the reasoning task after background variables were considered. Specifically, high levels of control
beliefs were associated with better cognitive performance. More work is needed to identify mediational
processes linking control beliefs and cognitive performance for various age groups and to determine
whether some cognitive processes are more controllable than others.
The developmental period of midlife has re- lent. It is also a time when multiple and diver-
ceived little attention relative to the earlier peri- gent demands are placed upon individuals from
ods of infancy, childhood, and adolescence and career, social, and community commitments and
the later period of old age (Brim, 1992; Lach- from children and parents (cf. Lachman &
man & James, 1997b; Willis & Reid, 1999). James, 1997a). In fact, due to the joint demands
Whereas some research has addressed midlife of a younger and an older generation on adults
issues of menopause (e.g., Lennon, 1982; in midlife, this time period has been labelled the
Matthews et al., 1990) and the notion of a sandwich generation (e.g., Roots, 1998). Be-
midlife crisis (e.g., Davidson, 1979; Rosenberg, cause of these types of experiences, in her pio-
Rosenberg, & Farrell, 1999), less attention has neering work on middle age, Neugarten de-
been paid to cognitive changes during this time. scribed midlife as ‘‘a period of maximum capac-
Midlife is a potentially rich area for investi- ity and ability to handle a highly complex envi-
gation of cognitive processes for a number of ronment and a highly differentiated self’’ (1968,
reasons. For example, a good deal of knowledge p. 97). This is consistent with implicit theories
has been accumulated by this point in life (cf. of midlife depicting middle-aged adults as com-
Ackerman & Rolfhus, 1999), yet the sharper petent and productive (Lachman, Lewkowicz,
declines associated with old age are not preva- Marcus, & Peng, 1994). Thus, it seems likely
*
This project was funded by National Institute on Aging (NIA) Training Grant T32 AG00204 and the John D.
and Catherine T. MacArthur Foundation Network on Successful Midlife Development (MIDMAC).
Address correspondence to: Lisa M. Soederberg Miller, Department of Psychology, Brandeis University,
Waltham, MA 02454-9110, USA. E-mail: Lmiller@brandeis.edu.
Accepted for publication: January 25, 2000.
70 L.M. SOEDERBERG MILLER AND M.E. LACHMAN
that this developmental time period would pro- Bandura, Elliot, & Lewkowicz, 1992). In inter-
vide ample opportunities for adults to maintain vention research, cognitive-behavioral strategies
or sharpen various cognitive skills. are used to modify beliefs about control over
Some empirical work is consistent with this memory as well as physical exercise (Lachman
notion of peak performance during midlife. The et al., 1997). Thus, a long-term goal of this line
bulk of the research on intellectual functioning of research is to identify factors in midlife that
in midlife is based on data from the Seattle Lon- could potentially remediate or even prevent cog-
gitudinal Study (e.g., Schaie, 1984, 1996; Willis nitive declines in later life.
& Schaie, 1999), which in part support Neugar-
ten’s (1968) notion that midlife is a time of peak Control Beliefs and Age
performance. This work shows that adults A sense of control refers to the belief that indi-
achieve peak performance on complex higher viduals feel able to affect their performance
order abilities such as inductive reasoning, spa- (Abeles, 1990; Rodin, 1990), or more specifi-
tial orientation, and vocabulary during midlife. cally, that individuals are responsible for their
However, processes such as perceptual speed outcomes because of their own efforts (Rodin,
and numerical ability show declines beginning Timko, & Harris, 1985). Weisz (1983) and oth-
early in midlife. Aside from this research, his- ers (e.g., Skinner, 1996) have identified two key
torically, most work assessing cognitive perfor- components of control called competence and
mance in midlife has been indirect, as a data contingency. Competence is often described in
point between elderly- and young-adult compar- terms of one’s judgments about his or her ability
ison groups. to achieve a goal (sometimes referred to as self-
Consequently, little is known about the fac- efficacy). Within a coping framework, Pearlin
tors that contribute to high levels of perfor- and Schooler (1978) referred to this belief as
mance during midlife. In particular, we were mastery, suggesting that this is an important
interested in control beliefs as a potential predic- psychological resource that is available to help
tor of cognitive performance. General control individuals cope with stress and strain. Contin-
beliefs have been linked to a number of out- gency refers to the belief that one’s actions will
comes such as health, life satisfaction, and well- lead to intended consequences, also called out-
being (e.g., Lachman & Weaver, 1998a), and come expectations (Bandura, 1977). Related to
domain-specific control beliefs have been linked this construct is the notion of perceived con-
to cognitive performance (e.g., Grover & Hert- straints (Lachman & Weaver, 1998a), which
zog, 1991; Riggs, Lachman, & Wingfield, 1997; refers to the extent to which there are factors
Stine, Lachman, & Wingfield, 1993; for a re- that are believed to interfere with reaching
view, see Miller & Lachman, 1999). Although goals. Within these and other frameworks, a
individual differences in control beliefs are re- strong sense of control (higher mastery and
lated to cognitive performance in young and el- lower perceived constraints) has been repeatedly
derly adults (e.g., Stine et al., 1993), little work linked to higher levels of well-being (e.g., Al-
has directly examined this relationship among bert et al., 1995; Lachman, Lyons, & Staudin-
middle-aged adults. One goal of the present ger, 1999; Rodin, 1990).
study was to examine whether control beliefs are Some have argued that, with advancing age,
related to cognitive performance during midlife. individuals begin to lose control over physical
Research linking control beliefs and cognition and cognitive abilities, which then leads to de-
could yield important data because information creases in perceived control. However, the data
about individual differences in control beliefs do not uniformly support this notion (Rodin,
during midlife may be useful for predicting or 1990; Rodin et al., 1985). Research has yielded
preventing negative outcomes in later life. Past results showing increases, decreases, and no
work has shown control beliefs to be modifiable differences in control beliefs with increasing age
and therefore they could be useful as an effec- (for reviews, see Lachman, 1986; Rodin et al.,
tive intervention mechanism (Lachman, Weaver, 1985). Brandstädter and Rothermund (1994)
 COGNITION IN MIDLIFE 71
have suggested that the data are equivocal be- Weaver, 1998b). Moreover, relationships be-
cause research has failed to take into account the tween control beliefs and behaviors or outcomes
accommodative process of goal adjustment that typically are stronger when using domain-con-
may buffer age-related declines. Still others gruent measures (Lachman & Weaver, 1998b).
have stated that the use of different types of However, in past work (Lachman, 1986) these
measures – domain-specificity versus general- relationships were more pronounced for the el-
ized and multidimensional versus unidimen- derly than for the young, which may indicate
sional – are responsible for the disparate find- that control beliefs are more differentiated for
ings (Lachman, 1986). the elderly.
Nevertheless, relatively few of these studies Nevertheless, generalized control measures
have included a middle-age group (Brandstädter were preferred in the present study because we
& Rothermund, 1994), and even fewer have ex- were interested in determining the extent to
plicitly addressed how this group differs from which generalized perceptions of control over
other age groups. Unfortunately, the little work one’s life affected cognitive performance. Thus,
that has investigated midlife has also yielded whereas domain-specific measures would be
inconsistent findings. In a study in which men in expected to highlight differences in control over
midlife were compared to younger and older cognition between the young and the elderly,
men, Lao (1975) found that men in midlife were general measures were expected to provide a
more internal than were younger men; however, more balanced approach to studying belief-per-
no differences were found between middle-aged formance relations in midlife that would more
and older males. Thus, contrary to the notion likely tap into the multiple demands placed on
that control beliefs decrease across the life span, adults during this time of life.
these data show that, for men at least, beliefs The research consistently shows that those
become stronger in middle adulthood and re- who feel they have greater control over their
main constant into old age. On the other hand, cognitive performance are able to achieve higher
among samples of men and women, lower levels levels of performance than those who do not
of memory self-efficacy have been reported for (Lachman & Jelalian, 1984; Lachman & Leff,
older and middle-aged adults relative to younger 1989; Riggs et al., 1997; Stine et al., 1993). For
adults (Hertzog et al., 1998; Hultsch, Hertzog, & example, Riggs et al. (1997) administered a
Dixon, 1987). Thus, the status of control beliefs speech processing task to older adults who were
in midlife remains unclear. either high (‘‘internals’’) or low (‘‘externals’’)
in perceived control. Participants were required
Control Beliefs and Cognitive Performance to listen to recorded passages in order to recall
Although it is uncertain whether control beliefs the text verbatim and were told to interrupt the
change with age, a positive relationship between flow of speech where they wanted in order to
beliefs and cognitive outcomes has been re- segment the text into recallable units. They
ported consistently in the literature. This link found that externals were more likely than were
can be investigated with either general control internals to make inaccurate predictions about
beliefs or domain-specific control beliefs, that the number of words they could accurately re-
is, those that specifically assess beliefs about call. These data suggest that individuals who are
cognitive performance. The use of general con- low in perceived control are poorer at monitor-
trol beliefs is typically preferred when multiple ing on-line memory processing.
domains are being considered. Similarly, when There are a number of other factors that po-
investigating a specific domain of functioning, a tentially mediate and moderate the relationship
specific measure of control beliefs is typically between control beliefs (domain-specific or gen-
preferred. One reason for this preference is be- eral) and cognitive performance. For example,
cause age differences are more likely to be individuals who believe they can affect their
found using domain-specific measures than by memory performance are likely to devote more
using generalized measures (Lachman & effort to solving memory problems (Bandura,
72 L.M. SOEDERBERG MILLER AND M.E. LACHMAN
1977) and may be more likely to apply suitable more (reasoning and speed) or less (short-term
strategies (e.g., Hertzog et al., 1998; Stine et al., memory and vocabulary) age sensitive (cf.
1993). Bandura (1977, 1997) has argued that Salthouse, 1991). These factors were selected to
beliefs are especially crucial to perseverance represent the multidimensionality and multi-
when individuals are faced with adversity and directionality inherent in cognitive abilities in
challenge, which could be the case when dealing adulthood (Horn & Cattell, 1967; Schaie, 1996).
with difficult cognitive tasks. At midlife, some To illustrate, age-related declines in reasoning
individuals report experiencing problems with ability are commonly found (e.g., Schaie, 1985),
cognitive functioning, especially memory even after specialized cognitive training (Willis
(Lachman, in press). This in turn can lead to & Schaie, 1986). Prominent age-related declines
increased distress and anxiety as well as de- have also been found on simple speeded tasks in
creased motivation to use adaptive strategies and which participants, for example, compare two
effort, all of which can interfere with effective strings of digits to determine, as quickly as pos-
performance (Lachman, in press). Although not sible, whether they are the same or different
assessed in the present study, cognitive perfor- (Salthouse & Babcock, 1991). Short-term mem-
mance can also influence beliefs (e.g., Grover & ory tasks which require individuals to passively
Hertzog, 1991), suggesting that this relationship hold recently encountered materials for a brief
is reciprocal in nature. period of time often fail to show age differences
Past research also suggests that control be- (cf. Smith & Earles, 1996). Lastly, age-related
liefs are associated with performance on some declines on vocabulary measures are seldom
cognitive tasks but not on others (Gold, Andres, found (cf. Salthouse, 1991) and, in fact, older
Etezadi, Schwartzman, & Chaikelson, 1995; adults sometimes show advantages in this area
Lachman & Jelalian, 1984; Seeman, McAvay, (e.g., Horn & Cattell, 1967; Schaie, 1996). One
Merrill, Albert, & Rodin, 1996). For example, could argue that the tasks that show the earliest
control beliefs have been found to be more re- declines are the most challenging for all individ-
lated to verbal than to nonverbal tasks (Gold et uals and therefore may show the greatest effects
al., 1995; Seeman et al., 1996). However, it is of control beliefs.
possible that the extent to which beliefs are re- Because background variables such as health,
lated to performance also depends on whether education, and age tend to be related to control
the outcomes are age-sensitive or age-insensi- beliefs and to cognitive performance (e.g., Lach-
tive tasks. That is, if declines on certain tasks man, 1991; Schaie, 1996) and, further, tend to
are keenly felt by middle-aged adults, control attenuate the relationship between control be-
beliefs may be relatively more important than liefs and cognitive performance (e.g., Miller &
they would be for tasks in which age-related Lachman, 1998), we analyzed the data both with
declines are less evident and successes are more and without controlling for background vari-
common. This is because age-sensitive tasks ables. Although these variables, particularly
presumably require more effort and therefore health, could be outcomes to be examined in
allow greater opportunity for motivational and their own right, our aim was to focus on beliefs
strategic influences. As mentioned above, these and performance while controlling for these fac-
factors have been implicated as possible media- tors. This decision was based on research show-
tors between beliefs and performance (Hertzog ing that health problems may result in lowered
et al., 1998; Riggs et al., 1997; Stine et al., 1993; sense of control (Lachman & Leff, 1989) as well
cf. Bandura, 1997; Miller & Lachman, 1999). as decreased cognitive functioning (Schaie,
In the present study, we explored cognitive 1990) and on research showing that younger and
performance and predictors of performance for more educated adults typically have higher con-
middle-aged adults relative to younger and older trol beliefs and better cognitive performance
adults within four areas of functioning that were (Lachman, 1991).
 COGNITION IN MIDLIFE 73
We also conducted group comparisons to in- METHOD
vestigate whether associations among these vari-
ables were similar for young, middle-aged, and Participants
older adults. Because the separate age group These data were a subset of the Midlife in the
samples were relatively small, these analyses United States (MIDUS) Survey conducted by the
John D. and Catherine T. MacArthur Foundation
should be considered as preliminary.
Network on Successful Midlife Development. This
Our hypotheses were that younger adults subset, the Boston In-Depth Study of Management
would outperform middle-aged and older adults Processes in Midlife, consisted of an intentional
on speeded tasks and that adults in midlife oversample (using random digit dialing) of 500
would outperform older adults on measures of adults in the Greater Boston metropolitan area. A
speed and reasoning. We also predicted that total of 429 participants completed the telephone
middle-aged and older adults would outperform and mail questionnaire portions of the MIDUS
survey, which was the criterion for inclusion in the
younger adults on vocabulary. We did not ex- Boston In-depth Study. Of the 391 who could be
pect to find any age differences on the short- reached by phone 6 months after the initial MIDUS
term memory tasks. Based on theoretical work survey, 302 (77%) agreed to participate in a three-
(Bandura, 1997) and empirical evidence (e.g., wave, short-term longitudinal study of life man-
Hertzog et al., 1998; Riggs et al., 1997) suggest- agement processes that included a face-to-face
ing that one’s beliefs can influence performance, interview. There were no significant differences
we expected control beliefs to predict perfor- between the participants and the nonparticipants on
any demographic variables.
mance on most if not all the cognitive outcomes Thus, the current sample consisted of 302
variables. We expected the relationship between noninstitutionalized, English-speaking adults be-
control and performance to be stronger for age- tween the ages of 25 and 75 (M = 47.8, SD = 13.1)
sensitive than age-insensitive tasks because age- who resided in the Greater Boston area. The sam-
sensitive tasks were thought to require more ef- ple was 41.1% female and roughly half of the par-
fort, which is believed to be a mediator between ticipants had a college degree or more education.
Participants were further screened for English as
beliefs and performance (Bandura, 1977).
their native language and for absence of stroke. Of
Past work has shown that health predicts con- the remaining 272 participants, 13 failed to com-
trol beliefs and cognitive performance such that plete all of the cognitive measures (these partici-
those who had more health problems were more pants did not differ in gender, education, or age
likely to show declines in control beliefs and from those who did complete the cognitive tests).
intellectual functioning (Lachman & Leff, The resulting sample (N = 259) consisted of 84
1989). Therefore, the relationships between con- young adults (ages 25-39; M = 32.6, SD = 4.1),
108 middle-aged adults (ages 40-59; M = 49.6, SD
trol beliefs and all measures of cognitive perfor-
= 5.0), and 67 older adults (ages 60-75; M = 65.7,
mance were expected to be attenuated when SD = 4.1). A chi-square test showed that the age
health (as well as age and education) was con- groups were comprised of comparable distribu-
trolled. It was predicted that control beliefs tions of males and females, Pearson P2(2, N = 259)
would be related to cognitive performance = .41, p = .82. Further, an Age × Sex ANOVA on
within all three age groups; however, the education level showed that education did not vary
strength of the associations between the age-sen- as a function of Age, F = 1.70, p = .18, or Sex, or
of a combination of the two, F < 1, for both.
sitive tasks and beliefs were expected to be most
evident for the middle-aged and oldest groups. Measures
This expectation was based on the assumption
that age-sensitive tasks would require more ef- Cognitive Measures
fort and effective strategies, particularly among Speed was assessed by the digit symbol substitu-
older individuals. These within-group predic- tion test of the Wechsler Adult Intelligence Scale
(WAIS; Wechsler, 1955) and a letter comparison
tions are tentative, however, given the limited
task (Salthouse & Babcock, 1991). Reasoning was
sample size. assessed by the Schaie-Thurstone letter series
(Schaie, 1985) and Raven’s Advanced Progressive
74 L.M. SOEDERBERG MILLER AND M.E. LACHMAN
Matrices (Raven, Raven, & Court, 1991). Short- the environment which limit one’s pursuits. This
term memory was assessed by the WAIS forward scale contained items such as ‘‘other people deter-
and backward digit spans as well as a counting mine most of what I can and cannot do’’ and
backward task requiring participants to count back- ‘‘there is little I can do to change the important
wards by sevens starting from a three-digit num- things in my life.’’ In order to facilitate conver-
ber. For the analyses of variance reported below, gence of the measurement model, we created three
scores on the component tasks of the speed, rea- indicators of control by parceling perceived con-
soning, and short-term memory factors were trans- straints items into two indicators. This was done
formed into z scores and then averaged to form a by randomly assigning scale items into one of two
composite. For vocabulary, which was assessed via parcels and then using each as a separate indicator
the WAIS vocabulary subscale, the z score trans- (for a detailed discussion on this procedure, see
formation was used alone (see Table 1 for means Kishton & Widaman, 1994). These two parcels had
and standard errors as a function of age group). We high reliability, as indicated by an internal consis-
expected these cognitive measures to form a four- tency estimate (coefficient " = .84). The coeffi-
factor model: speed, short-term memory, reason- cient " for the mastery scale was .70. Thus, to-
ing, and vocabulary. gether with mastery, we had three indicators of
control, with the mastery scale loading negatively
Health and the perceived constraints scales loading posi-
Health was assessed using: (a) a checklist of nine tively such that higher scores indicated lower lev-
acute illnesses experienced within the last 30 days els of perceived control.
on a six-point scale (1 = almost everyday, 6 = not Although this data set contained primarily gen-
at all); (b) the number of chronic illnesses, out of eralized measures of control beliefs, there were a
29, participants had been treated for within the last few items that tapped domain-specific beliefs spe-
12 months; and (c) the number of different pre- cific to cognitive abilities. However, the domain-
scription medications currently being taken. The specific measures were collected six months after
health factor was computed by summing the z the cognitive battery. Therefore, we used the gen-
scores of each subcomponent; therefore, a higher eralized measures of control because the model of
score indicates poorer health. interest specifies cognitive variables as outcomes.
Domain-specific measures are typically more sen-
Control Beliefs sitive than are generalized measures, often show-
Control beliefs were assessed via two seven-point ing stronger relationships with performance in the
scales: mastery and perceived constraints corresponding domain than do generalized mea-
(Lachman & Weaver, 1998b; Pearlin & Schooler, sures (Lachman, 1986). Thus, by using the gener-
1978). The mastery scale, designed to assess one’s alized measures, we created a more stringent test
beliefs about his or her ability to master the envi- of our model.
ronment, contained items such as ‘‘what happens
to me in the future mostly depends on me’’ and Procedures
‘‘when I really want to do something, I usually Measures were administered over two time inter-
find a way to succeed at it.’’ The perceived con- vals that were 10 to 12 months apart. Demographic
straints scale was designed to capture the extent to information of age and education, control beliefs,
which one perceives uncontrollable constraints in and health measures were collected via telephone
Table 1. Means (and Standard Errors) of Variables (in z score units) as a Function of Age Group.
Young adults Middle-Aged adults Older adults
Variables M (SE) M (SE) M (SE)
Health (lack) –.41 (.25) .03 (.20) .50 (.31)
Control (lack) –.17 (.07) .12 (.09) .03 (.10)
Speed .37 (.08) .10 (.08) –.64 (.10)
Reasoning .32 (.09) .09 (.09) –.62 (.10)
STM –.07 (.08) .13 (.08) –.18 (.05)
Vocabulary –.28 (.10) .19 (.09) .05 (.12)
Note. STM = short-term memory.
 COGNITION IN MIDLIFE 75
and mail questionnaires in 1995 as part of the did not differ from the older adults (M = .03, SD
MIDUS Survey. Cognitive measures were col- = .82). Further, the youngest and oldest groups
lected in participants’ homes during the second did not differ. These findings are consistent with
wave of the Boston Study in 1996. For the cogni-
those that show that middle-aged adults perceive
tive measures, participants were tested individually
in a quiet area of their home and received the for- a loss of control relative to the young, although
ward and backward digit spans, vocabulary test, some past work has found this to be true for the
counting backwards task, letter comparison task, elderly as well (Hertzog et al., 1998; Hultsch et
digit symbol substitution test, letter series test, and al., 1987; Lachman, 1991). However, these data
the Raven’s Advanced Progressive Matrices, in are not consistent with data showing increases in
that order. control beliefs for middle-aged adults relative to
young adults (Lao, 1975).
RESULTS AND DISCUSSION Performance Levels
In order to determine whether there were age
Control Beliefs differences on the four factors of cognitive abili-
A composite index of control beliefs was com- ties, we performed a one-way 3(Age: young,
puted by averaging a standardized measure of middle, old) ANOVA on each. In addition, post
the constraints scale with a standardized mea- hoc tests were conducted to test whether middle-
sure of the mastery scale multiplied by negative aged adults performed differently than younger
one. This yielded a summary variable represent- or older adults on these factors (see Figure 1).
ing a lack of control, consistent with our Significant effects of Age were found for all
LISREL model. To examine age differences in four analyses; however, the nature of the effects
control beliefs, a one-way 3(Age: young, mid- differed across cognitive abilities.
dle, old) ANOVA on control beliefs was con-
ducted. The results showed a marginally signifi- Speed
cant main effect of age, F(2, 268) = 2.98, p = The Age effect in this analysis was significant,
.052. To determine whether the middle-aged F(2, 250) = 28.68, p < .001. Specifically, the
adults differed from the other two age groups, a difference between young and middle-aged
Tukey’s post hoc test was performed. The re- groups approached significance (p = .06); how-
sults indicated that the young (M = –.17, SD = ever, both the young and middle-aged groups
.70) had significantly higher levels of control significantly differed from the older group. As
beliefs than the middle-aged adults (M = .12, SD predicted, performance decreased with age.
= 1.01), p < .05, but that the middle-aged adults
Speed
1.0 Reasoning
0.8 Short-Term Memory
Vocabulary
0.6
Mean z Score
0.4
0.2
0.0
-0.2
-0.4
-0.6
-0.8
-1.0 Young Middle Old
Fig. 1. Cognitive performance by age group.
76 L.M. SOEDERBERG MILLER AND M.E. LACHMAN
Reasoning term memory and vocabulary were not. In fact,
The Age effect on reasoning, F(2, 247) = 24.21, Figure 1 shows that adults in midlife, unlike
p < .001, was attributable to significant differ- younger and older adults, scored above the sam-
ences between the younger two age groups and ple mean (i.e., had positive z scores) on all four
the oldest. These data are consistent with past factors. Despite some evidence of perceptual
research showing that there are steep declines in slowing, this suggests that midlife is a time of
reasoning abilities after midlife (cf. Schaie, relatively strong performance across abilities.
1990).
Structural Equation Models
Short-Term Memory Structural equation models were used to explore
The results of the analysis on short-term mem- the nature of individual differences in the corre-
ory scores showed that Age was significant, F(2, lates of cognitive functioning across the entire
256) = 3.44, p < .05; however, this effect was age range (correlations among variables are pre-
attributable to the middle-aged adults perform- sented in Table 2). For all models presented
ing significantly better than did the older adults. here, we used LISREL 8.14 (Jöreskog & Sör-
The young and older groups did not differ. bom, 1996). Because multiple indexes of fit are
These data are consistent with past research preferable when conveying how well the data fit
showing no age differences between younger the hypothesized structural equation model (cf.
and older adults on short-term memory tasks (cf. Byrne, 1998), we chose the Root Mean Square
Smith & Earles, 1996). However, unlike past Error of Approximation (RMSEA), the Compar-
research, these data show that for the middle- ative Fit Index (CFI), and the Goodness of Fit
aged group there was a trend toward peak levels Index (GFI). The RMSEA is an index of fit that
of performance in this area. takes into account the error of approximation in
the population, with less than .05 reflecting a
Vocabulary good fit and values greater than .10 reflecting a
The analysis on vocabulary scores also yielded poor fit. The CFI reflects the degree of match
an effect of Age, F(2, 255) = 34.02, p < .001, between an independent model and the observed
which was due to lower performance levels of data with values greater than .90 reflecting a
younger adults compared to those of the two relatively good fit. The GFI is based on a com-
older groups. parison of the hypothesized model with no
model such that indexes close to 1.00 represent
Overall, these data are consistent with our hy- a good fit.
potheses by showing that speed and reasoning
were particularly age sensitive whereas short-
Table 2. Correlations among Background Variables, Control Beliefs, and Cognitive Variables.
1 2 3 4 5 6 7 8
1. AGE 1.00
2. EDUCATION –.04 1.00
3. HEALTH (lack) .16** –.11 1.00
4. CONTROL (lack) .08 –.16** .37** 1.00
5. STM –.08 .29** –.09 –.04 1.00
6. SPEED –.46** .28** –.18** –.06 .38** 1.00
7. REASONING –.41** .44** –.25** –.17** .51** .64** 1.00
8. VOCABULARY .14* .55** .01 –.10 .45** .25** .47** 1.00
Note. For health, control, short-term memory (STM), speed, and reasoning, z scores for the individual measures
were computed and then combined to yield a summary score.
* p < .05; ** p < .01.
 COGNITION IN MIDLIFE 77
Measurement Model ory factors. The path from control to speed,
The first step was to confirm our predicted four- however, was significant, t(34) = 1.87, p < .05,
factor structure (speed, short-term memory, rea- for a one-tailed test. Thus in general the data
soning, and vocabulary) of the cognitive mea- support the notion that higher levels of control
sures. The measurement model indicated a good are associated with better cognitive perfor-
fit to the data, P2(14, N = 272) = 20.16, p = .12, mance. However, our hypothesis that control
RMSEA = .04, CFI = .99, GFI = .98. The fit was beliefs are more highly related to age-sensitive
improved by allowing the error variance in the cognitive domains was only partially supported.
backward and forward digit spans to covary. Although reasoning, an age-sensitive ability, did
This modification (with a modification in