Social Network Characteristics and Cognition
in Middle-Aged and Older Adults
Ronald E. Holtzman, George W. Rebok, Jane S. Saczynski, Anthony C. Kouzis,
Kathryn Wilcox Doyle, and William W. Eaton
Bloomberg School of Public Health, Johns Hopkins University, Baltimore, MD.
We examined the relationship between social network characteristics and global cognitive status in a community-
based sample of 354 adults aged 501 and with Mini-Mental State Examination (MMSE) scores of 281 at baseline.
Multivariate analyses indicated that interaction in larger social networks related to better maintenance of MMSE
scores and reduced odds of decline to population-based lower quartile MMSE scores at follow-up 12 years later.
At follow-up, higher levels of interpersonal activity (more frequent contacts in larger social networks) and
exposure to emotional support independently related positively to MMSE. The findings suggest that interaction in
larger social networks is a marker that portends less cognitive decline, and that distinct associational paths link
interpersonal activity and emotional support to cognitive function.
of social integration or engagement, whereas functional features
refer to specific qualitative support functions (e.g., informa-
tional, instrumental, and self-esteem or emotional support;
Cohen & Wills, 1985). Those functions, and others that are not,
perhaps, expressly supportive, could promote cognitive health
in later life. For example, engagement might introduce life
experiences that are cognitively complex and stimulating.
Cognitive challenge is associated with higher cognitive
functioning (Hultsch, Hammer, & Small, 1993), and it might
create brain reserve (in the manner of education, e.g., Katzman,
1993) and delay the clinical expression of dementia (Verghese
et al., 2003; Wilson et al., 2002). Emotional support might
buffer against physiological stress (Seeman, Berkman, Blazer,
& Rowe, 1994; Uchino, Holt-Lunstad, Uno, Betancourt, &
Garvey, 1999), a possible risk factor for maintenance of
cognitive function (e.g., Lupien et al., 1994, 1998; Seeman,
McEwen, Rowe, & Singer, 2001; Seeman, Singer, Rowe,
Horwitz, & McEwen, 1997).
Other functions of engagement that benefit cognitive health
are also possible (see reviews relevant to functions in Barrera,
1986, Cohen, 1988, and Cohen & Wills, 1985), and some of
these may relate to cognitive stimulation and emotional
support. Examples include the provision of economic and
information resources; a sense of purpose, community, and
stability; and opportunities to develop self-efficacy or learn
coping strategies against stress.
In the present study, we examined longitudinally whether
interactions in larger social networks related to maintenance of
global cognitive status. Larger networks might provide more
opportunity for emotional support or diverse experiences with
the potential to produce cognitive stimulation or other functions
related to cognitive health. Given the available archival data,
we then used cross-sectional analyses to explicitly demonstrate
whether social networks might provide both emotional support
and the potential for cognitive stimulation as independent
benefits to cognition.
TRUCTURAL features of social networks such as size and
frequency of contact with others can be considered indices
We measured network size as the number of persons with
whom there was a personal relationship and with whom the
respondent kept in touch by phone or visit most of the time. In
our cross-sectional analyses, we determined the frequency of
those contacts and the amount of emotional support received.
Germane to the review that follows, we consider that these
measures reflect amounts of exposure to the potential benefits
of social networks to cognitive health.
Two recent community-based longitudinal studies relate
social network features and functions to incident cognitive
declines—that by Bassuk, Glass, and Berkman (1999), and that
by Seeman, Lusignolo, Albert, and Berkman (2001). A
comparison of these studies illustrates the use of measures that
include exposure levels, and their possible value in revealing
associations with cognition. In the study by Bassuk and
colleagues, an index incorporating the number and/or frequency
of network interactions or activities embodies levels of
exposure and predicted decline in global mental status over
12 years. In the study by Seeman, Lusignolo, and colleagues,
frequency of emotional support represents levels of exposure
and predicted decline on a composite of cognitive abilities over
7.5 years. However, a measure of network size with no index of
the extent of actual social interaction did not predict decline in
Seeman’s study, and a single binary (yes–no) item for
emotional support did not predict decline in Bassuk’s study,
possibly in part because these measures do not sufficiently
differentiate levels of exposure to beneficial network functions.
Seeman and colleagues note the issue of interactions and issues
related to exposures in the network as well as other differ-
ences in the studies such as sample characteristics, and the
measures of structural and functional features and cognitive
Other community-based longitudinal studies suggest that the
level of exposure is potentially important. Wang, Karp,
Winblad, and Fratiglioni (2002) found that daily–weekly, but
not less frequent, participation in social activities was
associated with reduced risk of dementia 6.4 years later.
Fratiglioni, Wang, Ericsson, Maytan, and Winblad (2000)
Journal of Gerontology: PSYCHOLOGICAL SCIENCES
2004, Vol. 59B, No. 6, P278–P284
Copyright 2004 by The Gerontological Society of America
by guest on October 21, 2015
found that frequent satisfying contacts with children and
relatives or friends contributed to a formula that related to
decreased risk of dementia over about 3 years.
Although measures with exposure levels might be more likely
to reveal associations with cognition, significant relationships
between binary indices of social engagement and cognition do
for dementia in Fratiglioni et al., 2000). In addition, a cross-
sectional analysis using older volunteers living independently in
the community (Arbuckle, Gold, Andres, Schwartzman, &
Chaikelson, 1992) and a recent case-control study using
dementia cases (Seidler, Bernhardt, Nienhaus, & Frolich, 2003)
on some measures. For example, Arbuckle and associates found
that level of support satisfaction directly linked to general
intelligence whereas support network size (absent an exposure
index) directly linked to performance on certain memory tasks.
In summary, research suggests that there are at least two
functions of social engagement that might be protective of
cognitive health in later life—that is, cognitive stimulation and
the emotional benefits of support. When significant associations
between measures of social ties that imply these or other
functions and cognition have not emerged, one reason might be
that levels of exposure were not included (e.g., Bassuk et al.,
1999, and Seeman, Lusignolo, et al., 2001; see also Fabrigoule
et al., 1995, and some of the measures in Fratiglioni et al.,
Given our available data, we performed longitudinal analyses
to determine whether interactions in larger networks related to
better maintenance of global cognitive status at follow-up about
12 years later. Again, exposures in larger networks might offer
more opportunity for emotional support, cognitive stimulation,
or other beneficial functions. At follow-up, we used measures
that captured (1) more frequent contacts in larger networks—
a plausible marker for high exposure to diverse experiences and
resources—suggesting functions of engagement that are
distinctive of emotional support in the strictest sense (e.g.,
informational or cognitive stimulation), and (2) levels of
exposure to emotional support. We hypothesized that these
measures would each positively and uniquely relate to
cognition because they probe exposure levels and denote
putatively distinct functions. Although we could not ascertain
that those functions were causally related to cognition, the
demonstration of associational paths would suggest that
We used baseline data at Wave 1 (1981) and follow-up data
an average of 12.4 years later during Wave 3 (1993–1996) of
the Epidemiologic Catchment Area (ECA) survey at the
Baltimore site. Wave 2 was conducted in 1982. The study
was approved by the Committee on Human Research at the
Bloomberg School of Public Health, Johns Hopkins University.
Background and methodology of the ECA survey have been
described in detail elsewhere (Eaton & Kessler, 1985). Of 4,238
ability sampling methods and 3,481 interviews successfully
completed, 1,920 were available for reinterview at Wave 3
(73% of survivors; see Eaton et al., 1997). Of 881 participants
who met the age and Mini-Mental State Examination (MMSE;
Folstein, Folstein, & McHugh, 1975) criteria for inclusion in the
data analyses (age ? 50 and MMSE ? 28 at Wave 1), we used
354 in the longitudinal data analyses (420 were lost to follow-up
and 107 had missing longitudinal data, including 88 with
missing Wave 3 MMSE scores). Of these 354 individuals, about
5.4% had missing cross-sectional data at Wave 3 and were
deleted from the analyses for which data were missing (see
n values in Table 1). Table 1 compares the criterion age–MMSE
assessed and nonassessed participants on the variables used in
the study. In summary, of some of the statistically significant
comparisons, the assessed participants were younger, had more
years of formal education, included a greater percentage of
females, and had higher baseline MMSE scores than the
Table 1. Comparison of Assessed and Nonassessed Participants
Aged 50þ with MMSE Scores of 28þ
(n ¼ 354)
(SD, range)29.1 (0.8, 28–30)
MMSE W326.5 (2.8, 10–30)
Network sizebW1 6.8 (2.2, 0–10)
(n ¼ 527)
28.9 (0.8, 28–30)****
6.2 (2.4, 0–10)***
(n ¼ 471)
66.0 (9.2, 50–91)d****
9.6 (2.8, 0–17)****
(n ¼ 526)
9.6 (1.5, 9–24)d****
(n ¼ 517)
1.2 (1.4, 0–9)
(n ¼ 513)
Mean age W1
Gender (% female)
Race (% white)
6.4 (2.2, 0–10)
61.3 (6.9, 50–81)
10.4 (2.8, 0–17)
9.3 (0.9, 9–17)
9.9 (2.0, 9–22)
1.1 (1.2, 0–7)
.9 (1.2, 0–6)
19.2 (5.2, 6–27)
(n ¼ 341)
7.3 (1.9, 1–10)
(n ¼ 345)
13.8 (3.4, 2–20)
(n ¼ 345)
Frequency of contactbW3—k
W3 (% lifetime)
Alcohol use disordern
W3 (% lifetime)
Notes: W1 and W3 refer to measures taken at Waves 1 and 3; MMSE ¼
Mini-Mental State Examination.
There were missing data for some assessed participants on frequency of
contact and interpersonal activity (present n ¼ 345) and emotional support
(present n ¼ 341); see the Methods section. For nonassessed participants,
Wave 3 data are not reported and comparisons not attempted because of drop-
outs and missing data (see Methods section). In these cases, ns are given in
specific notes for the number of scores available.
an ¼ 19;
Whitney tests conducted because of heterogeneity of variance yielded p values
of at least this magnitude;
ghigher scores reflect more dysphoria;hn ¼ 65;ihigher scores reflect more
support;jn ¼ 28;kn ¼ 30;ln ¼ 28;mn ¼ 28;nfrom a derived variable at
* v2p , .05; *** p , .001; **** p , .0005.
bhigher scores reflect more engagement;
cn ¼ 28;
ehigher scores reflect more disability;
fn ¼ 68;
SOCIAL NETWORK CHARACTERISTICS AND COGNITION
by guest on October 21, 2015
Measures of Cognition and Social Networks
We computed cognitive change by using total MMSE scores
(0–30). The MMSE item asking which floor of the building was
not included at Wave 3, and a scoring accommodation was
made. At both waves, of the items to spell world backward and
to count serial sevens, the one producing the highest total
MMSE score for the participant was used. In addition, at both
waves, participants were asked to name two main streets nearby
and not location by county. The assessed social network factors
differed at the two waves in that frequency of contact and level
of emotional support measures were available at Wave 3 but
not at Wave 1. At both waves, the introductory instructions
stipulated that the social network items referred to persons with
whom the respondent had a personal relationship and kept in
touch most of the time, and that the researchers were
particularly interested in the previous 6 months. Respondents
were cautioned to not include persons they did not know well.
Network size Wave 1.—Two survey items asked (1) the
number of relatives and family members outside the household
and (2) the number of friends and neighbors with whom the
respondent kept in touch by phone or visits. The categories of
response for each of these items were 0, 1, 2–3, 4–5, 6–10, and
11þ; these were given scores of 0–5, and summed (range of
possible scores ¼ 0–10).
Network size Wave 3.—This was assessed as at Wave 1.
Frequency of contact Wave 3.—Two additional items at
Wave 3 asked the frequency of contact by phone or ‘‘getting
together,’’ scored 0–5 (never, less than once/month, about once/
month, few times/month, few times/week, and most every day)
for each of the two categories of relationships (relatives or
family and friends or neighbors; range of possible scores ¼
Emotional support Wave 3.—These items asked how much
respondents believed spouse or partners, relatives, and friends
cared about them, could rely on them for help for a serious
problem, and could relax—be themselves with them. The items
were scored 0–3 for increasing support each from spouse or
partner, other relatives, and friends, and the items were summed
across the three items and the relationship categories. There
were 190 participants with missing data for the items relating to
emotional support from a spouse or partner, and these items
were scored as zero. There were 195 persons in the sample who
were not currently married. No currently married persons had
missing data for these items unless they were not living with
their spouse (one case). The summed scores had a possible
range of 0 to 27.
Cognition at baseline.—To the extent that low MMSE
scores reflect diminished capacities that result in smaller
networks, the central issue of the contribution of larger
networks to the maintenance of normal cognitive function
could be obscured. Thus, we included only persons who scored
28 or above on the MMSE at baseline in the analyses.
According to population-based norms established with data
from the five-site ECA survey (Crum, Anthony, Bassett, &
Folstein, 1993), 28 is the mean and median score for our sample
mean baseline age (as well as our sample mean age plus mean
By comparison with our selection criteria, Seeman, Lu-
signolo, and colleagues (2001) used the upper tertile of scores
on the Short Portable Mental Status Questionnaire (SPMSQ;
Pfeiffer, 1975) and delayed recall of a short story as cognitive
criteria (they had physical criteria as well) in constituting
a subsample of relatively high-functioning persons from
a community-based sample aged 70–79 years (SPMSQ criterion
¼ 6 of 9 correct; see Berkman et al., 1993, the MacArthur
Studies of Successful Aging). Bassuk and associates (1999)
studied noninstitutionalized persons aged 65þ years old and
excluded persons with SPMSQ scores , 7 at the beginning of
an assessment interval from analyses of decline in that interval.
Control covariates.—The covariate measures in the analyses
were cerebrovascular disease or risk (CVD), age, education,
depressive symptomatology at testing, race, gender, physical
disability, and alcohol use disorder (abuse or dependence).
Previous research relates these covariates to the network
predictors or cognition; for example, links to cognition obtain
with CVD (Logroscino, Kang, & Grodstein, 2004; Prencipe et
al., 2003), sociodemographic variables (Graham et al., 1997;
O’Connor, Pollitt, Treasure, Brook, & Reiss, 1989), and
depressive symptomatology (Dufouil, Fuhrer, Dartigues, &
We categorized CVD as being present if the respondent
indicated having ever had either high sugar levels or diabetes,
stroke, high blood pressure, or heart trouble at Wave 3. We
assessed physical disability from nine items extracted from the
ECA survey at Waves 1 and 3 relating to basic activities such as
bathing, dressing, use of arms to reach, and use of fingers to
grasp or handle—each scaled from 1 to 3 such that 1 ¼
performance of the task without difficulty, 2 ¼ performance
with difficulty, and 3 ¼ inability to perform. Five participants
had one disability item response of ‘‘don’t know,’’ and we
assigned these responses a value of 2. We measured dysphoria
at both Waves with three items on the General Health
Questionnaire (Goldberg & Hillier, 1979) that asked respond-
ents to self-assess feelings of unhappiness and depression,
hopelessness, and worthlessness over the past few weeks
(scored 0 to 3 for each of the three items indicating absent, as
much as usual, more than usual, much more than usual). The
internal consistency of these items was a ¼ .74 at Wave 1 and
a¼.62 at Wave 3. We derived the presence of lifetime alcohol
use disorder from the Diagnostic Interview Schedule—
a structured interview administered by trained laypersons—
(see Robins, Helzer, Ratcliff, & Seyfried, 1981), which was
based on Diagnostic and Statistical Manual of Mental
Disorders, third edition, revised (DSM-III-R; American
Psychiatric Association, 1987) criteria at Wave 3.
There were 226 participants who reported having (had) one
or more indicators of CVD. There were 26 participants who
indicated a history of stroke, with 20 of these reporting that the
most recent occurrence was more than 1 year prior to the Wave
3 interview, 5 reporting 6–12 months prior, and 1 reporting an
indeterminate recency. The mean (standard deviation) Wave 3
HOLTZMAN ET AL.
by guest on October 21, 2015
MMSE score for the 20 cases was 26.7 (2.2) compared with
26.5 (2.8) for those participants without stroke. The mean Wave
3 MMSE for the 5 cases was 24.6 (3.6). There were 26 persons
who had alcohol use disorder. At Wave 1, 307 participants were
without disability as were 241 at Wave 3. At Wave 1, 159
participants were nondysphoric, as were 189 at Wave 3.
We conducted the multivariate analyses by using simulta-
neous linear or logistic regression. We used Huber–White–
sandwich robust standard errors in the linear models because of
heteroscedasticity in the residuals. We conducted all analyses
by using the STATA 7.0 statistical package (Stata Corporation,
College Station, TX).
Baseline Network Size and Cognitive Change
The mean MMSE at Wave 1 was 29.1 (SD¼.8; range¼28–
30). At Wave 3 the mean was 26.5 (SD¼2.8; range¼10–30).
MMSE change (?MMSE) was M ¼ ?2.6 (SD ¼ 2.7, with
a range of ?19 to þ2).
We conducted the longitudinal analyses to determine
whether interactions in larger baseline networks predicted
interwave ?MMSE. We computed change in MMSE as
residualized change with baseline MMSE scores entered as
a predictor of Wave 3 MMSE scores.
The covariates were ?physical disability and ?dysphoria,
MMSE at baseline, lifetime presence of alcohol disorder and
CVD status as of Wave 3, age, education level, gender, and race
(White or non-White). We entered disability change as
increased (n ¼ 59) or decreased (n ¼ 13), with unchanged as
the reference (n ¼ 282); dysphoria change was entered as
increased (n¼43) or decreased (n¼61), with unchanged as the
reference (n¼250). Change indicated that the interwave scores
differed by 2þ points (a difference of approximately 1þ SDs).
We also included a term for ?network size to assess its
association with ?MMSE in the interwave interval.
The interwave decrease in MMSE scores was statistically
greater than zero, paired t(353) ¼ 18.03, p , .0005. The
bivariate and fully adjusted associations of the predictors with
?MMSE are given in Table 2. In the adjusted model, a linear
effect obtains for baseline network size (p¼.006; effect size¼
.06). The beta weight for network size indicates that its
contribution to the explained variance is relationally small. For
example, age accounts for about 3.2 times the variance that
network size does. The model also shows that less increase–
more decrease in interwave network size relates to decreased
Wave 3 MMSE (p ¼ .03; effect size ¼ ?.06), indicating an
independent linear ‘‘concurrent’’ association between these two
We used logistic regression with an end-point MMSE cutoff
score ? 26 (n ¼ 128) to further illustrate the relationship
between baseline network size and maintenance of MMSE.
This is the population-based lower quartile cutoff score for our
sample mean education and end-point age (Crum et al., 1993).
In the fully adjusted model, for a given network size, the odds
of a score ? 26 were 84% those for that size ?1 (odds ratio or
OR ¼ .84, p ¼ .01). The odds of a score ? 26 for a given
interwave change in network size were 115% those for a change
with one unit more increase–less decrease (OR¼1.15, p¼.02;
complete models for these and subsequent summarized results
are available on request from George W. Rebok).
Eighteen Wave 3 MMSE scores could be considered outliers,
that is, ?fQ1 ? [1.5 (Q3 ? Q1)]g; MMSE ? 20. A reanalysis
of the data in a linear regression excluding these scores showed
the reported associations for maintenance of MMSE with
baseline network size (p ¼ .02) and network size change (p ¼
.02). A reanalysis excluding scores , 24 (the often-used
indicator of possible dementia; n ¼ 45) revealed a significant
positive linear association between baseline network size and
maintenance of MMSE (p ¼ .04).
Network Measures and Cognitive Function at Wave 3
The longitudinal analyses established that a relationship
exists between interactions in larger social networks and
maintenance of MMSE. However, we were not able to specify
which function(s) of those interactions might benefit cognitive
health. We had measures of frequency of contact and levels of
emotional support in addition to network size at Wave 3, and
we used these data to assess the possibility of distinct
independent functions of social networks that contribute to
cognitive health. As explained earlier, we hypothesized that
more frequent contact in larger networks would reflect
Table 2. Bivariate and Adjusted Associations With MMSE Change
B SE B
B SE B
Baseline network size
0.13 .06.11* 0.18 .06.14**
Increase at W3
Decrease at W3
Increase at W3
Decrease at W3
Notes: For categorical variables, the indicator term(s) are given and com-
pared with the reference class, which for change terms is absence of change.
A positive coefficient indicates better maintenance of Mini-Mental State Ex-
amination (MMSE) scores. W1 and W3 refer to measures taken at Waves 1
and 3; ? ¼ change.
aPredictors are entered with baseline MMSE.bEquation for the adjusted
model is F(13, 340) ¼ 5.92, p , .0005, R2¼ .27.cLess increase–more de-
crease in interwave network size relates to decreased Wave 3 MMSE.
* p , .05; ** p , .01; *** p ? .001; **** p , .0005.
SOCIAL NETWORK CHARACTERISTICS AND COGNITION
by guest on October 21, 2015
a function relative to cognition that is independent of emotional
support. In order to test this hypothesis, we created the variable
of interpersonal activity by summing network size and
frequency of contact. Note that, whereas the network size
variable was constructed to include only those persons with
whom there was contact most of the time, it did not stipulate the
frequency of interaction. Considering minimum and mean
values of frequency and size across tertiles of interpersonal
activity scores, we found that higher tertiles of this variable had
more frequent contacts and larger networks.
Table 3 contains the correlations between all the network
measures at Wave 3, and those measures with cognition. We
conducted a series of linear regression analyses by using the
network measures to predict cognition, adjusting for age,
gender, education, race, CVD, alcohol use disorder, physical
disability, and dysphoria. Tertiles of physical disability and
dysphoria each represented the smallest difference in ns
between the reference classes of no disability and no dysphoria,
and two successively higher levels of each.
In an initial series of these analyses, each regression model
contained a single network measure entered as a continuous and
a categorical variable (the latter, tertiles with the smallest
differences in ns) to detect linear and nonlinear associations
between the network measure and MMSE. For example, the
network size models were (1) MMSE¼ network size (continu-
ous)þcovariates and (2) MMSE¼network size (categorical)þ
covariates. We report the results for the categorical variables for
the top tertile only. In these and subsequent analyses, all of the
models were significant (ps , .0005).
In the initial series, frequency of contact was the only
network measure that did not significantly predict MMSE:
continuous model Fs(11, 333) ¼ 7.57 and 8.37, frequency of
contact bs ¼ .10, ps ? .051; categorical model Fs(12, 332) ¼
6.71 and 7.36, frequency of contact bs¼.07, ps ? .21. (Two p
values were obtained for each continuous and categorical
network measure because there were two alternative criterion
tertile arrangements for the physical disability covariate,
resulting in somewhat different p values for the network
features, depending on which arrangement was entered.) Thus,
frequency of social contact by itself might not provide diverse
experiences with the potential for cognitive stimulation (e.g.,
frequent contact could be with one person), or ensure emotional
support, or substantively tap other functions of social in-
teraction that relate positively to cognitive health (see
Fratiglioni et al., 2000, for a comparable finding for mere
frequency of contact).
In contrast, network size was a significant predictor of
MMSE scores: continuous model Fs(11, 342)¼7.05 and 6.53,
network size bs¼.15, ps¼.004; categorical model Fs(12, 341)
¼ 6.29 and 5.85, network size bs ¼ .13 and .12, ps ? .03. So
was interpersonal activity: continuous model Fs(11, 333) ¼
8.41 and 9.09, activity bs ¼ .14, ps ? .006; categorical model
Fs(12, 332)¼7.78 and 8.59, activity bs¼.15, ps ? .009. So too
was emotional support: continuous model Fs(11, 329) ¼ 7.07
and 7.83, support bs¼.15, ps ? .005; categorical model Fs(12,
328) ¼ 6.35 and 7.11, support bs ¼ .18 and .19, ps ? .004.
Then, in the next series of analyses, we paired interpersonal
activity and network size each with emotional support to assess
independent paths with cognition in adjusted models. That is,
the paired interpersonal activity and support models were (1)
MMSE ¼ interpersonal activity (continuous) þ emotional
support (continuous) þ covariates, and (2) MMSE ¼ in-
terpersonal activity (categorical) þ emotional support (categor-
ical) þ covariates. We likewise paired network size and
The results of these analyses showed that, when we paired
network size and emotional support as continuous measures,
they were not both significantly related to MMSE scores—
continuous model Fs(12, 328) ¼ 7.31 and 8.07; network size
bs ¼ .10, ps ? .08, and support bs ¼ .11, ps ¼ .06. They also
were not both significant when paired as categorical variables—
categorical model Fs(14, 326) ¼ 5.84 and 6.53; network size
bs¼.09, ps ? .14, and support bs¼.16, ps ? .02. However, the
top tertiles of interpersonal activity and emotional support were
both significant when these features were paired as categorical
variables: Fs(14, 320) ¼ 6.33 and 7.09; activity bs ¼ .13 and
.12, ps ? .04, and support bs ¼ .16 and .17, ps ? .01; effect
size¼.26 for activity and .35 for support, with, in these models,
MMSE SD ¼ 2.6. When we paired them as continuous
variables, activity and support were both significant given one
of the disability tertile arrangements—F(12, 322) ¼ 7.09;
activity b ¼ .11, p ¼ .047 and support b ¼ .12, p ¼ .04—but
not the other arrangement—F(12, 322)¼7.75; activity b¼.10,
p ¼ .06 and support b ¼ .12, p ¼ .04. The results for the
categorical variables indicate that more frequent contact in
larger networks (most marked in the top tertile of interpersonal
activity) and higher levels of emotional support have in-
dependent positive associations with cognition, as we hypoth-
esized. We repeated the analyses in a subsample excluding
MMSE scores , 24, and there was no evidence to support the
The longitudinal models showed that interactions in larger
social networks at Wave 1 related to better maintenance of
MMSE at Wave 3 and reduced odds of decline from the
population-based median cutoff score (at minimum) to a lower
quartile score. This association could reflect benefits of greater
exposure to variety or novelty in resources or experiences,
increased opportunities for emotional support, or other
functions. We did not have the Wave 1 measures necessary
to partition out effects associated with emotional support
longitudinally. We cannot estimate the extent to which the
Table 3. Whole Sample Correlations Between Network Measures
and Wave 3 MMSE
Network Size þ
Network size þ
Notes: MMSE ¼ Mini-Mental State Examination.
Any differences in p values when Spearman rho correlations are used are
given in specific lettered notes:ap , .01;bp , .06;cp , .25;dp , .05.
* p , .05; ** p , .01; *** p , .001; **** p , .0005.
HOLTZMAN ET AL.
by guest on October 21, 2015
decline in MMSE scores is due to regression to the mean in the
The association between baseline size and MMSE change
manifested independent of a significant concurrent interwave
change in both size and MMSE. This concurrent change might
represent counterdirectional or bidirectional effects involving
network size and cognition. In this regard, although baseline
MMSE scores indicated that the sample was initially high
functioning, it is possible that some participants were in decline
at Wave 1 if we consider underlying cognitive capacities that
did not register as lower MMSE scores at Wave 1 (related
issues are addressed in, e.g., Coyle, 2003; Geerlings, Jonker,
Bouter, Ader, & Schmand, 1999; and Katzman, 1995), which
might affect baseline size. The conservative position is that
interaction in a larger network is a marker that portends less
decline in global cognitive function as measured by the MMSE.
There were significant independent effects of education, and
race, in addition to age, on the maintenance of cognitive
function in the whole-sample linear and logistic regression
models (see Table 2 for the linear model). Higher levels of
education might create brain reserve and then better mainte-
nance of cognitive function (Katzman, 1993). In addition, lower
levels of education and non-White race index lower socioeco-
nomic strata, and, as reviewed in Seeman and Crimmins (2001),
socioeconomic status is inversely related to a broad spectrum of
health outcomes (including cognitive health; see Lynch,
Kaplan, & Shema, 1997), with paths of effect that have not
been completely identified.
The results of the cross-sectional analyses confirmed our
hypothesis that more frequent contacts in larger networks and
level of emotional support independently relate positively to
cognitive health. The results of Arbuckle and colleagues (1992)
and Seidler and colleagues (2003), in their respective ways, also
imply the existence of independent paths linking emotional
It is important that our subsample cross-sectional analyses
that excluded MMSE scores , 24 did not reveal independent
paths. Thus, our partitioned social network functions might
reflect causes or effects that distinguish normal from morbid
cognitive ability, and not functioning within a normal range.
However, the MMSE is itself meant to detect possible clinical
cognitive dysfunction and not gradations of normal abilities, so
that it would be worthwhile to determine whether our network
characteristics relate to these normal gradations. Interestingly,
using a battery of cognitive tests in a cross-sectional analysis on
their baseline, relatively high-functioning sample, Seeman,
Lusignolo, and colleagues (2001) found an independent (of
emotional support) positive association with the frequency of
social demands or conflict, which they interpret as possibly
indexing complex social interactions. Similarly, as reported
earlier, the cross-sectional analyses of Arbuckle and associates
(1992) support independent paths using measures of general
intelligence and memory in persons living independently in the
community. More finely tuned cognitive measures in the
present study might also yield a greater effect for network size
than that obtained in the longitudinal analysis (see Table 2).
Whether more frequent phone or visit contacts in larger
networks produce or reflect better cognitive functioning is open
to question. If causal of better cognition, the benefit might
derive from experiencing more variety, novelty, or challenge
from events (e.g., trips to museums or other cities, new
recreational or church activities; see, e.g., Murrell, Norris, &
Chipley, 1992), or from handling informational or interactional
complexities that arise more often or prominently with frequent
encounters in larger networks (e.g., scheduling dates and
events, reaching for the best way to communicate thoughts,
anticipating the responses of others). Situations such as these
conceivably stimulate cognition. Significantly decreased risk
for the clinical expression of dementia has been associated with
higher levels of participation in cognitive activities such as
reading and playing board games (Verghese et al., 2003; Wang
et al., 2002; Wilson et al., 2002). We speculate as to the
mechanism(s), and direction, of effect for more frequent contact
in larger networks, but if it involves higher levels of cognitive
stimulation, then our results would be consistent with those
longitudinal findings, considering, too, the apparent contribu-
tion of MMSE scores , 24 to our results.
As we noted previously, there is a possible link between
emotional support, decreased physiological stress (cardiovas-
cular or neuroendocrine response), and cognitive health. In
addition, positive effects of support on lifestyle or behaviors
(e.g., eating, smoking, exercise, adherence to medication
regimens) might contribute to cognitive health (see review
pertinent to social support and health behaviors, and biopsy-
chosocial model in Seeman & Crimmins, 2001).
Given the long prodromal period before the expression of
clinical symptomatology in Alzheimer’s disease, longitudinal
studies with relatively short follow-up intervals, let alone cross-
sectional studies, are particularly vulnerable to a directionality
confound (see, e.g., Fabrigoule et al., 1995 and Friedland et al.,
2001). However, counterdirectional effects do not refute the
existence of distinct pathways in our study.
It is noteworthy that the assessed persons were significantly
younger, more highly educated, more physically able, and had
significantly higher MMSE scores and larger networks at
baseline than the nonassessed persons who met our age and
MMSE criteria. The generalizability of our results may thus be
In summary, we found evidence that interaction in larger
networks at baseline was positively related to maintenance of
global cognitive function about 12 years later. We also showed
that more frequent contact in larger networks has a distinctive
positive path of association with cognition, apart from an
emotional support function. Research that more closely
specifies these associations will help to inform strategies for
practical preventive interventions with aging populations.
Jane S. Saczynski is now at the Laboratory of Epidemiology,
Demography, and Biometry, National Institute on Aging. Kathryn Wilcox
Doyle is now with Psychological Counseling Services.
This research was supported by the National Institute of Mental Health
under Grants T32-MH18834, MH47447, and T32-14592. A partial report
of this study was presented as a poster at the annual meeting of the
American Psychological Association, August 2002, Chicago, IL.
We thank Elizabeth Johnson and Jeannie-Marie Sheppard for providing
statistical expertise pertinent to some of the analyses.
Address correspondence to George W. Rebok, Department of Mental
Health, Bloomberg School of Public Health, Johns Hopkins University, 624
North Broadway, Baltimore, MD 21205. E-mail: email@example.com
SOCIAL NETWORK CHARACTERISTICS AND COGNITION
by guest on October 21, 2015
REFERENCES Download full-text
American Psychiatric Association. (1987). Diagnostic and statistical
manual of mental disorders (Rev. 3rd ed.). Washington, DC: Author.
Arbuckle, T. Y., Gold, D. P., Andres, D., Schwartzman, A., & Chaikelson,
J. (1992). The role of psychosocial context, age, and intelligence in
memory performance of older men. Journal of Psychology and Aging,
Barrera, M. (1986). Distinctions between social support concepts,
measures, and models. American Journal of Community Psychology,
Bassuk, S. S., Glass, T. A., & Berkman, L. F. (1999). Social disengagement
and incident cognitive decline in community-dwelling elderly persons.
Annals of Internal Medicine, 131, 165–173.
Berkman, L. F., Seeman, T. E., Albert, M., Blazer, D., Kahn, R., & Mohs,
R., et al. (1993). High, usual and impaired functioning in community-
dwelling older men and women: Findings from the MacArthur
Foundation Research Network on Successful Aging. Journal of Clinical
Epidemiology, 46, 1129–1140.
Cohen S. (1988). Psychosocial models of the role of social support in the
etiology of physical disease. Health Psychology, 7, 269–297.
Cohen, S. & Wills, T. A. (1985). Stress, social support, and the buffering
hypothesis. Psychological Bulletin, 98, 310–357.
Coyle, J. T. (2003). Use it or lose it—do effortful mental activities protect
against dementia? New England Journal of Medicine, 348, 2489–2490.
Crum, R. M., Anthony, J. C., Bassett, S. S., & Folstein, M. F. (1993).
Population-based norms for the Mini-Mental State Examination by age
and educational level. Journal of the American Medical Association,
Dufouil, C., Fuhrer, R., Dartigues, J.-F., & Alperovitch, A. (1996).
Longitudinal analysis of the association between depressive symptom-
atology and cognitive deterioration. American Journal of Epidemiology,
Eaton, W. W., & Kessler, L. G. (Eds.). (1985). Epidemiologic Field
Methods in Psychiatry: The NIMH Epidemiological Catchment Area
Program. Orlando, FL: Academic Press.
Eaton, W. W., Anthony, J. C., Gallo, J., Cai, G., Tien, A., & Romanoski,
A., et al. (1997). Natural history of Diagnostic Interview Schedule/
DSM-IV major depression. The Baltimore Epidemiologic Catchment
Area follow-up. Archives of General Psychiatry, 54, 993–999.
Fabrigoule, C., Letenneur, L., Dartigues, J. F., Zarrouk, M., Commenges,
D., & Barberger-Gateau, P. (1995). Social and leisure activities and risk
of dementia: A prospective longitudinal study. Journal of the American
Geriatrics Society, 43, 485–490.
Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975). ‘‘Mini-Mental
State.’’ A practical method for grading the cognitive state of patients for
the clinician. Journal of Psychiatric Research, 12, 189–198.
Fratiglioni, L., Wang, H.-X., Ericsson, K., Maytan, M., & Winblad, B.
(2000). Influence of social network on occurrence of dementia: A
community-based longitudinal study. Lancet, 355, 1315–1319.
Friedland, R. P., Fritsch, T., Smyth, K. A., Koss, E., Lerner, A. J., & Chen,
C. H., et al. (2001). Patients with Alzheimer’s disease have reduced
activities in midlife compared with healthy control-group members.
Proceedings of the National Academy of Sciences of the United States
of America, 98, 3440–3445.
Geerlings, M. I., Jonker, C., Bouter, L. M., Ader, H. J., & Schmand, B.
(1999). Association between memory complaints and incident Alz-
heimer’s disease in elderly people with normal baseline cognition.
American Journal of Psychiatry, 156, 531–537.
Goldberg, D. P., & Hillier, V. F. (1979). A scaled version of the General
Health Questionnaire. Psychological Medicine, 9, 139–145.
Graham, J. E., Rockwood, K., Beattie, B. L., Eastwood, R., Gauthier, S., &
Tuokko, H., et al. (1997). Prevalence and severity of cognitive
impairment with and without dementia in an elderly population.
Lancet, 349, 1793–1796.
Hultsch, D. F., Hammer, M., & Small, B. J. (1993). Age differences in
cognitive performance in later life: Relationships to self-reported health
and activity life style. Journal of Gerontology: Psychological Sciences,
Katzman, R. (1993). Education and the prevalence of dementia and
Alzheimer’s disease. Neurology, 43, 13–20.
Katzman, R. (1995). Can late life social or leisure activities delay the onset
of dementia? Journal of the American Geriatrics Society, 43, 583–584.
Logroscino, G., Kang, J. H., & Grodstein, F. (2004). Prospective study of
type 2 diabetes and cognitive decline in women aged 70–81 years.
British Medical Journal, 328, 548–551 (abridged). Full text version
posted at http://bmj.com/cgi/doi/10.1136/bmj.37977.495729.EE).
Lupien, S., Roch Lecours, A., Lussier, I., Schwartz, G., Nair, N. P. V., &
Meany, M. J. (1994). Basal cortisol levels and cognitive deficits in
human aging. The Journal of Neuroscience, 14, 2893–2903.
atrophy and memory deficits. Nature Neuroscience, 1, 69–73.
Lynch, J. W., Kaplan, G. A., & Shema, S. J. (1997). Cumulative impact of
sustained economic hardship on physical, cognitive, psychological, and
social functioning. New England Journal of Medicine, 25, 1889–1895.
Murrell, S. A., Norris, F. H., & Chipley, Q. T. (1992). Functional versus
structural social support, desirable events, and positive affect in older
adults. Psychology and Aging, 7, 562–570.
O’Connor, D. W., Pollitt, P. A., Treasure, F. P., Brook, C. P. B., & Reiss,
B. B. (1989). The influence of education, social class and sex on Mini-
Mental State scores. Psychological Medicine, 19, 771–776.
Pfeiffer, E. (1975). A short portable mental status questionnaire for the
assessment of organic brain deficit in elderly patients. Journal of the
American Geriatrics Society, 23, 433–441.
Prencipe, M., Santini, M., Casini, A. R., Pezzella, F. R., Scaldaferri, N., &
Culasso, F. (2003). Prevalence of non-dementing cognitive disturbances
and their association with vascular risk factors in an elderly population.
Journal of Neurology, 250, 907–912.
Robins, L. N., Helzer, J. E., Ratcliff, K. S., & Seyfried, W. (1981). National
Institute of Mental Health Diagnostic Interview Schedule: Its history,
characteristics and validity. Archives of General Psychiatry, 38, 381–
Seeman, T. E., Berkman, L. F., Blazer, D., & Rowe, J. W. (1994). Social
ties and support and neuroendocrine function: the MacArthur Studies of
Successful Aging. Annals of Behavioral Medicine, 16, 95–106.
Seeman, T. E., & Crimmins, E. (2001). Social environment effects on
health and aging. Integrating epidemiologic and demographic ap-
proaches and perspectives. In M. Weinstein, A. I. Hermalin, & M. A.
Stoto (Eds.), Population health and aging. Strengthening the dialogue
between epidemiology and demography (pp. 88–117). New York: The
New York Academy of Sciences.
Seeman, T. E., Lusignolo, T. M., Albert, M., & Berkman, L. (2001). Social
relationships, social support, and patterns of cognitive aging in healthy,
high-functioning older adults: MacArthur Studies of Successful Aging.
Health Psychology, 20, 243–255.
Seeman, T. E., McEwen, B. S., Rowe, J. W., & Singer, B. H. (2001).
Allostatic load as a marker of cumulative biological risk: MacArthur
studies of successful aging. Proceedings of the National Academy of
Sciences of the United States of America, 98, 4770–4775.
Seeman, T. E., Singer, B. H., Rowe, J. W., Horwitz, R. I., & McEwen, B. S.
(1997). Price of adaptation—allostatic load and its health consequences.
MacArthur Studies of Successful Aging. Archives of Internal Medicine,
Seidler, A., Bernhardt, T., Nienhaus, A., & Frolich, L. (2003). Association
between the psychosocial network and dementia—a case-control study.
Journal of Psychiatric Research, 37, 89–98.
Uchino, B. N., Holt-Lunstad, J., Uno, D., Betancourt, R., & Garvey, T. S.
(1999). Social support and age-related differences in cardiovascular
function: An examination of potential mediators. Annals of Behavioral
Medicine, 21, 135–142.
Verghese, J., Lipton, R. B., Katz, M. J., Hall, C. B., Derby, C. A.,
Kuslansky, G., et al. (2003). Leisure activities and the risk of dementia
in the elderly. New England Journal of Medicine, 348, 2508–2516.
Wang, H.-X., Karp, A., Winblad, B., & Fratiglioni, L. (2002). Late-life
engagement in social and leisure activities is associated with a decreased
risk of dementia: A longitudinal study from the Kungsholmen Project.
American Journal of Epidemiology, 155, 1081–1087.
Wilson, R. S., Mendes de Leon, C. F., Barnes, L. L., Schneider, J. A.,
Bienias, J. L., Evans, D. A., et al. (2002). Participation in cognitively
stimulating activities and risk of incident Alzheimer disease. Journal of
the American Medical Association, 287, 742–748.
Received April 15, 2003
Accepted July 14, 2004
Decision Editor: Margie E. Lachman, PhD
HOLTZMAN ET AL.
by guest on October 21, 2015