Journal of Gerontology: PSYCHOLOGICAL SCIENCES Copyright 1999 by The Gerontological Society ofAmerica

1999, Vol. 54B, No. 4, P231-P239

A Model of Loneliness in Older Adults

Bronwyn S. Fees,1 Peter Martin,23 and Leonard W. Poon4

'School of Family and Human Services, Kansas State University, Manhattan.

2
Department of Human Development and Family Studies, Iowa State University, Ames.

Downloaded from https://academic.oup.com/psychsocgerontology/article-abstract/54B/4/P231/693063 by guest on 15 June 2020

3
German Center for Aging Research, University of Heidelberg, Germany.
4
Gerontology Center, University of Georgia, Athens.

Loneliness and physical health status in older adults have been correlated strongly but the predictive direction is unclear. This
study examined the relationship between personality, cognition, social network, and age modeled as predictors of loneliness
in older Americans. Self-assessed health mediated the relationship. The sample consisted of 208 independently living individ-
uals 60 to 106 years of age from the southern region of the United States. Model comparison revealed health did not mediate
the relationship significantly but that self-reported loneliness itself mediated between personal characteristics and perceived
health. Results indicate anxiety, frequency of telephone contact, and age, but notfrequency of face-to-face contact with
others or cognitive functioning, affect perceived loneliness. Perceived loneliness mediates the effects of anxiety, frequency of
telephone contact, and age on self-assessed health. Feelings of loneliness decrease one's evaluation of physical well-being.

health status and self-rated economic condition as mediators of

L ONELINESS is a pervasive issue among elderly adults, who
-/ often face a loss of committed intimate relationships and
communication with others, which are valued by the Western
the influence of demographic characteristics on loneliness.
The model we propose shares predictors with each of these
culture (de Jong-Gierveld, 1987), a decline in health, and a de- models; however, we believe the literature supports the inclu-
cline in personal resources. Developing a theoretical model of sion of personal traits as well as interpersonal relationships as
loneliness that represents the experiences of elderly persons in determinants of loneliness. Our model was based on a larger
everyday life has been a focus of research by gerontologists for conceptual framework that assesses successful adaptation in old
several decades. Loneliness is a concept that relies on compar- age (Poon et al., 1992). We proposed that mental health prob-
isons. It is defined as a sentiment that is experienced when one's lems in later life are, in part, dependent on individual character-
lifestyle (state) is deprived of the relationships desired and cur- istics including personality, level of cognitive functioning, level
rent relationships are seen as inadequate in comparison to those of social support (e.g., social network), and physical health (see
of the past, to those anticipated in the future, or to those pos- Figure 1). Physical well-being served as a mediator between
sessed by other people (Lopata, 1995; Weiss, 1973). The pur- constructs because decline in health remains a dominant issue
pose of this study was to examine predictors of loneliness and among older adults, compared to the younger population, po-
its relationship with self-assessed physical well-being. tentially limiting interaction in stimulating relationships, espe-
Forty percent of the elderly population has experienced some cially among the oldest.
form of loneliness according to data from Europe and the Martin, Hagberg, and Poon (1997) tested this conceptual
United States (Weeks, 1994). Weeks (1994) further suggested model across cultures by analyzing data from American and
that although this percentage has been relatively stable over the Swedish centenarians. Personality (conceptualized as anxiety),
last 25 years, it may be worse than it appears. This rate has been physical health, and social support were strong predictors of
cause for concern given that suicide, physical ailments, and de- loneliness in Americans, whereas social support and cognition
pression have been outcomes associated with the presence of predicted loneliness for Swedish centenarians. Data were ana-
loneliness in the elderly population (Creecy, Berg, & Wright, lyzed using Partial Least Squares Estimation (LVPLS) Soft
1985). Elderly persons have been regarded as particularly vul- Modeling (Falk & Miller, 1992).
nerable because they are considered at high risk for "experience In light of recent research suggesting stressors in life may
of change and loss" (Lopata, 1995, p. 572). have an adverse physiological affect on immune system func-
Predictors of loneliness vary not only with transitional life tioning (see review by Kiecolt-Glaser & Glaser, 1992), we ex-
events but also with increasing maturity (Dugan & Kivett, amined a second hypothesis in which loneliness served simul-
1994; Russell, 1996; Russell, Peplau, & Ferguson, 1978; taneously as a predictor for physical well-being and a mediator
Weeks, 1994). Several different theoretical models of loneliness for the remaining constructs (see Figure 2).
including older adult samples have been proposed. De Jong- The focus of the present analyses was to determine whether
Gierveld (1987) hypothesized that demographic characteristics, the model employed by Martin and colleagues (1997) was pre-
living arrangements, and personality characteristics predicted dictive for a more inclusive age range of older Americans using
loneliness. Creecy and colleagues (1985) included demographic structural equation modeling with simultaneous solutions and
characteristics, self-assessed health status, and income as pre- to further examine the relationship between physical well-being
dictors. Both authors used indicators of social involvement and loneliness. Chronological age was added as a predictor to
(e.g., social network, social activity) as mediating variables. the model to examine its effect. Each construct will be reviewed
Mullins, Elston, and Gutkowski (1996) proposed perceived separately in the following paragraphs.

P231
P232 FEESETAL

Downloaded from https://academic.oup.com/psychsocgerontology/article-abstract/54B/4/P231/693063 by guest on 15 June 2020

Figure 1. Conceptual Model of Loneliness in Older Adults. Figure 2. Conceptual Model of Physical Weil-Being in Older Adults.

Personality thinking and imagining situations beyond one's own situation.

Research has established a strong relationship between feel- Conversely, a high level of cognition may be a source of frus-
ings of loneliness and personality characteristics, particularly tration in that one is able to think of activities to do, but is un-
anxiety. Anxiety has been characterized by emotional instabil- able to pursue those goals. No relationship between cognition
ity, threat sensitivity, suspiciousness, guilt, low integration, and and loneliness was found in the American centenarians; how-
tension (Cattell, Eber, & Tatsuoka, 1970). Although anxiety ever, cognitive functioning was negatively related in Swedish
often manifests itself in reduced physical health, men and centenarians (Martin et al., 1997).
women who reported being lonely also reported being anxious
(often being treated for nervousness) and feeling depressed Social Network
(Berg, Mellstrom, Persson, & Svanborg, 1981; Russell, Peplau, A number of research studies have suggested that a decline
& Cutrona, 1980; Russell et al., 1978). An individual with a in or absence of social support is predictive of loneliness.
low self-concept (correlated with anxiety) may not have satis- Russell (1996) found that frequency of contact and number of
factory relationships with others and may not seek out new re- family members were not strongly predictive of loneliness;
lationships that indirectly affect loneliness (de Jong-Gierveld, however, the perceived quality of recent relationships was. Type
1987). Social phobia, apprehension, and fear of embarrassment of living arrangement has also been determined to affect loneli-
in public have been related to a decrease in interaction leading ness (de Jong-Gierveld, 1987). In a model of loneliness based
to loneliness as well (Weeks, 1994). upon persons 25-75 years of age, living with a partner signifi-
cantly and negatively predicted loneliness whereas being single
Cognitive Functioning significantly and positively predicted loneliness.
Decreased efficiency in mental processing has been observed Frequency of contact does appear to be predictive when the
in elderly adults, yet few studies include cognitive functioning type of relationship is considered. Mullins and Dugan's (1990)
directly as a predictor of loneliness. De Jong-Gierveld (1987) survey of residents in urban congregate housing determined
took a "cognitive processes" approach by examining respon- that frequency of contact with neighbors and friends, but not
dents' subjective evaluations and interpretations of their per- with family, was important to reducing the feelings of loneli-
sonal experiences rather than using objective measures of par- ness, as was the quality of the relationships with friends. In a
ticipants' intellectual functioning. Subjective evaluations of separate study, Dugan and Kivett's (1994) survey of rural inde-
relationships strongly predicted loneliness. Berg and colleagues pendently living elderly persons revealed that infrequent visits
(1981) found no differences in an objective measure of intellec- with family members (siblings) predicted loneliness as did the
tual functioning, verbal ability, between lonely and not lonely loss of a spouse. Lack of friends was an important predictor of
participants. Neither of these studies examined the effect of loneliness in Swedish 70-year-olds (Berg et al., 1981).
fluid intelligence, the ability to think logically and to reason ab- Because loneliness may be realized by a lack of communica-
stractly, on loneliness. tion with others (de Jong-Gierveld, 1987), we chose to measure
Level of cognitive functioning may affect loneliness in sev- social networks by the frequency of contact individuals had
eral ways. It may serve as a buffer against feelings of loneli- with others, either through visits or over the telephone. Each
ness; that is, as physical strength and coordination decline and contact represented an opportunity for communication and rela-
limit activity, one can remain cognitively inquisitive and alert, tionship building.
A MODEL OFLONEUNESS P233

Age ducted at the University of Georgia (N = 262). The multidisci-

Although a decline in physical health is common with age, re- plinary study collected data on three cohorts of cognitively in-
search literature has been inconclusive regarding the role of age tact, independently living individuals: sexagenarians, octoge-
in predicting loneliness among older adults. College students narians, and centenarians. Participant inclusion criteria included
have been found to report higher levels of loneliness than older a minimum score of 20 on the Mini-Mental Status Exam
adults (Russell, 1996). However, among rural adults 65 years of (Folstein, Folstein, & McHugh, 1975) given at the beginning of
age and older, Dugan and Kivett (1994) found that over half the interview/testing session and a Stage 2 or higher level on
(68%) of the sample reported social loneliness "sometimes" or the Global Deterioration Scale (Reisberg, Ferris, De Leon, &

Downloaded from https://academic.oup.com/psychsocgerontology/article-abstract/54B/4/P231/693063 by guest on 15 June 2020

"quite often." The authors concluded this rate was similar to that Crook, 1982). Data from each of the three cohorts were aggre-
of the general population of persons aged 65 years and older. gated in the following analyses.
Creecy and colleagues (1985) concluded that age had an in- Participants were recruited primarily from the state of
direct effect on loneliness that was mediated by social activity Georgia by the University of Georgia at Athens (UGA) Survey
and social fulfillment, explaining less than 1% of the variance Research Center. Interviews and testing of centenarians were
in loneliness. Similar results were reported by de Jong-Gierveld conducted at the participant's place of residence. Groups of
(1987), who found that age did not have a significant direct ef- 6-10 sexagenarians and octogenarians met in a common place
fect on loneliness in persons 25-75 years old. However, others to be interviewed and tested. Two thirds of the sample were fe-
suggest a positive relationship between age and loneliness male and approximately 72% were White. Between cohorts,
(Fischer & Phillips, 1982). Examination of single cohort sam- the level of formal education decreased with an increase in age.
ples does suggest that loneliness and age are related. Berg and Widowhood increased with age. The majority of the sample re-
colleagues (1981) found that 19% of their Swedish sample of ported "good" or "excellent" health (see Table 1).
septuagenarians reported "some" or frequent loneliness. A cross-tabular analysis of the sociodemographic character-
Approximately one third of a sample of centenarians in the istics of cases with missing data (n = 54) revealed that these
United States reported loneliness "sometimes" or "often," as cases differed from the included cases on age and marital sta-
did 44% of Swedish centenarians (Martin et al., 1997). A na- tus. More centenarians and widowed adults were excluded than
tional Swedish study in which the proportion of persons report- expected and fewer sexagenarians and married adults were ex-
ing loneliness increased from 20% in sexagenarians to over cluded than expected. No group differences were found for sex,
40% in octogenarians was cited by Berg and colleagues (1981). education, race, or income.
Our intent was to further examine the direct effect of age on
physical health and loneliness and the effect on each construct Variables and Instruments
when mediated by the other. Three subscales of anxiety were used from the Sixteen
Personality Factor questionnaire (16PF; Cattell et al., 1970).
Self-Assessed Physical Health The three first-order factors were measured on a normed scale
Both domestic and international studies have found that poor from 1 (low) to 10 (high) and included (1) emotional stability,
self-assessed physical health (Mullins et al., 1996; Mullins, (2) apprehension, and (3) tension. High scores reflected high
Johnston, & Anderson, 1988; Mullins & Mushel, 1992; anxiety. Reported test-retest reliability (1 to 7 days) for each of
Wenger, Davies, Shahtahmasebi, & Scott, 1996) and number of the scales, respectively, were .79 to .82, .72 to .83, and .81 to
chronic illnesses (Russell, 1996) correlated positively and .90 (Cattell et al., 1970).
strongly with loneliness in elderly adults. Lonely elderly men Cognition was measured with three subtests from the
and women had more negative assessments of health and feel- Wechsler Adult Intelligence Scale-Revised (Wechsler, 1987):
ings of fatigue than did elderly men and women who were not picture arrangement, block design, and arithmetic. Each mea-
lonely. Lonely participants also complained of backaches, sure was continuous; high scores reflected high cognitive abil-
headaches, and nonspecific nausea (Berg et al., 1981). A pro- ity. The average reported split-half reliability coefficient for pic-
gressive decrease in hearing has also been associated with feel- ture arrangement was .74; block design and arithmetic
ings of loneliness in elderly persons (Dugan & Kivett, 1994). exceeded .80 (Kaufman, 1985).
Whether poor health predicts lonely feelings or loneliness pre- Social Network was measured with two questions taken
dicts self-assessed poor health is not clear. Self-assessed physical from the Older Americans Resources and Services Procedures
health was found to mediate the effects of disability and educa- (OARS; Fillenbaum, 1988) Interaction scale: "About how many
tion on loneliness (Mullins et al., 1996). Physical health mediated times did you talk to someone—friends, relatives, or others—
the effect of anxiety on loneliness in American centenarians on the telephone in the past week" and "How many times dur-
(Martin et al., 1997). Researchers have also concluded that lone- ing the past week did you spend some time with someone who
liness may be predictive of mental health. Bazargan and Barbre does not live with you; that is, you went to see them or they
(1992) found loneliness explained variance in self-evaluations of came to visit you, or you went out to do things together." Both
memory loss in older Black adults. These conclusions provide measures were scored from 3 (once a day) to 0 (not at all).
support for analysis of both models as proposed. Reported reliability coefficient for the scale was .56
(Fillenbaum, 1988).
METHODS Although we intended to measure the extent of involvement in
social relationships, preliminary confirmatory factor analyses in-
Participants and Procedure dicated that we were, in fact, measuring two unique constructs.
The sample for this analysis (n = 208) was drawn from Phase The first was telephone interaction, which may be a dominant
1 of the Georgia Centenarian Study (Poon et al., 1992) con- means of communication among those who are not indepen-
P234 FEESETAL.

Table 1. Participant Characteristics

Sexagenarians Octogenarians Centenarians Total Group

Characteristics (n = 82) (n = 79) (n = 47) (n = 208)
Age Range (years) 60-^9 79-89 99-106 60-106
Mean (SD) 64.96(2.80) 82.59 (2.45) 100.79 (1.52) 79.75 (14.00)
Gender (n)
Male 33 26 11 70

Downloaded from https://academic.oup.com/psychsocgerontology/article-abstract/54B/4/P231/693063 by guest on 15 June 2020

Female 49 53 36 138
Race(«)
White 57 61 31 149
African American 25 18 16 59
Education (n)
0-4 years 4 3 8 15
5-8 years 6 14 9 29
Some high school 11 15 6 32
Completed secondary 15 7 5 27
Business/trade school 5 8 4 17
1-3 years of college 10 10 6 26
Completed college 13 5 5 23
Graduate school 18 17 4 39
Marital Status (n)
Single 2 0 4 6
Married 50 29 1 80
Widowed 19 46 39 104
Divorced 9 4 2 15
Separated 2 0 0 2
Missing 0 0 1 1
Subjective Health (n)
Poor 5 1 4 10
Fair 10 21 14 45
Good 45 40 27 112
Excellent 22 17 2 41

dently mobile or are separated from family and friends by great past few weeks did you ever feel very lonely or remote from
distances. Visiting suggested some level of independent mobil- other people," with four values ranging from 1 (not at all) to 4
ity, in which case there may be less reliance on the telephone (often). This item was part of the negative affect balance scale
than for less able-bodied persons or for those who live farther with a reported test-retest reliability of .81 (Bradburn, 1969).
from family and friends. Additionally, deterioration in hearing The final measure was a subscale score from the Philadelphia
may actually discourage telephone communication and increase Geriatric Center Morale Scale (PGC; Lawton, 1975) labeled as
reliance on visiting, if possible. Each question, therefore, be- "lonely dissatisfaction." This subscale was composed of six di-
came a single-item indicator reflective of separate constructs; chotomously scored items from which an aggregated subscore
high scores reflected frequent interaction. Age, in years, was was derived. Items represented "the extent to which an individ-
self-reported. ual feels lonely and dissatisfied with life" (Sauer & Warland,
Two indicators from the OARS (Fillenbaum, 1988) served as 1982, p. 223). Cronbach's alpha was reported as .85 (Lawton,
indicators for the latent construct Physical Illness. The first 1975) for the six-item scale. Measures were recoded so that a
measure assessed current perceived overall health, "How would high score reflected greater loneliness.
you rate your overall health at the present time," from 3 (excel- Reliability of the measures within the model may be esti-
lent) to 0 (poor). The second indicator was a comparative indi- mated from the factor loadings; that is, the factor loading is a
cator of health, "Is your health now better, about the same, or measure of the validity of a construct. Because the square root
worse than it was five years ago," from 2 (better) to 0 (worse). of reliability is its validity, validity squared is an estimate of the
Measures were recoded so that high scores reflected illness. reliability of the measure (Bollen, 1989). The presence of error
Reported reliability for the items was .74 (Fillenbaum, 1988). is calculated as part of model estimation. Refer to Table 2 for
Loneliness was measured with three indicators. The first, estimated reliabilities within the model.
"Do you find yourself feeling lonely quite often, sometimes, or
almost never" (OARS; Fillenbaum, 1988), was coded as 0 Data Analysis
(often), 1 (sometimes), and 2 (seldom). The item was a part of A covariance matrix was generated using a listwise proce-
the Affective dimension of social support, with a reported relia- dure (see Table 2). Data were analyzed using the structural
bility of .71. The second measure was taken from the Bradburn equation modeling procedures in LISREL VIII (Joreskog &
Affect Balance Scale (BABS; Bradburn, 1969): "During the Sbrbom, 1994) with maximum likelihood estimation.
A MODEL OF LONELINESS P235

Table 2. Correlation Matrix, Means and Standard Deviations for Measurement Model (n = 208)

Indicator 1 10 11 12 13 14
1. Emotional stability 1.00
2. Apprehension -.42 1.00
3. Tension -.39 .46 1.00
4. Picture arrangement .10 -.11 .02 1.00
5. Block design .16 -.14 .03 .64 1.00
6. Arithmetic .17 -.23 .02 .58 .64 1.00

Downloaded from https://academic.oup.com/psychsocgerontology/article-abstract/54B/4/P231/693063 by guest on 15 June 2020

7. Visiting friends/relatives .01 -.02 .13 .13 .09 .08 1.00
8. Telephone contact .09 -.00 .02 .20 .25 .18 .31 1.00
9. Age .01 -.04 -.15 -.48 -.52 -.43 -.04 -.24 1.00
10. Self-health rating -.26 .22 .17 -.24 -.29 -.30 -.09 -.17 .22 1.00
11. Health problems -.07 .11 .00 -.13 -.17 -.14 -.05 -.04 .24 .29 1.00
12. Lonely .32 -.24 -.18 .15 .14 .23 -.01 .12 -.12 -.34 -.07 1.00
13. Feeling lonely/remote .22 -.17 -.18 .15 .10 .04 .08 .22 -.21 -.32 -.05 .48 1.00
14. Lonely dissatisfaction .28 -.20 -.23 .08 .16 .17 .10 .27 -.18 -.35 -.12 .34 .50 1.00
Mean 5.28 5.12 5.01 5.56 15.06 9.29 2.16 2.48 79.75 1.12 1.06 1.65 3.53 11.04
Standard Deviation 1.91 1.88 1.87 4.54 10.61 4.05 .74 .76 14.00 .77 .64 .57 .92 1.24

Table 3. Measurement Model: Standardized Factor Loadings Table 4. Correlations Among Latent Variables

Latent Construct and Indicators Factor Loading Estimated Reliability Latent Variable
Personality 1. Physical Illness —
Emotional stability -.64 .41 2. Loneliness -.55 —
Apprehension .66 .44 3. Anxiety .38 -.50 —
Tension .62 .38 4. Cognition -.40 .24 -.19 —
5. Visiting -.11 .09 .04 .12 —
Cognition
6. Telephone Contact -.19 .31 -.03 .27 .31 —
Picture arrangement .76 .58
Block design 7. Age .27 -.26 -.10 -.61 -.04 -.24 —
.84 .71
Arithmetic .75 .56 Note: n = 208
Visiting Friends/Relatives 1.00 1.00
Telephone Contact 1.00 1.00
pared. First, a nonmediated model was tested in which the direct
Age 1.00 1.00 path coefficient from the mediator to the outcome was fixed to
Physical Health zero and all other paths were estimated (Baron & Kenny, 1986).
Self-health rating .78 .61 Second, a fully recursive, mediated model was run to assess me-
Health problems .37 .14 diation. Total effects were decomposed in the later model.
Loneliness
Feeling lonely (OARS) .60 .36 Model A: Loneliness as Outcome
Feel very lonely/remote (BABS) .74 .55 The focus of thisfirstset of analyses was to examine the direct
Lonely dissatisfaction (PGC) .67 .45 and mediated effects of each construct on loneliness. Results of
"Validity (factor loading) squared is an estimate of the reliability of the indi- the nonmediated model indicated a reasonable fit of the model to
cator (Bollen, 1989). x2 (60) = 88.11,/? < .05. the data, x2 (60) = 88.11,/? < .05, GFI = .95, AGFI = .90.
Regression coefficients were significant between Anxiety and
Physical Illness (standardized coefficients reported), 7 = .48, t =
RESULTS 4.16, as well as Anxiety and Loneliness, 7 = .60, t = 4.57. Higher
An advantage of structural equation modeling was the use of levels of anxiety were related to higher levels of physical illness
multiple indicators of latent constructs. Multiple indicators re- and higher levels of loneliness. Age significantly predicted
flected the specific domain of content defined by the latent vari- Physical Illness, 7 = .23, t = 2.03, and Loneliness, 7 = .36, t =
able, allowed estimation of measurement error, and simultane- 3.16, suggesting the older the individual, the higher the evaluation
ous estimation of parameters in the model. Preliminary of physical illness and the greater loneliness experienced.
confirmatory factor analyses were conducted using LISREL Telephone contact negatively predicted Loneliness, 7 = -.23, t =
VIII (Joreskog & Sorbom, 1994) to estimate the measurement 2.73. Cognition and Visiting were not predictive of either criterion
model with significant indicators of the latent constructs. The variable.
proposed measurement model and regression model were con- According to Baron and Kenny (1986), mediation is present
ducted concurrently (see Table 3 for factor loadings and Table 4 if the significant coefficients between the latent exogenous
for correlations among latent variables). (Anxiety and Age) and the latent endogenous (Loneliness) are
To examine the direct, indirect, and total effects of the latent reduced (partially mediated) or become nonsignificant (fully
exogenous constructs on loneliness, nested models were com- mediated) when the mediating path is present and significant in
P236 FEESETAL.

Downloaded from https://academic.oup.com/psychsocgerontology/article-abstract/54B/4/P231/693063 by guest on 15 June 2020

Figure 3. Structural Model of Loneliness in Older Adults. Figure 4. Structural Model of Physical Weil-Being in Older Adults.

Table 5. Model A: Decomposition of Effects of Personal Table 6. Model B: Decomposition of Effects of Personal
Characteristics on Loneliness Characteristics on Physical Illness

Physical Illness Loneliness Loneliness Physical Illness

Variables Direct Indirect Total Direct Indirect Total Variables Direct Indirect Total Direct Indirect Total
Anxiety .35*** 25*** .42*** .13 55*** Anxiety 55*** .55*** .11 24*** 35***
Cognition -.22 — -.22 .22 -.08 .14 Cognition .14 — .14 -.28** .06 -.22
Visiting Others -.07 — -.07 -.02 -.03 -.04 Visiting Others -.04 — -.04 -.05 -.02 -.07
Telephone Contact -.06 — -.06 -.21*** -.02 -.23*** Telephone Contact -.23*** — -.23*** .04 -.10** -.06
Age .15 — .15 .28*** .05 .34*** Age 24*** 24*** .00 .15** .15
Physical Illness .37* .37* Loneliness 44*** .44***
R2 .28 .50 R2 .40 .40

Note: n = 208. Standardized regression coefficients shown. Note: n - 208. Standardized regression coefficients shown.
*p < .05; **p < .01; ***p < .001. *p < .05; **p < .01; ***p < .001.

the model (in this analysis, the direct path between Physical Effects were decomposed in the mediated model (see Table
Ulness and Loneliness). The mediated model, x2 (59) = 76.66, p 5). The total effects of Anxiety, Telephone Contact, and Age on
>.05, GFI = .95, AGFT = .91, was a significant improvement in Loneliness were significant. However, although the direct path
fit over the nonmediated model, A \ 2 (1)= 11.45, p < .001 (see coefficient from Anxiety to Physical Illness was significant, the
Figure 3). However, the mediational path between Physical indirect effect on Loneliness was not (t = 1.81). No significant
Illness and Loneliness barely reached significance (t = 1.97). indirect effects were found.
Several coefficients changed as a consequence of mediational
path estimation. A reduction in the standardized regression coef- Model B: Loneliness as Mediator
ficient between Anxiety and Loneliness from .60 to .42 was ob- The second set of analyses reversed the direction of the arrow
served, which remained significant (t = 3.20). The coefficient between Loneliness and Physical Illness, making Loneliness
from Anxiety to Physical Illness decreased slightly to .48 from the mediating variable. As expected, path coefficients and the fit
.35 and also remained significant (t = 3.45). The path coefficient of the nonmediated model were identical to that of Model A, \ 2
was reduced to nonsignificance between Age and Illness (t = (60) = 88.11, p < .05, GFI = .95, AGFI = .90.
1.44), and the coefficient between Age and Loneliness remained The fit of the mediated model was, again, identical to the fit
significant (t = 2.59). The path coefficient between Telephone of Model A's mediated version; however, the mediational path
Contact and Loneliness dropped slightly (.23 to .21) but re- in Model B was highly significant (standardized beta = .44, t =
mained significant. The effects of Cognition and Visiting on 3.31; see Figure 4). When compared to the nonmediated model,
Loneliness were unaffected by the addition of the mediational the standardized coefficient between Anxiety and Illness was
path. The mediated model accounted for 28% of the variance in reduced to nonsignificance, y = .11, t = .89. The coefficient be-
Physical Illness and 50% of the variance in Loneliness. tween Anxiety and Loneliness decreased only slightly from .60
A MODEL OFLONEUNESS P237

to .55 (t = 4.34). The path from Age to Illness dropped from .23 a significant total effect on feelings of loneliness; the older the
to .00 and was not significant (t = .02). Age to Loneliness (t = participants were, the more likely they were to report loneli-
3.00) and Telephone Contact to Loneliness (t = 2.78) remained ness. Our results also suggest the effect of old age on serf-eval-
virtually unchanged. Cognition emerged as a negative predictor uation of health is effectively mediated by loneliness. With ad-
of Physical Illness, 7 = -.28, t = 2.41. Visiting was not predic- vancing age, feelings of loneliness affect perceptions of health
tive. The mediated model accounted for 40% of the variance in rather than self-perceived health affecting feelings of loneliness.
Loneliness and 40% of the variance in Physical Illness. Health status does appear to decline with age: centenarians
Total effect of Anxiety on Physical Illness was significant (t = reported proportionally more "poor" and "fair" health whereas

Downloaded from https://academic.oup.com/psychsocgerontology/article-abstract/54B/4/P231/693063 by guest on 15 June 2020

3.45). In contrast with Model A, the indirect effects of Anxiety, sexagenarians and octogenarians reported higher percentages of
Telephone Contact, and Age on Physical Illness were significant "good" and "excellent" health. An exception to the trend, how-
(see Table 6). Although Cognition had a significant regression ever, is that fewer octogenarians report poor physical health,
coefficient, neither it nor Visiting had significant effects. leading us to ask what is the effect of cohort as a moderating
variable rather than age as a predictor. Additional analyses need
DISCUSSION to examine how well this modelfitsthe data for each cohort.
The models examined in these analyses were unique with re- Consistent with the literature, a predictive positive relation-
spect to previous models in that they examined the effects of ship is present between anxiety and feelings of loneliness (e.g.,
the level of cognitive functioning and personality on loneliness Berg et al., 1981; de Jong-Gierveld, 1987); greater anxiety is
mediated by self-assessed physical health. Loneliness as a me- reflected in greater feelings of loneliness. However, like age,
diator was modeled as well, given research suggesting per- when loneliness was the mediator, the effect of anxiety on phys-
ceived inadequate relationships affect physical well-being. ical well-being was totally indirect. High levels of anxiety in-
One of the most provocative findings of these analyses is the crease feelings of loneliness, which in turn decrease assessment
relationship between loneliness and self-evaluation of health. of well-being. The outcome of anxiety may be both poor feel-
Elevated feelings of loneliness not only predict poor subjective ings about relationships and health.
health evaluations but also transform the effects of increasing Although an anxious personality may serve a positive func-
age and anxiety on health. These findings are consistent with tion (e.g., motivation for action or learning), high and continu-
the psychoneuroimmunology literature which suggests that ous anxiety (characterized by emotional instability, apprehen-
one's psychological characteristics, including evaluations of sion and tension) heightens sensitivity to any feelings of illness
personal relationships, affect one's physical health and, in this or may manifest itself in concrete physical illness. Inadequate
case, one's feelings about physical health. Kiecolt-Glaser, relationships may be the consequence of a lack of stimulating
Garner, Speicher, Perm, and Glaser's (1984) analyses of loneli- interaction or in overly cautious, serf-conscious behavior, char-
ness in medical students found loneliness to be negatively re- acteristic of an anxious personality. These results suggest that
lated to immune functioning. the linkage by which anxiety affects interaction and perceptions
Although the relationship between health status and loneliness of health needs to be explored more fully.
in Model A was in the expected direction (poor subjective health The difference in effect between the two social network vari-
status predicts greater feelings of loneliness), a stronger effect was ables is noteworthy. Participating in frequent telephone conver-
anticipated given Martin and colleagues' results (1997) for sations not only reduced the loneliness perceived by partici-
American centenarians and the significant change in path coeffi- pants but also affected perceived health indirectly and
cients for anxiety and age. Self-assessed health did not serve as a significantly. Visiting with someone other than a spouse or
strong mediator of the effects of the other constructs. Several pos- roommate had no predictive relationship with either outcome
sible explanations exist for the apparent discrepancy. First, it may variable. We measured frequency of contact rather than with
be a consequence of the difference in estimation procedures be- whom one had contact, specifically. Our results, therefore, are
tween LVPLS and LISREL. Partial least squares uses component somewhat discrepant with previous research by Russell (1996)
analyses (such as principal component analyses) to estimate the and the conclusions of Marangoni and Ickes (1989), in which
regression coefficients, whereas the maximum likelihood estima- loneliness in older adults was found to be related weakly to the
tion procedure uses simultaneous solutions. Second, the sample number of persons in the social network and average frequency
represents a broader age range including individuals in the "Third of social contact but strongly related to perceived quality of
Age" (approximate age range of 60-75 years) and "Fourth Age" contact. These results suggest frequency of contact, via the tele-
(persons 85 and older; Baltes & Baltes, 1998), which are theo- phone, is important. Older adults may be more limited in their
rized to be distinctly different groups in terms of physical and psy- physical mobility or lack the means to travel independently.
chological functioning. A third potential explanation for the rela- The telephone, then, becomes a dominant and readily accessi-
tively weak relationship between the two may be the relative ble mechanism (used by all ages) to remain in contact with dis-
homogeneous nature of our sample on health; mat is, about two tant family members, such as children, and friends, allowing
thirds of the participants in this study were reporting good or ex- adults to maintain an emotionally intimate relationship. Once
cellent health. Consequently, physical illness may have a greater again, it is important to note the characteristics of the sample.
mediational effect for those who experience poorer health. All participants were living independently. Most centenarians
However, given the near significance of die coefficient from health in this sample were widowed, and more than half of the sexage-
to loneliness, we believe the results deserve further examination narians were married as were about one third of the octogenari-
by cohort (age groups) with a more heterogeneous health sample. ans; therefore, for many younger participants, the presence of a
Contrary to the conclusions of de Jong-Gierveld (1987) and spouse may limit the reliance on the physical presence of out-
Creecy and colleagues (1985), our results indicate that age has siders for interaction.
P238 FEESETAL.

A second possible explanation for the discrepancy is that re- (1998) suggest, we are challenged to create a cultural support
spondents, as a whole, were expressing more emotional loneli- system for the Fourth Age that allocates more resources to com-
ness rather than social loneliness (Weiss, 1973). The amount of pensate for the increasing dysfunctionality that occurs with age.
face-to-face contact with others (i.e., visiting) would be less rel- All professionals that care for or interact with older adults must
evant than having an intimate confidant, a relationship that be alert to the level of functioning and health in the individual.
could be maintained over the telephone. This explanation is Model replication relies on further analyses specifically exam-
consistent with the strong positive relationship between anxiety ining the moderating effects of cohort and of residential status,
and loneliness given that anxiety was in part assessed by emo- that is, dependent care versus independent living, and following

Downloaded from https://academic.oup.com/psychsocgerontology/article-abstract/54B/4/P231/693063 by guest on 15 June 2020

tional stability (Cattell et al., 1970). individuals over time.
ACKNOWLEDGMENTS
Limitations and Implications
We recognize several limitations in this analysis. The first is Funding for this research was provided by NIH Grant R01-43435.
in the manner in which loneliness was assessed. Participants The authors would like to thank Dan Russell, PhD, for his review and com-
were asked, in three separate measures, if they felt lonely. As ments on this manuscript, and acknowledge a reviewer's suggestion to test the
model with loneliness as the mediating variable.
such, the definition of loneliness was left to the individual; mul-
tiple interpretations of loneliness may be represented in the Address correspondence to Dr. Peter Martin, German Center for Research
on Aging, University of Heidelberg, Bergheimer Str. 20, 69115 Heidelberg,
analysis. The feelings of the participants appeared genuine, re- Germany. E-mail: [email protected]
flecting their perception of their personal status. The nature of
the question, however, does not allow us to tease out whether REFERENCES

the type of loneliness experienced was dominantly social or Baltes, P. B., & Baltes, M. M. (1998). Savoir vivre in old age. National Forum,
emotional (Marangoni & Ickes, 1989). Second, with regard to 78, 13-18.
model evaluation, our chi-square statistic was significant in the Baron R. M, & Kenny, D. A. (1986). The moderator-mediator variable distinc-
tion in social psychological research: Conceptual, strategic, and statistical
nonmediated model. Examined alone, this would indicate a considerations. Journal ofPersonality and Social Psychology, 51, 1173-1182.
poor fit of the model to the data (Bollen, 1989). However, given Bazargan, M., & Barbre, A. R. (1992). Self-reported memory problems among
the AGFI and GFI indices of fit (above .90) and a critical N of the Black elderly. Educational Gerontology, 18, 71-82.
200 (Hoelter, 1983), we believe our models are not only theo- Berg, S., Mellstrom, D., Persson, G., & Svanborg, A. (1981). Loneliness in the
Swedish aged. Journal of Gerontology, 36, 342-349.
retically but also statistically defensible. Bollen, K. (1989). Structural equations with latent variables. New York: Wiley
The statement that "correlation does not prove causation" & Sons.
(Bollen, 1989, p. 52) holds true for this model. The data in this Bradburn, N. M. (1969). The structure ofpsychological well-being. Chicago:
analysis represent one time period within the study, although Aldine.
future analyses with the data will take a longitudinal perspec- Cattell, R. B., Eber, H. W., & Tatsuoka, M. M. (1970). Handbook for Sixteen
Personality Factor Questionnaire (16PF). Champaign, IL: Institute for
tive. As Bollen suggests, models can only approximate reality, Personality and Ability Testing.
which is what we have attempted to do. We are attempting to Christian, E., Dluhy, N., & O'Neill, R. (1989). Sounds of silence: Coping with
build a foundation for further model building which more accu- hearing loss and loneliness. Journal of Gerontological Nursing, 15, 4-9.
rately reflects the experiences of older Americans living inde- Creecy, R. E, Berg, W. E., & Wright, R. (1985). Loneliness among the elderly:
A causal approach. Journal of Gerontology, 40, 487-493.
pendently. By the same token, the "lack of a correlation does de Jong-Gierveld, J. (1987). Developing and testing a model of loneliness.
not disprove causation" (Bollen, 1989, p. 52). Consequently, Journal of Personality and Social Psychology, 53, 119-128.
future data collection and analyses that are able to isolate, if Dugan, E., & Kivett, V. R. (1994). The importance of emotional and social iso-
possible, variables from other agents may indeed find our initial lation to loneliness among very old rural adults. The Gerontologist, 34,
model to be a valid representation of reality. Obviously replica- 340-346.
Falk, R. F., & Miller, N. B. (1992). A primer for soft modeling. Akron, OH:
tion is necessary. University of Akron Press.
The generalizability of this study is limited by the character- Fillenbaum, G. G. (1988). Multidimensional functional assessment of older
istics of the sample. As noted, all participants were living inde- adults: The Duke Older Americans Resources and Services procedures.
pendently and lived in the southern region of the United States; Hillsdale, NJ: L. Erlbaum Associates.
Fischer, C. S., & Phillips, S. L. (1982). Social characteristics of people with
consequently, an analysis of institutionalized older adults or small networks. In L. A. Peplau & D. Perlman (Eds.), Loneliness: A source-
adults from different regions of the country may produce re- book of current theory, research and therapy (pp. 21-39). New York: Wiley
sults different from those presented here. & Sons.
Despite these shortcomings, the implications of this study 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.
are clear. Loneliness does affect one's perceptions about health. Journal of Psychiatric Research, 12, 189-198.
As recognized by Christian, Dluhy, and O'Neill (1989), profes- Hoelter, J. W. (1983). The analysis of covariance structures: Goodness-of-fit in-
sionals who work with older adults must be attentive to this re- dices. Sociological Methods and Research, 11, 325-344.
lationship in their clients and keep in mind that alleviating lone- Joreskog, K. G., & Sorbom, D. (1994). Windows LISREL 8.14. Chicago, IL:
liness may be one way of affecting overall health. Among Scientific Software International Incorporated.
Kaufman, A. S. (1985). Wechsler Adult Intelligence Test-Revised. In J. V.
elderly persons living independently, loneliness is more preva- Mitchell, Jr. (Ed.), The ninth mental measurements yearbook (9th ed., pp.
lent with increasing age. Anxiety is particularly detrimental by 1694-1703). Lincoln, NE: The University of Nebraska Press.
increasing perceived loneliness and thus perceived ill-health. It Kiecolt-Glaser, J. K., Garner, W., Speicher, C. E., Penn, G., & Glaser, R. (1984).
is clear too that the telephone is a fundamental tool for older Psychosocial modifiers of immunocompetence in medical students.
adults living independently to maintain relationships. Increasing Psychosomatic Medicine, 46, 7-14.
Kiecolt-Glaser, J. K., & Glaser, R. (1992). Psychoneuroimmunology: Can psy-
sensory impairment, particularly hearing loss, therefore, is of chological interventions modulate immunity? Journal of Consulting and
concern to effective telephone usage. As Baltes and Baltes Clinical Psychology, 60, 569-575.
A MODEL OF LONELINESS P239

Lawton, M. T. (1975). The Philadelphia Geriatric Center Morale Scale: A revi- Reisberg, B., Ferris, S. H., De Leon, M. J., & Crook, T. (1982). The global de-
sion. Journal of Gerontology, 30, 85—89. terioration scale for assessment of primary degenerative dementia.
Lopata, M. (1995). Loneliness. In G. L. Maddox, R. C. Atchley, J. G. Evans, C. American Journal of Psychiatry, 139, 1136-1139.
E. Finch, D. F. Hultsch, R. A. Kane, M. D. Mezey, & E. C. Siegler (Eds.), Russell, D. (1996). UCLA loneliness scale (Version 3): Reliability, validity, and
The encyclopedia of aging (pp. 571-572). New York: Springer Publishing factor structure. Journal of Personality Assessment, 66, 20-40.
Company. Russell, D., Peplau, L. A., & Cutrona, C. (1980). The revised UCLA loneliness
Marangoni, C, & Ickes, W. (1989). Loneliness: A theoretical review with im- scale: Concurrent and discriminant validity evidence. Journal of Personality
plications for measurement. Journal of Social and Personal Relationships, and Social Psychology, 39, 472-480.
6, 93-128. Russell, D., Peplau, L. A., & Ferguson, M. L. (1978). Developing a measure of
Martin, P., Hagberg, B., & Poon, L. W. (1997). Predictors of loneliness in cen- loneliness. Journal of Personality Assessment, 42, 290-294.

Downloaded from https://academic.oup.com/psychsocgerontology/article-abstract/54B/4/P231/693063 by guest on 15 June 2020

tenarians: A parallel study. Journal of Cross-Cultural Gerontology, 6, 1-22. Sauer, W. J., & Warland, R. (1982). Philadelphia Geriatric Center Morale Scale. In
Mullins, L. C , & Dugan, E. (1990). The influence of depression, and family D. J. Mangen & W. A. Peterson (Eds.), Research Instruments in Social
and friendship relations, on residents' loneliness in congregate housing. The Gerontology, (Vol. 1, pp. 22-228) Minneapolis: University of Minnesota Press.
Gerontologist, 30, 377-384. Wechsler, D. (1987). Wechsler Adult Intelligence Scale, Revised. New York:
Mullins, L. C, Elston, C. H., & Gutkowski, S. M. (1996). Social determinants Psychological Corporation.
of loneliness among older Americans. Genetic, Social, and General Weeks, D. J. (1994). Review of loneliness concepts, with particular reference to
Psychology Monographs, 122, 453-473. old age. International Journal of Geriatric Psychiatry, 9, 345-355.
Mullins, L. C, Johnston, D. P., & Anderson, L. (1988). Social and emotional Weiss, R. S. (1973). Loneliness: The experience of emotional and social isola-
isolation among elderly: A conceptual view of loneliness. Danish Medical tion. Cambridge, MA: MIT Press.
Bulletin: Journal of the Health Sciences, Gerontology: Special Supplement Wenger, G. C , Davies, R. D., Shahtahmasebi, S., & Scott, A. (1996). Social
Series, 6, 26-29. isolation and loneliness in old age: Review and model refinement. Aging
Mullins, L. C, & Mushel, M. (1992). The existence and emotional closeness of and Society, 76,333-358.
relationships with children, friends and spouses: The effect of loneliness
among older persons. Research on Aging, 14, 448-470.
Poon, L. W., Clayton, G. M., Martin, P., Johnson, M. A., Courtenay, B. C ,
Sweaney, A. L., Merriam, S. B., Pless, B. S., & Thielman, S. B. (1992). The
Georgia Centenarian Study. International Journal of Aging and Human Received January 5, 1998
Development, 34, 1-17. Accepted December 22, 1998