Effects of Social Support and Self-Esteem on Depressive Symptoms in Japanese Middle-Aged and Elderly People

We examined the relationship among social support, self-esteem, and depression. The subjects were 1,116 Japanese community-dwelling adults aged between 40-79, who were the first wave participants of the National Institute for Longevity Sciences-Longitudinal Study of Aging (NILS-LSA). Exploratory and confirmatory factor analyses were performed on the Rosenberg's self-esteem scale that supported the superiority of the bi-dimensional structure of the scale marked by self-confidence and self-deprecation subscales. The subsequent causal analyses, using structural equation modeling, demonstrated that social support reduced depressed affect through an increase in self-confidence and a decrease in self-deprecation. By contrast, social support did not show a direct effect on depressed affect. The findings suggest the importance of esteem-improving elements of social support in reducing depressive symptoms. J Epidemiol, 2000 ; 10 : S63-S69.


INTRODUCTION
Research over the past several decades has shown that social relationships affects psychological well-being.In related studies, social support, psychological and material resources provided from relationships 1), has been demonstrated to have a favorable impact on various subjects such as infertile couples2), arthritis patients 3), and unemployed persons 4).The question of how social relations influence the psychological distress of elderly people has also attracted considerable attention, much of it in relation to research on caregiving s.In fact, gerontological researchers are interested in social support because of its potential usefulness for care, continued health, and sustained independent living 6), which are the issues pertinent to elderly people.
Unfortunately, previous findings on the effects of social support for elders are inconsistent.Studies have found both positive 7) and negative 8) effects of social support on elderly psychological well-being.One study addressing the inconsistency demonstrated that moderate emotional support increased internal control of elderly people to cope with stressful situations, whereas excessive support decreased their internal control and could result in a negative outcome 9).This suggests the importance of the internal process to mediate the relationship between social support and psychological well-being.In addition, the association between psychological well-being and social support is rather weak 10,11), which also requires a mediational model to be tested to understand how interpersonal relationships affect well-being.
Several studies suggested that self-esteem may be an important mediator 3, 7).Self-esteem, a positive or negative attitude toward the self 12), has been conceptualized both as a causal variable that determines other psychological phenomena and/or as an outcome variable influenced by life events and interpersonal feedback 13).One study insisted that low selfesteem increases a person's vulnerability to depression 14), while others suggested that interpersonal relationships predict subsequent changes in self-esteem 10,15) The esteem-enhancing facet of social support refers to a generalized appraisal which makes individuals believe that they are cared for and valued and that others are available to them in times of need 16).This effect of social support is termed the of the time) to 3 (most or all of the time), so that higher scores on the scale represented higher levels of depressive symptoms (positively worded items were reverse-scored).
Self-esteem: The Rosenberg's self-esteem scale (RSE)12), one of the most frequently used measures of self-esteem, was employed.The Japanese-translated version of the RSE scale 21) comprised 10 items, the same as the original.This scale, though designed for adolescents, is considered to be an appropriate measure for all age groups including the elderly 22).The subjects responded to the items using 4-point options (response range: 1=strongly disagree, 4=strongly agree).Higher scores in positively worded items (items 1, 3, 4, 7 and 10) indicated higher self-esteem, whereas higher scores in negatively worded items (items 2, 5, 6, 8 and 9) indicated lower self-esteem.

Social Support:
Social support was measured using a 10-item scale developed in Japan 23).This scale asked subjects whether there was someone whom they regarded as a support resource.An instruction given to the subjects restricted the support resources to friends and other acquaintances (Family support was measured using a different scale in the NILS-LSA, though not examined here).All ratings were made on 5-point options so that higher scores indicated high support perception (response range: 1=none, 5=many).Each item probed one of the following basic functional properties of social support: tangible support, informational support, emotional support, and companionship 24, 25).

Procedure and Statistical Analyses
All statistical analyses were conducted using the Statistical Analysis System (SAS) release 6.12 26).The FACTOR procedure of the software was used to test the factor structure of the study scales, except the CES-D scale, on which a confirmatory factor analysis was performed via the CALIS procedure.Preliminary data analyses examined correlations between social support, self-esteem, and depression constructs and their age or gender specific difference with the CORR and the TTEST procedures, respectively.Causal analyses of the constructs were performed using structural equation modeling which was also available in the CALIS procedure.

FactorAnalyses of the Study Scales
For the CES-D scale, a confirmatory factor analysis was conducted as a four-factor structure of the scale (consisting of depressive affect, positive affect, somatic and retarded activity, and interpersonal) has been reported previously 19,27).As a result, indices indicated a good fit to the data, which suggested the factorial validity of the scale (Goodness of Fit Index: GFI=.94,Adjusted Goodness of Fit Index: AGFI=.92).However, only depressed affect was adopted in the subsequent causal analyses.The other factors were excluded from the analyses because of the following: 1) the measures of positive affect and interpersonal problem are less responsive for social factors 28); 2) Items for somatic complaints on self-report deppression scales result in higher scores, which are misinterpreted as representing more depression in the elderly population 29).Cronbach's alpha of depressed affect was .76 for the sample, indicating the acceptable reliability of the construct.
Performing a principal component analysis on the RSE scale, two factors were retained by the criterion of eigenvalues (greater than one).The two factors together accounted for a total of 54 % of the variance before being rotated.The subsequent rotation using an oblique promax method generated a pattern matrix (see Table 1), which indicated the emergence of two unambiguous constructs -self-confidence and self-deprecation.Self-confidence comprised all positively worded items, whereas self-deprecation comprised all negatively worded items.The inter-factor correlation was -.32, which suggested moderate and negative association between the two constructs.Cronbach's alphas were .80 and .72 for self-confidence and self-deprecation, respectively.Furthermore, the following confinnatory factor analysis indicated that this two-factor model fitted the data better than the one-factor model, in which all items were forced to load on a single construct (in the two-factor model; GFI=.97,AGFI=.96,AIC=84.49, and in the onefactor model; GFI=.82,AGFI=.72,AIC=776.66.The difference between the models was; *df=1, A X 2=694.17,p<.0001).
All of the findings appeared to indicate the bi-dimensionality of the scale.Finally, a principal component analysis was performed on the social support scale.Though the scale consisted of some sub-functions of social support described above, the analysis demonstrated that this scale had a uni-dimensional structure, which accounted for about 57 % of the observed variance.The Cronbach's alpha for the sample was .91.

Age and Gender Differences of the Constructs
The mean scores of the constructs were compared between the middle-aged (40-59 years) and the elderly (60-79 years), as well as between men and women (Tables 2-1 and 2-2).No constructs indicated age-related differences.On the other hand, three of four constructs indicated significant differences by gender.On social support and self-deprecation, women demonstrated significantly higher scores than did men.Selfconfidence in men was higher than in women.
Bivariate analyses showed that all constructs were correlated significantly in the expected direction, regardless of age or gender (Table 3-1 and 3-2).More social support was associated with higher self-confidence, lower self-deprecation, and lower depressed affect.Moreover, higher depressed affect was associated with lower self-confidence and higher self-deprecation.
Additionally, multicollinearity between social support, selfconfidence, and self-deprecation on depressed affect was examined.The SAS program 26) provides an option to assess the tolerance level, the strength of inter-relationships among the regressor variables in the model.The tolerance level of the constructs was in the range of .85 and above, indicating that multicollinearity was not a problem in this data set.

Examination of Causal Models
Based on the results obtained from the preliminary analyses above, a structural equation model was tested in which depressed affect was predicted directly from social support and indirectly through self-esteem.Some items were chosen from each construct and were entered into the model with the contents and factor loadings taken into account (see Table 4 for the items used in the model).Furthermore, the analysis was conducted with the effect of gender partialled out, since genderbased differences were found among some constructs.
The results are shown in Figure 1.This model produced an  As predicted, more social support was related to higher selfconfidence and lower self-deprecation.Higher self-deprecation was related to higher depressed affect while higher self-confi-dence was related to lower depressed affect.The standardized estimate from self-deprecation to depressed affect (.38) was higher than that from self-confidence to depressed affect (-.14).When small and similar interactive paths between the two selfesteem constructs (-.12 and -.08) were neglected, the difference indicated the strong effect of self-deprecation on depressed affect against the inverse effect of self-confidence.
Contrary to expectations, social support had no significant direct relation to depressed affect.In this regard, social support appeared to influence depressed affect only when mediated by self-esteem.
A slight modification of the model in which the direct effect (the path from social support to depressed affect) was omitted improved the overall model fit (GFI=.99,AGFI=.98,AICY2.32), but the improvement was not statistically significant (odf==1, A X 2=.18, n.s.).

DISCUSSION
Preliminary factor analyses indicated the predominance of the two-factor model over the one-factor model in the RSE scale.Despite its initial conceptualization and common use as a uni-dimensional scale, the same two-factor structure of the RSE scale has been demonstrated to be appropriate zz.3Q31).This study reconfirmed the two-factor structure of the scale with Japanese middle-aged and elderly people as in western or younger samples reported in previous studies 22,30,31).
However, the result that self-confidence consisted of only positively worded items, whereas self-deprecation consisted of only negatively worded items is arguable.For example, scales with both positively and negatively worded items could generate an artificial factor structure due to the subjects' "carelessness" in the form of agreeing with items with which they truly  disagree or vice versa 30).To address this issue, we identified potential "careless" samples who responded to more than six of the ten items with the highest level of agreement (i.e., circled 4) or the highest level of disagreement (i.e., circled 1), following the method used in a previous study 30).The reanalysis excluding the samples resulted in the same two-factor structure as shown in Table 1.Furthermore, an additional analysis was conducted to investigate an alternative causal model between social support, self-esteem, and depressed affect in which selfesteem was comprised of one-factor.In this case, the fit indices (GFI=.95,AGFI=.90,AIC=156.04)indicated that the alternative model was inapplicable compared to the two factor model (*df=4, * X2=164.55,p<.0001).On the basis of these findings, we believe the multi-dimensional use of the RSE scale to be preferable.
The causal analyses using the structural equation modeling demonstrated that more social support decreased depressed affect through self-esteem.In addition, consistent with previous research 30), self-deprecation was more potent in depressed affect than in self-confidence.However, a series of studies using interview-methods indicated that self-esteem reflected in negative comments was related to the onset of depression, whereas self-esteem reflected in positive comments was related to the recovery from depression 32,33).This might imply that our findings represented the effects of self-confidence and selfdeprecation, each of which actually influenced a different stage in the course of depression.Future refinement of the model using longitudinal data would help to clarify the methodological issues and make progress in understanding the developmental process of depression.
It was an important contribution of the study that social support did not show a significant direct effect on depressed affect.A previous study demonstrated that the correlation between family support and psychological distress was reduced when other related variables (personality, leisure activities, physiological measures, etc.) were partialled out 11).Additionally, although the uni-dimensional structure of the support scale used in our study prevented us from making detailed examinations, it has been suggested in a study that the esteem-improving elements of social support have a dominant effect on emo-tional well-being relative to other belonging or tangible support 34) .Thus, our findings, focusing particularly on peer support, extended prior research that supported the importance of selfesteem as a mediator in the relationship between social support and psychological well-being.
The effects of social support on self-esteem may be a controversial issue of the results in this study.A long-term follow-up study showed that perception of social support in early adolescence predicted the increment of self-esteem at an interval of four years 35).However, other explanations must be obtained for the causal relationship during shorter period of time as examined in the present study.In this regard, Markus and Kunda insisted that a distinction should be made between the working self-concept (tentative and situationally responsive self-concept) and the core self-concept (enduring self-concept) and found the former to vary in response to the temporary social environment 36).As self-esteem is one of the central motives self-concept theory credits for the protection and maintenance of one's self-picture 31), the results of our study suggested the same malleability of working self-esteem affected by interpersonal relationships.Yet the early description of social support may oversimplify the complexities of social relationships, just defining it as an information to lead an individual to reaffirm his sense of personal worth 31), the implications must be reconsidered in the light of recent views regarding the self.
Nevertheless, since the dual nature of self-concept has rarely been the focus of empirical work 36), self-esteem also remains open for discussion, including its interactive properties with the social environment.Despite the limitations, the findings emerged from this study added several specific points vis-a-vis the relationship among social environment, self-concept, and psychological well-being.

Figure 1 .
Figure 1.Path diagram showing the effect of social support and self-esteem on depressed affect.

Table 1 .
Factor analysis of the RSE.

Table 2 -
1. Means and standard deviations of the constructs by age group.

Table 2 -
2. Means and standard deviations of the constructs by gender.
in depressed affect.Effective indicators ranged from .34 to .89, which also suggested the appropriate fit of the model.

Table 4 .
Items entered into the structural equation model.