Testing the Higher-Order Factor Structure of a Social Isolation Scale Among Older Adults: Implications for Theory and Measurement

Backgrounds: Social isolation has been widely studied as a multidimensional construct that represents a major risk factor for health and wellbeing among older adults.

One critical issue is whether social isolation is more appropriately conceptualized as a superordinate construct (social isolation is manifested by its dimensions) or an aggregate construct (social isolation is formed by its dimensions). To date, researchers have not carefully considered and tested the relationships between dimensions and the higher-order construct of social isolation. Empirical research, however, can yield very different results depending on the way in which a particular construct is conceptualized.

The practice of theorizing dimensions as part of broader, more general (i.e., higher-order) construct has become more prevalent in social work. A crucial advantage of these constructs is that researchers can depict and test holistic models of complex phenomena in parsimonious ways.

The purpose of this study was to test the conceptualization of social isolation as an aggregate construct. Its factorial and concurrent validity were examined using the Multiple Indicator Multiple Cause (MIMIC) model.

Methods: We used the wave 2 data from NSHAP survey. The data were collected from 2010 to 2011 with a national representative sample of 3,196 older adults aged from 62 to 91. 11 items were selected and categorized into four dimensions: lack of intimate relationships, disconnectedness from families, disconnectedness from friends, and lack of social engagement. Outcome variables include loneliness (3-item UCLA loneliness scale), perceived lack of social support, and depression (11-item CES-D). The MIMIC model was constructed using SPSS Amos. The proportional structural effect was adopted to test the validity of the measure. An aggregate construct is considered valid when the latent construct can fully mediate the effects of its dimensions on outcome variables.

Results: The MIMIC model reports satisfactory model fit: X²(65)=778.1, CFI=.98, RMSEA=.059, SRMR=.036. However, the modification indices suggest that a correlation between lack of intimate relationships and lack of companionship can significantly improve the model fit. An alternative three-dimension model that treats lack of intimate relationships as a separated construct yields acceptable model fit: X²(66)=886.3, CFI=.98, RMSEA=.062, SRMR=.064, and no disproportionate effect. The four-dimension model reported associations with loneliness, lack of support, and depression as r=.25, r=.78, and r=.15. The three-dimension model reported slightly lower association with loneliness (r=.20) and depression (r=.12), but higher association with lack of support (r=.90).

Implications: There is a dilemma regarding the inclusion of lack of intimate relationships as a component of the social isolation construct. Lack of intimate relationships contributes to predicting loneliness and depression, but does not statistically fit into an integrated construct of social isolation. The two models also imply different conceptualizations of social isolation: minimal contacts/connections or isolation from the society. Based on the findings and enlightened by Weiss’s typology of emotional vs social loneliness, we suggest to conceptualize social isolation as the opposite of social integration and construct a separated measure to represent the lack of intimate relationship.