Psychological Reports, 2009, 105, 849-856. © Psychological Reports 2009
REVISITING THE FACTORIAL VALIDITY OF THE REVISED UCLA LONELINESS SCALE: A TEST OF COMPETING MODELS IN A SAMPLE OF TEACHERS1 MARC DUSSAULT, CLAUDE FERNET, STÉPHANIE AUSTIN, AND MATHIEU LEROUX Laboratoire de Recherche en Comportement Organisationnel Département des Sciences de la Gestion Université du Québec à Trois-Rivières Summary.—Over the years, various studies addressing different populations have consistently raised concerns about the unidimensionality of the Revised UCLA Loneliness Scale. In the present study, the factorial structure of the scale was examined by comparing it to alternative models. In a sample of 1,157 French Canadian teachers, results of confirmatory factor analysis support a three-factor model solution. Support for the invariance of this model across sexes and teaching levels (i.e., elementary and high school) was also obtained.
Since its introduction, the Revised UCLA Loneliness Scale ������� has become one of the most commonly used measures of loneliness. This self-report scale, developed by Russell, Peplau, and Cutrona (1980), is composed of 20 items that deal with general or overall feelings of loneliness (Allen & Oshagan, 1995). The concept of loneliness as unidimensional implies that feeling lonely is undifferentiated in nature, and therefore experienced and understood in the same way by all people. Although many studies have confirmed the internal consistency of the Revised UCLA Loneliness Scale (see Russell, et al., 1980; Cuffel & Akamatsu, 1989; De Grâce, Joshi, & Pelletier, 1993; Hartshore, 1993; Rotenberg, Shewchuk, & Kimberley, 2001; Hughes, Waite, Hawkley, & Cacioppo, 2004), others have suggested that not all the items load onto a single underlying factor, which raises doubts about the scale’s unidimensionality (Hawkley, Browne, & Cacioppo, 2005). Given these conflicting results, the internal structure of the Revised UCLA Loneliness Scale has been the subject of much investigation in recent decades. Among the authors who have conducted exploratory factor analyses yielding multiple solutions for loneliness, Hojat (1982) studied a sample of college students and obtained a solution with five factors: isolation, people, sharing with others, absence of an intimate person, and sociability. Another study conducted by Hays and DiMatteo (1987), in a sample of college students, also proposed five factors: perceived social isolation, Address correspondence to Marc Dussault, Département des Sciences de la Gestion, Université du Québec à Trois-Rivières, 3351 boul. des Forges, C.P. 500, Trois-Rivières, Québec, Canada, G9A 5H7 or e-mail (
[email protected]). This article was prepared with the financial support of the Social Sciences and Humanities Research Council of Canada. 1
DOI 10.2466/PR0.105.3.849-856
850
M. DUSSAULT, ET AL.
unavailability of empathic confidants, unavailability of supporting per sons, lack of social integration, and lack of meaningful social relationships. Two other studies conducted in college students (Austin, 1983; Adams, Openshaw, Bennion, Mills, & Noble, 1988) supported a three-factor solution underlying the Revised UCLA Loneliness Scale. Austin (1983) labelled the factors intimate others, social others, and belonging and affiliation. Another three-factor solution was developed by Adams, et al. (1988), with the factors psychological loneliness, psychosocial loneliness, and social loneliness. In addition, three studies obtained a bi-factorial solution reflecting the wording of the Revised UCLA Loneliness Scale items (Zakahi & Duran, 1982; Knight, Chisholm, Marsh, & Godfrey, 1988; Wilson, Cutts, Lees, Mapungwana, & Maunganidze, 1992). Knight, et al. (1988) investigated a sample of adults from the general population of New Zealand and identified the factors as positively and negatively worded items. Wilson, et al.’s analysis (1992) of high school and college students revealed a two-factor solution: the intimate other and the social other. Zakahi and Duran (1982) studied a sample of college students and identified factors related to intimate other and social network. Most studies using confirmatory factor analysis have obtained threefactor solutions. For instance, in a study in college students, Hartshore (1993) showed that a three-factor model, similar to Austin’s (1983), presented the best factorial solution, even though the model provided a poor fit to the data (e.g., AGFI = .80). The first factor comprised negative��������� ly worded items and the second and third factors comprised positively worded items. Note that a 4-point Likert scale ranging from 1: Strongly disagree to 4: Strongly agree was used instead of the original anchored 4-point scale (from 1: Never to 4: Often). Recently, in a sample of college students, Hawkley, et al. (2005) also concluded that Austin’s (1983) threefactor model provides the best factorial solution. The first factor, composed of negatively worded items������������������������������������������������� and labelled Isolation, reflects feelings of rejection and aloneness. The second factor, composed of positively worded items and labelled Relational Connectedness, corresponds to feelings of familiarity. The last factor, consisting of positively worded items and dealing with feelings of group identification, was labelled Collective Connectedness. In line with the above findings, the present study evaluated the factorial structure of the Revised UCLA Loneliness Scale. A sample of teachers was chosen because loneliness appears to be particularly prevalent in this population (Smith & Scott, 1990). This contrasts with past research, which has typically investigated college students and therefore adds little to the understanding of how the factorial structure of the Revised UCLA Loneliness Scale generalizes to other populations. Three alternative models were
REVISED UCLA LONELINESS SCALE
851
tested: (M1) a one-factor model; (M2) a model that includes a second-order factor and two first-order factors corresponding to positive and negative item wording; and (M3) a three-factor model based on Hawkley, et al. (2005) and Hartshore (1993). It is important to underscore that only M1 and M2 are in line with the unidimensional view of loneliness. Method Participants A random sample of 1,157 Canadian teachers was selected for the study, of whom 560 worked in an elementary school and 597 in a secondary school. Participants were 810 women and 347 men with an average age of 43.8 yr. (range = 23–72). Measure A French Canadian validated version (De Grâce, et al., 1993) of the Revised UCLA Loneliness Scale was used to assess teachers’ perceptions of loneliness. This scale is composed of 20 items, of which 10 are worded positively (e.g., “I feel in tune with the people around me”) and 10 negatively (e.g., “I lack companionship”). All items were scored on a 4-point scale ranging from 1: Never to 4: Often. Cronbach coefficient alpha for this sample was .89. Analyses Confirmatory factor analyses were conducted to test the factorial structure of the scale. The first model tested was a one-factor model and the second was a three-factor model. This second model is a secondorder factor model comprising two factors that reflect positive or negative wording. Thus, one factor represents the 10 positively worded items and another represents the 10 negatively worded items. An alternative threefactor model, similar to Hawkley, et al.’s (2005), was also tested. The first factor, composed of 10 negatively worded items,���������������������������� reflects feelings of rejection and aloneness; the second factor, composed of five positively worded items, corresponds to feelings of familiarity; and the last factor consists of five positively worded items dealing with feelings of group identification. Analyses were performed with EQS 6.1 (Bentler & Wu, 1995). According to Hu and Bentler (1999), the use of relative fit indices such as the standardized root mean square residual (SRMR), the comparative fit index (CFI), and the root mean square error of approximation (RMSEA) is strongly recommended to evaluate a model’s goodness-of-fit. The standardized root mean square residual is the standardized difference between observed and predicted covariance. A value of 0 indicates perfect fit, and values of .08 or less are desirable (Hu & Bentler, 1999). The comparative fit index (Bentler, 1990) reflects model improvement that
852
M. DUSSAULT, ET AL.
rests upon a continuum from the null model (variables are completely uncorrelated) to a model that completely accounts for the associations among variables. According to Schumacker and Lomax (1996), models with a comparative fit index higher than .90 present a satisfactory fit to the data. The root mean square error of approximation (Browne & Cudeck, 1993) estimates the lack of fit in a model compared to a perfect (saturated) model. Browne and Cudeck (1993) suggest that a value of .05 indicates a close fit and values up to .08 represent reasonable errors of approximation in the population. Results Confirmatory Factor Analysis Four models were tested by confirmatory factor analysis. First, the independence model testing the hypothesis that the variables are uncorrelated was rejected (χ2 = 10227.39, df = 190, p
