The reliability of an instrument’s test scores is important to examine and may vary across studies that administered the test with a certain protocol to certain participants on certain occasions (Thompson, 1992). However, to gain some insight in the degree of reliability that can be expected for a given instrument, a Reliability Generalization (RG; Vacha-Haase, 1998) study can be conducted. Using the meta-analytic RG methodology, researchers can examine the range of reliability scores of a particular instrument that have appeared in the literature, can compute an estimated mean reliability, and can examine whether characteristics of the instrument itself, the study, or the context of administration explains variation in the observed reliability scores. A high mean reliability score with low variability across studies would considerably increase our confidence in that particular instrument.
In the present study, we focused on loneliness, which is the unpleasant feeling that occurs when a discrepancy is perceived between the actual and desired levels of the quantity or quality of one’s relationships (Perlman & Peplau, 1981). More specifically, we focused on a multidimensional loneliness measure that distinguishes between loneliness in relationships with peers and with parents. The Loneliness and Aloneness scale for Children and Adolescents (LACA; Marcoen, Goossens, & Caes, 1987) is a 48-item scale intended for use in the age range of 10 to 19 years. In addition to measuring the two relation-specific types of loneliness, the LACA assesses a person’s attitudes towards aloneness, including aversion to being alone and affinity for being alone.
The analyses are based on 79 studies from 1987 to 2014, reporting Cronbach’s alphas for 92 samples (k). As the scale has been developed in Belgium, most studies (k = 65) came from this country, though data from 10 other countries (including Argentina, Canada, China, Greece, Spain, and The Netherlands) could also be included. In all, data from 41,076 children (k = 35) and adolescents (k = 57) were incorporated.
To account for dependency, as several studies included more than one sample and reported on more than one reliability estimate, we conducted a multilevel meta-analysis (Van den Noortgate & Onghena, 2003). Results showed high estimated mean reliabilities (i.e., alphas of .80 or above), with minimal variability (i.e., SDs around .05) for the four subscales of the LACA. In the final model, few of the potential moderators selected proved to be significant predictors of the variation in reliability scores. For parent-related loneliness, samples with higher sampling quality tended to yield higher Cronbach’s alphas. In addition, alphas for three of the four LACA subscales were somewhat lower in adolescents as compared to children.
It is important to realize that the results of the RG work cannot be generalized beyond the kind of samples used in the analyses. More demanding challenges to the instrument under scrutiny should be applied by including groups that can be expected to have lower alphas, such as particular clinical groups or immigrant youth. Pending such future work, the present study seems to support the use of the multidimensional LACA in children and adolescents sampled from the general population.Contact: Marlies.Maes@ppw.kuleuven.be
Link to paper > Contact >