|
Article Excerpt
The author used Rasch measurement to examine the reliability and validity of 382 Korean university students' scores on the Marlowe-Crowne Social Desirability Scale (MCSDS; D. P. Crowne & D. Marlowe, 1960). Results revealed that item-fit statistics and principal component analysis with standardized residuals provide evidence of MCSDS's unidimensionality. Six (of 33) items displayed differential item functioning for gender.
**********
Over the years, researchers have been concerned about the possibility that self-report instruments could be contaminated by respondents answering questions in ways that influence whether they are perceived favorably or unfavorably. Socially desirable responding in self-reports on sensitive behaviors such as delinquency, violence, and drug abuse interferes with accurate inferences of self-report scores (Fraboni & Cooper, 1989; King & Bruner, 2000). Social desirability bias, which is defined as "the tendency for individuals to portray themselves in a generally favorable fashion" (Holden, 1994, p. 429), has been studied since the 1950s (e.g., Edwards, 1957). One concern, however, is that Edwards's scale was derived from the Minnesota Multiphasic Personality Inventory (MMPI) and was associated with psychopathology. When Edwards's scale with items drawn from the MMPI is administered to college students, Crowne and Marlowe (1960) noted that "the meaning of high social desirability scores is not at all clear" (p. 349). Accordingly, the Marlowe-Crowne Social Desirability Scale (MCSDS; Crowne & Marlowe, 1960) was developed to minimize "pathological implications" (p. 349) and tested on college students. The MCSDS consists of 33 true-false statements in which 18 items describe socially approved behaviors but infrequent behaviors (e.g., "I never hesitate to go out of my way to help someone in trouble"), and the other 15 items refer to socially disapproved but frequent behaviors (e.g., "I like to gossip at times").
Since its development, the MCSDS has been described and adopted in more than 1,000 diverse studies (Beretvas, Meyers, & Leite, 2002). Although this scale has been widely used to measure one form of response bias or "faking good," because of its length, several short forms of the MCSDS (Ballard, 1992; Reynolds, 1982; Strahan & Gerbasi, 1972) have been developed and used. According to Barger (2002), for example, Reynolds's forms were cited in 128 studies and Strahan and Gerbasi's forms were cited in 145 studies in the 1990s.
In addition to the variety of uses to which the MCSDS has been put, there has also been some concern that the scale represents a unidimensional structure (called "social desirability" or "need for approval"). Several researchers have investigated the dimensionality of the full and short forms of the MCSDS using factor analytic techniques, and their findings have been contradictory (Barger, 2002; Collazo, 2005; Fischer & Fick, 1993; Leite & Beretvas, 2005; Loo & Thorpe, 2000; Paulhus, 1984; Ramanaiah & Martin, 1980; Ramanaiah, Schill, & Leung, 1977). In most of the studies investigating the factor structure of responses to the MCSDS, it was found that the MCSDS did not fit a one-factor model, suggesting a multidimensional structure. Recently, Barger found that the MCSDS did not fit the one-factor model using the Satorra-Bentler robust maximum likelihood estimation (Bentler, 1995). Leite and Beretvas also conducted confirmatory factor analysis using the mean and variance-adjusted weighted least squares estimator (L. Muthen & Muthen, 1998) and have found two subscales, thus raising questions about the unidimensional structure of the MCSDS.
A number of problems regarding the use of linear factor analytic techniques on dichotomous data have been noted (Meara, Robin, & Sireci, 2000; B. Muthen, 1978; Olsson, 1979). The dimensionality of a matrix of phi coefficients, for example, may differ from the dimensionality of the underlying continuous variables (Bernstein & Teng, 1989; Hambleton & Rovinelli, 1986). Because responses to the MCSDS are in a true-false format, the linear factor analysis may distort the underlying structure of dichotomous data (Bock, Gibbons, & Muraki, 1988; McDonald & Ahlawat, 1974).
Many of the problems posed by factor analysis procedures with dichotomous items can be avoided with the use of item response theory models or, as illustrated in this study, with the use of the Rasch model (Rasch, 1960). Unlike the traditional factor analytic techniques, Rasch modeling places both the difficulty of items and the ability of persons on a common scale. The graphical representation (item-person map) of the conjoint distribution of the person and item estimates on the same scale provides evidence about whether the instrument is appropriate for a given sample. The lack of spread of item difficulty estimates on the item-person map suggests that some of the items may be redundant. Furthermore, although the models of factor analysis work under the assumption of normal distribution of the data, Rasch models make no such assumption (Slinde & Linn, 1979). It should be noted, however, that the importance of item-level analysis (i.e., focusing on the quality of the individual items, rather than on the whole test) has not been properly recognized in the published research of the MCSDS, mostly because previous research on the MCSDS is based on traditional methods of factor analysis.
Because the Rasch model is based on the assumption of unidimensionality, deviations from Rasch model data fit are considered evidence of multidimensionality (Linacre, 1998; E. V. Smith, 2002; R. Smith & Miao, 1994; Wright, 1996). Given the frequent use of the MCSDS in many different fields and the conflicting findings about whether the MCSDS represents a unidimensional or a multidimensional construct, it is imperative to examine the dimensionality of the scale with the use of appropriate statistical techniques (Collazo, 2005; Leite & Beretvas, 2005); therefore, the purpose of this study was to examine the psychometric properties of the MCSDS by using the Rasch model with a college student sample.
METHOD
Participants
The participants were 382 (248 undergraduate and 134 graduate) students from several education classes at a large private university in the Republic of Korea. Participants received extra course credit for participating....
|
