Using a Rasch model to examine the utility of the South Oaks Gambling Screen across clinical and community samples

Abstract

The South Oaks Gambling Screen (SOGS: [Am. J. Psychiatr. 144 (1987) 1184]) is one of the most widely used measures of gambling problems in epidemiological studies and clinical evaluations. In the current paper, we were able to examine the SOGS using a Rasch model with data obtained from a representative community sample and a large clinical sample. The SOGS demonstrated significant stability across community and clinical samples despite the sample differences in gambling behaviors and demographic characteristics. In the clinical sample, we demonstrated the significant agreement between DSM-IV- and SOGS-based estimates of each person's level of gambling problem severity. However, the relative severity of DSM-IV and SOGS symptoms suggests that the measures tap somewhat different and overlapping regions of the latent continuum. We estimate that the DSM-IV reliably separates three levels of gambling problem severity and provides corresponding cut scores for a SOGS scale composed of 15 sample-invariant items. Recommendations for a relaxed cut score on the DSM-IV and reduced set of SOGS items are discussed.

Introduction

The South Oaks Gambling Screen (SOGS: Lesieur & Blume, 1987) is one of the most widely used measures of gambling problems in epidemiological studies and clinical evaluations. The SOGS has been used internationally for both clinical and research purposes in a variety of settings Abbot & Volberg, 1991, Ladouceur, 1991, Lesieur & Blume, 1990, Lesieur & Blume, 1991, Martinez-Pina et al., 1991, Volberg & Steadman, 1989. The SOGS operationalizes gambling problems by focusing on seven components derived from a diagnostic model using DSM-III criteria: family disruption, job disruption, lying about gambling wins and losses, defaults on debts, borrowing from others to relieve a desperate financial situation caused by gambling, borrowing from illegal sources, and committing an illegal act to finance gambling. Responses to the SOGS items are believed to be related to a latent construct of gambling problem severity. By developing items that reflect severe levels of problems, the SOGS serves as a screen for pathological gambling. Consistent with the DSM-IV definition, within this region of the continuum of gambling problems, there is “persistent and recurrent maladaptive gambling behavior characterized by an inability to control gambling, leading to significant deleterious psychosocial consequences: personal, familial, financial, professional, and legal” (APA, 1994).

Despite its widespread use, several researchers have identified limitations of the SOGS. First, the SOGS has been criticized for over-diagnosing pathological gambling relative to DSM-IV-based assessments in general populations Culleton, 1989, Dickerson & Hinchy, 1988, Ferris et al., 1999. As has been done with screening measures for other disorders (cf. Santor, Zuroff, Ramsey, Cervantes, & Palacios, 1995), researchers have attempted to solve this problem by adjusting cut scores Duvarci et al., 1997, Ladouceur, 1991, Pasternak & Flemming, 1999, Volberg & Steadman, 1989 but have not evaluated psychometrically the ability of the SOGS to make discriminations within different ranges of gambling problems. Indeed, before shifting cut scores on a measure designed to make categorical distinctions, it is first necessary to identify the levels of problem gambling at which particular items are most effective at mapping varying regions of the continuum of gambling problems and pathology (e.g., Embretson, 1996).

A second problem of the SOGS is that it does not require consideration of the type of problems that are endorsed, just that a sufficient number of problems are endorsed to meet a cut-off criterion. Thus, scores on the SOGS have been used as an additive index of problems without consideration of the relationship between the severity of particular item contents and overall gambling problem severity. For example, extensive weight (50% of SOGS items) is given to items inquiring about different types of borrowing to support gambling activity Cox et al., 2000, Stinchfield, 2001, with less emphasis given to features of pathological gambling such as returning to gamble in order to win back money lost on a previous day (i.e., chasing; Lesieur, 1984). The problem of using an index of symptom counts without consideration of the severity of the particular symptoms is not unique to the SOGS (e.g., DSM-IV) and is not necessarily problematic. However, when different indexes disagree (e.g., SOGS and DSM), it is unclear which to trust. Some indexes may represent more or less severe manifestations of problems and thus comparing the number of people who have exceeded a given threshold may not be appropriate as one threshold may reflect a higher severity than the other. Therefore, the anchoring of scales on a common metric is necessary before indexes measuring the construct of gambling problem severity can be compared.

Modern testing theory provides a framework for appropriately addressing the types of concerns regarding the SOGS that are discussed above. For example, using methods based in item response theory (IRT), we can use a statistical (Rasch) model that estimates the relative severity of each problem, the order in which symptoms are likely to be endorsed and how individual levels of gambling problems combine with the severity of each problem to produce SOGS total scores Andrich, 1988, Wright & Stone, 1979. By modeling item responses on a latent continuum, we can address the two major criticisms of the SOGS by gaining information about the relative severity of individual symptoms and we can evaluate the equivalence of diagnostic thresholds using a common metric. Such methods transcend the utility of more classical approaches that are not able to model the precision of measurement across levels of problem severity, and therefore are limited when seeking to model item-level differences on the SOGS across groups with different levels of gambling problems (e.g., community and clinical sample).

Using a Rasch model approach also allows for comparisons with other measures such as the DSM-IV, which are thought to identify a similar latent construct. Using DSM-IV-based measures, Volberg (1997) as well as Beaudoin and Cox (1999) found that the DSM-IV and SOGS were significantly correlated (Volberg: r=.71, P<.001; Beaudoin & Cox: r=.69, P<.001). By fitting DSM-IV and SOGS responses to a Rasch model, we can (a) compare agreement of DSM-IV- and SOGS-based independent estimates of each person's level of gambling problem severity to determine if the measures tap the same latent construct; (b) compare the relative severity of the DSM-IV and SOGS symptoms to determine if they tap unique or overlapping ranges of the latent continuum of gambling problem severity; (c) estimate the magnitude of increases in problem severity with each additional problem endorsed on the DSM-IV and SOGS; (d) compare the stability of scales across samples with different levels of gambling problem severity (e.g., clinical and community samples); and (e) anchor scales on the continuum and compare the precision of each instrument as a function of different levels of gambling problem severity to determine if differential prevalence rates reflect measurement imprecision or a sorting of individuals at different levels of problem severity. This information is vital to the further development of the SOGS and will increase our understanding of the precision with which we assess gambling severity.