The binary-choice, 25-item Michigan Alcoholism Screening Test (MAST; Selzer, 1971) is probably the most widely used test of its type for adults (Parsons, Wallbrown, & Myers, 1994). Its popularity is related in part to speed of administration (approximately 5 minutes) and relative ease of scoring (see Radzid, Freeman, & Mackenzie, 1999). The MAST has been used sporadically as an assessment instrument with adolescents (e.g., Kadis, Malca-Vila, McNiel, & McClendon, 1990), but to our knowledge (based on a computer-assisted literature search) there have been no attempts to modify the item content of the adult test to render it more appropriate for adolescents. Certain MAST items refer to "wife" or "work," which may be foreign to the life experiences of younger test-takers. Moreover, there are items dealing with drug-related physical deterioration (e.g., liver problems) that may not be discriminating for young persons. Thus, the aim of the present study was to develop a modified version of the MAST with item conten t appropriate for adolescent respondents.
The modification of the MAST basically involved making item content consistent with the life experiences of adolescents. Changes included substitution of "family member or significant other" for "wife" (item 3), "boyfriend/girlfriend" for "wife"' (item 11), and "school" for "work" (items 14 and 16). Data were then collected from 201 adolescents referred to an outpatient treatment center for evaluation of possible substance abuse: 145 males and 56 females, ranging in age from 12 to 19 years (M = 16). There were 171 Caucasians, 28 Hispanics, and 2 Native Americans in the sample. Administration of the adolescent MAST was part of a comprehensive series of assessments for chemical dependency problems. The psychometric goal of the current investigation was the same as in any initial approach to test construction: an internally consistent, homogeneous scale (see Nunnally, 1978). Accordingly, several itemetric indexes of homogeneity were computed, including coefficient alpha, the average interitem correlation, split- half reliability, and item-total correlations. Additionally, factor analytic procedures were used to ascertain the tenability of a single-factor solution.
The items comprising the adolescent version of the MAST are presented in Table 1, together with respective item means and item-total correlations. Item 7 is not included because the weighted scoring procedures inexplicably use a weight of zero for this item; hence, it is not scored (see Selzer, 1971). The average interitem correlation was .12, the split-half reliability was .65 and the alpha coefficient was .68. These values are adequate for basic research, but are below the standards for tests used for applied decision-making (Nunnally, 1978). Next, items with low endorsement or low item-total correlations were deleted (items 3, 5, 18, and 25). Additionally, item 24 had poor overall relationships with the other items and was also removed. These deletions increased alpha to .73, an acceptable value.
From a factor analytic perspective, item homogeneity is evinced by the emergence of a unitary factor with essentially equal paths from each observed variable (item) to the latent variable (see Hoyle & Smith, 1994). It is important to note that this approach should not supplant traditional techniques (above) for evaluating internal consistency. Maximum likelihood estimates (Amos; see Arbuckle, 1997) failed to confirm a unitary factor for the modified MAST. A subsequent exploratory principal components analysis suggested that both the 24- and 19-item scales are factorially complex. For the longer scale, nine components were extracted based on eigenvalue 1 and scree-test criteria, accounting for 63% of scale variance. Using the same criteria, seven components were extracted for the shorter scale, accounting for 64% of the variance. …