Despite the long-standing and widespread use of the MIA, no summary of psychometric research on its reliability and validity has been published, with the exception of a manual on research on the German version of the scale (Ehlers & Margraf, 1993). Although some publications have had as their avowed purpose examination of the psychometric properties of the MIA, in others such information is buried in reports with another primary aim. Accordingly, a review of the MIA’s psychometric features is overdue, and the first purpose of the present paper is to provide such a distillation. The second purpose is to add to the psychometric database in areas where little information on the MIA’s performance is available, in particular, the MIA’s convergent, discriminant, and criterion-related validity with reference to diagnosticians’ severity ratings for anxiety disordera and to their diagnosis of agoraphobia.

In a PsychInfo search we located 16 papers in addition to the original validation study (Chambless et al., 1985) in which at least one psychometric property of the MIA was examined. We selected papers published in English, Spanish, or French (the only languages the authors can read), but papers could include data collected with translated versions of the MIA.² In Table 1, we summarize the results of 10 papers in which data concerning the internal consistency and/or convergent and discriminant validity of the MIA are reported. Using meta-analytic methods for summarizing correlational data (Rosenthal, 1991), we calculated mean reliability and validity coefficients weighted by sample size. Participants in these studies included student and community subjects and patients with panic disorder with agoraphobia or other anxiety disorders. Studies in Table 1 were conducted in the United States (n = 2), Australia (n = 3), Sweden (n = 3), Canada (Anglophone n = 2; Francophone n = 1), and the Netherlands (n = 1).³

As can be seen in Table 1, internal consistency data were available for an aggregated sample of 1,279 respondents. As indicated by Cronbach’s α of ≥ .93, the MIA scales are highly internally consistent. Test-retest reliability has been reported for several samples. For two samples of agoraphobic patients Chambless et al. (1985) reported test-retest reliability coefficients of .86 for Avoidance Accompanied and .90 for Avoidance Alone over a period of 8 days, whereas the coefficients were .75 and .89, respectively, over a period of 31 days. Over a 42-day interval, Stephenson, Marchand, and Lavallée (1997) reported reliability coefficients of. 75-.76 for a student sample. Finally, Rodriguez, Pagano, and Keller (2007) reported that these scales were remarkably stable over a 5-year period with reliability coefficients of .76 for Avoidance Accompanied and .83 for Avoidance Alone for a sample of patients with panic disorder with agoraphobia. Thus, the available data indicate that test-retest reliability is excellent over short periods and very good even over very long periods.

Investigations of the construct validity of the MIA have involved studies of its factor structure, its convergent and discriminant validity, and its criterion-related validity. Four groups of authors have examined the internal structure of the MIA via factor analysis in American, Australian, Canadian, and Dutch samples of agoraphobic outpatients (Arrindell, Cox, Van der Ende, & Kwee, 1995; Cox, Swinson, Kuch, & Reichman, 1993; Kwon, Evans, & Oei, 1990; Rodriguez et al., 2007). Sample sizes ranged from 124 to 216. Although the results are not entirely consistent, the most common finding (Arrindell et al.; Cox et al., Rodriguez et al.) is a three-factor solution representing avoidance of public places, open spaces, and enclosed spaces.

As reported in Table 1, the convergent validity of the MIA with other self-report measures of agoraphobia has been examined in five studies for Avoidance Accompanied and six studies for Avoidance Alone, most commonly via correlations with the Fear Questionnaire Agoraphobia Scale (Marks & Mathews, 1979). In aggregated samples of over 600 participants, the weighted average convergent validity coefficients were large for both scales,⁴ although considerably larger for Avoidance Alone than Avoidance Accompanied (.80 vs. .55, respectively). This is to be expected, as the majority of items on the Fear Questionnaire Agoraphobia Scale, the measure with which Avoidance Accompanied was correlated, concern avoidance of situations when alone. Few data are available on convergent validity with measures of avoidance other than self-report. Chambless et al. (2002) correlated Avoidance Alone with agoraphobia severity ratings of interviewers following the Structured Clinical Interview for DMS-III-R (Spitzer, Williams, Gibbon, & First, 1989). In a sample of 22 patients with a diagnosis of panic disorder with agoraphobia, the correlation was large and statistically significant, r = .54. In contrast, testing the correlations of single Avoidance Alone items with the corresponding tasks on a behavioral approach test, Kinney and Williams (1988) found inconsistent results. The correlations ranged from small and nonsignificant (-.18) to very large and statistically significant (-.84). The median correlation of −.38, although medium in size, was not statistically significant with a sample size of only 37. The modest average agreement between self-report questionnaire items and behavioral tests may reflect the well-known limited correlation between methods of measurement (e.g., Achenbach, Krukowski, Dumenci, & Ivanova, 2005) or the limitations of correlating a measure of how much a respondent might avoid something on average (e.g., driving) versus a very specific test in a high-demand situation (e.g., driving on this particular road, on this particular day, at this particular time, when being observed by a research assistant; see Mischel, 2004). Additional data on the MIA’s convergent validity with measures other than self-report would be desirable.

Construct validity requires more than the assessment of convergent validity. A valid measure needs not only to be correlated with those measures to which it should show a relationship (e.g., other measures of the same construct), it needs to not be correlated, or to be less correlated, with measures designed to represent different constructs (Campbell & Fiske, 1959). As expected, the MIA scales show low and nonsignificant discriminant correlations with measures of distinct constructs such as scales tapping psychoticism and lying (Arrindell et al., 1995; Chambless et al., 1985). However, this sets a fairly low bar for discriminant validity. More challenging, and more important for the clinical use of the scales, is discriminant validity versus measures of other phobias. In Table 1 we also summarize the results of studies in which authors reported the correlations of the MIA scales with Fear Questionnaire Blood/Injury and Social Phobia scales (Marks & Mathews, 1979). The weighted average discriminant validity coefficients (total N = 468) approach medium in size but are notably smaller than the large convergent validity coefficients for both Avoidance Accompanied (.29 vs. .55, respectively) and Avoidance Alone (.28 vs. .80, respectively). This pattern of correlations provides support for the construct validity of the MIA as a measure of agoraphobia. However, in light of the small number of studies on discriminant validity of the MIA, additional data on this topic should be gathered.

In a handful of studies researchers have examined another important indicator of construct validity—criterion-related validity or known-groups validity (Berle et al., 2008; Chambless et al., 1985; Craske, Rachman, & Tallman, 1986; Stephenson et al., 1997). In this research, scores for both Avoidance Accompanied and Avoidance Alone have been shown to be higher in samples with agoraphobia versus (a) community or student samples (Chambless et al., Study 1; Craske et al.; Stephenson et al.), (b) patients with social phobia (Chambless et al., Study 2; Craske et al.), and (c) patients with panic disorder without agoraphobia (Berle et al.). Consistent with the discriminant validity findings summarized in the previous paragraph, these data bolster the argument that the MIA is a measure of agoraphobia rather than of phobia or anxiety disorders more generally. These findings suggest that the MIA may have utility as a screening instrument for research on agoraphobia. However, from these data, it is not possible to tell how effectively the MIA might be used in such a role, as the authors failed to report measures of diagnostic accuracy such as sensitivity (true positive rate) and specificity (true negative rate).

Hoyer, Becker, Neumer, Soeder, and Margraf (2002) addressed this gap in an epidemiologic study of young women in Dresden, Germany. Participants were given a structured diagnostic interview and a modified version of the German translation of the MIA (Ehlers & Margraf, 1993) in which avoidance when alone and when accompanied were not distinguished. Relatively few participants were diagnosed with agoraphobia (36 of 1,873 participants who completed the MIA). The MIA performed well in distinguishing those with agoraphobia in the total sample, with a cutoff score of 1.50 yielding the best combination of sensitivity (.78) and specificity (.85). When the sample was restricted to those who received at least one diagnosis (36 with agoraphobia, 215 with another diagnosis) the same cutoff score (1.50) emerged as the best, with sensitivity of .78 and specificity of .76. Given the fairly small number of people with agoraphobia and the restricted nature of the sample (women between the ages of 18 and 24) additional research on the MIA’s diagnostic accuracy is highly desirable.

The purposes of the remainder of this paper are (a) to provide additional data on the internal consistency and the convergent and discriminant validity of the MIA scales via correlations with diagnosticians’ clinical severity ratings of agoraphobia and other anxiety disorders, respectively; and (b) to further examine the MIA’s potential utility as a screening tool for agoraphobia via receiver operating curve analysis. This analysis yields not only a test of significance but also reports the sensitivity and specificity of cutoff scores with reference to diagnoses of agoraphobia assigned on the basis of a reliable structured diagnostic interview, the Anxiety Disorders Interview Schedule for DSM-IV (Brown, DiNardo, & Barlow, 1994).

On the whole, the MIA demonstrated excellent psychometric properties in the present research as well as in the prior literature. Internal consistency as evinced by Cronbach’s α has been consistently high across investigations, as have convergent validity correlations with other self-report and interviewer measures of agoraphobia. Evidence of discriminant validity vis-à-vis measures of other anxiety disorders is also positive in all investigations with one exception. In the present study, the discriminant validity correlation of Avoidance Accompanied with interviewers’ ratings of severity of social phobia was substantially smaller than the convergent validity coefficient with severity of agoraphobia (.37 vs. .54, respectively) but this difference failed to be statistically significant (p = .08). In contrast, Avoidance Accompanied convergent validity coefficients were larger than discriminant validity coefficients where severity of generalized anxiety disorder and specific phobia were concerned. For Avoidance Alone, evidence for construct validity was uniformly solid: The convergent validity coefficient with severity of agoraphobia ratings was significantly greater than the discriminant validity coefficients for Avoidance Alone with severity of generalized anxiety disorder, social phobia, and specific phobia. These data support the MIA as a specific measure of severity of agoraphobia rather than of phobic and anxiety disorders in general, and this is especially true for Avoidance Alone.

Where criterion-related validity was concerned, in past investigations the MIA has discriminated between respondents with agoraphobia and those with social phobia (Chambless et al., 1985; Craske et al., 1986) and panic disorder without agoraphobia (Berle et al., 2008). In the present study, we replicated Berle et al.’s findings by demonstrating that the MIA Avoidance Alone scale successfully discriminated between panic disorder patients with and without a diagnosis of agoraphobia. Additionally, we conducted analyses of sensitivity and specificity, finding a score of 1.61 was the optimal operating point; that is, yielded the highest sum of sensitivity (.87) and specificity (.73). These data suggest that when employing this score as a cutoff point for screening patients with panic disorder for a likely diagnosis of agoraphobia, the user will have relatively few false negatives (13%) but will incur a higher risk of false positives (27%). At 1.61, our optimal operating point was higher than the comparable point Hoyer et al. (2002) identified (1.5). Since Hoyer et al. used a modified version of the German MIA eliminating the distinction between Avoidance Alone and Avoidance Accompanied, it is possible that the discrepancy is due to the different versions of the instrument. In addition, Hoyer et al. sought a cutoff point for discriminating community participants with a diagnosis of agoraphobia from other community participants in general or from community participants who met criteria for some disorder other than agoraphobia—a different discrimination from that in the present study. Nonetheless, it is also possible that the present findings of the optimal operating point are simply not stable. Although the findings replicated across two subsamples of the present sample, additional replications are required. For the present, researchers using the MIA as a screening tool would do well to consider whether their samples more closely resemble those in the current study or those in Hoyer et al.’s study.

A limitation of the present research is that all data on reliability and validity come from a sample of treatment-seeking participants with a diagnosis of panic disorder. Thus, these findings may not extend to mixed samples of patients or people from the community. This concern, however, is mitigated by other studies on the MIA’s psychometric properties yielding positive results with a variety of populations including students, community samples, and patients with a range of anxiety disorders.