This article provides a brief review of the concepts of clinical significance and Reliable Change index. If pretest and posttest treatment scores are statistically significantly different, what does this tell programevaluators about how individuals fared in the treatment? what is a clinically significant change?
This article provides a brief review of the concepts of clinical significance and Reliable Change index. If pretest and posttest treatment scores are statistically significantly different, what does this tell programevaluators about how individuals fared in the treatment? what is a clinically significant change?
This article provides a brief review of the concepts of clinical significance and Reliable Change index. If pretest and posttest treatment scores are statistically significantly different, what does this tell programevaluators about how individuals fared in the treatment? what is a clinically significant change?
A Review of Clinical Significance, Reliable Change,
and Recommendations for Future Directions WISE RCI CS Edward A. Wise Mental Health Resources Memphis, Tennessee The purpose of this article is to provide a brief review of the history, development, and current status of the concepts of clinical significance (CS) and the reliable change index (RCI). I ad- dress issues regarding the development, criticisms, and applications of CS and RCI. I review the use of normative data, cutoff points, formula adjustments, and the comparative validity of various RCI methods. An examination of the convergence of multiple domains and multiple measures demonstrates ways to further develop the concepts of reliable change and CS. Finally, I make some recommendations and implications for future research and the development of as- sessment tools. If pretest and posttest treatment scores are statistically signif- icantly different, what does this tell programevaluators about howthe individual fared in the treatment? Do patients have to exceed p < .05 for them to have obtained a clinically signifi- cant change? For that matter, what is a clinically significant change? Jacobson, Follette, and Revenstorf (1984) identified several problems interpreting traditional psychotherapy out- come studies. These include the use of pretest and posttest group scores that provide no information about treatment ef- fects for specific individuals, the use of statistical signifi- cance tests that have little clinical relevance, the inability to determine the proportion of treated individuals that improve, and the fact that clinical significance (CS) is not consistently defined or utilized as an adjunct to statistical significance testing. Jacobson, Follette, and Revenstorf (1984) called for agreed upon conventions for determining what constitutes improvement [and] consensus as to what is meant by clini- cal significance (pp. 338339). CS Jacobson and Truax (1991) discussed the problem of differ- entiating statistical from CS and stated that judgments re- garding clinical significance are based on external standards provided by interested parties in the communities. The clinical significance of a treatment refers to its ability to meet standards of efficacy set by consumers, clinicians and re- searchers (p. 12). Jacobson, Follette, and Revenstorf (1984) propose[d] that a change in therapy is clinically significant when the client moves from the dysfunctional to the func- tional range during the course of therapy (p. 340), thereby providing a definition of CS. Jacobson, Follette, and Revenstorf (1984) operationalized the dysfunctionalfunc- tional continuum as the following: 1. Does the level of functioning at post-test fall outside the range of the dysfunctional population, where range is defined as extending to 2 standard deviations above (in the direction of functionality) the mean for that population? 2. Does the level of functioning at post-test fall within the range of the func- tional or normal population, where range is defined as begin- ning at two standard deviations belowthe mean for the normal population? 3. Does the level of functioning at post-test sug- gest that the subject is statistically more likely to be in the functional than in the dysfunctional population; that is, is the post-test score statistically more likely to be drawn from the functional than the dysfunctional distribution? (p. 340) Jacobson and Truax further clarified the choice of norms by stating the following: First, when norms are available, either 2 or 3 [as defined pre- viously] is often preferable to 1 as a cutoff point: In choosing between 2 and 3, when functional and dysfunctional popula- tions overlap, 3 is preferable to 2; but when the distributions are nonoverlapping, 2 is the cutoff point of choice. When norms are not available, 1 is the only cutoff point available: To avoid the problem of different cutoff points from study to JOURNAL OF PERSONALITY ASSESSMENT, 82(1), 5059 Copyright 2004, Lawrence Erlbaum Associates, Inc. study, 1 should be standardized by aggregating samples from study to study so that dysfunctional norms can be estab- lished. (p. 14) A normative change meeting either criterion 1, 2, or 3 would be considered CS. (An example of this procedure may be found in the Appendix.) In calculating these cutoff points, Jacobson, Follette, and Revenstorf (1984) envisioned the use of both functional and dysfunctional distributions. Although Jacobson, Follette, and Revenstorf proposed a return to the functional distribu- tion as signifying CS, they also acknowledged that it is not always possible to identify an appropriate normative group (p. 342) and that in some circumstances the return to normal criterion may be too stringent. For example, individuals treated in an inpatient psychiatric unit would not reasonably be expected to return to the normal distribution on complet- ing stabilization, although they may experience enough change to be safely discharged. In continuing to develop the concept, Jacobson and Truax (1991) also suggested using confidence bands around the CS cutoff points to further re- duce measurement error. CRITICISMS OF CS Wampold and Jenson (1986) were the first to raise the point that The assumption of two distinct distributions may be appro- priate for some clinical problems [but] a viable alternative assumption is that there is one population, which contains the functional and dysfunctional individuals, and the dysfunc- tional individuals are found in one of the tails of the distribu- tion of scores for that population. (pp. 302303). Hsu (1996) also criticized the derivation of the cutoff scores, stating that the cut scores provide no information about the probability that a client belongs to a Functional or Dysfunc- tional group (p. 373) because the two standard deviations from each of these groups are unequal. Hence, the arithmeti- cal mean between these scores might not be the true cut point differentiating one sample fromthe other. Kendall and Grove (1988) proposed that viewing symptoms on a continuum rather than as a bimodal distribution would solve the distribu- tion problem. However, Jacobson, Follette, and Revenstorf (1986) argued that irrespective of the population distribution, there were two samples of interest: those seeking treatment and those who do not. Jacobson et al. stated, As long as two distinguishable groups exist, it is possible to identify a point at which an individual is equally likely to be a member of ei- ther group. These are the necessary and sufficient conditions for establishing the cutoff points (p. 309). Tingey, Lambert, Burlingame, and Hansen (1996a) re- ported similar problems regarding the use of normative sam- ples in defining CS. Tingey et al. noted 1) an inability to identify and use relevant normative samples across studies; 2) the restriction of the social validation meth- odology by the use of only one dysfunctional and one func- tional sample; and 3) the lack of a procedure to determine the distinctness of samples (p. 110). Rather than use a return to normal criterion, Tingey et al. proposed the innovative use of multiple samples organized around a rational or empirical continuum corresponding [to] low to high levels (p. 111) of the variable being mea- sured. Establishing a functional to dysfunctional continuum addressed Wampold and Jensons (1986) concerns regard- ing the assumption of a bimodal distribution. Using this method, to be classified as CS, an individual must move from one category on the normative continuum to another but not necessarily return to normal. Tingey et al. used the Jacobson and Truax (1991) formula for determining cutoff points to be used in defining adjacent samples: where X 1 , s 1 , X 2 , and s 2 specify the means and standard devi- ations of two different samples. Tingey et al. clarified that the average normative sample variances be used in calculating the reliable change index (RCI) and that the normative sam- ples be demonstrated to be distinct, as evidenced by a one-tailed t test (p < .05) and Cohens d .50. Tingey et al. then went on to provide an illustration using the Symptom Checklist90Revised (SCL90R; Derogatis, 1983) and combined the SCL90R normative data (Mildly Symptom- atic and Severely Symptomatic) with two other distinct sam- ples comprised of Asymptomatic and Moderately Symptom- atic individuals. (Examples of CS and RCI calculations using the Tingey et al. [1996a] data may be found in the Appendix for the interested reader.) Noting that the use of confidence bands around the cutoffs resulted in unclear classifications and imposed an additional criterion to pass on an already conservative standard, they recommended only using cutoff scores and RCI to measure reliable CS change. Martinovich, Saunders, and Howard (1996) generally welcomed Tingey et als. (1996a) extension but noted sev- eral problems. These included Martinovich et al.s observa- tions that any underlying psychometric problems with the measures could be exacerbated by use of the method, and the method did not solve the problem of identifying and discriminating between functional and dysfunctional groups; they also noted difficulties involved with making distinctions involved with using Criterion 3 with nonnormal distributions (Jacobson, Follette, & Revenstorf, 1984). Follette and Collahan (1996) reiterated their position that the return to normal criterion is the necessary one and criticized the Tingey et al. method of determining if two distributions were unique and the use of intermediate nor- mative data. Tingey, Lambert, Burlingame, and Hansen RCI CS 51 1 2 2 1 1 2 ( ) ( ) , s X s X Cutoff s s
(1996b) responded to Martinovich et al. by noting that mea-
surement error was inherent in the instruments, and the method did not increase such error; that the use of interme- diate normative categories reflected clinical realities of gradual change; and for a variety of reasons, disagreed that Jacobson and Truaxs criterion c was inappropriately used. In their response to issues raised by Follette and Callaghan (1996), Tingey et al. reiterated their support for a contin- uum of normative data from the functional to dysfunctional and provided additional rationales for their method of using t and d to distinguish such distributions. Kazdin (1999) stated that clinically significant change can occur when there is a large change in symptoms, a me- dium change in symptoms, and no change in symptoms (p. 332). Kazdin (1999) indicated that symptom change may not be the gold standard on which to base clinical change and when symptomchange is the criterion, a continuumis appro- priate. Kazdin (1999) also observed that One may wish to judge treatments on the extent to which they change symp- toms, but the results could be quite different if other criteria were used, such as impairment, quality of life, or impact on others (p. 336). Kazdin (2001) also stated that it is not clear that what we refer to as clinically significant on most mea- sures or by most criteria reflects genuine or, indeed, any dif- ferences in the everyday lives of the clients (p. 456). Kazdin (2001) questions the reliance on symptom reduction as the sole or primary criterion, and the need to match the criteria and measures to the clinical problems, treatment goals, and lives of the clients (p. 455). Kazdin (2001) cogently ar- gued that it is difficult to find evidence to show that passing a threshold or entering a range means the client is better in any way that affects daily functioning or that a failure to pass this threshold means otherwise (p. 461). RCI Recognizing that a CS result could be obtained that was unre- liable, Jacobson, Follette, and Revenstorf (1984) introduced a second statistical requirement to ensure that any obtained CS change was reliable. Jacobson, Follette, and Revenstorf termed this the reliable change index (RCI). RCI was devel- oped to account for measurement error and was originally defined as where X 1 = pretest score; X 2 = posttest score; SE s r xx = 1 1 ; s 1 = the standard deviation of control group, normal population, or pretreatment group; and r xx = the testretest reliability. If RCI 1.96, then it is likely that the change was reliable (p <.05). In general, the less reliable the instrument, the greater the difference required to achieve a statistically reliable change. RCI was designed to account for the magnitude of CS change and measurement error, such as when an individual crosses the threshold from the dysfunc- tional to functional population, but the amount of change may be minimal or unreliable. According to the Jacobson, Follette, and Revenstorf criteria, for a treatment effect to be considered a reliable and significant change, it must pass a two stage process in which (a) it must be proven to be statisti- cally reliable (RCI) and (b) the individual must pass from the dysfunctional to the functional distribution (CS). Jacobson, Follette, and Revenstorf noted that if these criteria were adopted, psychotherapy would look less effective (p. 350) due to the very conservative nature of the criteria. Jacobson, Follette, Revenstorf, Baucom, et al. (1984) later supported the use of the criteria by stating that If a client changes to a degree that is both outside the range that would be explained by measurement error and places him or her with greater likelihood in the functional popula- tion this client can be considered unequivocally as a treat- ment success. (p. 498) Using these criteria, one could then classify each individual in a treatment outcome study as Recovered (passed both CS normative and RCI criteria), Improved (passed RCI criteria alone) Unchanged/Indeterminate (passed neither), or Deteri- orated (passed RCI in the negative direction). Examples of the Jacobson and Truax (1991) method of calculating CS and RCI is found in Appendix. CRITICISMS OF THE RCI Christensen and Mendoza (1986) noted that the RCI formula as originally proposed was based on an individuals obtained pretest score with inherent measurement error and not a true pretest score. To correct for this error, Christensen and Mendoza proposed the use of the S diff in place of the denomi- nator standard error (SE) to reflect the amount of difference which one could expect between two scores, obtained on the same test by the same individual, as a function of measure- ment error alone (p. 307) where S SE diff = 2 2 ( ) . Christensen and Mendoza also noted that the magnitude of difference required using this denominator is even more stringent than the original RCI formula. Hageman and Arrindell (1993) believed that this formula improved the ac- curacy of both the pretest and posttest scores, the resulting pretestposttest difference scores, and hence the RCI. Jacob- son et al. (1986) acknowledged that Christenson and Mendoza were correct and adopted their formula. In further highlighting the methodological complexities of measuring the RCI, Speer (1992) stated that The more devi- ant the initial scores and the less reliable the instrument, the greater the regression based improvement that may occur (p. 403). Although Speer noted that regression to the mean is not universal [and] occurs only when the correlation be- 52 WISE 1 2
RCI , X X SE
tween amount of change and the initial score is negative, he
went on to state that determining whether or not it is a factor that requires attention is an empirical matter and should not be ignored (p. 403). Speer (1992) clarified that the larger issue, as cited by Wampold and Jenson (1986), is whether one views the dys- functional population as separate from but overlapping with the functional population or whether they represent a sample from the extreme end of the population distribution. The more extreme the scores, the greater the chance of the scores regressing to the mean and influencing RCI. Speer (1992) then introduced the EdwardsNunnally method to minimize the risk of improvement rates capitalizing on regression to the mean (capitalizing on error of measurement), in those cir- cumstances in which regression to the mean is demon- strated (p. 404). Rather than using the uncorrected pretest score, Speer used an estimated true score based on the ob- tained pretest score to minimize regression to the mean. This correction results in more conservative recovery rates than the previously revised RCI formula. Additionally, when re- gression to the mean is not demonstrated, this formula would produce low estimates of improvement and lower reliability among the more dysfunctional scores, thereby creating dis- parities in classification rates between this formula and the Jacobson and Traux (1991) method. HagemanandArrindell (1993) pointedout that regression tothemeanrepresents aproblemonlyinas far as it is causedby measurement unreliability (p. 695). Hageman and Arrindell (1993) went ontoclarifythat Speers (1992) findingof aneg- ative correlation between initial score and change is indeed proof of the presence of regression to the mean, however not necessarily of a problematic kind (p. 695). That is, a negative correlation may also be due to true individual change and not necessarily measurement error. Hageman and Arrindell (1993) summarized Speers (1992) correction as only ad- dressing error in the pretest scores, noting that posttest scores were equally as subject to unreliability. Subsequently, Hageman and Arrindell (1993) proposed a correction to take into account error for both the pretest and posttest scores and adjusted for regression to the mean. Hageman and Arrindell (1999b) refinedtheir equationtoestimatetheunderlyingtrue scores instead of using observed scores in yet a further at- tempt to correct for regression to the mean ( ( ) ( ) ( ) / RC X X r M M r r indiv DD DD DD = + 2 1 2 1 1 SE 2 , where r DD = reliability of the difference scores). Speer (1999) later criticized Hageman and Arrindells (1999b) approach and discussed two relevant myths re- lated to RCI. Essentially, Speer changed his opinion and in- dicated that it was not necessary to adjust for regression to the mean and that in a pretestposttest design, unadjusted scores were acceptable. Speer went on, however, to state There is no agreement or consensus among methodologists about the ubiquitousness of regression to the mean, its ef- fects on d-scores [difference-scores], whether or not d-scores are biased and/or unreliable and whether or not d-scores re- ally need adjustment or correction in the analysis of two-wave data (p. 1209). COMPARISONS OF RCI METHODS In light of the statistical controversies surrounding the RCI, the construct validity of the methods and their relative accu- racies are of considerable importance. In one of the first em- pirical evaluations of change scores, Speer and Greenbaum (1995) examined four pretestposttest methods (Jacob- sonTruax, EdwardsNunnally, HsuLinnLord, and NunnallyKotsch) along with a hierarchical linear model (HLM). These pretestposttest methods used residualized pretreatment or difference scores, which were believed to be more accurate than the standard formula because of im- proved reliability. When Speer and Greenbaum omitted the pretestposttest method with the lowest average rate of clas- sification agreement, they found the average classification rate of agreement was 89%and the HLMwas less sensitive in classifying clients as Deteriorated compared to the difference score methods. Among the four pretestposttest difference score methods studied, Speer and Greenbaum Recommend[ed] use of the Jacobson and Truax (1991) pre-post difference method for the following reasons: (a) It avoids statistical problems associated with residualized true score adjustments; regression to the mean is neither inevitable nor, perhaps, as big a problem as previously thought; (b) it is computationally straightforward; and (c) there is a small litera- ture reporting change rates produced by this method. (p. 1047) McGlinchey and Jacobson (1999) compared the Jacobson andTruax(1991) methodwithone proposedbyHagemanand Arrindell (1999b). McGlinchey and Jacobson concluded that there were no substantial differences between the two meth- ods and that the complex computations used in the Hageman and Arrindell (1999b) method made the Jacobson and Truax formula the preferred method. Hageman and Arrindell (1999a) criticized the choice of norms used by McGlinchey and Jacobson, which Hageman and Arrindell (1999a) argued affectedthestandarderror of measurement andhencetheclas- sifications. Hageman and Arrindell (1999a) pointed out the importance of using the correct value for the standard error of measurement to be able to compare studies. McGlinchey, Atkins, and Jacobson (2002) subsequently compared five methods for determining RCI, including four used in the Speer and Greenbaum (1995) study as well as the Hageman and Arrindell (1999b) formula. In addition to ex- amining differential classification rates, McGlinchey et al. also studied the methods for predictive accuracy with respect to relapse. Whereas McGlinchey et al. found some differ- ences in classification rates, Hageman and Arrindell had the lowest number of Recovered patients and hence was the most conservative estimate. Nonetheless, All five methods sig- RCI CS 53 nificantly discriminated between participants who relapsed during the 2 years following therapy (McGlinchey et al., 2002, p. 541). McGlinchey et al. concluded that the evi- dence supports the Jacobson Truax method as a null method that has yet to be rejected by an alternative method of superior performance (p. 542). Bauer, Lambert, and Nielsen (2003) combined the meth- ods used by both Speer and Greenbaum (1995) and McGlinchey et al. (2002) to compare the classification ac- curacy of five RCI methods (JacobsonTruax [Jacobson & Traux, 1991]; EdwardsNunnally [Speer, 1992]; GulliksenLordNovick [Hsu, 1989]; HagemanArrindell [Hagemann & Arrindell, 1999b]; and the HLM approach). Bauer et al. found that the EdwardsNunnally and HagemanArrindell approaches demonstrated poor conver- gence with the other three methods. The Ed- wardsNunnally method tended to produce the most liberal Recovery rates, whereas the HagemanArrindell method produced the most conservative. The HLM method was noted to require more than two data points (pre- testposttest) and produced relatively low convergence with the other methods. Of the remaining two methods, Bauer et al. found little difference between the Jacob- sonTruax method and the GulliksenLordNovick method. Due to the widespread use of the JacobsonTruax method and the relative ease of calculation, Bauer et al. (2003) recommended the JacobsonTruax method. Thus, Speer and Greenbaum (1995), Speer (1999), Maassen (2001), McGlinchey et al. (2002), and Bauer et al. demon- strated the relative convergence of the JacobsonTruax method, and these authors called for a moratorium on creat- ing alternative RCI formulas to study the results derived from the Jacobson and Truax (1991) formula. Consensus appears to favor further study and development of the Ja- cobson and Truax approach. APPLICATIONS AND INNOVATIONS In a selected review of the RCI and CS literature, Ogles, Lunnen, and Bonesteel (2001) identified 74 published RCI-related articles in a 9-year period in the Journal of Con- sultingandClinical Psychology, rangingfrom3to14publica- tionsper year. Ogleset al. foundthat acrossall outcomestudies included, an RCI analysis was the most frequent method em- ployed. Manyprestigious journals have publishedspecial sec- tions related to the use of and debate about RCIs (e.g., Journal of Consulting and Clinical Psychology, Clinical Psychology: Research and Practice, Psychotherapy Research, Behaviour Research and Therapy, Behavioral Assessment, etc.). The method has been used with a variety of clients including adults, children, people with personality disorders as well as medical and neuropsychological patients (e.g., Dolan, Evans, & Wilson, 1992; Ferguson, Robinson, & Splaine, 2002; Ja- cobson, Roberts, Berns, & McGlinchey, 1999; Lunnen & Ogles, 1998; Sheldrick, Kendall, &Heimberg, 2001; Temkin, Heaton, Grant, Dikmen, & Sureyya, 1999). Jacobson and Revenstorf (1988) offered some method- ological refinements by noting that when multiple measures of similar constructs were used to assess treatment outcomes, summary statistics of mean recovery rates across the mea- sures could be computed to calculate CS and RCI. Alterna- tively, Jacobson and Revenstorf suggested a multivariate weighted composite score based on all measures that could be analyzed by separating the functional and dysfunctional populations, which would allow for the use of cutoff points. Discriminant functions were also mentioned when using multiple measures to classify individuals into functional or dysfunctional categories. When multiple measures are used to assess different constructs, however, Jacobson and Revenstorf advocated using the multiple measures and ac- cept[ing] the fact that no single index of clinically significant change will capture all components of the disorder under study (p. 140). Other innovative applications include Nietzel, Russell, Hemmings, and Gretters (1987) meta-analytic study of pa- tients with unipolar depression in which composite group meanscores were comparedtonormative groups todetermine CS change. Nietzel et al. also recommended a normative change of 1 SD instead of 2 SDs as the cutoff point for CS. Abramowitz (1998) used the RCI formula in a meta-analytic review of exposure therapy in the treatment of obses- sivecompulsive disorders and calculated a change score for each group instead of on each patient. Similarly, Sheldrick et al. (2001) employedtheRCI formulawithnormativecompari- son methods in the treatment of children with conduct disor- ders. Sheldricket al. derivedchange scores for eachgroupand then utilized group scores to determine if treated individuals werereturnedtonormal limits(1SD) todetermineCS. Seggar, Lambert, and Hansen (2002) developed a normative contin- uumusing the Beck Depression Inventory (Beck, 1987) simi- lar to that created by Tingey et al. (1996). Adding to the innovative work of Tingey (1989) and Tingey et al. (1996a), Wise (in press) extended the SCL90R normative contin- uumbyaddinga sample of 225outpatients attendinganInten- sive Outpatient Program, thereby adding a more severe population of outpatients. To address the multidimensional nature of clinical signifi- cance, the use of multiple measures or criteria have been use- ful. Ogles, Lambert, and Masters (1996) provided RCI graphs using various scales and instruments that facilitate the use of RCI by clinicians. Ogles, Lambert, and Sawyer (1995) dem- onstrated RCI convergence between a structured rating scale completed by clinicians and two client rated symptom mea- sures on a national sample of outpatients with depression un- der four different treatment conditions. Lunnen and Ogles (1998) studied the RCI using the perspectives of therapists, clients, andspouses torate symptomchange, alliance, andsat- isfaction. Beckstead et al. (2003) demonstrated the use of the RCI on five different instruments on the same population and 54 WISE found a 65% classification agreement rate between them. Becksteadet al. concludedthat different measures of different domains will produce different RCI classification results. Wise (in press) examined the effects of varying RCI criteria (1.96, 1.28, and .84, corresponding to 95%, 90%, and 80% confidence levels, respectively) with pretestposttest client symptom ratings, pretestposttest clinician-rated level of functioning, client satisfaction, and discharge to a lower level of care and demonstrated that although different measures of different domains produce different results, convergence of outcome classification rates can also be demonstrated. FUTURE DIRECTIONS AND RECOMMENDATIONS Jacobson and colleagues (Jacobson, Follette, & Revenstorf, 1984; Jacobson & Revenstorf, 1988; Jacobson et al., 1999; Jacobson & Truax, 1991) have repeatedly contended that consumers and clinicians expect to be as normal as their functional counterparts by the time therapy has ended (Ja- cobson & Revenstorf, 1988, p. 134) and that this return to normal criterion should be the definition of CS. The use of a return to normal criterion seems unrealistic for many clinical practice contexts. This is most evident in assessing the treat- ment effects of inpatient programs, partial hospitalization programs, and intensive outpatient programs. The compari- son may also not be appropriate for certain types of cases such as dual diagnosis disorders, personality disorders, psy- chiatric patients with comorbid medical problems, and so forth. For these patients, the natural course, chronicity, and recurrent waxing and waning of intensity of symptoms ar- gues against a return to normal criterion. In fact, literature re- views and meta-analytic studies of outpatients with depres- sion report that few achieve complete remission and despite significant improvements, some data indicate that many of these patients remain more depressed at the end of therapy than the general population (e.g., Hansen, Lambert, & Forman, 2002; Lecrubier, 2002; Nietzel et al., 1987; Robin- son, Berman, & Neimeyer, 1990; Westin & Morrison, 2001). Additionally, a functional return to normal may be expected to take considerably longer than symptomremission, and it is not clear if particular functional impairments remit faster or slower than others (e.g., return to work vs. increased social- ization). Because comparatively few patients actually achieve a full remission of depressive symptoms, and func- tional capacities are the last improvements, the return to nor- mal criterion appears unrealistic. The work of Norman, Sloan, and Wyrwich (2003), who conducted a review of the literature on minimally important differences (MID) for health-related quality-of-life instru- ments, is particularly relevant to the return to normal criteria. Based on 38 studies and 62 effect sizes, Norman et al. dem- onstrated that irrespective of the disease or instrument, .5 SD consistently detected reliable change in chronic medical pa- tients. This consistency corresponds almost exactly to the limit of human discrimination identified by Miller (1956) over 40 years ago (Norman et al., 2003, p. 588). Norman et al. also found that otherwise healthy people recovering from an episode of back or shoulder pain, referred to a therapist, with every expectation of complete recovery (p. 589) dem- onstrated a higher threshold for minimal change than those with chronic conditions. This indicates that the patients ex- pectations for full or partial recovery influence their discrim- ination of minimal change. This suggests that psychotherapy patients expectations regarding a return to normal outcome or a decrease but not elimination of symptoms would affect their judgments of change. Hence, Jacobson and Truaxs (1991) return to normal criterion may accurately reflect the perspective of the patient who has transient situational disor- ders but may not be accurate for those with more chronic psy- chological disorders such as Dysthymia, refractory or recurrent Major Depression, patients with underlying Axis II disorders, and so forth. Whereas some researchers noted pre- viously have recommended or utilized 1 SD 1 as a CS norma- tive change criterion, the MID work of Norman et al. provided support for lessening this criterion to .5 SD based on the patient population and suggests that corresponding RCI adjustments may also be appropriate. Note that the cost of lessening the standards for CS and RCI could be to increase the number of patients classified as Improved or Deteriorated. However, this would assume that the patients are normally distributed around a mean change of zero and that any change to the criterion would have a symmetrical effect. If, on the other hand, therapy is at all ef- fective (e.g., Nietzel et al., 1987; Robinson et al., 1990; Shadish et al., 1997; Smith, Glass, & Miller, 1980; Wampold, 2001), then the change would not necessarily be symmetrical and could result in more people classified as Im- proved rather than Deteriorated. In any event, such increases are likely to be relatively small. Additionally, varying the RCI allows the CS variables to have a greater influence in the determination of therapy out- comes in general and in individual improvement rates in par- ticular. In addition to varying the number of standard deviation units required to demonstrate normative CS change, a symptom scale might also be used by applying varying confidence levels to the RCI formula (e.g., 1.96, 1.28, and .84, corresponding to 95%, 90%, and 80% confi- dence levels, respectively). Such varying RCI confidence levels could be used in conjunction with real-world measures such as discharge to a lower level of care, clinician ratings, functional ratings, client ratings of coping, ratings by signifi- RCI CS 55 1 In reviewing the meta-analytic studies of psychotherapy out- comes, Wampold (2001) noted that the average improvement was reflected by an effect size (ES) of .80. Because a change of 1 SDcor- responds to an ES of 1.0, where .80 is considered to be a large ES, it would appear that a change of 1 SD is also a defensible indicator of CS. cant others, and client satisfaction (e.g., see Wise, in press). In this example, varying the RCI cutoff from 1.96 to .84 might be used with CS variables such as discharge to a lower level of care, clinician Global Assessment of Functioning Scale (GAF; American Psychiatric Association, 2000) rat- ings that were 1 SD closer to the functional distribution, cli- ent reports that treatment helped them cope more effectively with their problems, and so forth. Alternatively, an RCI cut- off of 1.96 might be used with less rigorous or unstandard- ized CS variables, whereas .84 might be used with standardized measures such as symptom rating scales. The use of multiple measures with varying levels of stringency allows for the comparative and convergent validity of treat- ment effectiveness within studies. Altering the RCI or CS criteria in some settings under some conditions using various measures might be more informative indicators of outcome. For example, when Wise (in press) reduced the RCI crite- rion from 1.96 to .84, the number of patients who Deterio- rated remained unchanged, whereas the number who Improved increased by 4%. However, when the same changes in RCI were used with Discharge to a Lower Level of Care as the CS variable, the number of Improved in- creased by 12%, the number who Deteriorated increased from 3% to 11%, and all of those reclassified came from the Indeterminant/Unclassified 2 group. Inspection of the charts of the Deteriorated group revealed that 92% of these patients had experienced significant psychosocial stressors in the 2 weeks before termination. Not only were the majority of the patients who were reclassified from the Indeterminant/Un- classified group, but the finding that 92% of those who Dete- riorated experienced psychosocial stressors just prior to discharge proved to be an important one with immediate, clinically relevant implications that otherwise would have been overlooked. Irrespective of what normative variables or cutoffs are uti- lized, RCI imposes an upper limit on the number of cases classified as Recovered, Improved, Indeterminant/Unclassi- fied or Deteriorated. Thus, if only 50% of a sample pass the RCI criteria for symptom improvement, individuals who demonstrate significant normative change on the CS variable of interest but fail to pass the RCI criteria, will not be in- cluded in these recovery rates. Additionally, although there is some differentiation between the Recovered and Improved categories in terms of CS, there is no such distinction within the Deteriorated classification. That is, the criterion of RCI change alone determines the classification of Deteriorated. Perhaps an additional category reflecting a reliably worse condition would be more accurate for those demonstrating only a negative RCI, whereas Deteriorated could be used to denote those who passed both the RCI and CS criteria in the negative direction. In any event, this degree of reliance on RCI seems to defeat the role and purpose of CS variables contributing to the classification of psychotherapy outcomes. There can be little doubt that traditional RCI recovery rates are extremely conservative psychotherapy outcome measures and that those who cross into the functional range are unequivocally treatment success[es] (Jacobson, Follette, Revenstorf, Baucom, et al., 1984, p. 498). De- veloping methods to further refine RCI and CS measures in the large proportion of patients in the Indeterminant/Unclas- sified range would be of considerable help in more accu- rately identifying and studying those who are not unequivocal treatment successes but who are nonetheless im- proving and on their way to a positive outcome as well as those who are not responding to treatment. Examining treatment response classifications from the psychopharmacology literature (Lecrubier, 2002) indicates that Remission is defined as a symptom reduction of 75% to 100% lasting 2 weeks to 6 months, whereas Recovery is de- fined as the same amount of symptom reduction lasting > 6 months. A positive or negative Response is defined as 50% symptomreduction or increase, and a Partial Response is de- fined as a 25% to 49% reduction or increase in symptoms. Table 1 shows one way of combining these definitions with RCI, CS, and Norman et al.s (2003) MID analyses in an ef- fort to further identify and classify those who would tradi- tionally be classified in the Indeterminant/Unclassified range. The gradations in Table 1 reflect the continuumof Re- coveryDeterioration as well as the degree of confidence that can be placed in the classifications. Additionally, note that in their meta-analytic review, Rob- inson et al. (1990) found posttreatment findings to be highly significantly predictive of follow-up findings. Another way to classify the Indeterminant/Unclassified group might be to analyze clinical variables that achieve either RCI or CS and 56 WISE 2 Statistically, this group shows No Reliable Change and histori- cally has been called Unchanged, No Change, and Unclassified. However, these people may have Improved or Deteriorated below the threshold of statistical detection; therefore, I refer to them as Indeterminant/Unclassified. TABLE 1 Proposed Terms to Classify Reliable Change Index and Clinical Significance Reliable Change Indexes Clinical Significance and Normative Change Criterion Confidence Levels (%) Change of 2 SDs Change of 1 SD Change of 0.5 SD 1.96 95 Recovered (+) Response Minimal (+) Response 1.28 90 Remitted (+) Response Minimal (+) Response 0.84 80 Improved (+) Partial Response Minimal (+) Response 0.84 80 Mildly Deteriorated () Partial Response Minimal () Response 1.28 90 Moderately Deteriorated () Response Minimal () Response 1.96 95 Deteriorated () Response Minimal () Response to classify these individuals as Improved (e.g., see Hansen, Lambert & Forman, 2002, and Wise, in press). Becausecontrol groupsandrandomassignment arenot typ- ically appropriate in naturalistic studies, unique methodolo- gies are often employed to accommodate clinically representative studies. Analyses of treatment outcomes that demonstrate RCI + CS results should be accorded greater weight inrecognizingtheir demonstratedeffectiveness. Simi- larly, individuals passing RCI or normative CS criteria with additional supporting real-world operationalized clinical measures or indicators (e.g., transfer to a lower level of care, functional ratings, clinician ratings, spouse ratings, etc.) are clearly passing a higher bar than traditional pretestposttest groupdesigns focusedonlyonstatistical significanceandsub- sequently should be accorded such relative status in acknowl- edging their empirically validated treatment effectiveness. Some clinically relevant measures may be indirectly re- latedtofunctional status (e.g., symptomseverity, GAFscale), whereas others reflect behavioral change (e.g., discharge to a lower level of care, returntowork, absenteeism, number of so- cial contacts, etc.). Althoughit has beendemonstratedthat im- provement in functional status takes longer than symptom improvement and is directly related to the dose and phase of treatment (Howard, Leuger, Maling, & Martinovich, 1993), there are few measures that evaluate both symptoms and the multiple domains of behavioral functional status. Although some measures assess familial, social, work, leisure, and health status (e.g., Sheehan, 1986; Ware, 1993; Weissman, Klerman, Paykel, Prusoff, &Hanson, 1974), they do not mea- sure all of these domains, are not in the public domain, are in- complete as stand alone outcome measures, or add too much time to be included in a pretestposttest design in addition to a stand alone symptom measure in real-world settings. Recog- nizingtheneedfor aninstrument that was practitioner friendly and assessed multiple domains, Lambert et al. (1996) pro- vided the OQ45, a brief 45-item instrument that assesses Symptom Distress, Interpersonal Relations, and Social Role Performance (employment, family, andleisure). Of particular relevance is the fact that Beckstead et al. (2003) demonstrated evidencesupportingtheconstruct validityof theOQ45.2cut- off scores for CS. Similarly, Grissom, Lyons, and Lutz (2002) reportedonanewinstrument, Treatment EvaluationandMan- agement, that is basedonthe dose andphase model of psycho- therapy. It is comprised of 92 items that assess Subjective Well-Being (emotional and physical health), Symptoms (de- pression, anxiety, somatization, substance abuse, etc.), Func- tional Disability (social, vocational and activities of daily living), Therapeutic Bond and Satisfaction With Care (work- ing alliance, understanding and trust), along with inconsis- tency and malingering indicators. Although it is too early to know the parameters of the normative database, the initial psychometrics are promising, and this instrument is notewor- thy for its emphasis on assessing multiple domains including alliance andcase mixadjustment variables withrelativelyfew items. Clearly there is a need for real-world measures designed to assess both symptoms and functional capacities in inter- vals or increments that can be expected to respond to psycho- therapy dosages, that are quantifiable, normed, and assess multiple functional domains. As mental health becomes inte- grated with medical care, these needs for multiple domain as- sessments will become even more evident. The RCI and CS methodology has withstood rigorous debate and survived stronger than originally conceived. Despite methodological limitations, studying RCI and CS has moved the outcomes paradigm from studying treatment groups to studying indi- vidual change within those groups. Similarly, assessment in- struments must move beyond symptom focus and evaluate individuals with respect to the complex broader domains of their functional, real-world, lives in which clinically signifi- cant change is operationalized. REFERENCES Abramowitz, J. S. (1998). Does cognitive behavior therapy cure obses- sivecompulsive disorder? A meta-analytic evaluation of clinical signifi- cance. Behavior Therapy, 29, 339355. American Psychiatric Association. (2000). Diagnostic and statistical man- ual (4th ed.). Washington, DC: Author. Bauer, S., Lambert, M. J., & Nielsen, S. L. (2003). Clinical significance methods: A comparison of statistical techniques. Manuscript submitted for publication. Beck, A.T., & Steer, R. A. (1987). Beck Drepression Inventory manual. San Antonio, TX: The Psychological Corporation, Harcourt Brace Jovanovich. Beckstead, D. J., Hatch, A. L., Lambert, M. J., Eggett, D. L., Goates, M. K., & Vermeersch, D. A. (2003). Clinical significance of the Outcome Ques- tionnaire (OQ45.2). The Behavior Analyst Today, 4, 7990. Christensen, L., &Mendoza, J. L. (1986). Amethod of assessing change in a single subject: An alteration of the RC index. Behavior Therapy, 15, 305308. Derogatis, L. R. (1983). SCL90R: Administration, scoring and proce- dures manual (2nd ed.). Towson, MD: Clinical Psychometric Research. Dolan, B. M., Evans, C., & Wilson, J. (1992). Therapeutic community treat- ment for personality disordered adults: Changes in neurotic symptomatology on follow-up. International Journal of Social Psychia- try, 38, 243250. Ferguson, R. J., Robinson, A. B., & Splaine, M. (2002). Use of the Reliable Change Index to evaluate clinical significance in SF36 outcomes. Qual- ity of Life Research, 11, 509516. Follette, W. C., & Callaghan, G. M. (1996). The importance of the principle of clinical significanceDefining significant to whom and for what pur- pose: A response to Tingey, Lambert, Burlingame, and Hansen. Psycho- therapy Research, 6, 133143. Grissom, G. R., Lyons, J. S., &Lutz, W. (2002). Standing on the shoulders of a giant: Development of an outcome management system based on the dose model and phase model of psychotherapy. Psychotherapy Research, 12, 397412. Hageman, W. L., &Arrindell, W. A. (1993). Afurther refinement of the reli- able change index by improving the prepost difference score: Intro- ducing the RC ID . Behaviour Research and Therapy, 51, 693700. Hageman, W. L., &Arrindell, W. A. (1999a). Clinically significant and prac- tical?: Enhancing precision does make a difference: Reply to McGlinchey and Jacobson, Hsu, and Speer. Behaviour Research and Therapy, 37, 12191233. Hageman, W. L., &Arrindell, W. A. (1999b). Establishing clinically signifi- cant change: Increment of precision and the distinction between individ- RCI CS 57 ual and group level analysis. Behaviour Research and Therapy, 37, 11691193. Hansen, N. B., Lambert, M. J., & Forman, E. M. (2002). The psychotherapy dose-response effect and its implications for treatment delivery services. Clinical Psychology: Science and Practice, 9, 329343. Howard, K. I., Leuger, R., Maling, M., & Martinovich, Z. (1993). A phase model of psychotherapy: Causal medication of outcome. Journal of Con- sulting and Clinical Psychology, 61, 678685. Hsu, L. M. (1989). Reliable changes in psychotherapy: Taking into account regression toward the mean. Behavioral Assessment, 11, 459467. Hsu, L. M. (1996). On the identification of clinically significant client changes: Reinterpretation of Jacobsons cut scores. Journal of Psychopathology and Behavioral Assessment, 18, 371385. Jacobson, N. S., Follette, W. C., & Revenstorf, D. (1984). Psychotherapy outcome research: Methods for reporting variability and evaluating clini- cal significance. Behavior Therapy, 15, 336352. Jacobson, N. S., Follette, W. C., &Revenstorf, D. (1986). Toward a standard definition of clinically significant change. Behavior Therapy, 15, 309311. Jacobson, N. S., Follette, W. C., Revenstorf, D., Baucom, D. H., Hahlweg, K., & Margolin, G. (1984). Variability in outcome and clinical signifi- cance of behavioral marital therapy: A re-analysis of outcome. Journal of Consulting and Clinical Psychology, 52, 497504. Jacobson, N. S., &Revenstorf, D. (1988). Statistics for assessing the clinical significance of psychotherapy techniques: Issues, problems and new de- velopments. Behavioral Assessment, 10, 133145. Jacobson, N. S., Roberts, L. J., Berns, S. B., & McGlinchey, J. B. (1999). Methods for determining the clinical significance of treatment effects: Description, application and alternatives. Journal of Consulting and Clin- ical Psychology, 67, 300307. Jacobson, N. S., & Truax, P. (1991). Clinical significance: A statistical ap- proach to defining meaningful change in psychotherapy research. Journal of Consulting and Clinical Psychology, 59, 1219. Kazdin, A. E. (1999). The meanings and measurement of clinical signifi- cance. Journal of Consulting and Clinical Psychology, 67, 332339. Kazdin, A. E. (2001). Almost clinically significant (p. < .10): Current mea- sures may only approach clinical significance. Clinical Psychology: Sci- ence and Practice, 8, 455462. Kendall, P. C., & Grove, W. M. (1988). Normative comparisons in therapy outcome. Behavioral Assessment, 10, 147158. Lambert, M. J., Hansen, N. B., Umphress, V., Lunnen, K., Okiishi, J., Burlingame, G., et al. (1996). Administration and scoring manual for the Outcome Questionnaire (OQ45.2). Wilmington, DE: American Profes- sional Credentialing Services. Lecrubier, Y. (2002). How do you define remission? Acta Psychiatrica Scandinavica, 106(Suppl. 415), 711. Lunnen, K. M., & Ogles, B. M. (1998). A multiperspective, multivariable evaluation of reliable change. Journal of Consulting and Clinical Psy- chology, 66, 400410. Maassen, G. H. (2001). The unreliable change of reliable change indices. Behaviour Research and Therapy, 39, 495498. Martinovich, A., Saunders, S., & Howard, K. I. (1996). Some comments on Assessing clinical significance. Psychotherapy Research, 6, 124132. McGlinchey, J. B., Atkins, D. C., & Jacobson, N. S. (2002). Clinical signifi- cant methods: Which one to use and how useful are they? Behavior Ther- apy, 33, 529550. McGlinchey, J. B., & Jacobson, N. S. (1999). Clinically significant but im- practical?: A response to Hageman and Arrindell. Behaviour Research and Therapy, 37, 12111217. Miller, G. A. (1956). The number seven plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63, 8197. Nietzel, M. T., Russell, R. L., Hemmings, K. A., & Gretter, M. L. (1987). Clinical significance of psychotherapy for unipolar depression: A meta-analytic approach to social comparison. Journal of Consulting & Clinical Psychology, 55, 156161. Norman, G. R., Sloan, J. A., & Wyrwich, K. W. (2003). Interpretation of changes in health-related quality of life: The remarkable universality of half a standard deviation. Medical Care, 41, 582592. Nunnally, J. C., & Kotsch, W. E. (1983). Studies of individual subjects: Logic and methods of analysis. British Journal of Clinical Psychology, 22, 8393. Ogles, B. M., Lambert, M. J., &Masters, K. S. (1996). Assessing outcome in clinical practice. Boston: Allyn & Bacon. Ogles, B. M., Lambert, M. J., & Sawyer, J. D. (1995). Clinical significance of the National Institute of Mental Health Treatment of Depression Col- laborative Research Program data. Journal of Consulting and Clinical Psychology, 63, 321326. Ogles, B. M., Lunnen, K. M., & Bonesteel, K. (2001). Clinical significance: History, application and current practice. Clinical Psychology Review, 21, 421446. Robinson, L. A., Berman, J. S., & Neimeyer, R. A. (1990). Psychotherapy for the treatment of depression: A comprehensive review of controlled outcome research. Psychological Bulletin, 108, 3049. Seggar, L. B., Lambert, M. J., & Hansen, N. B. (2002). Assessing clinical significance: Application to the Beck Depression Inventory. Behavior Therapy, 33, 253269. Shadish, W., Matt, G., Navarro, A., Siegle, G., Crits-Christoph, P., Hazelrigg, M., et al. (1997). Evidence that therapy works in clinically rep- resentative conditions. Journal of Consulting and Clinical Psychology, 65, 355365. Sheehan, D. V. (1986). The anxiety disease. New York: Bantam. Sheldrick, R. C., Kendall, P. C., & Heimberg, R. G. (2001). The clinical significance of treatments: A comparison of three treatment for conduct disordered children. Clinical Psychology: Science and Practice, 8, 418430. Smith, M., Glass, G., & Miller, T. (1980). The benefits of psychotherapy. Baltimore: Johns Hopkins University Press. Speer, D. C. (1992). Clinically significant change: Jacobson and Truax (1991) revisited. Journal of Consulting and Clinical Psychology, 60, 402408. Speer, D. C. (1999). What is the role of two-wave designs in clinical re- search? Comment on Hageman and Arrindell. Behaviour Research and Therapy, 37, 12031210. Speer, D. C., & Greenbaum, P. E. (1995). Five methods for computing sig- nificant individual client change and improvement rates: Support for an individual growth curve approach. Journal of Consulting and Clinical Psychology, 63, 10441048. Temkin, N. R., Heaton, R. K., Grant, I., Dikmen, S., & Sureyya S. (1999). Detecting significant change in neuropsychological test performance: A comparison of four models. Journal of the International Neuropsychological Society, 5, 357369. Tingey, R. C. (1989). Assessing clinical significance: Extension in methods an application to the SCL90R. Dissertation Abstracts International, 50(04B), 1659. Tingey, R. C., Lambert, M. L., Burlingame, G. M., &Hansen, N. B. (1996a). Assessing clinical significance: Proposed extensions to the method. Psy- chotherapy Research, 6, 109 103. Tingey, R. C., Lambert, M. L., Burlingame, G. M., &Hansen, N. B. (1996b). Clinically significant change: Practical indicators for evaluating psycho- therapy outcome. Psychotherapy Research, 6, 144153. Wampold, B. E. (2001). The great psychotherapy debate: Models, methods and findings. Mahwah, NJ: Lawrence Erlbaum Associates, Inc. Wampold, B. E., &Jenson, W. R. (1986). Clinical significance revisited. Be- havior Therapy, 17, 302305. Ware, J. E. (1993). SF36 Health Survey: Manual and interpretation guide. Boston: Nimrod. Weissman, M. M., Klerman, G. L., Paykel, E. S., Prusoff, B., & Hanson, B. (1974). Treatment effects in the social adjustment of depressed patients. Archives of General Psychiatry, 30, 771778. Westin, D., & Morrison, K. (2001). A multidimensional meta-analysis of treatments for depression, panic, and generalized anxiety disorder: An 58 WISE empirical examination of the status of empirically supported therapies. Journal of Consulting and Clinical Psychology, 69, 875899. Wise, E. A. (in press). Psychotherapy outcome and satisfaction methods ap- plied to intensive outpatient programming in a private practice setting. Psychotherapy: Theory, Research and Practice. APPENDIX To demonstrate the application of CS and RCI, Derogatiss (1983) SCL90R Global Severity Index (GSI) scale will be used as an example. Recall that Jacobson and Truax (1991) indicated that when two distributions are nonoverlapping, the cutoff point of choice is 2 SDs above the functional group. In this case, Derogatiss (1983) non-patient normal group obtained a M GSI = .31 with a SD =.31, whereas the Outpatient group had a M = 1.26 with a SD = .68. Hence, the cutoff point would equal 2 SDs above the nonpatient group, or (2 .31) + .31 (M) = .93. An individual whose pretreat- ment GSI score = 1.4 and posttreatment GSI = .80 would be considered to have obtained a CS change because they crossed the .93 threshold. To determine whether or not such a CS change is reliable requires calcu- lation of the Jacobson & Truax (1991) RCI formula: where X 1 = pretest score; X 2 = posttest score; ; s 1 = the standard deviation of control group, normal population, or pretreat- ment group; and r xx = the testretest reliability. Again, using the SCL90R Non-patient normal norms where the GSI SD = .31 and the GSI testretest correlation coefficient was reported by Tingey et al. (1996a) to be .939, Because 7.5 is 1.96, 7.5 results represent a reliable change. Alternatively, using normative cutoff points between adjacent samples, the following formula is used: where X 1 , s 1 , X 2 and s 2 specify the means and standard deviations of two dif- ferent samples. The sample means (and standard deviations) for the SCL90R GSI normative continuum gathered by Tingey et al. (1996a) and calculated by this formula are Asymptomatic = 0.19 (0.16), Mildly Symp- tomatic = 0.31 (0.31), Moderate Symptomatic = 0.79 (0.45), and Severely Symptomatic = 1.30 (0.82). The cutoff scores between the Asymptomatic and Mildly Symptomatic would be Similarly, computing the cutoff points for the Mildly Symptomatic:Moder- ately Symptomatic = 0.51 and Moderately Symptomatic:Severerly Symp- tomatic = 0.97. Tingey et al. (1996a) also required that the adjacent groups be differentiated by t < .05 and d >.50, which was demonstrated between these groups. Thus, an individual whose pretreatment GSI score was 1.4 and posttreatment GSI = 0.80 would be considered to cross the normative CS cutoff by passing the 0.97 cutoff. In this case, a normative shift from one group on the normative continuum to a less pathological normative group provides an alternative to the traditional CS method. To determine if this CS change is reliable, RCI must be calculated. Using these figures with the RCI formula cited previously, Then, because GSI X 1 = 1.4 and X 2 = 0.80, Because 3.0 1.96, the CS normative change in score from 1.4 to 0.80 is considered to be reliable. Edward A. Wise Mental Health Resources 1027 South Yates Road Memphis, TN 38119 E-mail: [email protected] Received June 15, 2003 Revised July 17, 2003 RCI CS 59 1 2
, X X RCI SE
1 1 xx SE s r 1 2 2 1 1 2 ( ) ( ) Cutoff . s X s X s s