Our first study in this series (
Williams & Butcher, 1989
) revealed considerable support for the validity of the Minnesota Multiphasic Personality Inventory (MMPI) standard scales in an adolescent treatment sample for both genders. However, from the early days of the use of the MMPI, researchers have stressed that configural interpretations of individuals' profiles provided much more useful information than interpretations based on the elevation of a single scale. Because of this, the code type approach to MMPI interpretation evolved (
Graham, 1987
). Some of the original work used complex rules for defining code types, but these approaches gradually fell out of use because only a small percentage of MMPI profiles in any given setting could be classified using these complex rules. As
Graham (1987)
pointed out, the current trend in the configural interpretation of the MMPI is the use of a simple two-scale approach to classification. The present study examined these current code type classification procedures for adolescents.
The code type approach was used with adolescents in the work of
Marks, Seeman, and Haller (1974)
. Using their sample of 834 adolescents in outpatient treatment,
Marks et al. (1974)
developed a set of rules for two-point code types and were able to classify all but 13 subjects. They then used therapist and self-report scales to provide empirical descriptors for the code types found in their sample. Unfortunately, the procedures used by
Marks et al. (1974)
were never replicated in subsequent studies with other large adolescent samples. Thus, for the last 15 years their adolescent code type descriptors remained the only source of MMPI code type descriptors based on responses from adolescents.
Questions have been raised about the
Marks et al. (1974)
adolescent descriptors because MMPI interpretations based on them sometimes varied from MMPI interpretations based on the traditional adult code type descriptors (
Graham, 1987
;
Williams, 1986
). This led to the awkward recommendation that both adolescent and adult code type descriptors be used for generating hypotheses about an adolescent's behavior (
Archer, 1984
,
1987
;
Graham, 1987
;
Williams, 1986
). This recommendation is problematic in cases when the adult and adolescent descriptors lead to significantly different impressions of an individual adolescent. The use of other assessment information (e.g., interviews, teacher reports, other testing) has been suggested as a method of resolving these discrepancies in MMPI interpretations for adolescents (
Graham, 1987
;
Williams, 1986
) in the absence of empirical data supporting the use of one set of descriptors over the other.
An additional problem with the adolescent code type descriptors provided by
Marks et al. (1974)
was revealed in a study of the clinical accuracy (determined by therapists) of the three most commonly used interpretive strategies for MMPI responses from adolescents (
Ehrenworth & Archer, 1985
). Narrative interpretations for 60 hospitalized adolescents were computer-generated utilizing one of three strategies: (a) adult norms and adult descriptors (adult interpretive approach), (b) adolescent norms and adolescent descriptors (adolescent interpretive approach), and (c) adolescent norms and adult descriptors (mixed interpretive approach). Surprisingly, the adolescent interpretive approach (
Marks et al., 1974
) was rated significantly less accurate than either the adult or mixed approaches.
The only other published study examining the most appropriate norms and descriptors for interpreting adolescent MMPI responding did not include an adolescent norms and adolescent descriptors condition. Instead,
Lachar, Klinge, and Grisell (1976)
compared computer-generated interpretations based on adolescent norms and adult descriptors with interpretations based on adult norms and adult descriptors, concluding that the former were the most accurate. However, this study did not address the relative accuracy of the adult descriptors compared with the adolescent descriptors (
Marks et al., 1974
) in interpreting adolescents' MMPI profiles.
Several reasons have been suggested to explain the differences that can be found between MMPI interpretations based on the adolescent descriptors (
Marks et al., 1974
) and interpretations based on the traditional adult code type descriptors. In fact,
Marks et al. (1974)
carefully described their sample characteristics and suggested that their descriptors might not generalize to adolescents differing from those in their sample. Their sample included only White subjects who were not mentally retarded or deficient but who were receiving some form of outpatient psychotherapy (minimum of 10 sessions) for some adjustment problem.
The age of the
Marks et al. (1974)
data set may contribute to the lower accuracy of their descriptors for contemporary adolescents. Their sample was collected between 1965 and 1970, and there have been important changes both in society and in diagnostic and assessment practices. A particularly dramatic societal change has been the increasing prevalence of alcohol and drug use among youth (
Clayton & Ritter, 1985
). Perhaps this substance use has contributed to changes in psychological symptoms and has led to the lower accuracy for today's youth of the MMPI code type descriptors from the earlier sample.
Some of the methodology for defining code types used by
Marks et al. (1974)
has been questioned by more recent reviewers (e.g.,
Archer, 1987
). For example, Marks et al.'s decision to study all possible code types, rather than just the most frequently occurring ones or those reaching a clinically significant elevation, may have resulted in heterogeneous groupings of subjects into code types. Subjects with invalid profiles (based on the Cannot Say, Lie, Infrequency, and Defensiveness validity scales) were not excluded from the code type groups, and subjects with infrequently occurring code types were reclassified into one of the more frequently occurring ones. Also, a very small number of subjects (
n
= 10
) was used to determine a code type. Although these procedures allowed for over 98% of their sample to be classified into a code type group (
Marks et al., 1974
), the utility of the code type descriptors may have been compromised.
A final possible explanation for the limitations of Marks et al.'s adolescent MMPI descriptors is problems with the adolescent norms on which the code types were based (
Graham, 1987
). Several recent studies, including a preliminary study using the MMPI Restandardization Project's normative data (
Williams, Graham, & Butcher, 1986
), a meta-analysis of adolescent samples from the general population (
Pancoast & Archer, 1988
), and our first study in this series (
Williams & Butcher, 1989
), reached similar conclusions about the limitations (i.e., the adolescent norms appear to underestimate psychopathology in adolescent clinical samples) of the currently used adolescent norms (
Marks et al., 1974
) and the need for more contemporary adolescent norms.
The main purposes of this study were to investigate some of the possible reasons for the problems noted above with the only available adolescent code type descriptors and to test contemporary, alternative code type classification procedures to determine if meaningful descriptors based on adolescents' responses could be found. The first question we addressed was whether our large, contemporary sample, made up of a sizable percentage of subjects hospitalized for substance abuse, would differ significantly in MMPI profiles from Marks et al.'s large, older sample of adolescents in outpatient treatment when the classification procedures developed by
Marks et al. (1974)
were used.
Next, we used more contemporary and consistent code type classification procedures to define code types. More specifically, we first eliminated from our code types any subjects with invalid profiles by using the exclusion criteria for the MMPI validity scales from our first study in this series (
Williams & Butcher, 1989
). MMPI profiles with Cannot Say (
CS
) scale raw scores greater than 10, Lie (
L
) scale adolescent
T
scores greater than 70, Infrequency (
F
) scale adolescent
T
scores greater than 90, and Defensiveness (
K
) scale adolescent
T
scores greater than 70 were eliminated from the analyses. This reduced the possible contamination of MMPI code types with deviant or invalidating response sets. In order to ensure that our two-point code types represented clinically significant scale elevations, we classified subjects into code types only if the two scales achieved an elevation of at least a
T
score of 65, rather than the traditional
T
score elevation of 70, which was identified as problematic by
Williams and Butcher (1989)
and by
Archer, Gordon, Giannetti, and Singles (1988)
. We kept our code types more homogeneous by
not
employing the
Marks et al. (1974)
rule for reclassifying subjects with infrequently occurring code types into more frequently occurring ones. In addition, we doubled the number of subjects required in the analyses for determining descriptors for the code types from the previous minimum of
n
= 10
(
Marks et al., 1974
). Once the code types were identified, we used the multimethod approach from our previous work on MMPI scale elevations to determine descriptors. Finally, we arrived at a judgment of the adequacy of our descriptors by comparing the current results with those of our previous study (
Williams & Butcher, 1989
) to see if our code type descriptors replicated across the multiple measures used here and with previous work from both adult and adolescent samples.
Method
Subjects
Subjects were 844 adolescents (492 boys, 352 girls) who were admitted to one of several inpatient or day treatment facilities or special schools in Minneapolis between November 1985 and December 1987.
1
Almost one quarter of the sample were from minority groups, in contrast to the completely White sample of
Marks etal. (1974)
. We used the same subjects and procedures as in our first study in the series (
Williams & Butcher, 1989
), and demographic and other characteristics of the sample and descriptions of the treatment facilities are provided in that study.
Instruments
The measures used in this study were Form TX of the MMPI, the Devereux Adolescent Behavior Rating Scale (DAB;
Spivack, Haimes, & Spotts, 1967
;
Ben-Porath, Williams, & Uchiyama, 1989
), the Child Behavior Checklist (CBCL;
Achenbach & Edelbrock, 1979
,
1983)
, and a Record Review Form (
Williams & Butcher, 1989
). Interrater reliability of the Record Review Form was determined by having two raters independently rate 14% of the cases. Kappas were computed as described in
Williams and Butcher (1989)
. All but one of the variables showed agreement exceeding the .05 level of significance.
Procedure
The MMPI was incorporated into the standard assessment battery at admission to the treatment facilities. Patients were included in the study after informed consent was obtained from their parents or guardians. In the special schools the school psychologist identified potential subjects who had achieved at least a fifth-grade reading level and wrote their parents offering a special psychological assessment at no cost. DAB ratings were obtained from treatment staff after approximately 10—14 days of observation. Treatment staff were instructed not to look at the patients' MMPI results until after they completed the DAB. Although guaranteeing absolutely blind DAB ratings would have been preferable (i.e., by removing the MMPI from the patients' hospital records), this was not possible in the naturalistic settings. However, we are confident, given the close monitoring and observations of our research assistants, that the treatment staff cooperated with the procedures, understood the importance of independent DAB ratings, and were motivated to produce uncontaminated data. DAB ratings were not collected in the special schools because there were no treatment counselors. CBCL parent ratings were collected in the inpatient settings at the time of admission. CBCL ratings were not collected in the day treatment settings or special schools at the request of their staffs. Record Review Forms were completed by research assistants at discharge from the treatment facilities or during the end of the school year. The same 27 Record Review Form variables used in the first study in the series (
Williams & Butcher, 1989
) were selected as potential MMPI code type descriptors.
Results
Comparison of Adolescent MMPI Profiles Across Samples, Settings, and Time
In order to answer our first question about whether adolescent MMPI profiles differed significantly across samples, settings, and time, we compared our sample of 844 adolescents with the 834 adolescents in the
Marks etal. (1974)
study using the original code type classification rules, with the exception of reclassifying infrequently occurring code types. The 45 code types and frequencies for each code type for the Marks et al. sample were taken directly from the previous work (
Marks et al., 1974, p. 146
). Rather than reclassifying the infrequently occurring code types in the original study, we decided to compare them with ours to determine if the frequencies of these code types had changed. We then classified our sample into Marks et al.'s 45code types using their classification rules (
Marks et al., 1974
, pp. 144—147). As in their previous study, we combined genders to determine frequencies.
Table 1
presents the frequencies for both samples.
A chi-square analysis comparing the two samples across the 45 code types was significant,
χ
2
44,
N
= 1,678
= 115.4,
p
<= .000
. We next used separate
Z
tests to determine which code types accounted for the significant differences. The Bonferroni correction for multiple comparisons (
Grove & Andreasen, 1982
) was used to determine the significance level of
p
<= .001
. The null hypothesis could be rejected in only 3 of these 45 comparisons. As indicated in
Table 1
, our sample had significantly fewer subjects classified as 1-3/3-1 or 7-8/8-7 and more subjects classified as 2-9/9-2, which was Marks et al.'s only code type with a frequency of zero. The most frequently occurring code types in both samples, using these code type classification rules, were the 4-9/9-4 code type, followed by the 2-4/4-2, 3-4/4-3, and 4-6/6-4 code types. Although we found a few statistically significant differences, the code type frequencies in both samples were not very different. These comparisons led to our conclusion that the problems with the Marks et al. adolescent descriptors described in our introduction probably were not due to unique characteristics of their sample and justified exploring other code type classification rules, which we proceeded to do as described below.
Determination of Code Types and Descriptors in the Current Sample
Table 1
also presents the frequencies of code types in our sample when the more contemporary classification rules described in the introduction were used. We combined genders in these analyses because gender- and age-corrected adolescent
T
scores were used and because sample size did not allow for separate analyses by gender. A total of 435 profiles had at least two scales elevated at or above an adolescent
T
score of 65, met the validity criteria described in
Williams and Butcher (1989)
, met the other classification criteria, and were grouped into a code type. Of these subjects, 202 met our frequency criterion of greater than or equal to 20 subjects for further analysis to determine code type descriptors: 59 as 2-4/4-2; 34 as 4-6/6-4; 26 as 4-7/7-4; 21 as 4-8/8-4; and 62 as 4-9/9-4. Although 32% of the total sample (60% of the valid profiles) were classified into code types, only 24% of the total sample (28% of the valid profiles) occurred with sufficient frequency for the descriptor analyses.
Descriptors for these code types were determined using procedures similar to those described by
Williams and Butcher (1989)
for determining scale descriptors, except that an age- and gender-corrected adolescent
T
score greater than or equal to 65 was used to define the code type groups and a combined gender sample was used for the analyses. The subjects in the code type groups were contrasted with the remaining subjects in the sample using either; tests or chi-square analyses to determine statistical significance. The
t
tests were used for the variables assumed to be continuous (DAB and CBCL scales), and chi-square analyses were used for the categorical variables (Record Review Form variables). Of the 725 subjects with valid profiles, there were 527 DABs, 476 CBCLs, and 709 Record Review Forms available for the code type descriptor analyses. This reduced the code type sample sizes to a range of 19—45, with our smallest sample almost twice as large as the smallest sample in the
Marks et al. (1974)
study. The probability level was set at .01 as suggested by
Green (1982)
for MMPI code type descriptor studies. Because a combined gender sample was used, the CBCL narrow-band scales, which are different for the genders, could not be used as potential descriptors. Rather, the broad-band Externalizing and Internalizing Behavior Problem scales were used as potential descriptors. What follows is an initial list of our predictions by code type and our findings. We based the predictions on the code type descriptors found in
Graham (1987)
and on our previous findings on the relationships of these criterion measures with the MMPI scales (
Williams & Butcher, 1989
).
2-4/4-2.
We predicted that the DAB Acting Out Behaviors (AOB), Neurotic/Dependent Behaviors (NDB), and Heterosexual Interests (HI) scales, the CBCL Externalizing and Internalizing scales, and 16 Record Review Form variables describing acting-out behaviors, depression, suicidal behaviors, and so forth, would be related to the 2-4/4-2 code type. The only significant relationship found for the 2-4/4-2 code type was with the Record Review Form variable of depression, which was more characteristic of the 2-4/4-2 subjects than others:
χ
2
1
= 7.47,
p
<= .006
(
n
of comparison group = 649,
n
of code type= 58). None of the DAB or CBCL scales reached significance.
4 6/6-4.
We predicted that the AOB, NDB, and HI scales of the DAB, the CBCL Externalizing scale, and six Record Review Form acting-out variables would be related to the 4-6/6-4 code type. As predicted, the 4-6/6-4 code type group was significantly higher on the DAB scales. For the AOB scale: comparison group,
M
= 75.0,
SD
= 21.8,
n
= 502
; code type,
M
= 87.9
,
SD
= 21.7
,
n
= 25
;
r
525
= -2.87
,
p
<= .002
. For the NDB scale: comparison group,
M
= 22.3
,
SD
= 6.7
,
n
= 502
; code type,
M
= 26.6
,
SD
= 5.7
,
n
= 25
;
t
525
= -3.14
,
p
<= .001
. For the HI scale: comparison group,
M
= 17.7
,
SD
= 7.2
,
n
= 502; code type
,
M
= 2l.5
,
SD
= 6.2
,
n
= 25;
t
(525) = -2.60
,
p
<= .005
. None of the CBCL scales or Record Review Form variables reached significance.
4-7/7-4.
We predicted that the AOB, NDB, and HI scales of the DAB, the CBCL Externalizing and Internalizing scales, and 13 Record Review Form variables including acting-out behaviors, tension/nervousness, and substance abuse would be related to the 4-7/7-4 code type. The 4-7/7-4 code type was related significantly to only one of the predicted DAB scales, the NDB scale: comparison group,
M
= 22.3
,
SD
= 6.6
,
n
= 503; code type
,
M
= 26.3
,
SD
= 8.2
,
n
= 24;
t
(525) = -2.88
,
p
<= .002
. Neither of the CBCL scales reached significance. Two predicted Record Review Form variables reached significance: Running away was more typical of the code type,
χ
2
= 6.46
,
p
<= .01
(
n
of comparison group = 681,
n
of code type = 26), as was impulsivity,
χ
2
1
= 7.84,
p
<= .005
.
4-8/8-4.
Our predictions for this code type included the AOB, PB, NDB, and HI scales of the DAB, the CBCL Externalizing scale, and 18 Record Review Form variables including acting-out behaviors, poor social skills, social withdrawal, and so forth. None of the DAB or CBCL scales reached significance for the 4-8/8-4 code type. The only significant Record Review Form variable was a history of sexual abuse, which was more characteristic of this code type:
χ
2
1
= 6.86,
p
< .009
(n of comparison group = 686,
n
of code type = 21).
4-9/9-4.
We predicted that subjects with the 4-9/9-4 code type in our sample would have higher scores on the AOB and HI scales of the DAB, the CBCL Externalizing scale, and 11 Record Review Form acting-out and impulsive variables. Only the Record Review Form variable of being sexually active was characteristic of this code type:
χ
2
1
= 6.33,
p
<= .01
(
n
of comparison group = 645,
n
of code type = 62).
Discussion
One of the most striking findings of this study is that the relative frequency of code types across samples, settings, and time is quite consistent, as revealed in the comparison of the present sample with the
Marks et al. (1974)
sample. The few differences in code type frequencies that were found could be attributed to setting differences. The 1-3/3-1 and 7-8/8-7 code types are more frequently found among outpatients, as in the
Marks et al. (1974)
sample, and the 2-9/9-2 code type is more common among substance abusers, as in the present sample. It is impressive that even with all of the factors differentiating these two samples, adolescent psychopathology, as reflected in MMPI code type frequencies in large samples, is relatively consistent.
Some evidence of empirical validity was found for the five code type groups studied. The 4-6/6-4 code type had the most support for its empirical validity, and these subjects seemed to demonstrate the most serious psychopathology. They were much more likely to engage in acting-out behaviors, including sexual acting out, and were more easily upset, overly dependent, and clinging. The mixed internalizing/externalizing dimension of the 4-7/7-4 code type was verified by its relationship to the NDB scale of the DAB and its association with running away and impulsivity. Because in our earlier validity study of the DAB we found a strong relationship between the NDB scale and substance abuse (
Williams, Ben-Porath, Uchiyama, Weed, & Archer, in press
), it can be inferred that both 4-6/6-4s and 4-7/ 7-4s are more likely than others to have substance abuse problems as well. Depression was found to be a feature of the 2-4/4-2 code type group, although there were no acting-out descriptors obtained. The 4-8/8-4 group was more likely to have a history of sexual abuse. Most surprisingly, the 4-9/9-4 group was not different from the comparison group on any measures except being sexually active. Although many of these descriptors can be found both in the
Marks et al. (1974)
code type descriptors and the adult code type descriptors provided by
Graham (1987)
, many expected descriptors for each of the five code types studied were notably absent. For example, both the adolescent(
Marks et al., 1974
) and adult (
Graham, 1987
) code type descriptors for 4-9/9-4 include a variety of acting-out variables. Even though our criterion measures included similar acting-out variables that were found to be differentially associated with the MMPI scales (
Williams & Butcher, 1989
), they were not found to be related to the 4-9/9-4 code type in this study.
Some limitations of the present study require comment. Because this study was conducted in community treatment facilities, rather than university-based research settings where controlled research is more acceptable, we were unable to obtain ratings for all subjects (e.g., the schools and day treatment center staff were reluctant to give permission to collect CBCLs, patients were discharged from hospitals before staff knew them well enough to provide DAB ratings). Thus, we did not have all measures for all subjects. Furthermore, we lost some of the richness of the CBCL with a combined gender sample because we could not use the gender-specific CBCL scales. However, we had reason to expect that some code types (i.e., 2-4/4-2 and 4-7/7-4) would be related to both the Externalizing and Internalizing scales, whereas the others would be related only to the Externalizing scale. This was not confirmed.
This study does suggest future directions for research on the MMPI-2 with adolescents. We found that any problems with the adolescent code type descriptors were not reasonably attributed to the characteristics of the
Marks et al. (1974)
sample. In fact, the present results suggest consistency in adolescents' MMPI profiles across samples, settings, and time. The five code types studied in the present study also were among the most frequent code types in the
Marks et al. (1974)
study. However, even when we used different code type classification rules that were consistent with more contemporary adult studies and our adolescent scale descriptor study (
Williams & Butcher, 1989
), we were unable to replicate fully either the adolescent code type descriptors found in the previous study (
Marks et al., 1974
) or the descriptors found in the adult literature (e.g.,
Graham, 1987
).
Obviously, further research on the external correlates of MMPI code types with adolescents is needed to clarify the utility of this approach with adolescents. The recently proposed solution to the norm problems (i.e., using an adolescent
T
score cutoff of 65 instead of 70) demonstrated only limited validity for the code type descriptors obtained when using this cutoff. Perhaps the MMPI Restandardization Project's current plan to develop new norms for adolescents will provide a solution to the norm problem, thus enhancing our ability to find meaningful code type descriptors for adolescents.
Another problem in classifying adolescent MMPI profiles into code types that requires further investigation involves the relative frequency of Scale 4 (Psychopathic Deviate) in this population. Actually,
Greene (1988)
noted that Scale 4 is a prominent elevation in adult clinical samples as well. However, the frequency of Scale 4 elevations in adolescent clinical samples and the fact that Scale 4 was empirically developed using older subjects may be factors contributing to the somewhat limited validity of the code type approach with adolescents. The code type approach is based on a comparison of subjects classified into a code type with the rest of the sample (i.e., the comparison group). However, the comparison group, because of Scale 4's dominance in clinical samples of adolescents, may not differ sufficiently from the various code type groups to allow for the derivation of meaningful descriptors for adolescents. Until the validity of the code type approach for adolescents is demonstrated in future studies, we recommend caution when using a code type approach to MMPI interpretations for adolescents, and we suggest the use of the scale descriptors presented in
Williams and Butcher (1989)
as an alternative.
References
