Behavioral Health Research Centre of the Southwest, 6624 Gulton Court NE, Albuquerque, NM 87109 and
1 Division of Government Research, University of New Mexico, Albuquerque, NM 87131, USA
Received 10 August 1999; in revised form 17 August 2000; accepted 17 September 2000
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Because the AUI was developed based on a multiple-condition theory of alcoholism (Horn et al., 1990), the present study used this multidimensional model by initially identifying distinctive profiles (typologies) of AUI, using the six broad, second-order scales by cluster analysis. These profiles were then evaluated for predictive validity with respect to drunk-driving recidivism and other driving outcomes.
Researchers and clinicians, over many years, have attempted to classify people with alcohol-use disorders into subtypes in order better to understand disease aetiology and treatment matching, as well as for prediction (National Institute on Alcohol Abuse and Alcoholism, 1996). Babor (1996) reviewed typologies from the nineteenth century to the present, reporting that modern typological theorists have incorporated great complexity into their models and applied data-based empirical approaches (Allen, 1996) for deriving new typologies. These recent typologies are exemplified by Morey and Skinner's (1986) hybrid model and Cloninger's (1987) neurobiological learning model. An approach that uses psychometric measures such as AUI has certain advantages and conveniences, because the validity, reliability, norms, and domains of interest for the tests have been pre-determined. The tests are also practical, easy to use, economical, and suitable for screening a large number of clients as a group (National Institute on Alcohol Abuse and Alcoholism, 1996).
Typologies of DWI offenders have been based on various data sources, including psychometric measures for personality, driver records, criminal record data, and clinical experience (Filkins et al., 1973; Fine et al., 1975
; Struckman, 1975
; Homel, 1980
; Sutker et al., 1980
; Scoles et al., 1984
). Donovan and Marlatt (1982) identified five subtypes through hierarchical cluster analysis of driving-attitudinal, personality, and hostility measures from 172 offenders. Donovan et al. (1985) also investigated recidivism for DWI through subsequent 3-year driving records, but found no significant differences among the subtypes. Arstein-Kerslake and Peck (1985) identified two sets of typologies through K-means analysis based on psychometric variables. Their typologies included personality and attitudinal scales and non-psychometric variables, such as driving and criminal records and intake interviews from 7316 offenders. Unlike Donovan et al. (1985), they found significant differences among the clusters in subsequent 4-year accident and conviction risks. Although meaningful typologies were achieved by these investigators, none of these studies does what the current study attempts to do, namely to investigate the AUI domains, which measure different features of alcohol involvement (Horn et al., 1990
).
Furthermore, among the many instruments available and in use for DWI screening, only three the Michigan Alcoholism Screening Test (MAST), the RIA Self Inventory (RIASI), and the Lovelace Comprehensive Screening Instrument (LCSI) have been evaluated for predictive validity. RIASI was developed by the Research Institute on Addictions to evaluate first-time DWI offenders. The instrument showed some predictive validity in reporting differential scores based on prior treatment and DWI recidivism (Nochajski et al., 1996). The LCSI (Lapham et al., 1996
) demonstrated good predictive validity for first-time offenders based on overall weighted scale scores and 3-year follow-up recidivism analysis (unpublished data).
The AUI was used by the Lovelace Comprehensive Screening Program (LCSP), Albuquerque, New Mexico (NM), to screen DWI offenders from April 1989 to March 1991. Offenders' traffic records were available from the NM Traffic Safety Bureau up until the end of March 1997. The availability of AUI data coupled with the traffic records provided a unique opportunity for determining how well the AUI could predict drink-driving recidivism.
Readers are referred to the Methods sections of the preceding article in this issue (Chang et al., 2001), where we described the LCSP, the offender population, and data regarding the AUI scale scores in this population.
The objectives of the present study were to: (1) define typologies of DWI offenders using the AUI; (2) determine the predictive validity of the AUI; (3) link typologies of DWI offenders with specific alcoholic populations. The present study was the first to establish typologies using the AUI for a DWI population and to compare them with alcoholic typologies established by other investigators (Morey et al., 1984; Donat, 1994
) who used the same instrument. These results may prove important, since they could lead to faster identification of problem-drinking drivers, better matching of treatment to DWI offenders, and further reduction of this serious societal problem.
![]() |
METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
NM DWI Citation Tracking System File
The NM DWI Citation Tracking System is part of the driver's licence master file system operated by the Motor Vehicle Division, and it has been used to manage the administrative licence revocation process since July 1984. When a citation for DWI is issued in NM, the police agency reports the citation to the Motor Vehicle Division, which enters it into the tracking system. The system is also used to help manage the scheduling of hearings in which administrative revocations are appealed. The results of court actions are also entered into the system.
NM Driver History File
The NM Driver History File is also part of the driver's licence master file system operated by the Motor Vehicle Division. All convictions for drivers licensed in the state are entered into the system. The file is used to administer the licensing laws that require the suspension or revocation of a driver's licence for certain offences or for an accumulation of points. Most convictions are kept for 3 years for administrative purposes. However, the DWI convictions are kept for 25 years. Complete historical records for 5 years following an offender's first DWI were obtained for the present study.
NM Vehicle Level Accident File
All police agencies in NM use the same form to report traffic crashes involving injuries or at least $500.00 in property damage. All police-reported crashes on public roads and meeting the $500.00 damage threshold are included in the file. Crashes on private property, e.g. drives and car parks, are not included, regardless of severity. A staff member of the Transportation Statistics Bureau, NM State Highway and Transportation Department, enters extensive data on each crash that occurs on a public road. Data have been entered for all crashes, all vehicles, and all people involved since July 1977.
Data quality and comprehensiveness
The LCSP staff separately evaluated the quality and comprehensiveness of the driving records retrieved from the databases identified above. LCSP independently maintained computer files for convicted DWI offenders when they were referred by the Court. A match between the LCSP computer files and the statewide databases is one indicator of data quality. Additionally, certain information was contained in more than one database, e.g. a crash-related DWI would appear in both the NM Citation and Tracking File and the NM Vehicle Level Accident File, if it occurred on a public road and involved damage of more than $500.00. Cross-checking between those databases was another way of ensuring quality. Three separate runs to check quality were performed. One matched the LCSP convicted DWI offender computer files with the NM DWI Citation and Tracking File, the second matched the LCSP convicted DWI offender computer files with the NM Driver History File, and the third cross-matched crash-related DWIs in the NM DWI Citation and Tracking File and the NM Vehicle Level File for traffic crashes.
Statistical analysis
Cluster analysis was performed to identify profiles of alcohol use among DWI offenders. Life-table analysis was used to compare recidivism curves (time to first arrest) among cluster members. The MantelHaenszel (1959) trend test was used to examine the association between number of arrests (0, 1, 2+) and clusters of severity. Logistic regression was used to identify risk factors/predictors for recidivism.
Cluster analysis
The six broad, second-order AUI scales were selected for the cluster analysis in this study. Compared with the first-order, primary scales, the second-order scales have greater measurement depth and higher reliabilities in the DWI sample. The third-order scale was not used, because of its item redundancy with the second-order scales. As in the construction of the AUI, different scales consisted of different numbers of questions and different responses could contribute different scores to the total. To ensure that equal differences relative to the overall ranges of the scales had equal weights, all scales were standardized (Graham, 1990; Everitt, 1993
) before clustering.
The SAS FASTCLUS procedure was first used for a preliminary cluster analysis to produce 40 clusters. The SAS CLUSTER procedure was then performed using Ward's minimum-variance method and squared Euclidean distances to hierarchically cluster the preliminary clusters. The same process was repeated to produce clusters 412. Because data were highly skewed toward zero, the FASTCLUS procedure could avoid a few individuals with high scores who were classified in clusters with very few observations (five or less). Ward's minimum-variance method was chosen, because other investigators have used it (Morey et al., 1984; Rychtarik et al., 1997
) and it has been evaluated to be more powerful in comparison with other approaches (Morey and Blashfield, 1981
; Morey et al., 1984
).
To determine the number of clusters in the sample, a split-sample replication stopping rule was used (Overall and Magee, 1992; Rychtarik et al., 1997
). The entire sample was split randomly into two subsamples. Solutions of clusters 412 were produced separately for the subsamples. The maximum number of near-identical clusters produced by the procedure in the two subsamples was determined to be the final number. The analysis followed the phases recommended by Morey et al. (1984). First, the stopping rule adopted ensured the replicability of the clusters by a
statistic (McIntyre and Blashfield, 1980
; Fleiss, 1981
). Second, the clusters were validated externally by examining their sociodemographic characteristics and driving outcomes. A reasonable presumption was that the basic characteristics of the clusters would be very different if they represented different subpopulations. A cross-validation process was not possible, however, due to lack of other samples.
Life-table analysis
Life-table analysis was used to compare driving outcomes among the clusters. Driving outcomes were subsequent DWIs, traffic convictions, and crashes since the initial DWI referral. Time to the first event was the follow-up time in months. The curves were compared by a log-rank 2 test (Kalbfleisch and Prentice, 1980
).
MantelHaenszel test
The MantelHaenszel 2 test (Mantel and Haenszel, 1959
) was used to test for linear association between alcohol involvement in the order of the clusters and rearrest severity in terms of subsequent number of DWIs (0, 1, 2+) after the initial arrest.
Logistic regression
To better facilitate use of AUI for DWI screening, potential predictors for recidivism were examined by logistic regression. Assuming AUI was the instrument used, and screeners had demographic variables (gender, ethnicity, age, education, and marital status) and the blood-alcohol concentration (BAC) value at arrest, the relationships between whether they were rearrested (0, 1) and these variables were entered into the logistic regression model. For each variable (e.g. ethnicity), a reference group was chosen (e.g. non-Hispanic white). The likelihood of being rearrested was expressed as 1, then the likelihood of other groups (e.g. Hispanic and Native American) relative to the reference group was predicted by the model.
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Profiles of the clusters
Profiles of the clustering results are shown in Fig. 1. The cluster analysis defined six clusters, ordered by the third-level score ALCINVOL, which indicated general alcohol involvement. The upper and lower 95th confidence intervals of the data (mean score ± 2SD) were also plotted to indicate data spread.
|
Cluster 2 (Alcohol-Preoccupation Type) was characterized by high OBSESSED scores, and slightly elevated RECPAWAR scale scores. This cluster represented 14% (n = 236) of the population.
Clients classified in cluster 3 (Enhanced Type) had high scores on the ENHANCED scale. This group represented 22% (n = 367) of the population.
Clients classified in cluster 4 (Enhanced-Disrupt Type) had reported high scores on the ENHANCED scale and high DISRUPT1 and DISRUPT2 scores. This group represented 9% (n = 144) of the population.
Cluster 5 (Anxious-Disrupt Type) was distinguished by high DISRUPT scale scores. The RECPAWAR and ANXCONCN scale scores were also elevated, compared with other cluster groupings. Clients in this group represented 3% (n = 55) of the population.
Clients in cluster 6 (High-Profile Type) were notable for their high scores across all drinking-related problems. All scales were significantly higher than overall mean scores for the entire population. Clients in this group represent a small portion of the DWI clients (1%, n = 17).
Demographic characteristics of the clusters
Demographic characteristics of the clusters are presented in Table 1. There were no gender differences among the clusters. The three most severe groups (Enhanced-Disrupt type, Anxious-Disrupt type, and the High-Profile type) contained a higher proportion of Native Americans. Older offenders were more likely to be of the Alcohol-Preoccupation type (cluster 2) or the Anxious-Disrupt type (cluster 5). Single offenders were more likely to be of the Enhanced type or the Enhanced-Disrupt type (clusters 3 and 4). Offenders of the Alcohol-Preoccupation type (cluster 2) were less educated than those in the other clusters. Offenders of the Enhanced and Enhanced-Disrupt types were more likely to be employed (clusters 3 and 4). The mean ALCINVOL scores were 1.65, 2.41, 5.00, 8.85, 17.65, and 30.12 for the six clusters, respectively.
|
|
|
|
|
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Typologies and treatment implications
Cluster 1 (Low-Profile Type), by far the largest cluster (with 50% of the sample), was composed of clients who reported minimal involvement in alcohol use or abuse patterns. However, their rate of DWI rearrest within 5 years of 22%, though the lowest of all clusters, is still high, which suggests that many of the clients falling into cluster 1 were defensively under-reporting their involvement.
Cluster 2 members of the Alcohol-Preoccupation type were notable for their more compulsive and continuous daily drinking style, and this group was more likely to be older and less educated than members of the other groups. According to Horn et al. (1990), high OBSESSED scores indicate a pressing preoccupation with the use of alcohol, and medium scores point to potential risk for liver disorder and other alcohol-related illnesses. Very high-scoring members of this group may drink throughout the day over many weeks with little or no time between drinking periods, hide bottles, sneak drinks, dwell on thoughts about drinking, and drink to go to sleep at night (Horn et al., 1990). The DWI offenders had lower overall scores than the normative population; therefore, the extent to which they are similar is uncertain. Persons with high scores in this group may require detoxification and in-patient treatment.
Clusters 3 and 4, Enhanced and Enhanced-Disrupt types, were among the most severe groups with respect to the percentage rearrested in the follow-up period. Twenty-nine per cent in cluster 3 and 40% in cluster 4 had at least one subsequent DWI. Members of this cluster reported drinking for sociability and felt that alcohol improved their overall functioning, causing them to be less shy, make friends more easily, and function better mentally. This group of clients had a high commitment to alcohol use. They were more likely to be younger and single, a high-risk group demographically for recidivism. Members of cluster 3 did not have elevated scores on any other scales, indicating that this group valued drinking socially and, though drinking in a risky manner, had not yet experienced many negative consequences from their drinking. This group would benefit from learning harm reduction strategies. Members of cluster 4 were more severe: their DISRUPT scores were high, indicating symptomatic drinking. Their DISRUPT2 scores were generally high, indicating less defensiveness about drinking. Thus, cluster 4 was most likely to disclose symptoms of alcohol abuse while continuing to be committed to drinking and involved in drinking and driving behaviour. A treatment goal of abstinence for this group would be essential, even though the more severe nature of the alcohol problem and commitment to use will make abstinence difficult. This group will benefit from cognitive-behavioural or motivational enhancement treatment, relapse prevention, close monitoring of their drinking and driving behaviour, and immediate implementation of negative consequences for continued drinking and driving. Although antidipsotropic drugs (disulfiram) may help, this group will be at higher risk for relapse and may drink on disulfiram.
Members of the Anxious-Disrupt type (cluster 5) reported many negative consequences from drinking. They acknowledged their drinking problems and readiness for treatment. High scores of ANXCONCN also indicate considerable anguish about drinking, including worry, depression, and guilt. Their ENHANCED score was higher than the OBSESSED score, which indicated that periodic vs continuous drinking was likely. Higher percentages of this group were Native American offenders. Twenty-nine per cent of them had at least one DWI in the 5-year follow-up, which was the second highest among the clusters, indicating that their acknowledgement and anxious concern about drinking were not enough to overcome the deterioration resulting from the use of alcohol. They are likely to be comprised of a high percentage with alcohol dependence, and total abstinence should be their treatment goal. Longer term treatment and drugs to help prevent relapse may be the most productive approach.
The High-Profile type, because of the small numbers involved, had variable and less consistent profiles. However, they tended to score high on all scales, especially the DISRUPT1 and ANXCONCN scales. Although only about one-quarter (24%) had one or more DWIs in the 5-year follow-up, 18% had two or more DWIs. Members of this cluster, representing a more severe expression of generalized alcoholism, may need more intensive or prolonged treatment, with strong emphasis on relapse prevention. However, due to the nature of the data spread skewed data distribution toward zero the algorithm for computer clustering trended to classify individuals with high scores in clusters with very few observations. With the restriction to lump small clusters of size 30 or less, the analysis has resulted in high uncertainty for the High Score group. Since the current study included only first-time DWI offenders, one approach that may further help determine the reliability of this cluster would be to repeat the entire analysis with a population of repeat DWI offenders.
Social/gregarious drinking nature of the DWI sample
Other than the Low-Score type, the cluster which contained 50% of the study population (n = 825), clusters 35 (Enhanced type, Enhanced-Disrupt type, and Anxious-Disrupt type) containing 34% (n = 566) of the population all had elevated ENHANCED scores. This high proportion of offenders having high ENHANCED scale scores indicated the social/gregarious drinking nature of the DWI population. This finding is understandable, because, if the offenders were drinking and driving, it might be for social reasons. Chang et al. (1996) have reported that about one-half of offenders drank at bars/lounges and 29% at private parties prior to their arrests. Furthermore, a high ENHANCED scale score indicated commitment to drinking, which explained their high rates of relapse/recidivism.
Recidivism as predicted by cluster
As the clusters increased in severity, the rates for recidivism for the clusters were not in the same order. Clusters 5 and 6, the most severe groups in alcohol involvement, had the second and third highest rates in DWI recidivism. This may be due to acknowledgement of their drinking problems and readiness for treatment of the cluster 5 members. It may be also due to the small sample size in cluster 6, which may have resulted in high uncertainty in estimating recidivism, or other factors, such as situational and personality/attitude factors, of the cluster members may have contributed to drink-driving as well. Situational factors, such as where and with whom they were drinking, personality/attitude factors, such as thrill and sensation-seeking, and lack of social responsibility, may be different among the clusters. This study has included drinking styles and severity of alcohol involvement in the analysis and demonstrated their association with drink-driving and recidivism. Future studies should also include those other factors and investigate their associations/interactions with DWI and recidivism.
Link with typologies of alcoholic populations
Morey et al. (1984), in their cluster analysis, identified three types of alcoholic groups, using AUI primary scales from a sample of 725 individuals seeking help for alcohol-related problems at the Clinical Institute, Addiction Research Foundation of Toronto, Canada. The clusters were defined as type A (early stage problem drinkers), type B (affiliative, moderate alcohol dependence drinkers), and type C (schizoid, severe alcohol dependence drinkers). Donat (1994) reported five clusters for his study of 217 individuals admitted to a private substance-abuse treatment programme in Central Virginia. In this study, cluster IV was identified as having an advanced stage of alcoholism (Donat, 1994). Results of the cluster analysis for the current study, using the second-order scales, showed some indirect resemblance with type C alcoholics reported by Morey et al. (1984) and cluster IV reported by Donat (1994). Comparison of those profiles showed that, although the scale scores for DISRUPT1 and DISRUPT2 increased in severity in the current study, other investigators also reported elevated scale scores for LCONTROL, ROLEMALA, DELIRIUM, and HANGOVER. DISRUPT1 includes many of the same item questions as LCONTROL, ROLEMALA, DELIRIUM, and HANGOVER, and both DISRUPT1 and DISRUPT2 are also highly correlated with those scale scores (Horn et al., 1990
). Although the scores were much lower, the resemblance was noticeable. If lower scores were primarily due to under-reporting, these results suggest a considerable level of similarity between the DWI sample and alcoholic populations.
Under-reporting
Under-reporting appears to be a problem. First, other studies conducted in this NM DWI population have found evidence of substantial under-reporting of alcohol use, drug use (Lapham et al., 2000), and criminal histories (Chang and Lapham, 1996
). Second, our results revealed differences among cluster groups that imply differences in severity. Even among the low-score cluster with no reported alcohol-related problems, the 5-year recidivism rates were still 22%, only 3% below the average rate (25%) for all first offenders. It is understandable that offenders may be motivated to under-report their alcohol-related problems. Screening was mandated by the court, and clients were informed that persons reporting serious alcohol-related problems or problem-drinking patterns would be ordered by the court to undergo treatment which involved considerable time and expense. It is also understandable that clients would not want to be labelled problem drinkers. Possible alternative explanations for low scores on the AUI followed by a second DWI arrest include poor test-taking motivation, inability to understand the questions, and inability of the instrument to identify problem drivers. These offenders may also constitute a group of early problem drinkers whose alcohol-related problems progressively increased in the follow-up period. Members of this group who are referred to treatment would do best in a non-confrontational treatment model that focuses more on motivational enhancement, rather than treatment per se.
Because offenders are likely to minimize their use of alcohol, screeners should use all available data in recommending treatment, including previous criminal/traffic records, blood-alcohol levels at arrest, and reports by significant others. Information gathered during the interview regarding the circumstances of the arrest, and any family, medical, personal, or legal problems may also indicate a need for treatment. In addition, a well-structured face-to-face interview can also minimize response bias and increase data validity. The AUI does not have a social-desirability or test-taking defensiveness-type scale, so the use of an additional instrument to test for defensiveness may enhance overall predictive validity.
Unmatched rates
The 7% unmatched rate between the LCSP population and the NM DWI Citation and Tracking File was primarily due to falsely reported names, birthdays, and social security numbers, presumably misreported by the offenders. The 15% unmatched rate for the NM Driver History File also resulted from other factors, such as plea bargaining that led to deletion of records, and loss of records by the Division of Motor Vehicles, etc. These match rates, however, were considered satisfactory. The quality of the NM Vehicle Level Accident File evaluated by matching crash records with crash-related DWIs provided in the NM DWI Citation and Tracking File is less satisfactory, with a 68% match rate. The discrepancies were presumably due to crashes not occurring on public roads, e.g. car park accidents and accidents with property damage less than $500.00 which would not have been included in the accident file. Lapham et al. (1998) have reported that about 25% of the DWI offenders were apprehended within half a mile of the drinking location, so some DWI arrests may have occurred on car parks. It is not known how many of these were crash-related.
False negative rates
Results of the logistic regression suggested risk factors for recidivism, assuming DWI screeners only had the AUI results together with the basic demographic variables (gender, ethnicity, age, education, and marital status) and the BAC. The risk factors identified were male gender, young age, less educated, high BAC, and members of clusters 3 or 4. Use of AUI together with those variables provided some predictive validity, as demonstrated in the Results section. It should be noted, however, without any risk factors, that the false negative rate was about 20% for clients. This limitation, however, is not unique to the AUI, and is in fact probably also universal for other screening instruments and programmes, due to the defensiveness of DWI subjects.
Lapham et al. (1997) evaluated four instruments the MacAndrews scale of the MMPI-2 (MAC), four scales of the AUI, the MAST and the Skinner's Trauma Scale (STS) to predict 4-year recidivism of the same DWI population. The above authors reported recidivism rates of 13.038.8% in groups of offenders with various levels of risk based on the predictors. In the present study, the rearrest rates of offenders scoring positive on the various predictors (i.e. the positive predictive power) ranged from 18% to 40%, depending on the predictor. Thus, the predictors we have available to date are far from ideal. Lapham et al. (1997) have discussed the contributing factors. The arrest rates are generally affected by various law enforcement activity over time, unavailability of out-of-state arrest data, and intervention of the screening programme itself.
Summary and recommendations
In summary, the cluster analysis of the AUI secondary scales produced six clinically meaningful groupings. Results further showed statistically significant relationships between the degree of clinical severity and recidivism rates. Trend analyses revealed that members of the two most severe clusters were more likely to recidivate repeatedly. The least severe cluster, having the least alcohol involvement, had the lowest 5-year recidivism rate, and one of the three clusters with high-scale scores, had the highest 5-year recidivism rate. This suggested positive predictive validity of the AUI. A notable limitation of the AUI when used for DWI screening is that low scores are most likely to be related to situational test-taking defensiveness. In-depth analysis of the profiles at the primary level may not be possible for all clients. Screeners should be cognisant of the low-scale score distributions in this population. Cut-off points for determining diagnostic patterns may need to be lowered significantly. In addition, to test interpretation and interviews by experienced counsellors, we recommend examining external indicators of alcohol misuse, such as blood-alcohol levels and interviews with significant others, as well as use of additional instruments to detect test-taking defensiveness.
![]() |
ACKNOWLEDGEMENTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
FOOTNOTES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
REFERENCES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Arstein-Kerslake, G. W. and Peck, R. C. (1985) A Typological Analysis of California DUI Offenders and DUI Recidivism Correlates. State of California DMV Department of Motor Vehicles 1191.
Babor, T. F. (1996) The classification of alcoholics: typology theories from the 19th century to the present. Alcohol Health and Research World 20, 614.
Chang, I. and Lapham, S. C. (1996) Validity of self-reported criminal offences and traffic violations in screening of driving-while-intoxicated offenders. Alcohol and Alcoholism 31, 583590.[Abstract]
Chang, I., Lapham, S. C. and Barton, K. J. (1996) Drinking environment and sociodemographic factors among DWI offenders. Journal of Studies on Alcohol 57, 659669.[ISI][Medline]
Chang, I., Lapham, S. C. and Wanberg, K. W. (2001) Alcohol Use Inventory: screening and assessment of first-time driving-while-impaired offenders. I. Reliability and profiles. Alcohol and Alcoholism 36, 112121.
Cloninger, C. R. (1987) A systematic method for clinical description and classification of personality variants. Archives of General Psychiatry 44, 573588.[Abstract]
Donat, D. C. (1994) Empirical groupings of perceptions of alcohol use among alcohol dependent persons: a cluster analysis of the Alcohol Use Inventory (AUI) Scales. Assessment 1, 103110.[Medline]
Donovan, D. M. and Marlatt, G. A. (1982) Personality subtypes among driving-while-intoxicated offenders: relationship to drinking behavior and driving risk. Journal of Consulting and Clinical Psychology 50, 241249.[ISI][Medline]
Donovan, D. M., Queisser, H. R., Salzberg, P. M. and Umlauf, R. L. (1985) Intoxicated and bad drivers: subgroups within the same population of high-risk men drivers. Journal of Studies on Alcohol 46, 375382.[ISI][Medline]
Everitt, B. (1993) Cluster Analysis. Edward Arnold, London.
Filkins, L., Mortimer, R., Post, D. and Chapman, M. (1973) Field Evaluation of Court Procedures for Identifying Problem Drinkers (Contract No. DOT-HS-031-2-303). National Highway Traffic Safety Administration, Washington DC.
Fine, E. W., Scoles, P. and Mulligan, M. (1975) Under the influence: characteristics and drinking practices of persons arrested the first time for drunk driving, with treatment implications. Public Health Reports 90, 424429.[ISI][Medline]
Fleiss, J. L. (1981) Statistical Methods for Rates and Proportions. John Wiley, New York.
Graham, J. R. (1990) MMPI-2: Assessing Personality and Psychopathology. Oxford University Press, New York.
Homel, R. (1980) Penalties and the Drink Driver: A Study of One Thousand Offenders, Research Report No. 7. Department of the Attorney General, New South Wales.
Horn, J. L., Wanberg, K. W. and Foster, F. M. (1990) Guide to the Alcohol Use Inventory (AUI). National Computer Systems, Inc., Minneapolis, MN.
Kalbfleisch, J. D. and Prentice, R. L. (1980) The Statistical Analysis of Failure Time Data. John Wiley, New York.
Lapham, S. C., Wanberg, K. W., Timken, D. S. and Barton, K. (1996) A User's Guide to the Lovelace Comprehensive Screening Instrument. The Lovelace Institutes, Albuquerque, NM.
Lapham, S. C., Skipper, B. J. and Simpson, G. L. (1997) A prospective study of the utility of standardized instruments in predicting recidivism among first DWI offenders. Journal of Studies on Alcohol 58, 524530.[ISI][Medline]
Lapham, S. C., Skipper, B. J., Chang, I., Barton, K. and Kennedy, R. (1998) Factors related to miles driven between drinking and arrest locations among convicted drunk drivers. Accident Analysis and Prevention 30, 201206.[ISI][Medline]
Lapham, S., Baum, G., Skipper, B. J. and Chang, I. (2000) Attrition in a follow-up study of driving while impaired offenders: who is lost? Alcohol and Alcoholism 35, 464470.
Mantel, N. and Haenszel, W. (1959) Statistical aspects of the analysis of data from retrospective studies of disease. Journal of the National Cancer Institute 22, 719748.[ISI][Medline]
McIntyre, R. M. and Blashfield, R. K. (1980) A nearest-centroid technique for evaluating the minimum-variance clustering procedure. Multivariate Behavioral Research 2, 225238.
Morey, L. C. and Blashfield, R. K. (1981) Empirical classifications of alcoholism: a review. Journal of Studies on Alcohol 42, 925937.[ISI][Medline]
Morey, L. C. and Skinner, H. A. (1986) Empirically derived classifications of alcohol-related problems. In Recent Developments in Alcoholism, Vol. 4, Galanter, M. ed., pp. 145168. Plenum Press, New York.
Morey, L. C., Skinner, H. A. and Blashfield, R. K. (1984) A typology of alcohol abusers: correlates and implications. Journal of Abnormal Psychology 93, 408417.[ISI][Medline]
Nochajski, T. H., Wieczorek, W. F. and Miller, B. A. (1996) Factors Associated with High Risk of Rapid DWI Recidivism for First Time Offenders. National Highway Traffic Safety Administration, Washington DC.
Overall, J. E. and Magee, K. N. (1992) Replication as a rule for determining the number of clusters in hierarchical cluster analysis. Applied Psychological Measurement 16, 119128.[ISI]
Rychtarik, R. B., Koutsky, J. R. and Miller, W. R. (1997) Profiles of the Alcohol Use Inventory. Alcoholism: Clinical and Experimental Research 21, 99A.
Scoles, P., Fine, E. and Steer, R. (1984) Personality characteristics and drinking patterns of high-risk drivers never apprehended for driving while intoxicated. Journal of Studies on Alcohol 45, 411416.[ISI][Medline]
Struckman, D. L. (1975) An Analysis of Drinker Diagnosis and Referral. SD: ASAP Analytic Study No. 5 1975. (Contract No. DOT-HS-045-1-061). National Highway Traffic Safety Administration, Washington DC.
Sutker, P. B., Brantley, P. J. and Allain, A. N. (1980) MMPI response patterns and alcohol consumption in DUI offenders. Journal of Consulting and Clinical Psychology 48, 350355.[ISI][Medline]