Measuring impaired driving behaviors of college students: development and reliability of the impaired driving assessment.

Abstract: The Impaired Driving Assessment (IDA) was developed to collect detailed drinking and driving information over a one-month period. Based on the Timeline Followback assessment, the IDA provides a comprehensive account of impaired driving to identify those at risk for impaired driving, as well as measure the effect of impaired driving prevention programs. Participants consisted of college students who reported drinking and driving at least twice during the previous month on a screening questionnaire. Results demonstrated that impaired driving behaviors reported on the IDA have good test-retest (N=41) and interrater (N=40) reliability, with Pearson correlation coefficients ranging from r = .76 to .96 and r = .73 to .94 respectively.

**********

Drinking and driving is a major public health problem in the United States, causing more than 300,000 injuries each year, nearly 16,000 deaths, and one alcohol-related fatality every 33 minutes (National Highway Traffic and Safety Administration [NHTSA], 1999). According to the Centers for Disease Control (1998), motor vehicle crashes are the leading cause of death for people aged 15-20. However, the alcohol-related fatality rates for 18-, 19-, and 20-year-olds are almost twice that of the population over 21 years of age (NHTSA, 1999). Research has shown that the combination of inexperience with drinking alcohol and limited driving experience lead young adults to be at a high risk for automobile crashes (Hingson et al., 1994).

The widespread use of alcohol among college students is a major health problem with a broad array of associated negative consequences (Wechsler & Kuo, 2000). Impaired driving can be viewed as the most serious of alcohol-related problems because of its potential for extreme negative outcomes such as driving under the influence (DUI) arrests, injuries, and deaths. Recent data from both the Core Institute Study (Presley et al., 1996) and Youth Risk Behavior Surveillance System (YRBSS) studies (CDC, 1997) reveal that more than 25% of college students surveyed reported that they had driven after drinking. Additionally, the Harvard College Alcohol Studies (CAS) have shown that impaired driving among college students has increased in frequency from 1994 to 1997 (Wechsler et al., 1998).

In recent years, numerous assessment methods have been employed to document the prevalence of alcohol consumption among college students. Past research has relied on the modification of existing measures such as the Michigan Alcoholism Screening Test (MAST), Alcohol Use Disorders Identification Test (AUDIT), and CAGE instruments to study college students (Cohen and Vinson, 1995; Heck and Williams, 1995; Samet et al., 1996; Seizer, 1971). However, other studies have developed assessment tools specific to college students such as the Harvard College Alcohol Study, Core, and College Alcohol Problem Scale (CAPS) (O'Hare, 1997; Presley et al., 1996; Wechsler et al., 1994). Because college students are distinguished from other groups in their widespread use of alcohol and the role alcohol plays in the college environment, existing instruments that collect information on alcohol consumption and associated negative consequences lack measurement specific to drinking and driving in this population.

The Timeline Followback method (TLFB) relies on an individual's ability to retrospectively reconstruct his or her behavior (e.g., impaired driving episodes) over a specific time interval. A calendar representing the specified time period is used to aid the individual with reconstruction of events (Sobell et al., 1980). Incorporated into the TLFB method are certain recall-enhancing techniques such as anchoring (marking salient events on the calendar) and identifying patterns of drinking behavior (Sobell et al., 1988). Advantages to using a TLFB approach include easy identification of specific episodes of target behaviors (i.e., impaired driving) and a detailed assessment of the amount and types of alcohol consumed.

Although originally developed as an assessment tool for alcohol consumption, the TLFB has been adapted for a wide range of health-related behaviors. The TLFB assessment has been used to collect data on other substances of abuse such as marijuana, cocaine, heroin, and inhalants as well as alcohol (Ehrman and Robbins, 1994). In a study by Weinhardt et al. (1998), a modified version of the TLFB to assess HIV-related sexual behaviors was found to have high (rho range = .86 to .97) test-retest reliability on multiple variables. The TLFB was similarly modified in a study to determine condom use among gay-bisexual male substance abusers (Crosby et al., 1996). For the purposes of this study, the TLFB calendar method was adapted and expanded in order to collect more detailed information on alcohol use and subsequent impaired driving behaviors among college students.

The TLFB assessment method has been shown to be a psychometrically sound method for obtaining retrospective estimates of alcohol consumption (Sobell and Sobell, 1992), but has never been used to assess drinking and driving behavior. Past reliability studies of the TLFB assessment with problem drinkers (Cohen and Vinson, 1995; Maisto et al., 1979; Sobell et al., 1996), normal drinkers (Sobell et al., 1988), and college students (Cohen and Vinson, 1995; Connors et al., 1985; Sobell et al., 1986; Weinhardt et al., 1998) have yielded high estimates of reliability (See Table 1).

Based on a modified version of the TLFB assessment method (Sobell et al., 1980), the Impaired Driving Assessment (IDA) attempts to enhance recall and retrospectively reconstruct impaired driving behaviors during a specific time interval. The IDA was developed to collect detailed, meaningful drinking and driving information that could further the study of alcohol use and impaired driving. If shown to collect reliable information on impaired driving behaviors, the IDA can be used to identify persons at high risk for impaired driving, as well as measure the effect of impaired driving prevention efforts. Successful implementation of the TLFB approach to assess several different health-related behaviors suggests that the IDA has the potential to collect more reliable and informative drinking and driving information from college students in contrast to more traditional and less specific alcohol assessment instruments.

The purpose of this study was to determine the reliability of impaired driving behaviors reported on the IDA from a college student sample. The reliability of these measures was evaluated in two separate studies. The first study attempted to establish the test-retest reliability (N = 41) of the IDA, and the second study measured the interrater reliability (N = 40).

METHODS

PARTICIPANTS

Undergraduate college students were initially screened and ultimately selected from a Department of Psychology subject pool from a large, public university for participation in the study. Approval from the Institutional Review Board (IRB) for protection of human subjects was obtained prior to data collection. At the beginning of each academic quarter, students participated in a screening process for several studies offering research participation credits. Students received credit for completing screening instrument administered during the first 2 weeks of each of five consecutive academic quarters, from the Winter Term in 1999 through the Winter Term of 2000. The investigators established an eligibility criterion for students who admitted to 2 or more drinking and driving episodes during the past month.

MEASURES

The screening questionnaire is a brief paper and pencil, one-page assessment instrument that collects self-reported information on alcohol use, drinking and driving behaviors, and negative consequences associated with drinking and driving (Schumacher et al., 2002). The IDA is a semi-structured assessment designed to assess the extent (e.g., number of impaired driving episodes) and nature of impaired driving (e.g., ounces of alcohol consumed and time consumed before driving) within the last 30 days. Administration of the IDA lasts from 10 minutes to an hour, depending on the participant's frequency of drinking and driving. A modified version of the TLFB method, the IDA uses a calendar method to collect detailed information about impaired driving and alcohol use behaviors. The TLFB procedure has already been established as a reliable measurement of alcohol consumption when administered through face-to-face interviews, over the telephone, or on the computer (see Table 1).

The first step in the administration of the IDA acclimated the participant to a calendar that spans the past 5 weeks. The participant was then asked to identify any anchor points (memorable events that may enhance recall; e.g., exams, holidays, etc ...) during this time frame. Once any anchor points were identified, the participant identified the days on which any alcohol was used and whether or not he or she drove after drinking. Finally, the calendar was used to identify those days during the past 5 weeks on which the participant was a passenger in a car of a driver who had been drinking. For each day on which the participant reported being a passenger of a driver who had been drinking, the student was asked how many times he or she was driven during that particular day (24-hour period).

The next step in the administration of the IDA involved the collection of more detailed information about each episode of drinking and driving reported by the participant. This information included the participant's body weight, the time at which he or she began drinking, drinking location, types and amount of alcohol consumed, times during which the participant drove after drinking, and any repeated patterns of drinking and driving within the same day. Because participants may drive after drinking on more than one occasion within the same day, a cumulative time line was created for each day on which the participant drove after drinking in order to accurately calculate the estimated BAC at each drinking and driving episode. Primary variables of interest collected from the IDA included days drinking, days as a passenger of a driver who had been drinking (DO for "driving other"), episodes of DO, days driving after drinking (DS for "driving self"), episodes of DS, total number of drinks consumed for all DS, average number of drinks per DS, and average blood alcohol content (BAC) per DS.

For the purposes of this study, one standard drink referred to 12 ounces of beer, or a 5 ounce glass of wine, or 1 1/2 ounces of liquor (Bailey, 1993). These alcoholic beverages were standardized through a formula multiplying the volume of the drink (measured in ounces) by the percentage of alcohol in each drink, which yields the amount of alcohol in each drink. Because, on average, most beers have 5% alcohol by volume, wines have approximately 12%, and 80 proof liquor has 40% alcohol, multiplying these percentages by the amounts listed above yields approximately 0.60 ounces of alcohol for each of these beverages (Bailey, 1993).

The participant's BAC for each DS episode was estimated by computer software designed by the National Highway Traffic Safety Administration called the BAC Estimator. This software is based on Widmark's (1932) original formula and was developed as part of a report to Congress on alcohol limits for drivers (U. S. Department of Transportation, 1992). The BAC Estimator provides an estimate of BAC based on an individual's gender, weight, number of standard drinks consumed, and the time span during which the alcohol was consumed. Because the types and quantity of alcohol consumed by students vary greatly, the drinks are standardized across types and amount.

PROCEDURE

The first portion of this study examined the test-retest reliability or the consistency of the IDA when administered under the same conditions to the same participant across repeated trials. For the test-retest reliability analysis, 41 participants completed an initial and second IDA session, with both sessions administered by the same interviewer. During both their test and retest, participants were asked to recall their drinking behavior during the past 5 weeks (35 days). Because follow-up assessments were given at least 2 and no more than 7 days after the initial assessment, information from the same 4-week (28-day) period of "real time" for each participant was used for the analyses.

The second portion of this study examined the interrater reliability of the IDA, which refers to the degree of consistency in results when the instrument is administered to the same participant by multiple interviewers. The interrater reliability analysis was conducted in a fashion similar to that of the test-retest except that the IDA was administered to each participant by two separate interviewers. A total of 40 participants completed an initial and second IDA session for the inter-rater analysis. The IDA administration procedure was identical for this analysis and for the test-retest study in all respects except for the multiple interviewers.

ANALYSIS

Reliability is often estimated based on variance, which is a measure of the differences among scores within a sample. Because perfect reliability is rare, researchers must define acceptable levels of reliability before ming an instrument for clinical purposes. Generally, below .50 represents poor reliability, .50 to .75 shows moderate reliability, and greater than .75 indicates good reliability (Gross-Portney and Watkins, 1993). Using the Pearson product-moment correlation coefficient, as a reliability estimate allows for comparison to past reliability studies of the TLFB approach that have also used this test statistic (see Table 1). In order to overcome any limitations of correlation as an estimate of reliability, the Intraclass Correlation Coefficient (ICC) was reported as an alternate measure of correlation because it is unbiased and more sensitive to changes in test-retest means than Pearson correlation coefficients (Maisto et al., 1990). The ICC represents the variance of interest divided by the sum of the variance of interest plus error (Shrout and Fleiss, 1979). The ICC reliability coefficient is calculated through analysis of variance methods, reflecting agreement among ratings (Gross-Portney and Watkins, 1993).

In order to establish the reliability of the IDA, each primary variable mentioned previously was correlated for students who completed both test and retest IDA assessments. When analyzing the DS and DO episodes, only those which participants reported as having occurred on the same real-time day across both sessions were included in the analysis. In order to determine the level of association between two variables, correlation coefficients were computed with SPSS software (SPSS, 1999). For the reliability studies, both Pearson product-moment correlations and ICCs were generated to compare each of the primary variables from the test and retest IDA assessments.

For both reliability studies, it was possible that those participants who reported pairs of zero data points could artificially inflate the correlation coefficients. For example, if a student reported no days of drinking and driving on either the test and retest IDA assessment, it could lead to a higher degree of reliability for the sample. Therefore, the extent to which students without zero-pairs of data were unreliable would not be as apparent when including the zero-pair data points. In order to account for this potential confounding factor, adjusted correlation coefficients were also calculated by excluding those zero-pair participants from the analyses.

Because perfect reliability is rare and not expected for this study, it was anticipated that there would be a discrepancy in the number of drinking and driving episodes between the first and follow-up IDA assessments. To account for any unmatched episodes of drinking and driving between tests 1 and 2, only the averages of the estimated BAC for all episodes of drinking and driving were used for this analysis.

RESULTS

The total number of students screened was 1,789, of which 61.3% were female, 55.8% Caucasian, and 37.8% African American, with an average age of 21.1 (SD = 6.4) years. Of those students that were screened, 247 (13.8%) reported driving after drinking at least twice during the last 30 days and were thus eligible to participate in the IDA reliability studies. Those students who met the eligibility criteria were 39.7% female, 71.8% Caucasian, with an average age of 21.1 (SD = 3.68) years. There were a significantly greater proportion of males and Caucasians among those students who met the eligibility criteria than among those who did not meet the criteria. A total of 97 students who met the eligibility criteria participated in an Initial IDA session and 81 (83.5%) returned for their second IDA assessment. The test-retest reliability analysis consisted of 41 students, and the inter-rater reliability analysis was conducted on 40 students. The mean interval between Tests 1 and 2 for both reliability studies was 5.10 days (SD = 1.87), ranging from 2 to 7 days.

TEST-RETEST RELIABILITY

Table 2 presents descriptive statistics and test-retest reliability correlation coefficients for impaired driving and alcohol use variables obtained from the IDA. Both unadjusted and adjusted reliability coefficients are presented in Table 2. As shown in Table 2, both Pearson's and ICC test-retest reliability coefficients demonstrated good reliability (i.e., r > .70), regardless of whether zero pairs were excluded from the analysis. Unadjusted Pearson correlation coefficients and ICCs ranged from .76 to .96. The variables drink days, total drinks, DS days, and DS episodes all had reliability coefficients above the .90 level.

INTERRATER RELIABILITY

Consistent with the test-retest analysis, both adjusted and unadjusted reliability coefficients were calculated in order to determine interrater reliability. Although the results shown in Table 3 show that both Pearson's and ICC interrater reliability coefficients were good for most variables, they were not as strong as those found in the test-retest analyses, as would be anticipated. Unadjusted Pearson correlation coefficients from the interrater analysis ranged from .73 to .94. However, the ICC for the variable of mean BAC was .62, indicating a moderate degree of reliability. When adjusting for zero-pairs of data, the ICC for this variable was even lower, at .54, yet still in the moderate reliability range. Only drink days and DO days yielded reliability coefficients greater than .90.

DISCUSSION

Results demonstrated that reports of recent impaired driving behavior by college students obtained using the IDA have good test-retest reliability. Findings are consistent with those of past studies examining the test-retest reliability of the TLFB method (see Table 1). This research represents the first application of the TLFB to drinking and driving behaviors. Consistent results of good reliability for all variables were also found when adjusting for those students who reported that they had not engaged in certain impaired driving behaviors.

Results from this research also showed that IDA reports of recent impaired driving behavior by college students have good interrater reliability. Correlation coefficients for most variables under study were in the good range of reliability. However, the adjusted Pearson's correlation coefficient and both adjusted and unadjusted ICCs for the mean BAC levels were below the good reliability cut-off of .70. All correlation coefficients did yield at least a moderate level of reliability for this variable.

As can be seen from Tables 2 and 3, the amount of impaired driving behaviors reported from the test-retest reliability sample were higher than the amount from the interrater reliability sample. The test-retest sample yielded a greater mean number of DS (8.3 vs. 3.7) and DO (6.3 vs. 3.3) episodes as compared to the interrater reliability sample. One possible explanation for this difference is that there were a greater proportion of males in the test-retest (57%) sample than in the interrater (40%) sample. Male gender has been seen in several studies to be associated with higher rates of driving after drinking alcohol (Wechsler et al., 1994). Other studies have shown that although college males were more likely to drive after drinking (44%) than females (30%), women were just as likely to be a passenger of a driver who had been drinking (Wechsler et al., 1998).

Due to the fact that the gender breakdown of the test-retest participants (57% male) more closely resembled those students who met eligibility criteria (60% male) than those participants from the interrater analysis (40% male), drinking and driving risk variables among the test-retest sample are believed to be most representative of a high risk student population (i.e., those who admitted to two or more drinking and driving episodes in the past month). Given this assumption, an examination of risk factor prevalence among this population revealed, on average, over 8 DS episodes and 6 DO episodes during the prior month. The combination of these two risk behaviors indicates that the college students in this sample were engaging in impaired driving behaviors approximately 14 times in the past month. Given the widespread use of alcohol among U. S. college students and associated negative consequences, the IDA represents a practical instrument in measuring the magnitude of impaired driving as a major public health concern for this population.

LIMITATIONS

There were some limitations to this research that should be addressed. First, impaired driving behavior data were all based on self-report and should therefore be viewed with caution. In this research, self-report could have biased the data by students' intentional or unintentional misrepresentation of their drinking and driving behaviors. However, numerous studies have established the validity of self-reported alcohol use (Embree and Whitehead, 1993; Midanik, 1988). One factor that may contribute to the underreporting of self-reported drinking and driving is social desirability. Because impaired driving is illegal and the IDA was administered as a face-to-face interview, students could have reported fewer drinking and driving behaviors in an effort to appear less deviant. In an attempt to control for this potential bias, participants were assured of confidentiality of their responses. However, the IDA, which was administered as a face-to-face assessment, yielded higher mean number of overall incidences of drinking and driving (m = 5.52, SD = 5.84) than the screening instrument (m = 4.46, SD = 4.82).

It is important for testing intervals to be far enough apart to avoid any memory effects (Gross-Portney and Watkins, 1993). The relatively short interval between the test and retest (m = 5.10 days, SD = 1.87 days) for the reliability studies could contribute to a lasting memory effect, artificially increasing the correlation coefficients of the IDA variables. Typically, TLFB reliability studies have used a longer interval between test and retest (i.e., 90 days) than those used for this study (Sobell et al., 1986, 1988, 1996). However, in a study by Weinhardt et al., (1998) on the TLFB method, time intervals with fewer than 10 days between tests were used and found to be an acceptable test-retest time interval. The shorter time interval was used for this study mainly because of the nature of the detailed information on recent drinking and driving behaviors sought by the IDA.

Despite these limitations, the current study shows that the IDA, when administered to college students, yields good reliability estimates comparable to those of other TLFB reliability studies. Taken as a conservative interpretation of these findings, the IDA is a reliable method for detecting impaired driving behaviors from college students in an urban campus setting.

RECOMMENDATIONS FOR FUTURE RESEARCH

The next logical direction for this research is to establish the validity of the IDA. Researchers attempting to determine the validity of self-reported health behaviors such as drug or alcohol use face many obstacles. For the validation of alcohol use, researchers have compared self-reports to in-field breath tests (Sobell and Sobell, 1975), reports of collaterals (Tucker et al., 1991), diaries kept by participants (Poikolainen and Karkkainen, 1983), and official records (Sobell and Sobell, 1975).

Many of these methods often employed to determine the validity of alcohol and drug use data would also be appropriate for the assessment of impaired driving information. More specifically, official records such as DUI arrests, automobile crashes, or hospitalizations would be an indicator of the severity of the problem. However, these are low base-rate consequences associated with impaired driving and most likely are much lower than actual rates of impaired driving among college students. This is evidenced by the fact that among the 247 students who met the eligibility criterion, only 18, or 7.2%, reported having ever been arrested for DUI, and only 13, or 5.2%, reported having ever attended treatment as a consequence of their drinking and driving. Perhaps the most feasible approach to the validation of the IDA would be through the use of daily diaries in which the participants chronicle their impaired driving over a specified time period. Results from this self-monitoring of impaired driving would then be compared with information obtained from the IDA.

In order to maximize this instruments' potential, the IDA should be delivered via the computer for widespread dissemination. Past studies have found the TLFB method to have high reliability when administered over the telephone or self-administered with a computer (Sobell et al., 1996). There are several advantages to delivering assessments such as the IDA over the computer rather than through face-to-face interviews. Major benefits of computerized assessments identified by therapists include increased credibility of the assessment; client enjoyment of participating in computer assessments more than paper-and-pencil measures, and assessments being quicker and easier to administer and score (Martin, 1995). Other advantages to using a computerized assessment over face-to-face interviews are that they have standardized protocols, they can deliver immediate feedback, and provide a non judgmental method of collecting confidential information (Sobell et al., 1996).

IMPLICATIONS FOR HEALTH EDUCATION

By 2005, the U.S. Department of Transportation wants to reduce the number of deaths in the United States caused by alcohol-related crashes to 11,000 (NHTSA, 1997) per year. In order to reach this goal, it is necessary to develop innovative programs to decrease the incidence of drinking and driving. Although past efforts to reduce impaired driving among college students have included environmental as well as individualized interventions, there is a lack of theory-based health education programs for preventing impaired driving in this population. Despite large-scale media and public awareness campaigns targeted toward college students, drinking and driving is still a major public health concern. This clearly indicates a need for effective health education to promote change in a population in which alcohol use and impaired driving are so pervasive.

The IDA was developed to enhance current assessment methods by generating more detailed information about impaired driving to develop comprehensive health education programs. The detailed nature of information obtained from the IDA such as drinking location and approximate BAC prior to impaired driving could assist health educators to formulate objectives for impaired driving prevention programs. Past approaches to the prevention of alcohol use have provided information about the general prevalence of physical, neurological, or family problems related to its use. When administered via the computer, the IDA could provide immediate feedback tailored specifically to an individual's current impaired driving behaviors. With ever-increasing internet use among college students, a web-based, educational program delivering personalized feedback based on the IDA has potential for health educators to reduce impaired driving in this high-risk population. Finally, the IDA could be used to develop more rigorous evaluation plans to effectively measure the achievement of program objectives.

HEALTH EDUCATION RESPONSIBILITY AND COMPETENCY ADDRESSED

Responsibility IV: Evaluating Effectiveness of Health Education Programs

Competency A: Develop plans to assess achievement of program objectives.

Subcompetency 5: Identify existing sources of health related databases.

Table 1. Reliability Studies of the Timeline Followback Method of
Assessment Across Various Subject Populations

    Study          Year   N          Population

Cohen et al.       1995   75   College problem drinkers (c)
Connors et al.     1985   48   College students (c)
Maisto et al.      1979   12   Outpatient alcoholics (c)
Maisto et al.      1979   12   Inpatient alcoholics (c)
Sobell et al.      1996   40   Outpatient problem drinkers (c)
Sobell et al.      1988   62   Normal Drinkers (c)
Sobell et al.      1986   80   College students (c)
Sobell et al.      1996   63   Outpatient problem drinkers (c)
Vuchinich et al.   1985   26   Inpatient alcoholics (c)
Weinhardt et al.   1998   58   College students (d)

                     r (a)      ICC (b)

Cohen et al.                   .38, .93
Connors et al.      .87, .97
Maisto et al.       .79, .92
Maisto et al.      -.13, .94
Sobell et al.       .77, .90
Sobell et al.       .70, .96
Sobell et al.       .76, .96
Sobell et al.       .78, .92
Vuchinich et al.    .85, .93
Weinhardt et al.    .84, .97

(a) Pearson product moment correlations were conducted for these
analyses
(b) Intraclass correlationcoefflcients
(c) Alcohol use variables
(d) HIV-related sexual behaviors

Table 2. Test-Retest Correlation Coefficients and Summary Statistics
for Impaired Driving Variables for All Participants (n = 41) from the
IDA

                   Test mean           Retest mean
Variable         ([+ or -] SD)        ([+ or -] SD)      r (a)

DS days         6.1 [+ or -] 5.3     6.0 [+ or -] 5.0     .96
DS episodes     8.3 [+ or -] 7.6     7.9 [+ or -] 6.8     .93
Mean BAC        0.1 [+ or -] 0.06    0.1 [+ or -] 0.07    .82
Total drinks   57.1 [+ or -] 61.3   53.7 [+ or -] 58.0    .93
Mean drinks     6.1 [+ or -] 2.8     6.1 [+ or -] 3.1     .83
DO days         3.2 [+ or -] 2.5     3.2 [+ or -] 2.9     .80
DO episodes     6.3 [+ or -] 6.6     5.7 [+ or -] 6.4     .76
Drink days      9.6 [+ or -] 6.0     9.2 [+ or -] 5.6     .96

Variable       Adjusted r (a) (n)   ICC   Adjusted ICC (n)

DS days             .95 (40)        .96       .95 (40)
DS episodes         .93 (40)        .92       .92 (40)
Mean BAC            .81 (40)        .81       .81 (40)
Total drinks        .93 (40)        .93       .93 (40)
Mean drinks         .82 (40)        .83       .81 (40)
DO days             .77 (37)        .80       .92 (37)
DO episodes         .74 (37)        .76       .72 (37)
Drink days             NA           .96          NA

Note. NA = not applicable, no zero-pairs in the data set,
ICC = intraclass correlation coefficient

(a) For all Pearson correlation coefficients, r, p < 0.01.

Table 3. Interrater Correlation Coefficients and Summary Statistics for
Impaired Driving Variables for All Participants (n = 40) from the IDA

                   Test mean           Retest mean
Variable         ([+ or -] SD)        ([+ or -] SD)      r (a)

DS days         3.1 [+ or -] 3.0     3.2 [+ or -] 2.7     .87
DS episodes     3.7 [+ or -] 3.5     3.7 [+ or -] 3.1     .84
Mean BAC       0.05 [+ or -] 0.03   0.06 [+ or -] 0.06    .73
Total drinks   18.3 [+ or -] 21.2   18.9 [+ or -] 23.4    .82
Mean drinks     3.7 [+ or -] 2.6     3.9 [+ or -] 3.0     .85
DO days         2.3 [+ or -] 2.4     2.4 [+ or -] 2.5     .94
DO episodes     3.3 [+ or -] 4.0     3.1 [+ or -] 3.4     .78
Drink days      6.2 [+ or -] 4.2     6.5 [+ or -] 3.8     .94

Variable       Adjusted r (a) (n)   ICC   Adjusted ICC (n)

DS days             .85 (35)        .87       .85 (35)
DS episodes         .81 (35)        .84       .80 (35)
Mean BAC            .67 (35)        .62       .54 (35)
Total drinks        .79 (35)        .82       .80 (35)
Mean drinks         .80 (35)        .85       .80 (35)
DO days             .92 (31)        .94       .92 (31)
DO episodes         .72 (31)        .78       .72 (31)
Drink days           NA (b)         .93        NA (b)

Note. NA = not applicable, no zero-pairs in the data set,
ICC = intraclass correlation coefficient

(a) For all Pearson correlation coefficients, r, p < 0.01.

REFERENCES

Bailey, W. J. (1993). Drug Use in American Society (3rd ed.). Minneapolis: Burgess Publishing Group.

Cohen, B. B., & Vinson, D. C. (1995). Retrospective self-report of alcohol consumption: Test-retest reliability by telephone. Alcoholism, Clinical & Experimental Research, 19, 1156-1161.

CDC (1997). Youth Risk Behavior Surveillance: National College Health Risk Behavior Survey. Morbidity Mortality Weekly Reports Centers for Disease Control Surveillance Summary, 46, 1-56.

CDC (1998). National Center for Health Statistics (NCHS)--Vital Statistics Mortality Data--1998, Multiple Cause of Death.

Connors, G. J., Watson, D. W., & Maisto, S. A. (1985). Influence of subject and interviewer characteristics on the reliability of young adults' self-reports of drinking. Journal of Psychopathology and Behavioral Assessment, 7, 365-374.

Crosby, G. M., Stall, R. D., Paul, J. P., Barrett, D. C., & Midanik, L. T. (1996). Condom use among gay/ bisexual male substance abusers using the timeline follow-back method. Addictive Behaviors, 21, 249-257.

Drinking and Driving: Facts. (1997). Washington, D.C.: U.S. Department of Transportation and Highway Safety.

Embree, B. G. & Whitehead, P. C. (1993). Validity and reliability of self-reported drinking behavior: dealing with the problem of response bias. Journal of Studies on Alcohol, 54, 334-344.

Gross-Portney; L. & Watkins, M. P. (1993). Foundations of Clinical Research: Applications to Practice. Appleton and Lange, Norwalk, CT.

Heck, R. J. & Williams, M. D. (1995). Using the CAGE to screen for drinking-related problems in college students. Journal of Studies on Alcohol, 5, 282-286.

Hingson, R., Heeren, T., & Winter, M. (1994). Lower legal blood alcohol limits for young drivers. Public Health Reports, 109, 738-744.

Maisto, S. A., Sobell, M. B., Cooper, A. M., & Sobell, L. C. (1979). Test-retest reliability of retrospective self-reports in three populations of alcohol abusers. Journal of Behavioral Assessment, 1, 315-326.

Maisto, S. A., McKay, J. R., & Connors, G. J. (1990). Self-report issues in substance abuse: state of the art and future directions. Behavioral Assessment, 12, 117-134.

Martin, G. (1995). OPTIONS: Target, systems update. Toronto, Addiction Research Foundation.

Midanik, L. (1988). Validity of self-reported alcohol use: a literature review and assessment. British Journal of Addictions, 83, 1019-1030.

National Highway Traffic Safety Administration (NHTSA). (1997). Partners in Progress: an Impaired Driving Guide for Action. Washington, D.C.: NHTSA, 1997 September. DOT HS 808 365, 1997.

National Highway Traffic Safety Administration (NHTSA). (1999).You drink and drive. You lose driving home the facts--about impaired driving.

O'Hare, T. (1997). Measuring problem drinking in first time offenders. Development and validation of the College Alcohol Problem Scale (CAPS). Journal of Substance Abuse Treatment, 14, 383-387.

Poikolainen, K. & Karkkainen, E (1983). Diary gives more accurate information about alcohol consumption than questionnaire. Drugand Alcohol Dependence, 11, 209-216.

Presley, C. A., Meilman, P. W., & Cashin, J. R. (1996). Alcohol and drugs on American college campuses: [he, consequences, and perceptions of the campus environment, Volume IV: 1992-94. Carbondale, IL: Southern Illinois University Press.

Samet, J. H., Rollnick, S., & Barnes, H. (1996). Beyond CAGE. A brief clinical approach after detection of substance abuse. Archives of Internal Medicine, 156, 2287-2293.

Schumacher, J.E., Usdan, S.L., McNamara, C.M., & Bellis, J. M. (2002). Screening for impaired driving risk among college students. College Student Journal, 36, (2), 180-187.

Seizer M. L. (1971). The Michigan alcoholism screening test: The quest for a new diagnostic instrument. American Journal of Psychiatry, 127, 1653-1658.

Shrout, P. E. & Fleiss, J. L. (1979). Intraclass correlations: uses in assessing rater reliability. Psychological Bulletin, 86, 420-428.

Sobell, L. C., & Sobell, M. B. (1975). Outpatient alcoholics give valid self-reports. Journal of Nervous Mental Disorders, 161, 32-42.

Sobell, L. C., Brown, J., Leo, G. I., & Sobell, M. B. (1996). The reliability of the Alcohol Timeline Followback when administered by telephone and by computer. Drug and Alcohol Dependence, 42, 49-54.

Sobell, M. B., Maisto, S. A., Sobell, L. C., Cooper, A. M., Cooper, T., & Sanders, B. (1980). Developing a prototype for evaluating alcohol treatment effectiveness. In L. C. Sobell, M. B. Sobell, and E. Ward (Eds.), Evaluating alcohol and drug abuse treatment effectiveness: Recent advances (pp. 129-150). New York: Pergamon.

Sobell, L. C. & Sobell, M. B. (1992). Timeline Followback: A technique for assessing self-reported alcohol consumption. In R. Z. Litten and J. Allen (Eds.) Measuring alcohol consumption: Psychosocial and biological methods (pp. 41-72). Humana Press: New Jersey.

Sobell, M. B., Sobell, L. C., & Klajner, E (1986). The reliability of a timeline method for assessing normal drinker college students' recent drinking history: utility for alcohol research. Addictive Behaviors, 11, 149-161.

Sobell, L. C., Sobell, M. B., Leo, G. I., & Cancilla, A. (1988). Reliability of a timeline method: assessing normal drinkers' reports of recent drinking and a comparative evaluation across several populations. British Journal of Addictions, 83, 393-402.

SPSS Base 9.0. (1999). Applications Guide and Computer Software. Chicago: SPSS, Inc.

Tucker, J. A., Vuchinich, R. E., Harris, C. V., Gavornik, M. G., & Rudd, E. J. (1991). Agreement between subject and collateral verbal reports of alcohol consumption in older adults. Journal of Studies on Alcohol, 52, 148-155.

U.S. Department of Transportation. (1992). Driving under the influence: A report to Congress on alcohol limits. Wechsler, H., Davenport, A., Dowdall, G., Moeykens, B., & Castillo, S. (1994). Health and behavioral consequences of binge drinking in college. JAMA, 272, 1672-1677.

Wechsler, H., Dowdall, G. W., Maenner, G., Gledhill-Hoyt J., & Lee H. (1998). Changes in binge drinking and related problems among American college students between 1993 and 1997: Results of the Harvard School of Public Health College Alcohol Study. Journal of American College Health, 47, 57-68.

Wechsler, H., & Kuo, M. (2000). College binge drinking in the 1990s: A continuing problem. Journal of American College Health, 48 (5), 199-210.

Weinhardt, L. S., Carey, M. P., Maisto, S. A., Carey, K. B., Cohen, M. M., & Wickramasinghe, S. M. (1998). Reliability of the timeline follow-back sexual behavior interview. Annals of Behavioral Medicine, 20, 25-30.

Widmark, E. M. E [Principles and applications of mediolegal alcohol determination] [German]. Berlin: Urban and Schwartzenberg, 1932.

Stuart L. Usdan, Ph.D., CHES is an Assistant Professor in the Department of Health Promotion, Education and Behavior in the School of Public Health at The University of South Carolina. Joseph E. Schumacher, Ph.D. is an Associate Professor, and Cecelia McNamara, Ph.D. is an Assistant Professor in the School of Medicine and Jeffrey M. Bellis, Ph.D. was an Assistant Professor in the School of Public Health, all at University of Alabama at Birmingham. Address all correspondence to Stuart L. Usdan, Ph.D., CHES, Assistant Professor, Department of Health Promotion, Education and Behavior, The University of South Carolina, School of Public Health, 800 Sumter St., #216, Columbia, SC, 29208; PHONE:[phone]; FAX: [phone]; E-MAIL: [e-mail].