The National Longitudinal Alcohol Epidemiologic Survey (NLAES) was a national probability survey sponsored by the National Institute on Alcohol Abuse and Alcoholism. In 1992, the U.S. Bureau of the Census administered face to face interviews with 42,862 respondents age 18 and older, mean age 44, residing in the non-institutionalized population of the contiguous states including the District of Columbia. 

Sample Design:

The multi-stage sampling approach used for NLAES was described by Massey et. al., (1989). Primary sampling units (PSUs) were stratified according to socioeconomic criteria and were selected with a probability proportional to their population size. Within PSUs, geographically defined secondary sampling units, referred to as segments, were selected systematically for the sample. The African American population was oversampled at this stage of the sample selection to secure adequate numbers for analytic purposes. Segments were then divided into clusters of 4-8 housing units, and all occupied housing units were included in the survey. Within each household, one randomly selected person age 18 and older was selected to participate. Young adults age 18-29 were oversampled at a ratio of 2.25:1.00 at this stage of sample selection to include a greater representation of this heavier drinking population subgroup. Weighting using SUDAAN (Shah, 1996) adjusted for the deliberate oversampling of African American and persons age 18-29 and accounted for the complex sampling design of NLAES. The household response rate for this representative sample of the U.S. population was 91.9% and the sample person response rate was 97.4%. The overall response rate was 90%. (A longitudinal follow up was initially planned but not completed because of financial constraints. A repeat cross sectional survey is currently being considered).

Outcome Measures

Driving after drinking was explored by asking respondents, "In your entire life, did you ever drive a car, motorcycle, truck or boat, or other vehicle after having too much to drink? Did that happen in the past 12 months?" Alcohol-related crash involvement was explored by asking respondents, "In your entire life, did you ever have a car, motorcycle, truck, boat, or other accident because of your drinking? Did that happen in the past 12 months?"

Predictor Variables

Measures of alcohol use and dependence were derived from the Alcohol Use Disorder and Associated Disabilities Interview Schedule (AUDADIS) (Grant and Hasin, 1992); a fully structured diagnostic psychiatric interview designed to be administered by trained interviewers who were not clinicians. 

The age of drinking onset was ascertained by asking respondents how old they were when they first started drinking, not counting small tastes or sips of alcohol. Drinking onset data was collected from respondents who were classified as current drinkers (persons who had consumed at least 12 drinks in the past 12 months) and former drinkers (persons who had consumed at least 12 drinks in any one year of their lives but not during the year prior to the interview). Only current and former drinkers were included in the analysis. In an independent test retest study before the full NLAES was conducted, the test retest reliability of the drinking onset variable was good with a Kappa of .72 (Grant, Harford, Dawson et. al., 1995).

The definition of lifetime alcohol dependence was based on the diagnostic criteria of the Diagnostic and Statistical Manual, Fourth Edition (DSM-IV) (American Psychiatric Association, 1994). The AUDADIS interview included an extensive list of symptom questions that operationalize the DSM-IV criteria for alcohol dependence. Diagnosis of alcohol dependence required that in any one year a respondent meet at least three of the following seven criteria for dependence: 1) tolerance, 2) withdrawal or avoidance of withdrawal, 3) persistent desire or unsuccessful attempts to cut down or stop drinking, 4) spending much time drinking, obtaining alcohol, or recovering from its effects, 5) giving up or reducing occupational, social, or recreational activities in favor of drinking, 6) impaired control over drinking and 7) continuing to drink despite a physical or psychological problem caused or exacerbated by drinking. Respondents were classified as showing a life time alcohol dependence diagnosis if they experienced an episode of dependence in the past year or at any time before the past year. The independent test retest study determined good reliabilities with Kappas of 0.76, and .0.73 for past year and prior to past year dependence diagnoses respectively (Grant, Harford, Dawson et. al., 1995). 
The following demographic and behavior variables were examined as potential confounders of the association between age of drinking onset and later behavior: current age, sex, race/ethnicity (white non-hispanic, black non-hispanic, hispanic, other), education, marital status, smoking status (current, former, never), illicit drug use status (current, former, never) and alcohol dependency (current, former, never).

Statistical Analysis

Like other reports of the NLAES, our statistical analyses were conducted using the SUDAAN statistical package (Shah, 1996) to account for the complex survey design and oversampling of NLAES in the estimation of both effects and their standard errors (the pattern of results was very similar with and without use of the SUDAAN weighting). We focused on respondents who reported drinking ever in their lifetime (n=27,081). The univariate associations between age of drinking onset and drinking and driving outcomes and demographic and background behavior characteristics was tested using a modified test of independence that adjusts for the sampling design. The test for independence in SUDAAN is based on the Wald statistic comparing observed and expected values and follows the strategy proposed by Koch, Freeman, and Freeman (1975) for analyzing complex survey data. The p-value for this test is based on transforming the Wald Statistic to an F-statistic where degrees of freedom represent the size of the contingency table and the number of PSU's and strata in the study design. Logistic regression explored whether age of drinking onset was associated with each of the drinking and driving outcomes, controlling for potential confounding from demographic and behavior characteristics including alcohol dependency. The overall significance of the relations between different ages of drinking onset and study outcomes in the logistic models were tested through a chi-square statistic comparing models with and without the set of indicator variables representing age of onset. We examined whether the potential association between age of drinking onset and drinking and driving outcomes persisted after controlling for alcohol dependence because of the established relationships between alcohol dependence and drinking and driving and between age of drinking onset and alcohol dependency (Grant, 1998). We also examined the potential relationships between age of drinking onset and drinking and driving behavior in the subset of respondents who were never alcohol dependent (n=21,713).