Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Am J Public Health. Author manuscript; available in PMC 2016 July 1.
Published in final edited form as:
PMCID: PMC4458165

Racial/Ethnic disparities in the self-reported number of drinks in 2 hours before driving becomes impaired



Studies of racial/ethnic disparities in driving under the influence (DUI) prevalence have not found a clear pattern of differences. This study utilizes data on self-reported impaired driving and the number of drinks the person states they can have in 2 hours before impairment to evaluate predictors of individuals’ impairment thresholds.


Data come from the 2000, 2005 and 2010 US National Alcohol Surveys with over-samples of black and Hispanic populations. Negative binomial models were estimated overall, by gender and for those who reported impaired driving. Analyses focus primarily on 8,553 respondents who drank alcohol and drove a car in the past year.


Models controlling for relevant available measures including bodyweight, socio-demographics and drinking patterns found impairment thresholds to be 30.3% (95% CI: 23%–38%) higher for black drinkers and 26.3% (95% CI: 18%–35%) higher for Hispanic drinkers as compared to white drinkers.


The greater number of standard drinks before perceived impairment reported by black and Hispanic drivers implies a likely relative under-report of impaired driving and potentially higher severity of impairment when driving relative to whites.

Keywords: Disparities, drink driving, drink size

Many drivers in alcohol-involved fatal crashes have no prior driving under the influence (DUI) arrests and it is estimated that only about 1% of DUI incidents result in arrests each year, highlighting the need for research supporting primary prevention in the general population.1 The proportion of drivers in fatal crashes with a 0.08 or higher blood alcohol content (BAC) has been stable in recent years at around 22%.2 Rates of past year DUI self-report in national surveys have been found to differ substantially by gender and across the three largest US race/ethnicity groupings. A study of rates in 2000 found the highest rates for white men (22%) compared to black (16.5%) and Hispanic (16.8%) men. Rates for women were lower, with white women also the highest (11.8%) compared to black (9.2%) and Hispanic (6.7%) women.3 A review of disparities in alcohol-attributable crash injuries found higher rates among whites in Fatality Accident Reporting System (FARS) data, but lower rates for whites in some studies of specific states and, in one case, for the 55 and older age group only, suggesting some inconsistency in disparity findings between self-report incidence and injury rates.4

The number of drinks that drinkers feel that they can consume in two hours before driving becomes impaired (here termed impairment threshold) is a measure that offers a unique window into drinkers’ subjective views on impairment—one that has not been previously analyzed in a population sample. This measure is relevant to both the interpretation of self-reported DUI and to the potential severity of DUI episodes. In particular, one might expect that groups with a higher mean impairment threshold would also show a lower self-reported prevalence of DUI at any given level of drinking, thus biasing analyses of differences between these groups. A higher impairment threshold group mean would also suggest that the severity of reported DUI episodes might be higher for that group. Findings among college students have indicated that estimated blood alcohol content (eBAC) and reports of subjective intoxication interacted such that DUI was more likely when students had higher eBAC but perceived less intoxication, emphasizing the importance of subjective impairment for DUI risk.5 Similarly, this risky situation would be more likely to occur among those with high impairment thresholds.

Studies using the National Alcohol Surveys (NAS) have indicated the importance of heavy occasions,6,7 for risk of self-reported DUI. The current study builds on previous analyses of subjective intoxication measures and drink alcohol content (drink strength) where some evidence for differences between black, Hispanic and white Americans has been found.8,9 Analyses showed that blacks reported needing fewer drinks to feel drunk, and male Hispanics reported more, compared to whites.8 Research on drink strength has found that black male drinkers have higher-strength drinks on average than white males both at home9 and in bars.10 Larger than standard drinks have also been reported in a study of US Hispanic groups, particularly for spirits drinks.11 Here, analyses focus on black, Hispanic and white drinkers, evaluating predictors of differences in impairment thresholds, defined with a drink strength-adjusted measure of the amount (in standard drink equivalents) that the respondent believes they could consume in two hours before their driving would be impaired, including racial/ethnic group indicators representing differences not accounted for by other measured factors. Drink strength adjustments improve the comparability of this measure across gender and race/ethnicity groups by addressing this potential source of bias in survey self-report measures.



Data are drawn from the three most recent NAS samples, nationally representative list-assisted random-digit-dialed telephone surveys of the US with oversamples of Hispanic and African Americans. The 2000 NAS included 7,612 respondents with a cooperation rate of 58% and the 2005 NAS had 6,919 respondents and a cooperation rate of 56%. The 2010 NAS had 6,855 cases in the landline sample and an overall cooperation rate of 52%. Nonresponse does not necessarily bias population estimates12,13 and these rates are consistent with those from recent US telephone surveys.14 Using Census data, all surveys are weighted to the general population of the US at the time they were conducted, taking account of nonresponse, age, sex, ethnic group and geographic area. Analytic models focus on the drink-strength adjusted number of drinks in two hours before driving becomes impaired (impairment threshold) and include 8,553 eligible respondents aged 20 and older who in the past year both drank and drove a car (impairment threshold was not asked of non-drinkers or non-drivers) and who had complete data for all measures. Measures involving additional missing values were past year marijuana use (323), bodyweight (396) and reporting 5+ days at least monthly in questions on decades heavy drinking during the 20’s, excluding 18 and 19 year olds (394).


Questions on impaired driving assessed past-year incidence of having “driven a car when you had drunk enough to be in trouble if the police had stopped you?” Subjective impairment threshold related to driving was assessed by “How many drinks do you think you can have, over a two hour period, before your ability to drive becomes impaired? By impaired we mean you had too much to drink to drive safely.” These questions were asked of all drinkers who drove a car in the past year. The reported number of US standard drinks (14g ethanol) was adjusted by the respondent’s estimated drink strength and was capped at 30 drinks.

Past-year alcohol volume and maximum drinks in a day were based on responses to the Graduated Frequency series,15,16 which, after asking the maximum17 on any day in the prior 12 months,17 asks separately about frequencies of each relevant quantity level: 12 or more drinks, 8–11, 5–7, 3–4 and 1–2 drinks in a day over the past year. Responses to questions on the beverage-specific frequency of drinking are used to estimate the number of days each beverage type was drunk in the prior 12 months. For each beverage type, respondents then report their relative frequency of 1–2, 3–4, or 5+ drinks and these responses were combined to calculated beverage-specific alcohol volume. Reported drinks were adjusted for estimated drink alcohol content utilizing estimates of beverage-specific drink alcohol content derived from two prior national studies of home drinks, and a separate study of bar and restaurant drinks in Northern California.10,18

Race and ethnic group indicators were created for white (non-Hispanic), black (non-Hispanic), Hispanic (of any race) and all others combined. Educational attainment was coded as less than high school, high school graduate only, some college and college graduate or higher. Marital status was coded never married vs. all other categories based on preliminary analyses. Religion was coded as Catholic vs. all other categories based on preliminary analyses. Employment status was included in preliminary models but was dropped because no significant effects were found. Income was converted to 2005 dollars using the mid-point of the categories asked in each survey year and then re-categorized into $0–20,000, $20,000 to 40,000, $40,000 to 70,000 and $70,000 or more, and income missing. State groupings by “wetness” of drinking culture were created in a previous study based primarily on the estimates of heavy occasion drinking and per capita apparent consumption of alcohol.19 Here a six-level categorization that resulted was simplified to Pacific Coast vs. all others regions based on preliminary analyses indicating no significant differences between other regions. A four-level family history of alcohol problems scale measures the degree of biological relatives’ involvement with alcohol problems (none, 2nd degree, 1st degree, both). Lifetime heavy drinking was assessed through questions on the frequency of 5+ drinking days in the respondents teens, 20’s and 30’s and operationalized with an indicator of monthly heavy drinking during the 20’s based on preliminary analyses. Indicators identified those reporting any marijuana use and any tobacco use in the past year. An indicator for those living in states where the per se blood alcohol content (BAC) limit for driving was 0.10 in 2000 was included (note that by 2005 all states had a limit of 0.08). To capture body water and other size–related influences, self-reported measures of bodyweight in pounds and height in inches were included in each model.


Mean values for any past year impaired driving among current drinkers and impairment threshold among current drinkers who drive were estimated using survey weights in Stata 13.20 Confidence intervals are presented but differences are not tested as testing these influences are the focus of the controlled models. Among past year drinkers who drove a car, the negative binomial models of impairment threshold included key predictor measures as described above and were estimated using survey weights.21,22 Negative binomial models were used because the dependent variable was a count with a skewed distribution and likelihood-ratio tests indicated over-dispersion. Models were estimated overall, for those who drove while impaired and for each gender group separately. To evaluate relationships across regions of the dependent variable distribution, quantile regression models were estimated, reporting results for the 0.25 and 0.95 models. These models implement weighted minimum absolute deviations estimation and utilize sample weights.20,23 The 0.5 quantile model evaluates deviations around the median of the distribution while other models are weighted to focus on other areas such as the 0.95 model where the weighted median will be greater than 95% and less than 5% of the responses. This allows us to evaluate differences in the size and significance of coefficients around differing impairment thresholds, say 3 versus 8, while still utilizing all of the observations. Predictors for those reporting low impairment thresholds may be different from those who report higher thresholds. Significance in the quantile regression models is evaluated using bootstrapped standard errors with 20 repetitions.


Mean values and group percentages for the analytic sample are presented in Table 1 for gender and race/ethnicity sub-groups. Group differences in potential predictors of self-reported impairment thresholds such as bodyweight, educational attainment, drinking patterns and drinking histories highlight the importance of evaluating group differences in the multivariable models. The prevalence of self-reported impaired driving and the mean impairment threshold overall, by gender and by gender and race/ethnicity group are presented in Table 2. A higher prevalence of past-year impaired driving is found among white respondents for both men and women. Distributional differences by ethnicity in the mean impairment threshold (in standard drinks after adjusting for drink sizes) are observed for both men and women, with black and Hispanic drinkers reporting higher numbers than white drinkers in nearly all years. Compared to white drinkers, both black and Hispanic drinkers had flatter distributions with more reports of higher numbers of drinks. Among Hispanics, 31.2% reported an impairment threshold of greater than 5 compared to 28% of black and 12.3% of white respondents. An impairment threshold of greater than 10 was reported by only 1.3% of white respondents but 5.1% of black and 7.7% of Hispanic respondents.

Table 1
Mean values and indicator percentages for variables included in models for current drinkers and relevant subgroups (by gender and race/ethnicity).
Table 2
Drinkers’ self-reported prevalence and mean, median and interquartile range (IQR) of the minimum number of drink alcohol-content adjusted drinks for alcohol impaired driving (impairment threshold) in the past year for gender and race/ethnicity ...

The estimated models, presented in Table 3, consistently show significant positive coefficients for black and Hispanic indicator variables as compared to the white reference group. Effect sizes in the negative binomial models indicated that black drinkers and Hispanic males reported impairment thresholds about 30% higher than the white drinker reference group, while Hispanic female drinkers were 17.6% higher. Quantile regression results indicated that those in the 0.25 area reported 0.325 more drinks among black drinkers and 0.2 drinks among Hispanic drinkers while those in the 0.95 area reported 2.121 additional drinks among black drinkers and 2.79 additional drinks among Hispanic drinkers. A number of predictors were found to be significant in the overall model, with some differences in significant coefficients across subgroup models. For example, men on average reported requiring more drinks to reach impairment than women. A number of alcohol measures were included in the models. The most consistent positive predictor was the past-year maximum number of adjusted drinks in a day, which was significant in all models. An indicator for having monthly 5+ days during the 20’s was significant overall and for men, but not among women or impaired drivers. Total alcohol volume was a positive predictor in the full sample, for women and in the quantile regression models, while the liquor volume coefficient was positive and significant in the full sample, for men and in the quantile regression models. Beer volume was generally not significant but was a significant negative predictor for women. Past-year marijuana use and the indicator for states with 0.10 per se BAC limit in 2000 were not significant in any models. Bodyweight was a significant positive predictor overall, for men and in the quantile models. Income and educational attainment were not significant predictors overall or in the gender-specific models, however, among impaired drivers and in the 0.95 quantile model, higher incomes were associated with lower numbers of drinks. An interesting contrast in the associations for educational attainment indicators was seen in the quantile regression results. In the 0.25 model emphasizing the association for those reporting lower impairment thresholds, a positive effects of education was seen, while in the 0.95 model emphasizing associations for those reporting higher impairment thresholds a strong negative effect was seen. Not surprisingly, having reported past-year impaired driving was a positive predictor and significant in all models.

Table 3
Negative binomial and quantile regression models predicting the reported number of drinks in 2 hours before driving becomes impaired a


The broadly supported finding of higher self-reported thresholds for drunk driving among Hispanic and black drinkers relative to white drinkers has important implications for DUI prevention and epidemiology. This difference is seen in both drinkers who report past year DUI episodes and those that do not, in men and women and in those reporting both lower and higher impairment thresholds in the quantile regression models, suggesting cultural differences in the meaning or understanding of impairment. These race/ethnicity effects persist after controlling for many relevant factors including demographic measures, body weight and height, current drinking patterns, drinking history and family-based genetic factors. Importantly, these differences suggest that comparisons of self-reported DUI rates could be biased across these groups due to higher perceived drinks-to-impairment thresholds among black and Hispanic drinkers. A related possible implication is that the severity of the DUI episodes reported may be greater for black and Hispanic than for white drinking drivers for the same reason.

While effective policies preventing impaired driving such as per se laws, checkpoints and random BAC testing, as well as the severity and certainty of penalties and the enforcement of these remain of primary importance, there is a role for public health information campaigns that publicize these policies so as to enhance deterrence.24 The influence of social norms has also been found to help determine the effectiveness of these polices,25 for example, an earlier study utilizing NAS data found that the passage of 0.08 per se laws reduced the reported number of drinks to feel drunk between 1979 and 2000.8 The acceleration of crash risk as BAC rises above 0.06, and especially after 0.10,24 suggests the importance of marginal increases in intoxication at higher levels, again serving to emphasize the relevance of the number of standard drinks consumed to crash risk. However, an experiment on providing information to drivers going to Mexico for the night through a “Know Your Limit” card did not reduce BACs measured on their return to the US; however, reminding drivers about the risks of DUI did lower returning BACs,26 suggesting that avoidance of penalties was more important than errors in tracking intake. Conversely, a recent survey of drivers who drink conducted in four areas found that black males overestimated the probability of being stopped and overestimated jail sentences for a DUI conviction,27 suggesting the importance of alternative strategies for further reducing DUI events for this group.

Our results supporting cultural differences in perceived amounts of intake associated with driver impairment, together with previous findings of stronger drinks for black and Hispanic drinkers,28 highlight potential avenues for culturally informed norm-based education and intervention efforts regarding drink choices. They emphasize too the need to reconcile culturally-determined perceptions of intoxication and impairment with more objective physiological impairment outcomes for a given gender and bodyweight. Further studies focused on race/ethnicity group differences in norms, attitudes and expectations regarding alcohol impairment, particularly in relation to driving risks, are need to understand more fully the differences identified in this study that appear to be broadly applicable to black and Hispanic drinkers at all levels of consumption.


This work was supported by the National Institute on Alcohol Abuse and Alcoholism (P50AA005595). An earlier version of this paper was presented at the Research Society on Alcoholism annual conference in 2013.


Contributor Statement

William C. Kerr was responsible for all aspects of the study including conceptualization, data coding and analyses and manuscript preparation. Thomas K. Greenfield was involved in the conceptualization and analyses as well as manuscript preparation.

Human Participant Protection

Study protocols were approved by the Public Health Institute Institutional Review Board, IRB# I11=019.


1. Ferguson SA. Alcohol-impaired driving in the United States: contributors to the problem and effective countermeasures. Traffic Inj Prev. 2012 Sep;13(5):427–441. [PubMed]
2. Voas RB, Torres P, Romano E, Lacey JH. Alcohol-related risk of driver fatalities: an update using 2007 data. J Stud Alcohol Drugs. 2012 May;73(3):341–350. [PubMed]
3. Caetano R, McGrath C. Driving under the influence (DUI) among U.S. ethnic groups. Accid Anal Prev. 2005 Mar;37(2):217–224. [PubMed]
4. Keyes KM, Liu XC, Cerda M. The role of race/ethnicity in alcohol-attributible injury in the United States. Epidemiol Rev. 2012 Jan;34(1):89–102. [PMC free article] [PubMed]
5. Quinn PD, Fromme K. Event-level associations between objective and subjective alcohol intoxication and driving after drinking across the college years. Psychol Addict Behav. 2012;26(3):384–392. [PMC free article] [PubMed]
6. Cherpitel CJ, Ye Y, Greenfield TK, Bond J, Kerr WC, Midanik LT. Alcohol-related injury and driving while intoxicated: a risk function analysis of two alcohol-related events in the 2000 and 2005 National Alcohol Surveys. The American Journal of Drug and Alcohol Abuse. 2010 May;36(3):168–174. [PMC free article] [PubMed]
7. Greenfield TK, Rogers JD. Alcoholic beverage choice, risk perception, and self-reported drunk driving: effects of measurement on risk analysis. Addiction. 1999 Nov;94(11):1735–1743. [PubMed]
8. Kerr WC, Greenfield TK, Midanik LT. How many drinks does it take you to feel drunk? Trends and predictors for subjective drunkenness. Addiction. 2006 Oct;101(10):1428–1437. [PubMed]
9. Kerr WC, Patterson D, Greenfield TK. Differences in the measured alcohol content of drinks between black, white and Hispanic men and women in a US national sample. Addiction. 2009;104(9):1503–1511. [PMC free article] [PubMed]
10. Kerr WC, Patterson D, Koenen MA, Greenfield TK. Large drinks are no mistake: glass size, not shape, affects alcoholic beverage drink pours. Drug Alcohol Rev. 2009 Jul;28(4):360–365. [PMC free article] [PubMed]
11. Caetano R, Mills BA, Harris TR. Hispanic Americans Baseline Survey (HABLAS) effects of container size adjustments on estimates of alcohol consumption across Hispanic national groups. J Stud Alcohol Drugs. 2012 Jan;73(1):120–125. [PubMed]
12. Groves RM. Nonresponse rates and nonresponse bias in household surveys. Public Opin Q. 2006;70(5):646–675.
13. Keeter S, Kennedy C, Dimock M, Best J, Craighill P. Gauging the impact of growing nonresponse on estimates from a national RDD telephone survey. Public Opin Q. 2006;70(5):759–779.
14. Curtin R, Presser S, Singer E. Changes in telephone survey nonresponse over the past quarter century. Public Opin Q. 2005 Apr;69(1):87–98.
15. Greenfield TK. Ways of measuring drinking patterns and the difference they make: experience with graduated frequencies. J Subst Abuse. 2000;12(1):33–49. [PubMed]
16. Greenfield TK, Kerr WC, Bond J, Ye Y, Stockwell T. Improving graduated frequencies alcohol measures for monitoring consumption patterns: results from an Australian national survey and a US diary validity study. Contemp Drug Prob. 2009 Fall-Winter;36(3/4):705–733. [PMC free article] [PubMed]
17. Greenfield TK, Nayak MB, Bond J, Ye Y, Midanik LT. Maximum quantity consumed and alcohol-related problems: assessing the most alcohol drunk with two measures. Alcohol Clin Exp Res. 2006 Sep;30(9):1576–1582. [PubMed]
18. Kerr WC, Patterson D, Koenen MA, Greenfield TK. Alcohol content variation of bar and restaurant drinks in Northern California. Alcohol Clin Exp Res. 2008 Sep;32(9):1623–1629. [PMC free article] [PubMed]
19. Kerr WC. Categorizing US state drinking practices and consumption trends. Int J Environ Res Public Health. 2010 Jan;7(1):269–283. [PMC free article] [PubMed]
20. Stata Statistical Software: Release 13.0 [computer program] College Station, TX: Stata Corporation; 2013.
21. Stata Statistical Software: Release 11.0 [computer program] College Station, TX: Stata Corporation; 2009.
22. Hilbe JM. Negative binomial regression. New York: Cambridge University Press; 2007.
23. Koenker R. Quantile Regression. New York, NY: Cambridge University Press; 2005.
24. Voas RB, Fell JC. Preventing impaired driving opportunities and problems. Alcohol Res Health. 2011;34(2):225–235. [PMC free article] [PubMed]
25. Snortum JR, Berger DE. Drinking-driving compliance in the United States: percpetion and behavior in 1983 and 1986. J Stud Alcohol. 1989;50(4):306–319. [PubMed]
26. Johnson MB, Clapp JD. Impact of providing drinkers with “know your limit” information on drinking and driving: a field experiment. J Stud Alcohol Drugs. 2011 Jan;72(1):79–85. [PubMed]
27. Sloan FA, Chepke LM, Davis DV. Race, gender, and risk perceptions of the legal consequences of drinking and driving. Journal of Safety Research. 2013 Jun;45:117–125. [PMC free article] [PubMed]
28. Kerr WC, Stockwell T. Understanding standard drinks and drinking guidelines. Drug Alcohol Rev. 2012 Mar;31(2):200–205. [PMC free article] [PubMed]