Drinking and Driving Among Whites and Hispanics on and off the US/Mexico border in California

Abstract

Objective:

To examine self-reported rates of driving under the influence (DUI) with and without arrest among border and non-border residents in California.

Methods:

Data were obtained from 1,209 adults 18 to 39 years of age resident in four counties in California: Imperial on the U.S./Mexico border; and Kern, Tulare, and Madera in California’s Central Valley. Households were selected using a list assisted sample. Data were collected on the phone or online and analyzed with a heteroskedastic ordinal generalized linear model.

Results:

Driving after drinking (11.1% vs. 6.5%; q=0.04) and the lifetime DUI arrest rates were higher for men than women (10.7% vs. 4%; q=0.001). In multivariable analysis driving after drinking and DUI arrests were not higher on the border, not higher among Hispanics than Whites, and among Hispanics, the rates were not higher among those located on the border. Income was positively associated with drinking and driving. Impulsivity was positively and significantly associated with both drinking and driving and lifetime DUI arrest.

Conclusion:

The null results suggest that DUI related risk behaviors may not be higher on the border than in other areas of California. There may be health related risk behaviors of higher prevalence in the border population than in other areas, but DUI related behavior may not be one of them.

Introduction

Driving under the influence of alcohol (DUI) remains a concerning public health problem in the U.S. and in California. In the U.S, in 2020, there were 11,654 deaths in motor vehicle crashes where at least one of the drivers was alcohol impaired (National Center for Statistics and Analysis, 2022). This represents an increase of 14.3% over the number of fatalities from 2019. Further, the 2019 Federal Bureau of Investigation Uniform Crime Report reported 658,902 thousand DUI arrests nationally or 286.8/100,000 population. About 26% of these arrests were among Hispanics (Federal Bureau of Investigations, 2019).

Law enforcement DUI arrest data for California have repeatedly showed higher DUI arrest rates for Hispanics compared to other ethnic groups (Daoud, 2021). A recent 18-year trend analysis (2005–2017) of these data among Whites and Hispanics showed a decline in rates among both groups (Caetano et al., 2020). Rates among Hispanics remained higher than among Whites throughout the years in the analysis but the difference in rates between both groups was reduced. For instance, in 2005 the rate per 100,000 population was two times higher among Hispanics than Whites (1,754.08 versus 874.59). In 2017 the rate among Hispanics was 1.7 times higher than among Whites (878.94 versus 507.00). Some do not think that Hispanics engage in DUI more than Whites, explaining these higher rates for Hispanics as a result of police profiling.

Data for rates of drivers under the influence of alcohol in road-side surveys have been more equivocal, with Hispanics showing higher rates than Whites in 1996 (7.5% vs. 2.3%), similar to Whites in 2007 (1.5% vs. 1.8%) and higher in 2013–2014 (1.9% vs. 1%) for drivers with Breath Alcohol Content (BrAC ) equal to or higher than .10 (Ramirez et al., 2016). Arrest rates and results from road-side surveys are not representative of the population, and thus are not ideal indicators of DUI. Arrest rates, are influenced by existing DUI laws, police enforcement of the laws, and driving and drinking behavior in the population. Roadside surveys are done at specific times of the day and in selected locations across the country.

Besides examining arrest statistics and rates in roadside surveys, the level of DUI in the population can also be assessed by survey self-reports of drinking and driving and of DUI arrest. In 2015, the National Survey on Drug Use and Health (NSDUH) reported a 9.9% twelve-month DUI rate for the population 16 years of age and older (Center for Behavioral Health Statistics and Quality, 2016). More recently, Fan et al., (2019) reported a twelve-month rate of “driving after drinking too much” of 3.9% based on analyses of data from the 2012–2013 National Epidemiologic Survey on Alcohol and Related Conditions III. These discrepancies in rates across surveys are due to differences in population sampled and also differences in question wording assessing respondents DUI-related experiences.

National rates cannot be easily applied to ethnic minorities, especially to a group as diverse as Hispanics. First, survey data do not show higher DUI rates for Hispanics compared to other ethnic groups, including Whites, as arrest rates do. For example, NSDUH trend data from 2002 to 2017 show that Whites had higher self-reported DUI rates than Hispanics for 2002–2004 (16.9% vs. 11.2%), for 2012–2014 (13.5% vs. 9.3%), and for 2016–2017 (10.1% vs. 6%) (Oh et al., 2020). Second, self-reported DUI rates among U.S. Hispanics vary by country of origin, being higher among Mexican Americans and South/Central Americans than among Cuban Americans and Puerto Ricans (Caetano et al., 2008). For instance, data from the 2006 Hispanic Americans Baseline Alcohol Survey show that the twelve-month rate of drinking and driving among men was 7.8% among Puerto Ricans, 6.2% among Cuban Americans, 17.3% among Mexican Americans, and 14.5% among South/Central Americans. Twelve-month self-reported DUI arrest rates were .8% for Puerto Ricans, .1% for Cuban Americans, 1.6% for Mexican Americans, and 5.6% for South/Central Americans.

DUI related research among Hispanics in the U.S./Mexico border area, most of whom are of Mexican origin, show contradictory results. Caetano et al., (2013) reported no statistically significant differences in drinking and driving rates between border and non-border areas. Rates among men and women in the border area were 11.1% and 2.9%, respectively. In the non-border area rates were 15.7% for men and 3.4% for women. Wallisch et al., (2017) reported higher rates for ever been arrested or stopped by police in border areas (13.2% in Laredo, 13.7% in Brownsville-McAllen, Texas) than off the border (9% in San Antonio, Texas). Higher DUI arrest rates closer to the border could be an effect of increased police activity in the area, which depending on the state can be under authority of state highway patrols, plus the border patrol, and local police departments. A recent year spatial analyses of DUI arrests in California indicated that Law Enforcement Reporting Areas (LERAs) closer to the border, as well as those with a higher percent of Hispanics in their population had higher rates of DUI arrests than other areas (Caetano et al., 2020). Arrests decreased with distance from the border among LERAs with less than 60% Hispanic population. However, this decrease could also reflect the association between unique risk behaviors present at the border such as crossing the border to drink in Mexico. Lange and Voas, (2000) documented that a third of late night drivers crossing back into the U.S. after visiting Tijuana had a positive blood alcohol content (BAC) and 8% had a BAC 0.08 or higher. Crossing the border to drink in Mexico is a high risk activity that has been associated with heavier drinking, alcohol-related problems, and DSM-5 alcohol use disorder (Cherpitel et al., 2015). It is possible therefore that this behavior also contributes to higher rates of drinking and driving in the border area.

Previous survey research assessing DUI in border and non-border areas was conducted in selected sites in Texas, which were relatively large urban areas (Wallisch et al., 2017), or contrasting a sample of the whole border with data from urban areas off the border in Texas (Houston) and California (Los Angeles) (Caetano et al., 2013). Both these studies focused on Mexican Americans only. The research reported in this paper has a different and more efficient design. First, it has closer matching between border and off border sites (both areas with mostly small towns in agricultural counties). Second, participants were in a younger and narrower age range (18–39) to maximize positive DUI reporting. Finally, a comparison group of White participants in the same age range was selected in the same areas. Altogether, these methodological characteristics strengthen internal validity for the cross ethnic comparison of Whites and Hispanics, and the comparison of border versus non-border locations.

Based on this design, on findings of previous alcohol use and availability on the border (Kelley-Baker et al., 2008, Lange and Voas, 2000), the analyses in this paper will test the following hypotheses: a) driving after drinking and lifetime DUI arrests will be higher on the border; higher among Hispanics than Whites, and within Hispanics, higher among those located on the border; and b) driving after drinking and DUI lifetime arrest will also be higher among border residents who cross the border into Mexico to drink. Results from these analyses will provide detailed information about DUI and its correlates among Whites and Hispanics that can be used in the prevention of this important public health and traffic safety problem with ethnic specific interventions.

Methods

Data are from a household survey of 1,209 White or Hispanic adults, 18 to 39 years of age interviewed in four counties in California: Imperial, Kern, Tulare, and Madera. Imperial County is agricultural and one of two California counties that share a border with Mexico, the other is San Diego County. Imperial county was selected for the study because San Diego, with a White population of 75% is not typical of other places along the U.S. Mexico border. In Imperial County, Hispanics are 85% of the population. Kern, Tulare, and Madera, the comparison sites, are all located in California’s Central Valley. They were selected because they, as Imperial County, have large Hispanic populations (between 56% and 62%), similar large agricultural rural areas punctuated by similarly small towns. Figure A1 shows a map of California with the study counties identified.

Households were selected randomly from a list assisted address-based sample (ABS) purchased from a commercial vendor. List assisted sampling differs little from samples developed using random digit dialing (Brick et al., 1995, Kempf and Remington, 2007, Tucker et al., 2002). These samples are constituted by records derived from postal addresses, built on the USPS Computerized Delivery Sequence File (CDSF). They contain all available residential mailing addresses covering almost 100% of all households in the U.S. Sample records have the mailing address and postal service relevant geographic information. This sample of addresses was overlaid with Census data and then appended by other household data, including names, phones, and residents gender.

All address-based records purchased for the project received an invitation postcard with a unique ID code, which described the study and directed potential respondents to the study website to learn more about the study and to submit an eligibility screener. Households were eligible if at least one of the residents was 18–39 years of age. If a household had more than one eligible respondent, all eligible respondents were listed and one of them was selected randomly to be interviewed. If eligible respondents selected to participate in the study entered their provided unique ID code in the online screening tool, they were rerouted to the web-based survey.

Surveys were conducted from June 2019 to May 2020 in both English and Spanish. Respondents who did not opt to respond to the questionnaire online were interviewed on the phone. These interviews were conducted by experienced interviewers working for the fieldwork agency contracted to conduct the survey. They were trained in the use of the questionnaire by the survey agency under supervision of the two co-directors of the project (Vaeth and Caetano). About 90% of the respondents opted to be interviewed in English or to respond to an online questionnaire in English, and 86% responded to the survey online. Respondents gave verbal consent to participate in the survey and received $20.00 remuneration. Written consent to participate in the research was not obtained from respondents because survey interviews were conducted on the phone, or the questionnaire was answered online. This study was approved by the Pacific Institute for Research and Evaluation Institutional Review Board (IRB00000630, IRB protocol number 1148122).

Both the survey overall cooperation rate and the response rate, using version #4 of the American Association for Public Opinion Research (American Association for Public Opinion Research, 2016), were different between the Central Valley and the border samples. The response rate estimates the percentage of respondents who completed an interview over the total number of eligible respondents in the sample. The cooperation rate estimates the percentage of households with positive contacts with survey personnel over all households contacted. Cooperation and response rates for the Central Valley were 95% and 72.5%, while on the border they were 12% and 7%. These low rates in Imperial County may have resulted from time overlap between the survey and the COVID-19 pandemic, which disproportionately affected this county, where 85% of the population is Hispanic. Responses to the survey invitation on the border may have been also affected by repressive immigration policies at the time by the U.S. federal administration.

Therefore, the sample should not be seen as representative of larger White and Hispanic groups in California. However, most important for the comparison between border location and the Central Valley, is internal validity, which was secured by group matching respondents in both locales on gender, age, and ethnicity. In other words, fieldwork was conducted so that the samples on the border and in the Central Valley would be constituted of 18–39 years old respondents and would have a similar percentage of Whites and Hispanics, and gender. Fieldwork also maintained uniformity of respondent selection and data collection, and analyses were conducted with controls for potential confounders. Analyses were conducted on data weighted to adjust the sample to known population distributions on gender, age, and ethnic identification (Hispanic, White) on the U.S. Census Bureau 2018 American Community Survey (American Community Survey (ACS) (census.gov) accessed on July 7, 2020).

Measurements

Location:

This variable identified sample respondents interviewed in the Central Valley of California (Kern, Madera, and Tulare counties), coded as “0” (reference) and those interviewed in Imperial County, on the border, coded as “1”.

Driving after drinking:

Respondents were asked if in the past 12 months they had “driven a car when you had drunk enough to be in trouble if the police had stopped you.” Answers were coded as “0” for “no” (reference) and as “1” for “yes”.

Lifetime driving under the influence (DUI) arrest:

Respondents were also asked if they had “ever been arrested for driving after drinking?” Answers were also coded as “0” for “no” (reference) and as “1” for “yes”.

Quantity and frequency of drinking:

Respondents were asked the number of days on which they had at least one drink of beer, wine, or liquor in the past four weeks or, among those who had not used alcohol in the previous four weeks, the past year. Those who had used alcohol were then asked on how many days they drank more than one, three or more, six or more, and nine or more drinks. A “drink” was defined as a 12-ounce can of beer, a 5-ounce glass of wine, or a 1.5-ounce shot of liquor. These data were fit using a validated log-logistic model of these responses (Gruenewald et al., 2003, Gruenewald et al., 2003) providing measures of drinking frequency (F), average drinking quantity (Q), volume (F*Q), and continued drinking volumes, V - F, a dose-response measure of heavier drinking related to problems (Gruenewald and Mair, 2015). Using these measures, F represents independent effects related to greater drinking exposures (drinking occasions) and V - F represents independent effects related to greater average drinks consumed per drinking occasion (two critical dimensions of dose-response) (Cunradi et al., 2020) (Cunradi et al., in press). In other words, V-F is the number of drinks consumed by respondents after they have the first drink, which makes it possible to distinguish the effects related to drinking frequencies to those related to greater volumes of drinking (Gruenewald and Mair, 2015). This approach also minimizes collinearity between measures of drinking frequency and drinking quantity. All drinking measures were scaled to a 365-day metric. Test retest reliabilities of these measures vary from r=0.65 for drinking quantities to r=0.85 for drinking frequencies (Gruenewald and Johnson, 2006).

Six drinks or more (heavy drinking):

Respondents were also asked on how many days they drank six or more drinks. Answers were recorded in integers representing the number of days reported by respondents and then recoded as a binary variable, with “0” for no days (reference) and as “1” for one or more days. A similar question (six or more drinks on one occasion) had good sensitivity (men, .87; women, .60) and specificity ( men .84; women, .92) for detecting risky drinking (Bradley et al., 2003, Bradley et al., 2007).

Drinking in Mexico:

Respondents were asked how many times they had visited Mexico in the past 12 months. Those that answered at least once were then asked on how many of those days they had had at least one drink of any alcoholic beverage. Answers were coded as “0” for no days (reference) and as “1” for one or more days.

Impulsivity:

This was measured with three items assessing respondents’ agreement with the following statements: I often act on the spur-of-the-moment without stopping to think; You might say I act impulsively; Many of my actions seem to be hasty (Cunradi et al., 2009). Four response categories ranged from “quite a lot” to “not at all” with scores ranging from 1 to 4 per item. Items were recoded 0 to 3 and also reverse coded, and scale values ranged from 0 to 9, with higher values indicating higher impulsivity. Cronbach’s alpha in the present data set was .86.

Perceived risk of selected behaviors:

This was assessed with five items asking respondents’ to evaluate the likelihood that “something bad would happen” if they drove over the speed limit, drove while drunk, drove without a seat belt, drank a lot, and had sex with someone they just met. Five response categories ranged from very likely to very unlikely. Scale values ranged from 0 to 25, with higher values indicating poorer risk perception. Cronbach’s alpha in the present data set was .88.

COVID shelter in place:

The survey was conducted partially before and partially during the adoption of COVID related shelter-in-place orders in California on 4/19/2020. Thus, an indicator of the time when survey respondents were interviewed or completed the questionnaire online was included in multivariable analyses. This variable was coded as “0” (reference) for those interviewed before the stay-at-home order and as “1” for those interviewed during the order.

Sociodemographic variables:

Gender.

A dichotomous variable was coded as men and women (reference).

Age.

This was coded as a categorical variable: 18–25, 26–29, and 30–39 (reference).

Ethnicity.

This was based on self-identification. This variable was coded as “0” if respondents were Hispanic and as “1” if they were White (reference).

Birthplace.

Respondents were asked if they were born in the United States or elsewhere (reference). Those born in the U.S. were coded as “0” and those born elsewhere were coded as “1”.

Income.

This variable had three categories: $20,000 or less (reference), $20,001 to $60,000, and $60,001 or more.

Employment.

This was a dichotomy coded as: “0” for currently working (reference) and as “1” for not currently working.

Religion.

This variable had four categories: Protestant, no religious preference, Catholic (reference), and other.

Marital status.

This is a 3-category variable: married (reference), separated/divorced/widowed, and single.

Statistical Analyses

Power analysis called for the enrollment of 1200 participants 18–39 years of age. Once the sampling design effect was considered, the effective N=825, with alpha at .05, would have 80% power to detect an odds ratio of 1.3 associated with a 1 standard deviation unit change in a given independent variable at mean values of other independent variables for events occurring with probability as low as .14 Hsieh, (1989). Analyses were conducted on weighted data using Stata’s 17.0 “svy” prefix (Stata, 2015). Multivariable analyses of lifetime DUI arrest and driving after drinking (Table 2) were conducted using Stata’s “oglm” command, an ordinal generalized linear analysis procedure that addresses sources of heteroskedasticity related to selected independent variables (Williams, 2009, Williams, 2010). These regressions were first run with a variable representing the main effect of survey interview time relative to the COVID shelter-in-place order in California (before and during the order). An interaction between this main effect and participant location (border versus Central Valley) was also assessed in the same model. This was because on 4/21/2020 the U.S./Mexico border was closed to “non-essential” travel between the two countries, which could affect survey respondents located on the border more than those in the Central Valley.

Table 2:

Multivariable heteroskedastic logistic regressions of driving after drinking and lifetime arrest for DUI on drinking, location, psychological and sociodemographic variables.

	Drinking and driving		Lifetime DUI arrest
Independent Variables	Odds ratio	95% CI	Odds ratio	95% CI
Hispanic (ref: White)	0.85	(0.50, 1.46)	1.06	(0.37, 2.99)
Border location (ref: Central Valley)	0.53	(0.03, 10.23)	0.54	(0.15, 1.89)
Drank 1+ days in Mexico (ref: None)	1.27	(0.50, 3.24)	0.15	(0.03, 0.73)
Border Location by Drank in Mexico(ref: Central Valley/No drinking in Mexico)	1.41	(0.26, 7.49)	8.54	(1.24, 58.61)
Border location by Hispanic ethnicity (ref: Central Valley/White)	0.84	(0.05, 13.23)	0.67	(0.15, 2.98)
Drinking frequency	1.08	(1.03, 1.14)	1.09	(1.02, 1.17)
Continued drinking volume	1.01	(1.00, 1.03)	0.99	(0.96, 1.02)
Six drinks on one occasion (ref: No)	1.61	(0.66, 3.96)	1.63	(0.75, 3.51)
Male (ref: female)	0.92	(0.62, 1.39)	2.36	(1.05, 5.28)
Age (ref:30–39)
18–25	1.20	(0.48, 3.02)	0.05	(0.01, 0.36)
26–29	1.68	(0.88, 3.24)	0.96	(0.34, 2.76)
Annual Income (ref: Less than $20,000)
$20,000–$60,000	2.58*a	(1.49, 4.46)	0.58	(0.17, 1.92)
$60,l001+	1.05	(0.42, 2.65)	0.71	(0.18, 2.81)
Marital Status (ref: Married)
Separated/Divorced/Widowed	3.36	(1.49, 7.58)	1.89	(0.44, 8.12)
Single	1.76	(0.74, 4.20)	2.32	(1.02, 5.26)
Religion (ref: protestant)
Catholic	1.72	(0.63,4.69)	0.97	(0.40, 2.33)
No religious preference	1.69	(0.84, 3.38)	1.10	(0.53, 2.25)
Other	1.54	(0.22, 10.87)	0.66	(0.14, 3.17)
Born in United States (ref: Born abroad)	3.75	(0.71, 19.81)	1.29	(0.36, 4.56)
Unemployed (ref: Employed)	1.53	(1.02, 2.28)	0.42	(0.17, 1.00)
Impulsivity scale	1.29*b	(1.10, 1.51)	1.33**e	(1.15, 1.53)
Risk perception scale	0.95	(0.90, 1.00)	0.94	(0.89, 1.00)
Heteroscedasticity terms	Coefficient	95% CI	Coefficient	95% CI
Overall yearly drinking frequency	0.15*c	(0.07, 0.24)	9.00E–04	(−0.07, 0.07)
Overall yearly drinking frequency squared	−0.01***d	(−0.02, −0.01)	−3.90E−03	(−0.01, 0.00)
Overall yearly drinking continued volume	0.03	(0.00, 0.06)	0.03	(0.00, 0.05)

Significance test with p values corrected for multiple tests with Benjamini–Yekutieli method in “a” thru and “e”

^*p<0.02

^**p<0.009

^***p<0.003.

Confidence intervals are not corrected.

Subsequent model development was done with all independent variables plus the two hypothesized interactions (border by ethnicity in hypothesis “a,” border by drinking in Mexico, hypothesis “b”) together in each of the two models in Table 2 (drinking and driving and lifetime DUI arrest). Statistical testing was conducted both in Tables 1 and 2 with corrections for multiple testing for correlated data. In Table 1 corrections were applied with Stata’s “qqvalue” command with Benjamini-Yekutieli option (Newson 2009). In Table 2 , corrections were done by applying Stata’s “smileplot” command also with Benjamini-Yekutieli option (Newson, 2012). Independent variables selection was based on previous findings in the literature (Caetano et al., 2018, Cherpitel et al., 2015). In both cases, the corrections are applied to the level of statistical significance not to the confidence intervals.

Table 1:

Percentage of respondents who drove after drinking and who reported a lifetime DUI arrest

		Drove after drinking	Significancea	Lifetime DUI arrest	Significancea
Men	(487)	11.1%	P p=.04a	10.7%	p=.001b
Women	(718)	6.5%		4.0%
White	(347)	8.6%	Not significant	7.5%	Not significant
Hispanic	(865)	8.1%		6.2%
Border	(539)	6.5%	Not significant	5.4%	Not significant
Central Valley	(677)	9.7%		7.6%

Unweighted Ns in parentheses; significance tests with p values corrected for multiple tests with Benjamini–Yekutieli method in “a” and “b”.

Results

Drinking and driving and lifetime DUI arrest rates by location, gender, and ethnicity

The rates for driving after drinking and for lifetime DUI arrest were higher for men than women (Table 1).

Correlates of drinking and driving and lifetime DUI arrest

Initial results showed that neither the main effect for COVID-19 shelter (drinking and driving: OR=1.23; p= 0.44; 95%CI=.69–2.20; lifetime DUI arrest: OR=1.13; p=0.78; 95%CI=.45–2.79) nor the interactions of COVID-19 shelter by location in the regressions on drinking and driving (OR=1.76;p= 0.37;95%CI=0.50–6.20) and for lifetime DUI arrest (OR=1.58;p= 0.67;95% CI=0.18–13.41) were significantly associated with the outcomes. These variables were dropped from the model for each of these two outcomes, which were then rerun with the two hypothesized interactions (location by ethnicity and location by drinking in Mexico). All models controlled for the effects of gender, age, birthplace, education, employment status, marital status, and religion.

There were positive and statistically significant associations between drinking and driving and an annual family income of $20,000 to $60,000 and a higher score on the impulsivity scale. The only correlate with a statistically significant positive association with lifetime DUI arrest was impulsivity.

Discussion

The associations proposed in hypothesis “a” and “b” were not confirmed. Driving after drinking and DUI arrests were not higher on the border, not higher among Hispanics than Whites, and among Hispanics the rates were not higher among those located on the border. Further, rates were not higher among border residents who drink in Mexico compared to non-border residents who do not drink in Mexico. The text below discusses these findings in more detail.

The results confirmed previous survey findings by Caetano et al., (2013) showing that DUI arrest and drinking and driving rates are not higher on the border compared to off border areas, in this case the Central Valley of California. This contradicts previous findings by Wallisch et al., (2017) for areas on and off the border in Texas. More broadly, the results herein suggest that the documented increased availability of alcohol in border areas may not affect drinking and driving rates, which would be true for the border population as a whole and also for border Hispanics compared to those in the Central Valley. A potential explanation for this null finding is the increased level of police activity in border areas. This is partly because of the constant presence of Border Patrol agents not only in areas close to the international line that separates the U.S. from Mexico, but in airports, highways, parks, and city streets and other areas within a “reasonable distance” of a border (100 miles) (Anthony, 2020).

Existing evidence shows that police activity to enforce DUI laws is associated with lower levels of this behavior (Fell et al., 2014, Lenk et al., 2021). There also were null findings for differences between Whites and Hispanics, which is not surprising. As stated in the Introduction, previous survey research data on self-reports of DUI related behavior, including self-reports of DUI arrest and of drinking and driving, do not show significant differences between Whites and Hispanics (Oh et al., 2020).

Correlates with significant associations with drinking and driving were income and a higher level of impulsivity. The latter also has a significant and positive association with lifetime DUI arrest. The income category with a positive association with drinking and driving is that for annual income between $20,000 and $60,000 compared to those with an annual income of less than $20,000. Individuals with higher income have a higher rate of car ownership and thus potentially more opportunities for drinking and driving. For instance, the 2011 National Household Travel Survey showed that 20.3% of household with less than $25,000 annual income had no cars compared to 2.3% of households with annual income above $25,000 (Bureau of Transportation Statistics 2011). Higher income is also associated with higher rates of drinking (Collins, 2016). However, the association between income, drinking and alcohol related harms such as DUI is complex. A higher income, as we see in the case of those with an annual income equal or higher than $60,001, does not necessarily imply that this group will have a higher DUI rate (Beard et al., 2016, Smith and Foster, 2014).

The associations with impulsivity are also important. Previous results in the literature have showed a positive association of this psychological construct with alcohol use disorder (Dick et al., 2010, Rubio et al., 2008), with hostility and drinking at higher risk contexts such as bars (Gruenewald et al., 2018, Treno et al., 2007). There have also been consistent positive associations with DUI and DUI related behavior (Fillmore and Van Dyke, 2020, McCarthy et al., 2012, Moan et al., 2013, Van Dyke and Fillmore, 2014). However, it is important to consider that impulsivity is a multidimensional concept with attentional, behavioral, and cognitive components and not all of them may be equally associated with DUI (Jakubczyk et al., 2013, Stamates and Lau- Barraco, 2017).

Finally, when considering the results above, the following limitations should be kept in mind: First, the survey had a low response rate in Imperial County, on the border. There are no data to compare respondents to non-respondents, but if the latter included more participants with a positive history of drinking and driving or DUI arrests, the results would be biased. Data were self-reported, which could be associated with under-reporting of DUI related events potentially seen as undesirable by respondents. This under-reporting could also be greater among Hispanics than Whites because of this former group minority status. Further, the cross-sectional nature of the data does not allow for inferences related to the direction of many of the associations described. Strengths of the study include a strong design with matched data collection locales in a large state, a focus on Whites and Hispanics, participant matching for age and ethnicity, and detailed measurement of drinking variables and drinking and driving.

In conclusion, results herein did not show differences between border and non-border areas on self-reports of drinking and driving and of lifetime DUI arrests. Broadly, this suggests that there may be health related risk behaviors of higher prevalence in the border population than in other areas, but DUI related behavior does not seem to be one of them.

Data availability statement:

Contact corresponding author (RC) if interested in using data.