Skip to main content
NCBI home page
Journal of Studies on Alcohol and Drugs logoLink to Journal of Studies on Alcohol and Drugs
. 2021 Nov 17;82(6):720–729. doi: 10.15288/jsad.2021.82.720

Rideshare Trips and Alcohol-Involved Motor Vehicle Crashes in Chicago

Christopher N Morrison a,b,*, Gabrielle D’Ambrosi a, Ava Kamb a, Kytt MacManus c, Andrew G Rundle a, David K Humphreys d
PMCID: PMC8819616  PMID: 34762031

Abstract

Objective:

Rideshare companies such as Uber and Lyft have substantially changed transportation markets in the United States and globally. The aim of this study was to examine whether ridesharing is associated with reductions in alcohol-involved crashes.

Method:

This case-series study used highly spatially and temporally resolved trip-level rideshare data and motor vehicle crash data from the Chicago Data Portal from November 2018 to December 2019. The units of analysis were motor vehicle crashes in Chicago. Events of interest were 962 crashes that police indicated were alcohol involved. The comparison group was 962 non–alcohol-involved crashes that occurred in the same census tract, matched 1:1. The exposure of interest was the density per square mile of rideshare trips that were in progress at the time of the crash, calculated using a kernel density function around the estimated route paths of active trips. A conditional logistic regression compared alcohol involvement to rideshare trip density while adjusting for matching and relevant time-varying covariates (taxi trips, precipitation, temperature, holidays).

Results:

Mean rideshare trip density was 69.0 per square mile (SD = 129.7) at the time and location of alcohol-involved crashes and 105.7 per square mile (SD = 192.6) at the time and location of non–alcohol-involved crashes. After controlling for covariates, the conditional logistic regression model identified that a standard deviation increase in rideshare trips per square mile at the crash location was associated with 23% decreased odds that the crash location was alcohol involved (odds ratio = 0.771; 95% confidence interval [0.594, 0.878]).

Conclusions:

Ridesharing may replace motor vehicle trips by alcohol-impaired drivers

Alcohol-involved motor vehicle crashes are a major cause of injury in the United States. More than one third of fatal motor vehicle crashes involve a driver who had consumed any alcohol (Bureau of Transportation Statistics, 2019), including a total of 10,142 (28%) deaths in 2019 (National Center for Statistics and Analytics, 2020). Total direct and indirect costs because of alcohol-involved crashes are estimated to be around $125 billion per year (Zaloshnja et al., 2013). Although the absolute risk that any single impaired driving event will result in a crash is low, these vast public health and economic costs arise because relative risks for crashing increase even at low levels of alcohol consumption (Moskowitz et al., 2000; Taylor & Rehm, 2012) and because impaired driving is pervasive in the United States (Lacey et al., 2009). Approximately 3% of adults report driving after drinking too much to drive safely in the previous 30 days (Centers for Disease Control and Prevention, 2017), and it is estimated that adults age ≥18 years drive while impaired more than 100 million times each year (Bergen et al., 2011; Jewett et al., 2015).

Changes to environmental conditions can affect incidence of alcohol-impaired driving and alcohol-involved crashing. These changes can arise because of deliberate preventive interventions that aim to reduce the injury burden—such as the enactment of minimum legal drinking age laws (Wagenaar, 2002) and maximum legal blood alcohol concentration laws (Fell & Scherer, 2017)—or because of changes that arise naturally, such as global decreases in alcohol consumption within populations. One such organic change that could theoretically affect alcohol-involved motor vehicle crash incidence is the emergence of ridesharing companies, including Uber (Uber Technologies, Inc.) and Lyft (Lyft, Inc.). These companies use smartphone applications to connect owner/operator drivers to prospective passengers using GPS locations. Rideshare services are available in 263 cities in the United States, and providers have facilitated more than 11 billion rides since operations began around a decade ago (Uber newsroom, 2018; Zaveri, 2018). Transportation industry reports and other sources suggest that rideshare drivers now contribute up to 13% of vehicle miles traveled in some U.S. cities (Balding et al., 2019), and use of taxies and public transportation has decreased markedly since ridesharing began (Clewlow & Mishra, 2017; Cramer & Krueger, 2016).

Theory and observations of consumer behavior provide good reason to expect that ridesharing will more substantially reduce alcohol-impaired driving—and, therefore, alcohol-involved motor vehicle crash incidence—compared with conventional private transportation (i.e., taxis) and public transportation (e.g., trains, buses; Deaton & Muellbauer, 1980). Rational consumers weigh multiple factors when making purchases, including price, convenience, amenity value (e.g., comfort, cleanliness), functionality (e.g., ease and breadth of use), habit, and reliability. Taxis mostly rely on chance curbside meetings to link drivers and passengers, whereas ridesharing companies make these connections algorithmically and can use “surge pricing” (variable pricing based on momentary local conditions) to balance supply and demand (Uber newsroom, 2013). More efficient deployment of driver resources will theoretically minimize operating costs for rideshare companies compared with taxi companies, leading to lower prices for consumers (Azevedo & Weyl, 2016). Ridesharing may therefore be more convenient, more reliable, and cheaper than taxis. By contrast, ridesharing is typically more expensive than public transportation because these public services often receive subsidies and benefit from economies of scale. Nevertheless, ridesharing may be more reliable, be more convenient, and have greater amenity value compared with public transportation (Perk et al., 2008). Assuming that drivers are not alcohol affected (per rideshare company policies) (Lyft, 2021; Uber, 2020), compared with these alternative forms of transportation ridesharing could replace more trips by alcohol-impaired drivers and could therefore produce greater reductions in alcohol-involved crashes. On the other hand, drinkers are more impulsive and weigh risks and benefits of impaired driving differently compared with non-drinkers (Sloan et al., 2014), and impacts of ridesharing on impaired driving could be inconsistent with predictions guided by rationalist theories of consumer behavior.

Some empirical studies support the motivating theory. In difference-in-difference analyses, ridesharing was associated with fewer alcohol-involved crashes in 540 California townships (Greenwood & Wattal, 2017) and fewer drunk driving arrests and fatal crashes in U.S. counties (Dills & Mulhol-land, 2018). A time-series analysis found that ridesharing was associated with fewer alcohol-involved crashes in two of four U.S. cities (Morrison et al., 2018). However, other studies found no such evidence (Brazil & Kirk, 2020). For example, a study in the 100 most populous U.S. counties detected no association between ridesharing and alcohol-involved crash fatalities (Brazil & Kirk, 2016).

A common limitation for these previous ecological studies is that most used crude dichotomous variables to measure the presence or absence of rideshare services within space-time units (e.g., county-months [Brazil & Kirk, 2016]; city-weeks [Morrison et al., 2018]). Authors have used statistical (Nazif-Muñoz et al., 2020) and methodological (Morrison et al., 2018) solutions to account for gradual increases in rideshare usage, but aggregation bias could still affect these analyses and lead researchers to erroneously conclude that there is no relationship. Other problems related to confounding (e.g., by unmeasured differences between space-time units) and bias (e.g., by different enforcement of laws between units) could further affect ecological analyses.

Individual-level studies can address some of the limitations of ecological study designs (Morgenstern, 1995). However, no studies of ridesharing and alcohol-impaired crashes have used this approach, perhaps because it has other considerable challenges. In epidemiologic terms, an individual-level study could take alcohol-involved crashes as the outcome, rideshare availability as the exposure, and individual motor vehicle trips as the units of analysis. The major difficulty comes when selecting a comparison group.

The counterfactual for a trip that ends in an alcohol-involved crash is a trip that does not end in an alcohol-involved crash, taken by the same person (either as a driver or a passenger), at the same time, from the same origin to the same destination. Because counterfactuals are unobservable, selection of trips that do not end in alcohol-involved crashes can be relaxed, provided researchers can demonstrate that other factors known to cause alcohol-involved crashes are balanced between event and comparison groups. Given the myriad individual and environmental causes of alcohol-involved crashes (e.g., Fell et al., 2020; Lipton et al., 2021), credible claims of exchangeability—where exchanging the exposure state of an alcohol-involved crash trip and a comparison trip would result in the same outcome distribution (Greenland & Robbins, 1986)—are unlikely.

The aim of this study was to assess associations between ridesharing and alcohol-involved motor vehicle crash incidence. To address the limitations of previous ecological studies, we conducted an individual-level analysis by assessing spatially and temporally varying rideshare trip volume for a sample of alcohol-involved crashes. To overcome obstacles related to non-exchangeability, we used a case-series design comparing alcohol-involved crashes with non–alcohol-involved crashes. Case-series studies are an efficient epidemiologic approach for identifying subsets of cases that are etiologically heterogenous for an exposure of interest (Begg & Zhang, 1994). The design is conceptually similar to a case-control study, except that all included units are “cases” and comparison is made between case subtypes. Case-series designs do not allow researchers to infer that an exposure causes the outcome. Rather, they identify conditions that contribute to subtypes that are theoretically caused by different mechanisms—in this study, the relative contribution of rideshare availability to alcohol-involved crashes compared with non–alcohol-involved crashes.

Method

Study design

This case-series analysis was set in the City of Chicago, IL, which has a population of 2.7 million and covers a land area of 234 square miles. The cases were motor vehicle crashes that occurred in Chicago between November 1, 2018, and December 31, 2019. The subtype of interest was alcohol-involved crashes, and the comparison group was non–alcohol-involved crashes. The main exposure was the density of rideshare trips that were in progress at the time and location that each crash occurred.

Two main groups of variables could confound associations between rideshare trip density and alcohol involvement in motor vehicle crashes: time-varying conditions (e.g., vehicular traffic flow, hour of day, temperature, precipitation) and time-invariant conditions (e.g., roadway and other area characteristics). We controlled statistically for the time-varying conditions by including these measures as covariates in a multivariable model. Taxi trip density served as a proxy for vehicular traffic flow. We controlled methodologically for the time-invariant conditions by randomly selecting non–alcohol-involved crashes that were matched to alcohol-involved crashes at a ratio of 1:1 within census tracts. This matching procedure accounts for known and unknown confounders within census tracts that contribute differently to the geographic distribution of alcohol-involved and non–alcohol-involved crashes, such as alcohol outlet density, roadway characteristics, and aggregate driver characteristics and behavior (Lipton et al., 2021).

Selecting a small subset of comparison group crashes was important because the approach used to calculate rideshare trip density and taxi trip density (described below) was computationally very burdensome, and it was not feasible to include all motor vehicle crashes in the analytic data set. We excluded crashes and rideshare trips that occurred in the census tract containing O’Hare International Airport because crashes and trips to that location are very likely to have systematically less alcohol involvement compared with other crashes and trips.

Data and measures

Definitions. The outcome was a binary indicator for whether a motor vehicle crash was alcohol involved. In accordance with the Model Minimum Uniform Crash Criteria, Fifth Edition (National Highway Traffic Safety Administration, 2017), the Chicago Police Department reports incident-level information for all motor vehicle crashes that occurred in the city in which a person was killed or injured or there was at least $1,000 property damage. These data are publicly available through the City of Chicago Open Data Portal (City of Chicago Data Portal, 2020), including crash date, crash time, location coordinates, and primary and secondary contributory causes. Alcohol-involved crashes were identified as crashes where attending officers judged the primary or secondary cause as “under the influence of alcohol/drugs (use when arrest is effected)” or “had been drinking (use when arrest is not made).” Crashes eligible for selection as controls were those where the attending officers did not code the primary or secondary cause as alcohol/drug involved as defined above. Crashes with missing latitude or longitude data were excluded, including 0.7% of alcohol-involved crashes and 0.7% of non–alcohol-involved crashes.

Rideshare and taxi trip density. The City of Chicago requires Transportation Network Providers (i.e., rideshare companies) (Chicago, Ill., Mun. Code § 9-115-210, 2020) and Taxicab Licensees (Chicago, Ill., Mun. Code § 9-112-210, 2020) to report trip-level data to the Commissioner. The City of Chicago Open Data Portal makes these data publicly available for rideshare trips from November 2018 and for taxi trips from January 2013. To protect confidentiality, the trip origins and destinations are masked to the centroid of a census tract or, if two or fewer unique trips occur in the same census tract and 15-minute time window, to the centroid of a Chicago Community Area polygon. Trip start and end times are rounded to the nearest 15 minutes. The trip length in seconds is included in the data set. To account for trips that lasted less than 15 minutes and had identical start and end times, we created new start and end times for each trip by mean centering the trip duration around the single recorded start/end time.

We calculated the space-time specific density of rideshare trips and taxi trips corresponding to each included alcohol-involved and comparison crash using the following four steps: (a) selecting trips that were in progress in Chicago when the crash occurred; (b) estimating the route path for the selected trips; (c) calculating the continuous trip density separately for rideshare trips and taxi trips at that time across the extent of the city; and (d) extracting the density of ride-share trips and taxi trips at the crash point.

First, we selected trips that had a start time before the crash time and an end time after the crash time. Trips were linked to crashes with replacement (i.e., with a many-to-one relation) because trips could coincide with the occurrence of multiple crashes.

Second, we estimated route paths for the selected crashes using the Closest Facility Analysis tool in ArcGIS. The tool uses origin and destination coordinates to solve for the least-cost trip route through a specified network. A street centerline file for Chicago streets—obtained from Cook County Central and maintained by the Cook County government—provided the underlying network, and we used roadway distance as the cost parameter (Cook Central, 2019). This approach produced a line file describing the shortest network distance between trip origins and destinations for each trip that was in progress at the time of the crash (Figure 1). Using the ArcPy module and the Python multiprocessing module, we automated this portion of the analysis over a 48-core computer with 256 GB of RAM. The Closest Facility analysis took more than 4 weeks of continuous processing to calculate the least-cost path for 6,289,967 trips.

Figure 1.

Figure 1.

Open in a new tab

Estimated route path and kernel density layer for rideshare trips that were active at the time of one crash. Crash location denoted by the red star. Raster value interpreted as density of active rideshare trips per square mile at the time of the crash.

Third, we used a kernel density function to calculate a raster layer describing the continuous density of rideshare trip lines and taxi trip lines across the extent of Chicago. A raster cell size of 250 feet and search radius (i.e., bandwidth) of 1 mile (5,280 feet) were used to produce the kernel density layer. The kernel density analysis took 3 days to complete using multiprocessing.

Fourth, we extracted the value of the raster cells at the crash point to measure the density of rideshare trip density and taxi trip density per square mile at the time of the crash. All analyses used the NAD83 Geographic Coordinate

System and the NAD83 Illinois East (US ft) (EPSG 3435) projected coordinate system.

Covariates. Other time-varying conditions were time of day, day of week, public holiday, temperature, precipitation, and month. Information on hourly temperature and daily precipitation were obtained from the National Oceanic and Atmospheric Administration for the city of Chicago (National Oceanic and Atmospheric Association, 2020). Temperature was a continuous variable measured at the hour each crash occurred; precipitation was a binary variable wherein crashes that occurred on days where there was any rain or snowfall were coded as 1, and were otherwise coded as 0. Government holidays were accessed from the United States Office of Personnel Management (2019). School holidays were assessed using the 2018–2019 and 2019–2020 Chicago Public Schools Calendars (Chicago Public Schools, 2018, 2019). Government and school holidays were combined to create a binary holiday variable.

Statistical analysis

Conditional logistic regression models estimated the odds of a crash being alcohol involved:

graphic file with name jsad.2021.82.720eq1.jpg (1)

where p = Pr(Y = 1) for the binary variable Y, which was coded 1 when a crash was alcohol involved, and 0 otherwise. β0 is an overall model intercept, and βk is a fixed effect that conditions on census tract k to account for associations that may arise as artefacts of the matching procedure (Pearce, 2016). βn is a vector of independent coefficients estimating the linear relationship between n independent variables X‘ (i.e., rideshare trip density and covariates). Model 1 included only rideshare trip density; Model 2 included only the covariates (taxi trip density, temperature, holiday, precipitation, time of day, day of week, and month); and Model 3 combined rideshare trip density and the covariates. Additional analyses included interaction terms between rideshare trip density and time of day (Model 4) and day of week (Model 5) to consider whether associations differed temporally.

We present parameter estimates for all included variables to help readers assess the statistical methods, but we caution that the coefficients for taxi trip density and the time-varying covariates are not interpretable (Westreich & Greenland, 2013). Model fit was assessed using McFadden's pseudo R-square, the Akaike Information Criterion (AIC), and Bayesian Information Criterion (BIC). The pseudo R-square approximates the proportion of variance explained by the independent variables. Absolute values for the AIC and BIC are not meaningful, but comparison of these values between models is informative. We interpreted lower AIC and BIC values as indications of better model fit. Statistical analyses were conducted using SAS V9.4 (SAS Institute Inc., Cary, NC). This study involved no human subjects.

Results

There were 136,598 motor vehicle crashes that occurred in Chicago between November 2018 and December 2019, of which 1,006 (0.7%) were ineligible for inclusion because of missing latitude or longitude data. A total of 962 (0.7%) included crashes met the definition of being alcohol involved. We randomly selected 962 non–alcohol-involved comparison crashes that were matched to the alcohol-involved crashes within census tracts, yielding a total analytic sample of 1,924 crashes. City of Chicago records indicate that 129,312,195 rideshare trips and 19,498,014 taxi trips occurred during the 14-month study period.

Table 1 shows the distribution of rideshare trip density, taxi trip density, and covariates for alcohol-involved and comparison group crashes. Figure 2 shows the geographic distribution of crashes within the Chicago census tracts. At the time of each crash, there were between 124 and 9,112 active rideshare trips and between 13 and 1,064 active taxi trips. Rideshare trip density per square mile was 69.0 (SD = 129.7) at the location of alcohol-involved crashes and 105.6 (SD = 192.6) at the location of comparison group crashes. Taxi trip density per square mile was 5.4 (SD = 19.8) at the location of alcohol-involved crashes and 9.6 (SD = 29.2) at the location of comparison group crashes.

Table 1.

Descriptive statistics for alcohol-involved motor vehicle crashes and non–alcohol-involved motor vehicle crashes

graphic file with name jsad.2021.82.720tbl1.jpg

Variable Alcohol-involved crashes (n = 962) M (SD) Non-alcohol-involved crashes (n = 962) M (SD) All crashes (n = 136,598) M (SD)
Vehicles for hire
 Rideshare trips/mile2 69.00(129.70) 105.60 (192.60) n.a
 Taxi trips/mile2 5.44 (19.79) 9.55 (29.20) n.a
Time-varying characteristics
 Temperature (° F) 48.27 (19.69) 50.62 (20.97) 50.72 (21.07)
N (%) N (%) N (%)
 Holiday 108 (11.23) 60 (6.24) 11,004 (8.06)
 Precipitation 150 (15.59) 176 (18.30) 24,904(18.23)
Time of day
 11:00 a.m. - 4:59 p.m. 121 (12.58) 351 (36.49) 51,855 (37.96)
 5:00 p.m. - 10:59 p.m. 307 (31.9) 284 (29.52) 38,364 (28.09)
 11:00 p.m. - 4:59 a.m. 437 (45.43) 90 (9.36) 13,900 (10.18)
 5:00 a.m. - 10:59 a.m. 97 (10.08) 237 (24.64) 32,479 (23.78)
Day of week
 Sunday 200 (20.79) 120 (12.47) 16,720 (12.24)
 Monday 109 (11.33) 135 (14.03) 19,332 (14.15)
 Tuesday 92 (9.56) 155 (16.11) 19,828 (14.52)
 Wednesday 92 (9.56) 133 (13.83) 18,861 (13.81)
 Thursday 108 (11.23) 134 (13.93) 19,794 (14.49)
 Friday 146 (15.18) 151 (15.70) 21,772 (15.94)
 Saturday 215 (22.35) 134 (13.93) 20,291 (14.85)
Month
 November 2018 71 (7.38) 67 (6.96) 9,406 (6.86)
 December 2018 63 (6.55) 67 (6.96) 9,981 (7.31)
 January 2019 53 (5.51) 80 (8.32) 9,081 (6.65)
 February 2019 64 (6.65) 57 (5.93) 8,586 (6.29)
 March 2019 70 (7.28) 75 (7.80) 9,710 (7.11)
 April 2019 66 (6.86) 60 (6.24) 9,403 (6.88)
 May 2019 63 (6.55) 85 (8.84) 10,666 (7.81)
 June 2019 54(5.61) 94 (9.77) 10,656 (7.80)
 July 2019 75 (7.80) 69 (7.17) 10,597 (7.76)
 August 2019 73 (7.59) 67 (6.96) 9,894 (7.24)
 September 2019 79 (8.21) 59 (6.13) 9,777 (7.16)
 October 2019 73 (7.59) 66 (6.86) 9,892 (7.24)
 November 2019 74 (7.69) 65 (6.76) 9,585 (7.02)
 December 2019 84 (8.73) 51 (5.30) 9,364 (6.86)
Open in a new tab

Notes: n.a. = not available.

Figure 2.

Figure 2.

Open in a new tab

Geographic distribution of motor vehicle crashes within Chicago census tracts; November 2018–December 2019

Table 2 presents the results from the logistic regression models that included fixed effects to condition on the spatial units in which alcohol-involved and comparison group crashes were matched. In the bivariate Model 1, an increase of 1 rideshare trip per square mile was associated with 0.5% decreased odds that a crash was alcohol involved (odds ratio [OR] = 0.995, 95% CI [0.994, 0.997]). The association was attenuated toward null in Model 3 after controlling for the covariates, but the confidence interval provided evidence in favor of rejecting the null hypothesis. Specifically, in Model 3, an increase of 1 rideshare trip per square mile was associated with 0.2% decreased odds that a crash was alcohol involved (OR = 0.998; 95% CI [0.996, 0.999]). Rescaling the associations from Model 3 demonstrates that a standard deviation increase in rideshare trip density was associated with 23% reduced odds that a crash was alcohol involved (OR = 0.771, 95% CI [0.594, 0.878]).

Table 2.

Conditional logistic regression models for the odds that a crash is alcohol involved (n = 1,924)

graphic file with name jsad.2021.82.720tbl2.jpg

Variable Model 1 OR [95% CI] Model 2 OR [95% CI] Model 3 OR [95% CI]
Vehicles for hire
 Rideshare trips/mile2 0.995 [0.994, 0.997] 0.998 [0.996, 0.999]
 Taxi trips/mile2 0.991 [0.984, 0.999] 1.000 [0.991, 1.009]
Time-varying characteristics
 Temperature (° F) 0.987 [0.976, 0.999] 0.987 [0.975, 0.999]
 Holiday 2.793 [1.824, 4.278] 2.682 [1.747, 4.117]
 Precipitation 0.621 [0.460, 0.839] 0.633 [0.468, 0.857]
Time of day
 11:00 a.m. - 4:59 p.m. [referent]
 5:00 p.m. - 10:59 p.m. 2.975 [2.224, 3.980] 3.110 [2.319, 4.170]
 11:00 p.m. - 4:59 a.m. 12.830 [8.997, 18.297] 11.854 [8.288, 16.954]
 5:00 a.m. - 10:59 a.m. 0.999 [0.701, 1.425] 0.996 [0.697, 1.422]
Day of week
 Sunday [referent]
 Monday 0.436 [0.283, 0.673] 0.433 [0.280, 0.670]
 Tuesday 0.506 [0.335, 0.764] 0.495 [0.327, 0.747]
 Wednesday 0.518 [0.338, 0.794] 0.521 [0.340, 0.799]
 Thursday 0.565 [0.375, 0.851] 0.560 [0.371, 0.846]
 Friday 0.618 [0.416, 0.917] 0.636 [0.427, 0.946]
 Saturday 1.175 [0.802, 1.721] 1.257 [0.855, 1.847]
Month
 January [referent]
 February 1.550 [0.838, 2.867] 1.502 [0.808, 2.794]
 March 1.683 [0.916, 3.093] 1.672 [0.906, 3.085]
 April 2.226 [1.095, 4.527] 2.178 [1.064, 4.456]
 May 2.018 [0.969, 4.203] 1.996 [0.957, 4.161]
 June 1.096 [0.486, 2.472] 1.084 [0.479, 2.455]
 July 3.230 [1.341, 7.781] 3.106 [1.285, 7.505]
 August 3.211 [1.378, 7.484] 3.148 [1.346, 7.360]
 September 3.239 [1.411, 7.431] 3.148 [1.346, 7.265]
 October 2.271 [1.131, 4.559] 2.211 [1.097, 4.458]
 November 1.653 [0.951, 2.870] 1.613 [0.925, 2.810]
 December 2.125 [1.230, 3.671] 2.069 [1.193, 3.589]
McFadden pseudo Ä-square 0.036 0.234 0.238
AIC 1,702.513 1,305.095 1,297.278
BIC 1,708.075 1,438.575 1,436.319
Open in a new tab

Notes: OR = odds ratio; CI = confidence interval; AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion.

The marginally higher pseudo R-square value and marginally lower AIC and BIC values for Model 3 compared with Model 2 indicate that rideshare trip density accounted for a very small proportion of overall variance in the odds that crashes were alcohol involved. Associations for Models 4 and 5 (Supplemental Table S2) show that interaction terms for time of day and day of week were mostly unrelated to the outcome, except that rideshare trip density was more strongly associated with decreased odds that crashes were alcohol involved on Tuesdays compared with Sundays (OR = 0.994, 95% CI [0.989, 0.999]). (Supplemental material appears as an online-only addendum to this article on the journal's website.)

Discussion

Alcohol-involved motor vehicle crashes are a major contributor to the injury burden in the United States (Bureau of Transportation Statistics, 2019). Ridesharing is altering public and private motor vehicle use in this country and globally, including, perhaps, by replacing some trips by alcohol-impaired drivers. This case-series study assessed associations between rideshare trip density and alcohol-involved motor vehicle crash incidents in the City of Chicago. Consistent with the hypothesis that ridesharing replaces impaired-driver trips, we found that the density of active rideshare trips near a crash site was associated with decreased odds that the crash was alcohol involved. Relative associations for each increase of 1 rideshare trip per square mile were small, but given the wide variation in rideshare trip density and the large volumes of alcohol-involved crashes and rideshare trips, absolute impacts could be considerable (Gruenewald et al., 2018).

These results provide evidence of etiologic heterogeneity with respect to ridesharing and alcohol involvement in motor vehicle crashes (Khoury & Flanders, 1996). Because inclusion in the study sample as an alcohol-involved or comparison group crash was conditional upon a crash having occurred, the parameter estimates measure the different associations between rideshare exposure and a crash being alcohol involved or non–alcohol involved. Begg and Zhang (1994) demonstrated that the measure of association from a case-series study is equivalent to the ratio of the relative risk for the occurrence of one case subtype (in our study, crashes with alcohol involvement) to the relative risk for the occurrence of the other case subtype (crashes without alcohol involvement).

Interpreted causally, our results either provide evidence that ridesharing reduced alcohol-involved crashes, or that ridesharing is associated with changes in all motor vehicle crashes of unknown direction and magnitude but with different effect sizes for alcohol-involved versus non–alcohol-involved crashes. Prior studies of ridesharing and all motor vehicle crashes have mixed results (Brazil & Kirk, 2016; Dills & Mulholland, 2018; Huang et al., 2019), so the correct interpretation is not clear. Further analyses using case-control or other epidemiologically rigorous designs will help disentangle these complex associations.

The theoretical mechanism linking ridesharing and motor vehicle crashes is predicated on rational choice theory and observations of consumer behavior. Many people will prefer ridesharing compared with public transportation (e.g., buses, trains) and commercial private transportation (e.g., taxis) because of relatively lower financial costs, relatively lower convenience costs, and/or favorable assessment of other drivers of consumer behavior, such as amenity and reliability. When faced with a choice between driving while alcohol impaired and taking alternate means of transportation, improved access to ridesharing could tip the balance away from impaired driving to a greater extent than conventional public and private transportation. Despite the limits of rationalist frameworks for explaining decisions by impaired drivers (Sloan et al., 2014), our results are consistent with this mechanism. Rideshare trip density was associated with decreased odds of alcohol-involved crashes, even after controlling for taxi trip density. Because we conditioned methodologically on geographic location and statistically on time of day, access to public transit was similar for alcohol-involved and comparison group crashes.

This research is an important advance compared with previously published studies of ridesharing and alcohol-involved motor vehicle crashes. Prior studies all used spatial ecological designs, in which outcomes of interest—such as counts of impaired driver crashes (Brazil & Kirk, 2016) and driving under the influence arrests (Dills & Mulholland, 2018)—were aggregated within space-time units. Access to ridesharing was mostly measured as binary variables indicating the presence or absence of any ridesharing services within these units. Spatial ecological designs have the advantage that data are readily available; however, the crude approach to measuring access to ridesharing could introduce aggregation bias that attenuates point estimates and inflates standard errors.

The small relative associations identified within our highly spatially and temporally precise design may not be detectable when aggregated across cities, counties, or states. Nevertheless, we caution against committing the atomistic error, in which individual- or event-level studies are erroneously generalized to the population level (Diez Roux, 2002). For example, greater rideshare density may coincide with decreased rideshare availability within geographic areas, because drivers are servicing other passengers. Innovative ecological designs are required to determine whether the identified associations are observable on aggregate.

This work has important public health implications. Ridesharing has increased rapidly over the last several years across the United States, including by approximately 6% in Chicago during the short 14-month study period (Table S1). This one part of the natural evolution of the transportation market appears to have public health benefits, and communities could benefit from lower alcohol-involved crash incidence simply by allowing rideshare companies to operate. Some municipalities have attempted to enhance these hypothesized effects by subsidizing rideshare trips for impaired drivers. For example, the Townships of Evesham and Voorhees provided free rideshare trips home from bars for local residents between 2016 and 2018, which reduced nighttime injury crash incidence by 38% and averted 371 injury crashes (Humphreys et al., 2020). Nevertheless, the impacts of these deliberate interventions are not fully clear. Over 17 weeks in 2017, Anheuser-Busch and Lyft provided more than 33,000 free trips from bar districts in Columbus, Ohio, but an observational evaluation found the intervention was not associated with reductions in impaired driving or motor vehicle crash incidence (Miller et al., 2020).

Important areas for further research are examining the benefits of ridesharing against possible adverse consequences, such as assessing increases in alcohol consumption (Teltser et al., 2021; Zhou, 2020) and increases in crash incidence for vulnerable road users (Morrison et al., 2021); assessing possible moderation by environmental conditions, including impaired driver laws (Nazif-Muñoz et al., 2020) and access to alternative forms of transportation; and assessing different effects according to driver characteristics (e.g., sex, age, history of arrest for driving while impaired).

Results of this study should be considered in light of some important limitations. A case-series design is less susceptible to problems related to non-exchangeability than a case-control design because the comparison group is effectively selected from within a small stratum of similar units (Greenland & Robbins, 1986), but the possibility remains that alcohol-involved crashes and non–alcohol-involved crashes could arise from different source populations. For example, there may be systematic differences in crash risks and impaired driving behavior for people with underlying health conditions (Dischinger et al., 2000). If these differences are also related to rideshare trip density, the associations of interest could be confounded by driver characteristics, unmeasured time-varying characteristics, and perhaps by environmental conditions that vary at scales other than census tracts.

Police assessment of driver impairment is known to be unreliable (Rubenzer, 2011), and some crashes may have been misclassified with regard to alcohol involvement. Nevertheless, such measurement error and selection bias are likely attenuate associations toward null, so are unlikely to have affected our main conclusions. Finally, our measure of rideshare trips density uses an estimate of trip route paths based on the shortest network distance. Systematic differences between estimated trips routes and driver behavior (e.g., because of local knowledge, traffic congestion) could bias results in either direction. We were unable to test alternate specifications of the rideshare trip density variable (e.g., including trips to O’Hare International Airport, selecting different bandwidth for the kernel density calculation) because of the sizable computational burden.

Alcohol-involved motor vehicle crashes have a considerable human and economic costs in the United States. Ridesharing has emerged over the last decade as technology with the potential to reduce this impact, albeit without clear empirical support. Despite the scarcity of scientific evidence, the National Academies of Sciences, Engineering, and Medicine Committee on Accelerating Progress to Reduce Alcohol Impaired Driving Fatalities recently recommended that regulators promote ridesharing as an alternative means of transportation to prevent impaired driving. Specifically, a recent report states, “Municipalities should support policies and programs that increase the availability, convenience, affordability, and safety of transportation alternatives for drinkers who might otherwise drive. This includes permitting transportation network company ridesharing, [and] enhancing public transportation options (especially during nighttime and weekend hours)” (National Academies of Sciences, Engineering, and Medicine, 2018, p. 233). The results of this study support that recommendation.

Acknowledgments

The authors thank Gregory Yetman of The Earth Institute, Columbia University, for technical advice regarding geoprocessing for this study.

Footnotes

This study received funding from the Centers for Disease Control and Prevention (R49-CE003094).

References

  1. Azevedo E. M., Weyl E. G. Matching markets in the digital age. Science. 2016;352:1056–1057. doi: 10.1126/science.aaf7781. doi:10.1126/science.aaf7781. [DOI] [PubMed] [Google Scholar]
  2. Balding M., Whinery T., Leshner E., Womeldorff E. Estimated percent of total driving by Lyft and Uber. 2019 Retrieved from https://www.urbanismnext.org/resources/estimated-percent-of-total-driving-by-lyftand-uber. [Google Scholar]
  3. Begg C. B., Zhang Z. F. Statistical analysis of molecular epidemiology studies employing case-series. Cancer Epidemiology, Biomarkers & Prevention. 1994;3:173–175. [PubMed] [Google Scholar]
  4. Bergen G., Shults R., Rudd R., & the Centers for Disease Control and Prevention Vital signs: Alcohol-impaired driving among adults—United States, 2010. Morbidity and Mortality Weekly Report. 2011 October 7;60:1351–1356. [PubMed] [Google Scholar]
  5. Brazil N., Kirk D. S. Uber and metropolitan traffic fatalities in the United States. American Journal of Epidemiology. 2016;184:192–198. doi: 10.1093/aje/kww062. doi:10.1093/aje/kww062. [DOI] [PubMed] [Google Scholar]
  6. Brazil N., Kirk D. Ridehailing and alcohol-involved traffic fatalities in the United States: The average and heterogeneous association of Uber. PLoS One. 2020;15:e0238744. doi: 10.1371/journal.pone.0238744. doi:10.1371/journal.pone.0238744. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Bureau of Transportation Statistics. Washington, DC: Department U.S. of Transportation; 2019. Fatalities by highest driver blood alcohol concentration (BAC) in highway crashes. [Google Scholar]
  8. Centers for Disease Control and Prevention. Atlanta, GA: Department U.S. of Health & Human Services; 2017. BRFSS prevalence & trends data. Retrieved from https://www.cdc.gov/brfss/brfssprevalence/index.html. [Google Scholar]
  9. Chicago, Ill., Mun. Code § 9-112-210. Duty to maintain real time records as to a chauffeur operating a taxicab. 2020 [Google Scholar]
  10. Chicago, Ill., Mun. Code § 9-115-210. Records and reports. 2020 [Google Scholar]
  11. Chicago Public Schools. Chicago, IL: Author; 2018. 2018–2019 CPS calendar elementary and high schools. Retrieved from https://www.nctq.org/dmsView/SY18-19Calendar_English. [Google Scholar]
  12. Chicago Public Schools. Chicago, IL: Author; 2019. 2019–2020 CPS calendar elementary and high schools. Retrieved from https://www.collinsacademy.org/pdfs/CPS%20Calendar%202019-20.pdf. [Google Scholar]
  13. City of Chicago Data Portal. Chicago, IL: City of Chicago; 2020. Traffic Crashes—Crashes [Data file] Retrieved from https://data.cityofchicago.org/Transportation/Traffic-Crashes-Crashes/85ca-t3if. [Google Scholar]
  14. Clewlow R., Mishra G. S. Disruptive transportation: The adoption, utilization, and impacts of ride-hailing in the United States. UC Davis Institute of Transportation Studies (Report No. UCD-ITS-RR-17-07) 2017 [Google Scholar]
  15. Central Cook. Chicago, IL: Cook County Government; 2019. Streets [Data file] Retrieved from https://hub-cookcountyil.opendata.arcgis.com/datasets/4569d77e6d004c0ea5fada54640189cf_5/data. [Google Scholar]
  16. Cramer J., Krueger A. Disruptive change in the taxi business: The case of Uber. American Economic Review. 2016;106:177–182. doi:10.1257/aer.p20161002. [Google Scholar]
  17. Deaton A., Muellbauer J. Cambridge, England: Cambridge University Press; 1980. Economics and consumer behavior. [Google Scholar]
  18. Diez Roux A. V. A glossary for multilevel analysis. Journal of Epidemiology and Community Health. 2002;56:588–594. doi: 10.1136/jech.56.8.588. doi:10.1136/jech.56.8.588. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Dills A. K., Mulholland S. E. Ride-sharing, fatal crashes, and crime. Southern Economic Journal. 2018;84:965–991. doi:10.1002/soej.12255. [Google Scholar]
  20. Dischinger P. C., Ho S. M., Kufera J. A. Medical conditions and car crashes. Annual Proceedings. Association for the Advancement of Automotive Medicine. 2000;44:335–346. [PMC free article] [PubMed] [Google Scholar]
  21. Fell J. C., Scherer M. Estimation of the potential effectiveness of lowering the blood alcohol concentration (BAC) limit for driving from 0.08 to 0.05 grams per deciliter in the United States. Alcoholism: Clinical and Experimental Research. 2017;41:2128–2139. doi: 10.1111/acer.13501. doi:10.1111/acer.13501. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Fell J. C., Scolese J., Achoki T., Burks C., Goldberg A., DeJong W. The effectiveness of alternative transportation programs in reducing impaired driving: A literature review and synthesis. Journal of Safety Research. 2020;75:128–139. doi: 10.1016/j.jsr.2020.09.001. doi:10.1016/j.jsr.2020.09.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Greenland S., Robins J. M. Identifiability, exchangeability, and epidemiological confounding. International Journal of Epidemiology. 1986;15:413–419. doi: 10.1093/ije/15.3.413. doi:10.1093/ije/15.3.413. [DOI] [PubMed] [Google Scholar]
  24. Greenwood B. N., Wattal S. Show me the way to go home: An empirical investigation of ride-sharing and alcohol related motor vehicle fatalities. Management Information Systems Quarterly. 2017;41:163–187. doi:10.25300/MISQ/2017/41.1.08. [Google Scholar]
  25. Gruenewald P. J., LaScala E. A., Ponicki W. R. Identifying the population sources of alcohol impaired driving: An assessment of context specific drinking risks. Journal of Studies on Alcohol and Drugs. 2018;79:702–709. doi: 10.15288/jsad.2018.79.702. doi:10.15288/jsad.2018.79.702. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Huang J. Y., Majid F., Daku M. Estimating effects of Uber ride-sharing service on road traffic-related deaths in South Africa: A quasi-experimental study. Journal of Epidemiology and Community Health. 2019;73:263–271. doi: 10.1136/jech-2018-211006. doi:10.1136/jech-2018-211006. [DOI] [PubMed] [Google Scholar]
  27. Humphreys D. K., Degli Esposti M., Williams F. M., Kondo M. C., Morrison C. Assessing the impact of a local community subsidised rideshare programme on road traffic injuries: An evaluation of the Evesham Saving Lives Programme. Injury Prevention. Advance online publication. 2020 doi: 10.1136/injuryprev-2020-043728. doi:10.1136/injuryprev-2020-043728. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Jewett A., Shults R. A., Banerjee T., Bergen G. Alcohol-impaired driving among adults—United States, 2012. Morbidity and Mortality Weekly Report. 2015;64:814–817. doi: 10.15585/mmwr.mm6430a2. doi:10.15585/mmwr.mm6430a2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Khoury M. J., Flanders W. D. Nontraditional epidemiologic approaches in the analysis of gene-environment interaction: Case-control studies with no controls! American Journal of Epidemiology. 1996;144:207–213. doi: 10.1093/oxfordjournals.aje.a008915. doi:10.1093/oxfordjournals.aje.a008915. [DOI] [PubMed] [Google Scholar]
  30. Lacey J., Kelley-Baker T., Furr-Holden D., Voas R. B., Romano E., Ramirez A., Berning A. Washington, DC: Department U.S. of Transportation, National Highway Traffic Safety Administration; 2009. 2007 National Roadside Survey of Alcohol and Drug Use by Drivers: Alcohol results (Report No. DOT HS 811 248) Retrieved from https://rosap.ntl.bts.gov/view/dot/1888/dot_1888_DS1.pdf? [Google Scholar]
  31. Lipton R., Banerjee A., Ponicki W. R., Gruenewald P. J., Morrison C. Impacts of confounding roadway characteristics on estimates of associations between alcohol outlet densities and alcohol-related motor vehicle crashes. Drug and Alcohol Review. 2021;40:239–246. doi: 10.1111/dar.13156. doi:10.1111/dar.13156. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Lyft. San Francisco, CA: Author; 2021. Zero-tolerance drug and alcohol policy. Retrieved from https://help.lyft.com/hc/en-ca/articles/115012926187-Zero-tolerance-Drug-and-Alcohol-Policy. [Google Scholar]
  33. Miller T. R., Courser M., Shamblen S. R., Lange J. E., Tippetts A. S., Ringwalt C. Efficacy and cost-effectiveness of subsidized ridesharing as a drunk driving intervention in Columbus, OH. Accident Analysis and Prevention. 2020;146:105740. doi: 10.1016/j.aap.2020.105740. doi:10.1016/j.aap.2020.105740. [DOI] [PubMed] [Google Scholar]
  34. Morgenstern H. Ecologic studies in epidemiology: Concepts, principles, and methods. Annual Review of Public Health. 1995;16:61–81. doi: 10.1146/annurev.pu.16.050195.000425. doi:10.1146/annurev.pu.16.050195.000425. [DOI] [PubMed] [Google Scholar]
  35. Morrison C. N., Jacoby S. F., Dong B., Delgado M. K., Wiebe D. J. Ridesharing and motor vehicle crashes in 4 US cities: An interrupted time-series analysis. American Journal of Epidemiology. 2018;187:224–232. doi: 10.1093/aje/kwx233. doi:10.1093/aje/kwx233. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Morrison C. N., Mehranbod C., Kwizera M., Rundle A. G., Keyes K. M., Humphreys D. K. Ridesharing and motor vehicle crashes: A spatial ecological case-crossover study of trip-level data. Injury Prevention. 2021;27:118–123. doi: 10.1136/injuryprev-2020-043644. doi:10.1136/injuryprev-2020-043644. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Moskowitz H., Burns M., Fiorentino D., Smiley A., Zador P. Washington, DC: US Department of Transportation, National Highway Traffic Safety Administration; 2000. Driver characteristics and impairment at various BACs (Report No. DOT HS 809 075) [Google Scholar]
  38. National Academies of Sciences, Engineering, and Medicine. Washington, DC: National Academies Press; 2018. Getting to zero alcohol-impaired driving fatalities: A comprehensive approach to a persistent problem. [PubMed] [Google Scholar]
  39. National Center for Statistics and Analysis. Washington, DC: Department U.S. of Transportation; 2020. Traffic safety facts research note: Overview of motor vehicle crashes (Report No. DOT HS 813 060) [Google Scholar]
  40. National Highway Traffic Safety Administration. Washington, DC: Department U.S. of Transportation; 2017. Model minimum uniform crash criteria. [Google Scholar]
  41. National Oceanic and Atmospheric Administration. Local climatological data station details. National Centers for Environmental Information. 2020 Retrieved from https://www.ncei.noaa.gov/products/land-based-station/local-climatological-data. [Google Scholar]
  42. Nazif-Muñoz J. I., Batomen B., Oulhote Y., Spengler J., Nandi A. State or market? How to effectively decrease alcohol-related crash fatalities and injuries. Journal of Epidemiology and Community Health. 2020;74:502–509. doi: 10.1136/jech-2019-213191. doi:10.1136/jech-2019-213191. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Pearce N. Analysis of matched case-control studies. BMJ. 2016;352:i969. doi: 10.1136/bmj.i969. doi:10.1136/bmj.i969. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Perk V., Flynn J., Volinski J. National Center for Transit Research (Report No. NCTR-776-07) Tallahassee, FL: State of Florida Department of Transportation; 2008. Transit ridership, reliability, and retention. Retrieved from https://www.nctr.usf.edu/pdf/77607.pdf. [Google Scholar]
  45. Rubenzer S. Judging intoxication. Behavioral Sciences & the Law. 2011;29:116–137. doi: 10.1002/bsl.935. doi:10.1002/bsl.935. [DOI] [PubMed] [Google Scholar]
  46. Sloan F. A., Eldred L. M., Xu Y. The behavioral economics of drunk driving. Journal of Health Economics. 2014;35:64–81. doi: 10.1016/j.jhealeco.2014.01.005. doi:10.1016/j.jhealeco.2014.01.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Taylor B., Rehm J. The relationship between alcohol consumption and fatal motor vehicle injury: High risk at low alcohol levels. Alcoholism: Clinical and Experimental Research. 2012;36:1827–1834. doi: 10.1111/j.1530-0277.2012.01785.x. doi:10.1111/j.1530-0277.2012.01785.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Teltser K., Lennon C., Burgdorf J. Do ridesharing services increase alcohol consumption? Journal of Health Economics. 2021;77:102451. doi: 10.1016/j.jhealeco.2021.102451. doi:10.1016/j.jhealeco.2021.102451. [DOI] [PubMed] [Google Scholar]
  49. Uber. San Francisco, CA: Author; 2020. Zero tolerance policy. Retrieved from https://www.uber.com/legal/en/document/?country=united-states&lang=en&name=zero-tolerance-policy. [Google Scholar]
  50. Uber Newsroom. San Francisco, CA: Uber, Inc; 2013. Getting more for less on uberX. Retrieved from https://newsroom.uber.com/us-dc/getting-more-for-less-on-uberx/ [Google Scholar]
  51. Uber Newsroom. San Francisco, CA: Uber, Inc; 2018. 10 Billion. Retrieved from https://www.uber.com/newsroom/10-billion/United. [Google Scholar]
  52. United States Office of Personnel Management. Policy, data, oversight. Washington, DC: Author; 2019. Retrieved from https://www.opm.gov/policy-data-oversight/pay-leave/federal-holidays. [Google Scholar]
  53. Wagenaar A. C. Effects of minimum drinking age laws: Review and analyses of the literature from 1960 to 2000. Journal of Studies on Alcohol. 2002;Supplement 14:206–225. doi: 10.15288/jsas.2002.s14.206. doi:10.15288/jsas.2002.s14.206. [DOI] [PubMed] [Google Scholar]
  54. Westreich D., Greenland S. The table 2 fallacy: Presenting and interpreting confounder and modifier coefficients. American Journal of Epidemiology. 2013;177:292–298. doi: 10.1093/aje/kws412. doi:10.1093/aje/kws412. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Zaloshnja E., Miller T. R., Blincoe L. J. Costs of alcohol-involved crashes. United States. Annals of Advances in Automotive Medicine. 2013;2010;57:3–12. [PMC free article] [PubMed] [Google Scholar]
  56. Zaveri P. CNBC; 2018. Sep 18, Lyft has now delivered 1 billion rides. Retrieved from https://www.cnbc.com/2018/09/18/lyft-hits-1-billion-rides.html. [Google Scholar]
  57. Zhou Y. Ride-sharing, alcohol consumption, and drunk driving. Regional Science and Urban Economics. 2020;85:103594. doi:10.1016/j.regsciurbeco.2020.103594. [Google Scholar]

Articles from Journal of Studies on Alcohol and Drugs are provided here courtesy of Rutgers University. Center of Alcohol Studies

RESOURCES