Off the Kuff Rotating Header Image


More on Uber and drunk driving

We have a study.

In this work, we investigate how the entry of the driving service Uber influences the rate of alcohol related motor vehicle homicides. While significant debate has surrounded the entry of driving services like Uber and Lyft, limited rigorous empirical work has been devoted to uncovering the social benefits of such services (or the mechanism which drives these benefits). Using a difference in difference approach to exploit a natural experiment, the entry of Uber into markets in California between 2009 and 2013, findings suggest a significant drop in the rate of homicides during that time. Furthermore, results suggest that not all services offered by Uber have the same effect, insofar as the effect for the Uber Black car service is intermittent and manifests only in selective locations. These results underscore the coupling of increased availability with cost savings which are necessary to exploit the public welfare gains offered by the sharing economy. Practical and theoretical implications are discussed within.

From the introduction:


Preliminary analysis conducted by Uber and several industry analysts suggest that introduction of Uber and other ride sharing services has a negative influence on DUI arrests. However, these studies have been questioned on several grounds: including involvement of Uber in the data analysis, methodological rigor (i.e. single city estimations), and the presence of confounding factors such as changes in city’s population, bar scene, and tougher enforcement.

Moreover, a limited understanding of the mechanisms by which such services influence the rate of DUIs exists. On one hand, it is plausible that the decrease in DUI is simply the result of availability of vehicles for hire and that patrons are willing to pay a price premium for such services. Insofar as it is often difficult to hire a taxi, based on time, location, or even the race of the patron (Meeks 2010), it is plausible that the presence of the platform mitigates these market inefficiencies by soliciting the driver electronically, thereby significantly reducing search costs (Parker and Van Alstyne 2005) and creating excess utility for the consumer. On the other hand, it is equally plausible that the effect is a result of both availability and cost. Drawing from rational choice theory (Clarke and Cornish 1985, Cornish and Clarke 2014) it is conceivable that individuals who make the decision to drive under the influence do so based on the costs associated with conviction, the cost of searching for and hiring a taxi, and the probability of being stopped by the police and/or striking another driver. This broad question: what is the impact of Uber’s introduction on alcohol related motor vehicle homicides in the local area and by what mechanisms, forms the core of the research investigated in this paper.

Empirically, we exploit a natural experiment, the introduction of the ride sharing service Uber into cities in the State of California between 2009 and 2014, to investigate the effect. Leveraging this econometric setup offers us several advantages. First, to the extent that the entrance of Uber is staggered temporally and geographically, we execute a difference in difference estimation to establish the effect. Second, Uber offers multiple services in each of the treated areas with varying price points (note that these services also enter at varying times and orders). On one hand, UberBlack, a town car service, offers transportation with a significant markup over taxicabs (~20% – ~30% price premium). On the other, the UberX service is a personalized driving service which offers significant discounts over taxis (~20% – ~30% price reductions). To the degree that each of these services identifies a different mechanism being at play (availability v. availability and price point), we are able to cleanly identify the dominant mechanisms at play. We test these using hand collected data from the California Highway Patrol (CHP) safety and crash dataset and a custom webscraper which indicates when each service entered a geographic area in California.

Results indicate that while the entry of UberX strongly and negatively affects the number of motor vehicle homicides which occur in townships, limited evidence exists to support previous claims that this occurs with the Uber Black car service as well (indicating that prior claims about the efficacy of Uber may have been overstated (Badger 2014)). Further, results indicate that the time for such effects to manifest vary is significant (upwards of 9 – 15 months). These results are robust to a variety of estimations (e.g. OLS, Poisson, and Quasi-Maximum Likelihood count models) and operationalizations. Finally, findings suggest an absence of a heterogeneous pre-treatment homicide trend in treated locations, indicating that the primary assumptions of the difference in difference model are not violated (Angrist and Pischke 2008, Bertrand et al. 2002). Further, results suggest no effect of Uber when surge pricing is likely in effect, thereby underscoring the importance of cost considerations. Economically, results indicate that the entrance of Uber X results in a 3.6% – 5.6% decrease in the rate of motor vehicle homicides per quarter in the state of California. With more than 1000 deaths5 occurring in California due to alcohol related car crashes every year, this represents a substantial opportunity to improve public welfare and save lives.

Emphasis mine, as that’s the key point. The paper overall is fairly technical and written for an academic audience, but you get the idea. We’ve discussed this issue before, and now that there appears to be some solid evidence behind the idea that Uber (and presumably Lyft) can help reduce the vehicular homicide rate in a city where those services are legal, it’s a factor that needs to be taken into account when they are debated. Sure, the people who are now using Uber to get themselves home after a night on the town could have called a cab in the past, but they didn’t. Uber is the option they prefer, and it comes with a benefit. Of course, Uber also comes with its own problems that haven’t been fixed yet, too, so this isn’t a debate ender. It’s just another point to keep in mind. Link via Streetsblog.

Murder by numbers 2013

The beginning of the new year means a look back at the homicide count for the previous year.

Homicides are up in unincorporated Harris County, where the Sheriff’s Office is reporting a nearly 20 percent uptick in 2013, preliminary year-end statistics show.

Killings in 2013 totaled 91 as of Tuesday – the second-highest tally in the past five years, and about a 19.7 percent increase from 2012, according to the Harris County Sheriff’s Office.

Authorities said the 2013 figure appears to have been driven by a cluster of cases involving multiple victims.

Sheriff Adrian Garcia cited a Nov. 9 case at a Cypress house party where two high school students were fatally shot and 19 others were wounded. He also recalled an incident Nov. 20 in which a gunman shot five people at a northwest Harris County apartment complex. Three died.

“We don’t see that as a particular pattern,” Garcia said of the multi-victim cases. “These are just circumstances that have occurred this year and we hope they never repeat themselves.”

In Houston, preliminary data showed the homicide count was down from 2012, which ended with 217.

As of Dec. 20, the Houston Police Department recorded 199 slayings compared with 207 for the same time last year, according to Homicide Division Capt. Dwayne Ready. HPD’s latest reports show about a 3.8 percent decrease.

If the 2013 total remains below 217, it would be the second lowest since 1965, when 139 people were killed, HPD officials said. The lowest since that date was in 2011, which had 198.


Violent crime overall has been trending down for several years, both nationally and locally. By and large, crime experts say that violent crime has been experiencing slight fluctuations rather than sharp increases and decreases.

Phillip Lyons, a criminal justice professor at Sam Houston State University, said those decreasing figures may now be leveling off, showing some stabilization in crime statistics.

“We are at that point, where it seems as though there is overall stability, and that obviously means there are going to be some places that are reporting higher numbers than last year and other places that are reporting lower numbers than last year,” Lyons noted. “It all essentially averages out to not much change.”

See here and here for the previous installments of this story. I basically agree with Prof. Lyons, there really isn’t much happening here. The uptick in unincorporated Harris County is likely just statistical noise. If it goes up for a few years in a row, that may be something. A one year bump that isn’t that big in absolute terms and even smaller in per capita terms is not.

Here’s the sidebar to the story with numbers from the past five years:

Annual number of homicides in Houston and unincorporated Harris County in recent years:

City of Houston:

2009: 287

2010: 269

2011: 198

2012: 217

2013: 199 (As of Dec. 20, 2013)

Unincorporated Harris County

2009: 93

2010: 74

2011: 69

2012: 76

2013: 91 (As of Dec. 31, 2013)

If unincorporated Harris is up over 100 for the next couple of years that may be worrisome, but again keep in mind that the overall population there is rising, too. This chart would be a lot more meaningful if it included the number of homicides per 100,000 residents, as that is a number that will be better to compare over time. Consider the statement above about how 199 murders in Houston would be the second lowest since 1965 when there had been 139. Well, the population in Houston in 1965 would have been less than half what it is today, so 199 murders in 2013 is therefore significantly less – back of the envelope, it would have been about 14 per 100,000 in 1965 (I’m assuming a population halfway between the 1960 and 1970 Census numbers, which would be about one million) but only about 9.5 per 100,000 in 2013 (assuming a population of 2.1 million). Putting it that way, the total number of homicides in Houston was probably as low as it has ever been in a much longer time frame. When was the last time you heard someone say that?

Murder by numbers

There were a lot fewer murders committed in Houston last year than in recent years.

HPD recorded 195 murders for the year as of Friday, a 27.5 percent decrease from the previous year’s total of 269. The preliminary figure doesn’t yet include the death of a 27-year-old woman who police believe was killed in her west Houston apartment on New Year’s Eve.

As long as the 2011 total remains below 200, it would be the lowest since 1965, when 139 people were killed, said HPD homicide division Capt. David Gott.

Harris County’s unincorporated areas also reported fewer murders. The 2011 preliminary total stands at 60 – down more than 7 percent from the 65 in 2010. Murder totals have continued to drop in unincorporated Harris County since 2009, when 87 were reported.

Hair Balls has a slightly different number for the city. The story primarily focuses on the comparison to last year, but if you look at the accompanying chart, what stands out to me is that the number of murders in Houston has dropped nearly fifty percent from the peak of 376 in 2006. Since the story doesn’t look that far back, it doesn’t mention this, which means it also doesn’t dredge up the association of that year with Hurricane Katrina evacuees and their supposed effect on the city’s crime rate. For the city of Houston, there’s a spike from 2005-07, and a trough this year, otherwise the annual number of murders is between 250 and 300. For unincorporated Harris County, outside of a one-time spike in 2009, the body count is basically the same, right at about 60, every year from 2004 through 2011. I point this out for two reasons. One is that it’s plausible to me that there isn’t much more we can do to affect the murder rate in a given year. It is what it is, and outside of larger societal trends that affect the crime rate in general, year to year variations are likely to be statistical noise. What that means is that if the number creeps back up to 250 or so for Houston in 2012, it doesn’t represent a failure of public policy, just a return to historic norms after an unusually slow year. That’s not going to stop the city from taking credit for the decline, nor should it. They do have some policies to point to as causes, and as such we may see a downward trend. But don’t be surprised if it goes up this year, and don’t spend too much time looking for a reason. These things do happen.

And two, in 2010 after the uptick in unincorporated Harris County murders was noted, County Commissioner Steve Radack was critical of Sheriff Adrian Garcia for not having enough patrols to suit him. I can only presume that after two years of normal numbers, including a dip in 2011 to the lowest level seen since 2004, that Radack will now be fulsome in his praise of the Sheriff for his restoration of law and order. Otherwise, his criticism from two years ago will have been shown to be little more than crass political haymaking, and surely that wasn’t the case. Right?

Play politics first, ask questions later

The number of homicides increased in unincorporated Harris County last year while declining in Houston.

With only a few hours left in the year, 86 murders had occurred in the county’s unincorporated areas in 2009, up from 69 in 2008. The county saw 62 murders in 2007.

Sheriff’s Office spokesman Thomas Gilliland said the department doesn’t know why murders have jumped, and at a time when violence is decreasing across the nation.

“We don’t know the reason or have empirical evidence as to why,” he said Thursday. “Common sense might say it’s the economy with people under a lot of stress, but we don’t know what causes people to commit homicide. ”

For the record, the corresponding numbers for Houston were 281 in 2009, 295 in 2008, and 353 in 2007.

Harris County Judge Ed Emmett said authorities will be analyzing the cases and looking at such factors as patrol and demographics.

“The sheriff and the district attorney and everybody in law enforcement obviously need to take a look at this and find out what’s going on,” Emmett said. “We need to know: ‘Where are the murders occurring? Is there a pattern?’”

Emmett figured a driver is a burgeoning population out in the county, fueled in part by more city dwellers moving outward.

I suspect Judge Emmett is correct, but we don’t have enough information to say. What exactly is the population of “unincorporated Harris County”, and by how much has it grown over the last couple of years? We’re talking about a fairly small number of homicides here, and with such a small number it may be that the per capita rate hasn’t changed much at all. It also may be the case that this is little more than statistical noise, and that the number of murders in unincorporated Harris County may decline this year even if the Sheriff and the DA do nothing different. Again, at least based on this article, we just don’t know enough to draw any conclusions. Not that that will stop people from doing so if they have an agenda to push.

Harris County Commissioner Steve Radack blamed at least part of the increase on a reduction in patrol, reducing police visibility in the unincorporated areas.

“The backbone of any police agency is patrol,” he said. “Dozens have been moved out of patrol and moved into internal affairs.”

Just so we’re clear here, Radack is apparently still mad about a decision Sheriff Garcia made in April to assign more investigators to the Office of Inspector General, to help clean up one of the many messes left behind by his predecessor, in this case a large backlog of unresolved complaints against jailers and deputies. Radack of course has no idea why the number of murders rose, or even if the rise in absolute numbers necessarily implies anything more than a teeny tick up in the murder rate, but that won’t stop him from trying to pin the blame for it on the Sheriff.

The thing is, Radack’s complaint is shaky at best on its face. Look at this FBI data table from 2008, which shows murder circumstances by relationship. Of the 7,918 murders in 2008 for which the identity of the murderer was known, 6,176 of them were committed by someone the victim knew – family member, friend, neighbor, or acquaintance. Only 1,742 homicides – 22% of the total – were committed by strangers. How much effect do you think increased police patrols would have had on those 6,000+ killings where the killer and victim knew each other? My guess would be not very much.

Now, this is a facile analysis as well, since “acquaintance” accounts for nearly half of the “someone the victim knew” total, and a fair number of those crimes might have occurred in situations where having more police around could have prevented them. A junkie and his dealer would be acquaintances, for example. There were also 6,268 homicides for which the killer was officially unknown, and surely a higher percentage of those would have been stranger killings. The Chron story talks about how a large fraction of murders in Houston are concentrated in a few areas (this is not unusual), and as far as that goes, beefing up police presence in those areas has a positive effect. But again, at least based on this article, we have no idea if similar hot spots exist in unincorporated Harris, and if so what the numbers look like if you separate out those areas from the rest. The point I’m making is simply that Radack’s comment isn’t particularly constructive, and that’s because it wasn’t meant to be.

UPDATE: Check the comments for some great numbers and maps from Jay Crossley of Houston Tomorrow. Thanks, Jay!