Do Reactions To Drug-Sniffing Dogs Say More About Drug Policy Than Privacy?

In Florida v. Jardines, the U.S. Supreme Court will determine whether the sniff of a trained narcotics dog at the front door of a person’s home constitutes a Fourth Amendment search. This is very exciting for privacy scholars because it presents two possible shifts in Fourth Amendment jurisprudence. First, the court might further expand Justice Scalia’s “magical places” rationale to reinforce that the home is a formidable privacy fortress, protecting all information from the detection of government agents unless that information happens to be visible to the human eye.

The second possibility — the one I root for — is that the court may choose to reopen the holding and reasoning of the previous dog sniff cases, Place and Caballes (which determined that dog sniffs conducted on a car and on luggage did not constitute a search.) Justice Stevens’ majority opinion in Caballes held that a dog sniff “discloses only the presence or absence of narcotics, a contraband item.” (Quoting Place.) While Justice Souter found this assumption deeply flawed, Stevens avoided wrestling with the false alert problem on procedural and evidentiary grounds.

Although respondent argues that the error rates, particularly the existence of false positives, call into question the premise that drug-detection dogs alert only to contraband, the record contains no evidence or findings that support his argument.

This is certainly fixable. Because the reasoning of the former precedents partially rests on the mythical infallible dog, it is high time (heh heh) the court ditch the old cases and start the dog sniff analysis anew in light of dog error rates.

In honor of the Court’s decision to take cert in Jardines, I recently conducted a short survey of 187 Brooklyn Law School students to understand the privacy intuitions of the next generation of lawyers. I asked the students whether contraband-detecting dog sniffs under three different scenarios should be considered an invasion of privacy in the absence of suspicion or probable cause. In the first scenario, the dog never generates a false alert. If the dog indicates that the area contains contraband, contraband will be found 100% of the time. This is the perfectly accurate dog (or, rather, even if it fails to detect contraband, it will never alert when contraband is not present.) The second dog has a 1% false alert rate. If the dog alerts, contraband will be found 99% of the time. The other 1% of alerts result in an unproductive search – there is no contraband present. The third dog has a 10% false alert rate.  If the third dog alerts, contraband will be found 90% of the time. It’s worth noting at the outset that all three of these fantasy dogs are much more accurate than real police dogs out in the field.  

Not surprisingly, the accuracy rates of the police dogs mattered to the students. Fewer than half thought that a sniff of a car from a perfectly accurate dog constituted a search, whereas two-thirds believed the sniff of the dog with a 10% error rate was an invasion of privacy.

Proportion of Respondents Finding an “Invasion of Privacy” When Dog Sniffs Are Used on a Motor Vehicle

Perfect Accuracy 40%
1% False Alert 51%
10% False Alert 66%

Moreover, the students who believed a dog sniff on a motor vehicle was not a search tended to change their mind when the same nose was pointed at a home.

Change in Views: Car Versus Home

Number finding no violation in car

Number finding no violation at home

% change

Perfect Accuracy




1% False Alert




10% False Alert




So far, no surprises. But there are some illuminating differences between student responses.

In addition to varying the survey design to see if question ordering made a difference (it didn’t), I also varied the type of contraband the dogs were trained to detect. Roughly one-third of the students responded to a drug-sniffing dog, one-third responded to a bomb-sniffing dog, and (for reasons that will be clear in a minute), one-third responded to a human cadaver-sniffing dog. This allows us to test whether common instincts about privacy are sensitive to the type of criminal investigation. Sure enough, when police tools are applied to drug enforcement investigations, the tools are regarded as more privacy-invading than when the same tools are applied to other types of criminal investigation.

Proportion of Respondents Finding an “Invasion of Privacy” When Dog Sniffs Are Used on a Motor Vehicle

Drugs Cadavers Bombs
Perfect Accuracy 56%** 30% 36%
1% False Positive 67%** 37% 50%
10% False Positive 82%*** 54% 63%

**= significant at the 1% level using a two-sample t-test in comparison to combined cadavers and bombs respondents
***= significant at the .1% level

Change in Views: Car Versus Home



Perfect dog



1% False Positive



10% False Positive



(Reporting the proportion of respondents who had found no invasion of privacy when the dog was used on a car who changed their position when the dog was used on the front door of a house)

As a matter of black letter law, the type of criminal investigation has not been a factor when courts decide whether state conduct constitutes a Fourth Amendment search outside the context of the special needs doctrine. (And the special needs doctrine typically concedes that a search IS taking place, and asks whether that search is “reasonable” in context.) And as a practical matter, making a distinction on the basis of the type of crime will not be fruitful. As Ryan Calo points out, a contraband-sniffing dog is just a first generation information-gathering tool. In the future, a single instrument (possibly a drone) will detect drugs AND bombs. (OK, maybe not cadavers.) If police conduct is intrusive, it should not evade designation as a search simply because it is employed to achieve relatively lofty goals.

In a footnote in his Caballes dissent, Justice Souter reserved judgment about the constitutionality of a dog sniff used to detect explosives in the wake of a risk of terrorism. As a group, law students do not make the distinction Souter anticipated between retrospective and prospective law enforcement. The use of police dogs to detect dead bodies was perceived to be as legitimate and privacy-respecting as the use of bomb-sniffing dogs. Moreover, Souter always considers a dog sniff to be a search, and carves out the Fourth Amendment exception for bombs only when there is special need. “Suffice it to say here that what is a reasonable search depends in part on demonstrated risk.”

I am concerned that the ill-advised War on Drugs is distorting the privacy debate. These distortions are hardening a distrust of technological advances in law enforcement, even when new technologies have the chance to improve criminal investigations and enhance privacy simultaneously. Our Fourth Amendment jurisprudence should stay focused on how law-abiding citizens are treated by their state and federal governments. As Daniel Solove’s influential article “Nothing to Hide” makes plain, privacy interests persist even if a government’s information-gathering program avoids putting a person through the full-blown hassle of arrest, seizure, or some other intrusive interference.

These privacy interests are especially strong for anyone who participates in legal but socially deviant behavior. So, when analyzing Fourth Amendment fact patterns, I like to imagine a law-abiding citizen who is driving around with socialist propaganda, pornography, and a case of snausages. Let’s call him Bob. If a police dog falsely alerts to the snausages, Bob’s political and pornographic materials will be observed. Will Bob experience an unconstitutional search under the various conditions I used in my survey?

The Perfect Dog

The perfect dog will never alert at Bob’s car or residence. Under these conditions, the “sui generis” reasoning of Place is perfectly defensible. Moreover, an expectation of privacy is indefensible. If the dog is truly infallible, a right to privacy is indistinguishable from a right to get away with crime.

Moreover, by eliminating human error and discretion, the infallible dog presents the opportunity to remove prejudice and other problems, especially if there is sufficient pressure to use the tool uniformly. Uniform enforcement is the true test of a criminal law’s legitimacy. This is what I call the “Senator’s daughter” test: if a law enforcement tool is deployed such that the Senator’s daughter is as likely to be arrested as any other offender, then criminal sanctions will adjust to better reflect the severity of the crime.

But this is a purely academic exercise— though new technologies might significantly reduce error, no screen or test will ever be error-free.

The 10% Dog

With that said, I share the law students’ instincts that even a 10% error rate raises serious privacy problems. In Caballes, Justice Stevens reasoned that a dog alert, even with the possibility of error, is sufficiently reliable to establish probable cause, which can then justify a warrantless search (either because the search is reasonable or because of exigent circumstances.)

In fairness, as compared to the reliability of evidence that supports a full probable cause warrant, Stevens is probably right. A recent article by Jeffrey Rachlinski et al. asked judges to estimate the chance that a determination of probable cause would result in a fruitful search; the judges estimated there would be a 65% chance that the search would produce contraband or incriminating evidence.

But probable cause warrants come from traditional suspicion-based investigations. The resources required to conduct a personalized investigation create a natural limit on how many unproductive searches can happen. We need not put up with so much error when new, more efficient investigative techniques are applied to society at large.

The “Senator’s Daughter” justification for police dogs depends on the even application of the technologies on everybody. Precisely because of this dragnet nature—because dogs could be deployed on every car and home—a program with a 10% error rate will ensnare too many people in false positive searches. Bob is put at too much risk of an unnecessary intrusion into his car’s trunk.

The 1% Dog

This leaves the Goldilocks dog—the dog that produces error, but only an amount that could be considered negligible. Current police practices rely very heavily on the hunches of individual police officers. Being human, they are prone to bias, self-interest, and randomness. Tal Zarsky’s article Governmental Data Mining and its Alternatives, a must-read for scholars thinking about this stuff, questions whether the old traditional practices really are normatively superior to new investigative practices. The 1% dog (or drone, or whatever it is we will be using in the distant future) can help police forces withdraw some of the discretion that puts some citizens at disproportionate risk of unnecessary searches.

I hope the Supreme Court takes the opportunity in Jardines to craft an opinion that recognizes the privacy risks of error rates without undermining the potential for promising new advances in crime detection.

2 Responses to “Do Reactions To Drug-Sniffing Dogs Say More About Drug Policy Than Privacy?”

  1. I enjoyed your article, but suggest that it contains a fallacy where you suggest: ‘If the dog is truly infallible, a right to privacy is indistinguishable from a right to get away with crime’.

    A police officer with an infallible dog may harass somebody by carrying out frequent searches or may discriminate against a category of people by selecting them for searches. That would constitute a breach of privacy that does not presuppose any crime.

    Note, the victim need not even be present for this to constitute harassment. If the police were to frequently visit my doorstep while I am away that would send an unwelcome message to my neighbours.

    To counter this the rule should be that you may only deploy an infallible dog when you already have due cause to conduct a search.

  2. You may wish to reconsider your presentation of false-positive rates. Under your scenarios, the number of false alerts is dependent on the reliability of the dog AND on the number of cases in which drugs are really present. A more realistic (and more usual) presentation of the false positive rate is based on the number of sniffs conducted.

    For example, suppose a million sniffs are conducted. In your 1% scenario, if there are 100 alerts then there will be 1 false positive (1% of 100) and 99 drug carriers; and if there are 10,000 alerts there will be 100 false positives. In the 10% scenario these figures rise to 10 and 1000, respectively.

    More usually, though, the false alert rate is given with respect to the number of tests, so if a million sniffs are conducted, then a 1% false alert rate would result in 10,000 false alerts — perhaps acceptable if there were 990,000 drug carriers among the 1 million sniffed, but disproportionate if there were only, say, 100. The 10% rate would be correspondingly more alarming.