On March 17, the New York Times ran a story about the troubling inconclusiveness of breast biopsy results. Some results are clear and generate a reassuring consensus among pathologists examining the tissue. But for “grey area” results, some doctors find reason for alarm and thus for more tests and procedures, while others believe that all is essentially well. In this column, I will compare this phenomenon of ambiguity-generating medical tests with a seemingly unrelated phenomenon that surrounds dog sniffs for narcotics. I will suggest that when a test’s results are less than definitive, we may have a reason to avoid taking the test in the first place, even when the test itself is relatively uninvasive or innocuous on its own.
A number of years ago, I remember hearing about a traveling van that would allow free preliminary screenings for breast cancer and possibly other sorts of cancers. My reaction to the idea of this van was strong and unequivocal—I would not take advantage of the free offer, because of something known in statistics circles as “Type I errors.”
Since screening devices are rarely even close to perfect, they will predictably generate two different kinds of errors (and their calibration can affect which error predominates). Type I errors, also known as “false positives,” will identify a target as having the condition in question even though she does not in fact have it. A highly sensitive metal detector, for example, might falsely indicate that a plastic zipper contains metal (perhaps because a tiny piece of metal once came into contact with the zipper). A highly sensitive test for cancer might likewise indicate that someone has cancer when she does not.
A second kind of error is a Type II error. This type—also known as a “false negative”—occurs when a screening device misses the presence of the condition for which it is screening in a particular target. The metal detector, for instance, might fail to identify the metal weapon that has entered the screening device.
Neither of these errors is desirable, of course, but we make calculated decisions to prefer one type over the other, depending on the context. In criminal trials, as a policy matter, we strongly favor Type II errors, and we express that preference by asking jurors to bring back a “not guilty” verdict even if they are convinced that the defendant is probably guilty of the crime charged, so long as the evidence does not yield confidence of his guilt beyond any reasonable doubt. This preference for Type II errors reflects the normative judgment that it is far better for a guilty person (or even ten guilty people) to go free than for an innocent person to be convicted and punished.
We make a different judgment call in civil cases and allow for close to numerical parity between Type I and Type II errors by applying a “preponderance of the evidence” standard, whereby plaintiffs prevail if the odds of their being right even slightly outweigh the odds of the defendants being right.
The cancer screening van that I mentioned earlier, as far as I could tell, would strongly favor Type I errors—false positives—if only to avoid liability for mistakenly giving someone with cancer a clean bill of health. As a result, my expectation, upon exiting the screening, would be to hear something like the following speech: “Our tests show that you may have cancer in one of several locations in your body. These tests are not conclusive, so there is no reason to panic or to be alarmed. Our recommendation is that you see a physician and show him or her these results so that your doctor can pursue additional tests or screenings, where appropriate. Such tests might include X-rays, fine-needle aspirations, or biopsies, etc.”
I could then choose either to ignore what I had just heard or subject myself to a battery of time-consuming, invasive, painful, frightening, and expensive tests. The one thing I could not do would be to go back to my previous state of blissful ignorance (in some relevant ways similar to my new state of high-anxiety ignorance). The screening would have robbed me of the option of remaining in that state.
Dog Sniffs and Justice Souter
What does any of this have to do with dog sniffs for narcotics? To understand the connection, we need some background in the Fourth Amendment and dog sniffs. In the most recent case on the topic, Rodriguez v. United States, the United States Supreme Court is currently considering whether lengthening a completed routine traffic stop for a brief narcotics dog sniff violates the Fourth Amendment right against unreasonable seizures. I discussed Rodriguez in an earlier column. The case focuses on the freedom from unreasonable seizures rather than unreasonable searches because the Court has already held, in Illinois v. Caballes, that a dog sniff of a car for contraband does not constitute a “search” of the car and therefore triggers no Fourth Amendment protection against unreasonable searches. The reason, explained the Court, is that exposing the fact that a car does (or does not) contain contraband reveals no private or personal matter in which anyone holds a reasonable expectation of privacy.
Justice Souter dissented from the Court’s opinion in Caballes. He argued in his dissent that dogs are not infallible and that therefore, on occasion, a dog will alert to the supposed presence of contraband and thereby lead police to perform an invasive search (either with a warrant or in the presence of an exception to the warrant requirement such as an exigent circumstance), and it will turn out that the person who experienced this invasive search was in fact not in possession of contraband. For this reason—the worry, in effect, about Type I errors in the dog sniff process—Justice Souter contended that a dog sniff should be classified as a Fourth Amendment search that should therefore occur only in the presence of some level of individualized suspicion (such as probable cause).
When I have taught Caballes in my Criminal Procedure Investigations course, I have been quite critical of Justice Souter’s analysis here. To determine whether a dog sniff for contraband is or is not a search, I have explained, the fact that dogs are fallible is irrelevant, as is the fact that a search will likely follow a dog’s positive alert.
No one is infallible, I would maintain, including human informants, and the question of how accurate or inaccurate dog sniffs might be, I would propose, goes to whether a dog’s positive alert to narcotics is enough, on its own (or maybe even coupled with other evidence), to justify a search, just as the proper question in considering the fallibility of human informants is whether the informant’s statement is enough, either alone or coupled with other evidence, to justify a search, not whether questioning a (necessarily fallible) human informant constitutes a “search” of the suspect being investigated (which it does not).
Though I still think this is the most logical way to think about Caballes and dog sniffs, I have become more sympathetic to Justice Souter’s position as I have thought about his concerns in the context of the uncertain medical screenings for cancer and other conditions that seem to be proliferating.
There is a momentum that gets going once we get a positive result from a test, whether it is a cancer screening with a high rate of false positives, or whether it is a positive alert for narcotics from a (fallible) dog working with a police officer. When the positive result comes, it is natural and expected that we will want to do further, and probably more invasive, tests, to find out whether or not there is “something there.”
Once a positive result comes back from a mammogram or a biopsy or a Prostate-Specific Antigen (PSA) blood test, it may feel affirmatively neglectful not to follow up with more invasive investigation. And likewise, once a dog alerts to narcotics, it may seem positively irresponsible for the police to ignore the alert and do nothing to find out whether in fact there are drugs (and in what quantity) in the car to which the dog has alerted.
If this is true, then Justice Souter may have a point. Just as many women might now choose to forgo mammography or other imperfect cancer screening (and just as many men might choose to forgo PSA tests), to avoid falling down the rabbit hole of biopsies and potentially unnecessary and disabling surgeries and treatments, Justice Souter may want us to forgo the dog sniffs, at least absent a level of individualized suspicion that would justify a real search, thereby avoiding a similar rabbit hole of highly invasive searches of property that will almost inevitably follow a positive alert from a dog.
The problem with screening devices with a high (or even a substantial but not especially high) frequency of Type I errors is that they seem much more innocuous than they actually are. They seem to be just the single test (perhaps a blood test or even a minor biopsy) or a dog’s quick walk around a vehicle, but they rarely if ever end there if the result is positive. Given this fact, Justice Souter (in the context of dog sniffs) and I and others (in the context of medical screenings) might want to treat that first, seemingly innocuous, step as the whole staircase that it may rapidly become and approach it with care.
For medical tests, this could mean avoiding many screenings in the absence of either symptoms or some other reason to believe there is a problem. And for dog sniffs, it could mean requiring some independent, articulable suspicion (or probable cause) beforehand. Such moves could potentially serve us well by reducing the considerable collateral harm caused by the (foreseeable) Type I errors occasioned when we apply screening devices promiscuously, in contexts in which we have no pre-existing reason to believe that there is anything to be found.