Tuesday, May 10, 2005

Battling Confirmation

About a week ago I promised that I would spend a little time talking about confirmation bias. Today, I shall demonstrate that I am a man of my word.

First, though, I'd like you to take a look at this post over at A Gentleman's C (if you haven't already). The part I especially like is the discussion of the Wason four-card task.
In this task, people are shown a deck of cards. Each card has a letter printed on one side and a numeral on the other. The task is to decide whether rule like "If a card has a vowel on one side, the number on the other side is even," is true. Four cards are dealt such that one consonant, one vowel, one even number and one odd number are showing. The person trying to test the rule is allowed to turn over two cards.
If you follow the link, there is a nice graphic to help you visualize the situation. But, for those of you too lazy to click the mouse button (what the hell is wrong with you anyway?), let's imagine that the four cards you see are "E", "K", "4", and "7". Now, before reading on, decide which two cards you would turn over in order to determine whether or not the "vowel on one side means even number on the other" rule is true.

Let me guess. You chose "E" and "4". Well, that's wrong. Here's why:
The first choice is correct, and consistent with the logical rule modus ponens, or "mode of affirmation." The second choice is incorrect, and represents the logical error of "affirming the consequent." Why is choosing the 4 an error? It is because the rule says nothing about what should be on the back of consonants, so either a vowel or a consonant could appear on the back of the even number and the rule could still be true. The card that should be selected is the odd number, because if a vowel appears on the back then the rule is disconfirmed. This represents the use of the rule modus tollens, or "mode of disconfirmation."
So, what, you ask, is the point of this little exercise? Two things. First, it's a great demonstration of confirmation bias. The selection of the "4" over the "7" perfectly illustrates the tendency to design tests that confirm pre-existing assumptions. Second, because almost everyone gets this test wrong, it helps to dispel the myth that confirmation bias is a problem of the slow-witted, and that the "intelligent" among us can easily avoid it. You can't -- no one can. And that is why it is an important phenomenon to be aware of.

I cannot think of a single example where an individual would bother testing a hypothesis without first forming an opinion as to its truthfulness. The fact that we care enough to test it means that we care about the outcome. This isn't a flaw; it is merely a function of human motivations. But, the fact that we care about the outcome means that we will have a tendency to skew results in a favorable direction. This tendency is damned hard to avoid.

The scientific community generally understands this dynamic and has developed systems to counteract it. When designing experiments to test a given hypothesis, it is not uncommon to consult with those who dispute the theory in question in order to ensure the possibility of disconfirmation. Peer review and experimental repetition also guard against the scientist's tendency to see the data that he wants to see. And, of course, a battery of statistical tools has evolved in order to objectively identify the significance of experimental results.

Other communities, being less familiar with the phenomenon, have been less successful in addressing it. Law enforcement, for example, is a particularly egregious offender. Typically, investigations begin with objective assessments of the emerging evidence. However, once a theory begins to coalesce, the pursuit of conviction frequently overwhelms any aspirations of impartiality. Similarly, crime labs are rarely independent agencies and regularly perceive themselves to be an additional tool of the prosecution. It is not uncommon for testing to be conducted by those fully aware of the DA's desired outcome. Few within this system are deliberately attempting to subvert justice. Yet, as it is currently constructed, such subversions are more common than anyone in the system is willing to admit.

But, it is unfair to single out law enforcement for its tendency to succumb to confirmation bias when it is so rampant in any pursuit of truth. Journalists, pundits, and bloggers regularly cherry-pick from the available data in order to tell the story as they see it. Advocates of various policy initiatives support their positions by presenting only the evidence that reflects favorably on their cause. And could we complete this discussion without at least tangentially referring to this administration's preinvasion assessment of the Iraqi WMD threat? The list is endless.

Sometimes these misrepresentations are deliberate, no question. However, I suspect that they usually are not. They might fail to perform the due diligence required to avoid these mistakes, but this failure is driven by ignorance and a desire for expediency, rather than malice. For the most part, these people aren't evil -- they are simply human. And that is why the fix must be systemic.

Why does confirmation bias occur? There isn't a definitive answer, but theories exist. Some suggest that negative processing is inherently more difficult than positive processing, and that we tend to select the path of least resistance. Disconfirmation requires a restructuring of our intellectual framework in order to accommodate observed data, while confirmation allows for easy assimilation. In other words, it's easier to believe what you already believe.

Regardless of its origins, it is an unavoidable component of human investigations. Awareness of the phenomenon is helpful, but is not enough to eradicate it completely. The only escape is systemic -- as demonstrated by much (but not all) of the scientific community. Anything less is lip service.

So, as you meander about the blogosphere, keep this principle in mind. We're all advocates of one sort or another. We (usually) aspire for truth -- but we are, in the end, only human. Anyone motivated enough to work in this medium has a stake in his conclusions. Some of us will successfully temper our bias and many others will not. In light of this fact, every reader would be well served by a healthy smattering of skepticism -- regardless of who they are reading.

Even if they are reading me.
Weblog Commenting and Trackback by HaloScan.com