Faculty » Steven E. Clark

My research is most broadly directed at questions about human memory:

How do we store information in memory?
How is that information represented?
How is that information retrieved?
How do we make decisions based on the imperfect information that we are able to retrieve from memory?
And how do we collaborate with other people in memory tasks?

Many of these questions are addressed within a research program that also seeks to understand the complications of eyewitness memory.

A current focus in my research is to understand how people who have witnessed a crime later make identification decisions. Consider the case in which a person witnesses a robbery. Days, months, perhaps even years later, that witness is presented with a lineup. How does that witness make a decision to identify or not identify someone from that lineup as the robber?

It has become glaringly apparent that witnesses can point someone out in a lineup, and say "that's him" with great confidence - and be wrong. Legal scholars have argued that the most frequent cause of false convictions in the United States is the result of inaccurate memory and mistaken identification. The astonishing number of convictions that have recently been overturned because of DNA evidence is also clear evidence that eyewitness can be very unreliable. The common thread that connects these DNA exonerations is the inaccurate testimony of witnesses (See the U.S. Justice Department's 1996 report "Convicted by Juries: Exonerated by Science" for an earlier review of these cases, or read "Actual Innocence" by Scheck, Neufeld, & Dwyer, 2000).

This raises the question: How can witnesses be so confident and yet so wrong? Why do witnesses misidentify innocent persons, and do so with such sincerity?

To address this question I have been working to develop a comprehensive model of the memory and decision processes which underlie eyewitness identification. The WITNESS model is a theory of eyewitness identification, formalized mathematically, and instantiated in a computer simulation program, in order to allow the model to generate quantitative predictions in the form of response probabilities that can be compared to data. The WITNESS model is a variation of a class of memory models called global matching models (Clark & Gronlund, 1996), so the model is an extension of decades of theoretical work.

Two things have become clear from this research: 1. Witnesses may have incomplete and fragmented memories of the crimes they have seen, and 2. Their fragmented memories don't stop them from picking someone from the lineup, anyway!

Our empirical and model-driven research suggests that witnesses use a decision rule that is rather lenient. In other words, people make identifications when they should be more conservative and say "I don't know". This research builds on the previous work by Gary Wells (1984) and others in showing that witnesses tend to use a relative judgment in making their identification decisions. An example of such a strategy is to pick the person from the lineup "who looks most like" the person who committed the crime. Of course, such a decision strategy is extremely risky, because it is almost always the case that someone in the lineup looks more like the criminal than the others. An innocent person in the lineup will look more like the "bad guy" than anyone else if the other people in the lineup are picked poorly, and can be ruled out easily.