The authors believe that an information-gain approach to the analysis and understanding of eyewitness identification data will help advance the literature. They believe their approach preserves a distinction between two different conditional probabilities and their directions. Experimental Research take the status of the suspect, culprit or innocent, and try to estimate the probability of certain eyewitness behavior. The other side is evidence evaluators who must consider the eyewitnesses’ behavior and derive from that the probability of the suspect’s status. The authors also see their approach as highlighting the role of base rates, or prior probability, which in research up until this point has been standardized to .50. The authors contend that the base rate is variable depending on what department is doing the lineups, and that this variability of the base rate can have a significant effect on the size of information gain. They make the point that eyewitness behaviors can be exonerating in terms of the direction of information gain. They feel this has been ignored.

Information Gain…How it works:

Basically, information gain is the absolute value of a difference score. The absolute value comes into play because the sign really just gives you the direction of the information gain, either incriminating or exonerating, not a negative gain in information.

The first part in the equation is prior probability. This is the proportion of the time that the suspect is actually the culprit. Past research has maintained that this is a fairly constant variable and should be valued at .50 for statistical purposes. The authors disagree and see this "base rate", as prior probability is also known, as a variable and should be treated as such. This is noted in the equation as p(S is Culprit). Starting with this proportion, which the authors range in value from 0.0 to 1.0, the posterior probability is subtracted.

The posterior probability is just referring to the postlineup probability that the suspect is the culprit. With the eyewitness behaviors taken into account, the prior probability is raised, lowered, or remains the same. This is denoted by p(S is Culprit|IDS) where IDS stands for identification of the suspect by the witness. So basically, the notation refers to the probability that the suspect is the culprit when the witness identifies the suspect. Three other posterior probabilities were looked at: p(S is culprit|IDfiller), where the witness identifies a filler as the culprit, p(S is culprit|"not there"), where the witness claims the culprit is absent from the lineup, and p(S is culprit|"don’t know"), where the witness doesn’t know whether or not the culprit is present. The difference in the prior probability and the posterior probability is then said to be the information gain. So, let’s say that after experiments, its found that for certain lineup conditions the probability that the suspect is the culprit when the witness identifies him as such is .90 (this would be p(S is Culprit|IDS)). If in a certain district, its found over time, that p(S is culprit) is .60, then the information gained when a witness identifies a suspect as the culprit is | .60 - .90 | or .30.

Information-gain Curves:

The authors used information gain curves to analysis the information gain over a variety of witness behaviors. Simultaneous lineups and sequential lineups were compared with each other to see which would offer the greatest information gain. It was found that Sequential lineups had the greatest gain in information when the suspect was identified and when a filler was identified with little difference when the witness arrived at a "don’t know" or "not there" decision.

Turning to the mismatch or match description lineup debate, information gain using the curves analyzed which would carry the greatest gain in information. The match description lineups provided the most information gain where the witness identified the suspect as the culprit and where the witness identified a filler as the culprit. The mismatch description lineup did poorly on those two with both lineup conditions performing about the same on the "don’t know" and "not there" conditions.

The authors see a witness identifying a filler as being a strong information giver. In real life applications, this is rarely given any thought at all. A witness that identifies a filler is dismissed as not credible and nothing is taken from that witness information, when, in fact, according to the authors, this could be a strong information gain in the exonerating direction for the suspect in question.

The authors feel like it is important to keep in mind the variability of the base rate when dealing with eyewitness behavior because the prior probability has a great effect on the information gain that is arrived at after the witness behavior is taken into account. Also, exonerating information can be taken from instances were witnesses identify a filler as the culprit.