In 1984, Ronald Cotton was convicted for rape and burglary. He was sentenced to life + 50 years. In 1995, he was released having served over 10 years in prison, when DNA evidence demonstrated that Booby Poole was guilty of the crime (Ronald Cotton, n.d.). The Innocence Project estimates that between 2% and 5% of prisoners have been falsely convicted. Australia had a prison population of 42,974 at 30 June 2018 (Prisoners in Australia, 2019), suggesting that over 2000 individuals may be serving time for crimes they did not commit. In the United States, 2.3M people were incarcerated in 2019 (Sawyer & Wagner, 2019) suggesting over 100,000 people may be falsely incarcerated. And as Richard Rosen, Ron Cotton’s lawyer points out, Bobby Poole went on to rape other women while the police were focused on Cotton, so Ron was not the only person to suffer as a consequence of the mistake.
The Cotton case is one of the best known examples of false eyewitness testimony. The victim Jennifer Thompson-Cannino not only identified Cotton in original photo and physical line ups, but in a second trial when asked whether it was Cotton or Poole who assaulted her, picked Cotton (Thompson-Cannino, Cotton & Torneo, 2009). However, there is another aspect of the case that has received relatively little attention. When Cotton was interrogated he provided a false alibi. Rather than report where he had been at the time of the crime, Cotton recalled where he had been the week before. Jurors have a strong tendency to assume that inaccurate alibis are a result of lying rather than memory error (Crozier, Strange, & Loftus, 2017, Culhane & Hosch, 2012).
Perhaps the most distressing aspect of the story is that it could have been easily avoided. Had the police officer involved asked where he had been the week before, Cotton would likely have realized his mistake. The police may then have been able to verify his alibi and his conviction could have been prevented.
But what are the right questions to ask? In this case, asking about the week before would have been decisive, but there are many questions one could ask. Where were you the day before and after the crime? Where were you the hour before and after the crime? You said you were running – when else were you running? Memory traces consist of many elements, but which ones are the most likely to generate false retrievals?
Unfortunately, current work in the memory literature is poorly equipped to answer these questions. The majority of work occurs within laboratory paradigms where aspects such as font colour, word frequency and semantic category are manipulated – aspects that are quite different from those a detective is likely to be interested in. Work in autobiographical memory relies on self report of selected memories or on events on public record. These kinds of memories might be expected to more faithfully capture the kinds of information that typical memories contain. However, these events tend to attract more attention and to be reinforced by subsequent retellings than typical memories (Neisser & Harsch, 1992). Furthermore, memory performance depends critically on the nature of the other events in memory which generate the background interference from which a target memory must be distinguished (Osth & Dennis, 2015). Current memory paradigms have no way of capturing the nature or extent of this interference in daily life.
A primary challenge for alibi generation research is establishing the ground truth of the real world events of interest. In a pilot study, we used a smartphone app to record data on participants (N=57) for a month prior to a memory test. The app captured their accelerometry continuously and their GPS location and sound environment every ten minutes. After a week retention interval, we presented participants with a series of trials which asked them to identify where they were at a given time. To respond they were given an interactive map on which four pins had been placed. One pin indicated the correct location and the others were distractors.
Participants were accurate 64% of the time (SD = 16%). Furthermore, our forced choice procedure allowed us to conduct a conditional logit analysis to assess the relative importance of different aspects of the events to the decision process. The figure to the right shows Bayesian estimates of the beta weights along with their 95% credible intervals. Similarity of location induced more errors than similarity of sound environments or movement types.
The Cotton example suggests that participants might also confuse days across weeks. To test this possibility, we created categorical time predictors. The figure below shows the Bayesian estimates of the conditional logit weights for this analysis along with their 95% credible intervals. There is strong evidence for this kind of error. In addition, people often confused weeks in general and also hours across days hours. Furthermore, model fits were not improved by adding a continuous time predictor.
These results allow us to provide preliminary recommendations to investigators. Asking where the suspect was the week before and after at the same time appears to be the most critical question. However, asking where they were at that time on the previous and subsequent days and more generally if they think they have the right week are almost as important. Similarly, asking at what other times they were near the place they provided would be worthwhile. However, based on these results, it would seem less important to ask about other times they were engaged in similar activities or during which they were exposed to a similar sound environment.
We should emphasise, however, that these are tentative suggestions designed mainly to illustrate how one would draw actionable conclusions rather than results that are ready for immediate adoption. They require replication and generalisation. What they do demonstrate is the power of big data approaches to answer critical questions that could not be addressed meaningfully in the past.
If you would like to learn more about this research feel free to read the preprint.