The crisis call experiment

This experiment shows the dangers of leaving the task of evaluating the transcript of a ‘disputed utterance’ to the jury.

The story

Early one morning, a young man returned home from his paper round. About twenty minutes later, he made a crisis (or emergency) call reporting his entire family were lying dead in the house.

You can listen to the crisis call below (1 minute), but please be warned, it is a young man reporting a real murder, and is quite distressing to listen to.

Some days later, it was alleged the young man had carried out the murders himself, and the call was a hoax. The following year, he was convicted and given a lengthy prison sentence.

There ensued a long series of appeals. During preparation for one of these it was noticed that the crisis call itself contained a confession by the young man that he had committed the murder himself.

You can listen to the section of the call containing the alleged confession here, and also to the alleged confession itself (the ‘section of interest’).

When you have listened in this ‘open’ condition, watch the very short video below to participate in a mini-experiment about the transcript.

Please note this mini-experiment is not at all like
the main experiment described below (in particular,
participants were not given the alternative transcription).
It is intended 
simply to give you a quick taste
of some issues in transcribing  audio like this.

The experiment


190 people from a wide range of demographic groups took the experiment online. They were randomly divided into two groups.


Participants heard the three pieces of audio that you have just listened to, and were asked what they thought was said in the section of interest. They were then given evidence about the case, through a series of ‘evidence points’. Group 1 heard a story similar to one above, suggesting the caller was guilty. Group 2 heard a parallel story suggesting the caller’s father was guilty.

At each evidence point they were able to listen again to the three recordings, and were asked ‘What do you hear now in the section of interest?’ – along with some other questions. At evidence point 4, they were given a specific transcript of the section of interest. Group 1 received the actual alleged transcript; Group 2 received another transcript that was poorly supported by the audio (not the alternative transcript given in the mini-experiment above). At evidence point 7 both groups were given ‘the full story’, including information that experts on both sides had agreed that the alleged transcript was inaccurate, and that the young man had been fully exonerated of the crime and released from prison.

Summary of key results

The graph below shows how many people in each group heard the alleged confession at each evidence point. The transcript was presented to Group A at Evidence Point 4.


Please click the image to enlarge. Recall the alleged confession was shown to Group A at Evidence Point 4.

Key points to note:

  • virtually no one heard the alleged confession before it was suggested
  • as soon as it was suggested to Group 1, over 30% of participants confidently ‘heard’ the alleged phrase
  • around half of these continued to ‘hear’ it even after being given the full story, including information that experts on both sides of this case had agreed the alleged confession had not been spoken, and that the young man has now been fully exonerated on other evidence
  • nearly as many in Group 2 ‘heard’ the confession at evidence point 7, when it was mentioned in the context the full story – along with information that experts on both sides of this case had agreed the alleged confession had not been spoken and the young man has been exonerated
  • for both groups, around 15% still ‘heard’ the alleged phrase at the end of the experiment and these were significantly more likely to say they thought the caller was guilty than other participants – even after having been told in the full story that he has been fully exonerated and released from prison

Note that participants were given far more information
and far more opportunity to listen repeatedly in controlled
conditions than is usual in current Australian
trials – where, naturally, no ‘full story’ can be given)

Some other important points

  • Group B did not accept their prime, which was deliberately chosen to be a poor match to the acoustics (to show that priming is not automatic; perception involves weighing information from both the acoustics and the context)
  • in both groups, those who did not accept the prime as a whole, were still demonstrably influenced by the prime in a range of more subtle ways (see full paper in reference below)
  • long before the transcript was presented, before hearing any information at all about the case, 20 of the 190 participants (10%) said they did not trust the caller; and these were twice as likely as the group as a whole to find him guilty at the end.

What does it all mean?

Importantly, none of this says anything about the guilt or innocence of the caller. This is academic research intended to demonstrate aspects of speech perception that are of general relevance to the legal system as a whole.

What it does is present several reasons why it is not valid to leave evaluation of a disputed utterance to the jury, as is recommended by the Australian legal system. More reasons are explored by the pact experiments.

For more detail on the experiment

Fraser, Helen, Bruce Stevenson, and Tony Marks. 2011. “Interpretation of a Crisis Call: Persistence of a Primed Perception of a Disputed Utterance.” International Journal of Speech Language and the Law 18 (2): 261–292.

All material on this site is Copyright © Helen Fraser 2013.
By all means share links to it but please do not copy or use without permission.