All right. Professor Speed, in this case, there has been testimony that some of the items of evidence such as the Bundy blood drops, a portion of the sample was sent for testing at one laboratory and another portion of the sample was sent to a second laboratory. Are you familiar with that testimony?
To what extent does dividing the sample and sending it to two laboratories or three laboratories and finding that they get identical results reduce the chance that error explains the result from a statistical standpoint?
The chance of--or chance that errors after the samples were sent would be reduced, but any errors that occurred prior to the samples being sent would not be protected by that multiple testing.
To what extent does obtaining the same results on different items of evidence reduce the chance of errors explaining the result?
Well, it's a similar issue. If there's a common source of error, then having consistency across different samples doesn't protect you against that source of error, but errors that will be separate to the samples.
Can you give an example of what "Common mode failure" is in the reliability literature?
Well, there's a nice example that I like to use from my study of nuclear reactor safety where people had what they regarded as independent safety systems that just happened to be wired through the same electrical switchboard; and there was a fire in that electrical switchboard and it put out both of these or a number of supposedly independent systems, because that was the common source back at the beginning. They're independent after they left the switchboard, but they were dependent through this common switching.
And how would you analogize that example to what has been testified to in this case?
Well, that having redundancy or replication will protect you against the impact of errors after the samples have been separated. But if there's an error that occurs at the time they were together or at the time before a particular sample was split and sent, then this extra redundancy doesn't help.
Let me ask you a hypothetical. Dr. Gerdes testified that given the manner the items of evidence were processed in this case, that, for instance, all the Bundy blood drops could have been cross-contaminated by Mr. Simpson's reference sample on the morning of June 14th, 1994. Would that be an example of common mode error?
And if the same error could affect all the swatches for a particular item, would sending the different swatches to different laboratories and getting identical results reduce the chance that the result is explained by that type of error?
And if the same error could have affected a number of different items whose results were identical, would that reduce the chance that the result is explained by that type of error?
Are you aware that Cellmark does not participate in external blind proficiency tests?
Are you aware that the Department of Justice laboratory that Gary Sims works for does not participate in external blind proficiency testing?
And are you aware, sir, that the Los Angeles Police Department does not participate in external blind proficiency testing for its DNA tests?
To your knowledge, sir, does the Los Angeles Police Department laboratory participate in any kind of proficiency testing which would help estimate the frequency of errors of the type discussed by Dr. Gerdes?
For instance, Professor, are there any proficiency tests addressing evidence collection, evidence processing before the DNA is actually extracted?
Now, what effect does the absence of external blind proficiency testing have on constructing a scientifically appropriate estimate of the laboratory's error rate?
Well, in the absence of such a test, you can't really do that. You don't have a suitable estimate of error rate.
KEY QUOTEIn your opinion, sir, is the error rate for forensic DNA testing conducted by the Los Angeles Police Department zero?
All right. As a professor of statistics, have you assessed the role that error plays in various kinds of processes?
And have you assessed the role that error rates play in processes in which people are involved?
And as a result of what you have read and what you have studied, what is your opinion about error rates in processes where human beings are involved?
Well, error is involved in all human activities and all processes, particularly complex processes, errors will occur. The only question is what is their rate.
KEY QUOTEThrough your studies on error rates and proficiency testing, Dr. Speed, is it expected that error rates from blind external tests will be higher or lower than for open tests?
Are you familiar with any literature which discusses the different expected error rates due from blind external tests as opposed to open tests?
And is this also a topic which is discussed among professors of statistics or statisticians in your field?
And as a result of what you've read, the result of discussions that you've had with other members in your field, what is your opinion on this particular subject?
Well, it's generally the case that error rates with blind tests are greater than error rates with open tests.
Well, it seems to me common sense is the explanation. I'm not sure that I can really say why it is, but it's the sort of thing that happens when somebody knows they're being tested, they can do a more careful job, a more thorough job, whereas if it's just part of routine processing, then it may from time to time be treated less carefully.
Are you familiar, Professor, with Dr. Robin Cotton's testimony regarding the comparative likelihoods as to how someone could be falsely implicated by a DNA match?
And what is her statement on that particular issue as you can best recall it, sir?
If you have it in front of you, I'd rather read it, as I didn't do such a good job in my memory last time, or I could paraphrase.
And what is--could you please tell us what Dr. Cotton's opinion is on this particular issue?
Well, she says that she agrees that if someone was falsely implicated by an RFLP DNA test, then it is more likely to have arisen by lab error than by this coincidental match.
KEY QUOTEIs there a scientific basis for knowing that the chance of erroneously concluding that Mr. Simpson is the source of the DNA tested is greater than the frequency cited by Drs. Weir, Cotton and Mr. Sims?
Well, do you have expert knowledge of the relative sizes of these two contributions?
And what is the expert knowledge you have on the relative sizes of these two different types of explanations, one being the frequency of a coincidental match and the other being a match which is explained by error?
Well, the frequencies of coincidental matches, apart from the ones that have very, very few not very helpful loci, have been 1 in millions, 1 in billions or 1 in trillions, whereas the frequencies of errors even of the best labs are of the order of 1 in 50, 1 in 100 or 1 in 200 for very, very good labs. So they are much, much bigger than the 1 in millions, 1 in billions and 1 in trillions and hence play a more important role in contributing to errors.
What is the importance of this difference to the statistical evaluation of the DNA evidence in this case?
Well, it really means that you must know about the errors and that you should really not be that concerned about the coincidental matches once it's reached a certain relatively small value. The errors will be much more important in this overall calculation.
The frequencies of coincidental matches... have been 1 in millions, 1 in billions or 1 in trillions, whereas the frequencies of errors even of the best labs are of the order of 1 in 50, 1 in 100 or 1 in 200 for very, very good labs.
Error is involved in all human activities and all processes, particularly complex processes, errors will occur. The only question is what is their rate.
In the absence of such a test, you can't really do that. You don't have a suitable estimate of error rate.
She says that she agrees that if someone was falsely implicated by an RFLP DNA test, then it is more likely to have arisen by lab error than by this coincidental match.