Expressing Conclusions

L8

Sally Clark

• had a baby boy who passed away at age 8 months- ruled at cot death

• around a year later she had another baby boy who also passed away at 9 months

• this was also ruled as a cot death until leading paediatrician at Leeds at the time spoke up

• he said it was a 1 in 73 million chance that it would happen twice and so there must be something going on here

  • gave the wrong statistics

  • multiplied the statistical chances of having one baby die of cot death by each other

  • however:

    • having one baby die makes you more likely to have another one die

    • the probability of a boy dying is much more than a girl

What is the role of the expert?

To aid the court with specialist knowledge beyond what the court could reasonably understand.

• ANALYSIS: systematic analysis of individual elements of speech e.g phonetic

• INTERPRETATION: compile analysis into an overall conclusion

What is the history of expressing conclusions?

In answer to the question “is it the same speaker?”

• used to be a binary yes or no

• then changed to a classical probability scale: positive side and a negative side

• then looked at posterior probability: it is highly likely that the suspect and the offender are the same person [given the evidence].

What are some issues with using a classical probability scale in expressing conclusions?

• We have a positive and a negative side

  • Positive = sure beyond reasonable doubt, very probable, possible

  • Negative = probably, likely

• there are less options on the negative side- meaning there us a bias towards a positive identification of speaker

• sure beyond reasonable doubt is a question for the jury not for us

• issue of possibility- possibility doesn’t tell you anything about probability

How does posterior probability work?

• you have:

  • probability - what the courts deal with

  • proposition - argument of events (either H_d or H_p)

  • evidence - try to express some notation

• we are expressing our conclusion with the probability of the proposition given the evidence

How can posterior probability present itself as an issue within the courts?

• Analogy

  • say we have 1% of population with a blood type, this is found at crime scene and matches our suspect

  • there is a 99% chance that a person would not have this blood type

  • does that mean they are guilty? NO

• Posterior probability can overstate the value of the evidence- it is one small part of an entire investigation

• it is up to the court to decide on guilt or innocence- all we can do is provide a likelihood that the speaker is the same

• providing anymore confidence can sway the jury on a fact that we can never conclude to be 100% true

UK Position Statement

• French and Harrison (IJSLL, 2007) produced this in response to CPS and PP

• “fundamental change in the role of the analyst and the evidence”

• no longer focusing on identification!

• say we do our comparison, we are making a judgement about consistency between two samples, if the samples are consistent then we can look at distinctiveness between speech patterns

Critical response to French and Harrisons UK Position statement

• we cannot do a categorial assessment of innocence

• can lead to ‘cliff-edge’ decisions about consistency (what does it mean for the samples to be consistent with each other?)

• serial ordering of consistency/distinctiveness

• ultimately falls short of numerical, likelihood ratio framework

• by saying the samples are not consistent, we are back to a binary conclusion

Likelihood Ratios

• there is an increasing pressure to move away from PPs

• improving the quality of forensic evidence: a “paradigm shift” Saks and Koehler

• we have to report on our reliability and validity of our methods (error rates) e.g have done [n] error rates and found [n] consistency

Formula for numerical likelihood ratios

Similarity

---

Typicality

What are LRs telling us?

whether the evidence supports prosecution or defence and the strength/weight 

of support 

May be that evidence is inconclusive so scales don’t tip at all

This is a pro of LRs- it tell us the strength of the evidence

If we have an LR of 20, what does that tell us?

evidence is 20x more likely assuming the suspect and offender are the same person

(if something is typical, then we will have a larger number at the bottom)

this is a numerical LR

Verbal LRs

we have a number- anything above 1 = H_P

anything below 1 = H_d

e.g 1000 = very strong evidence in favour of prosecution

1-10 = moderate

1-0.1 = limited evidence

< 0.001 = very strong evidence for defence

Why use LRs?

• they are a logical way to evaluate evidence

• separates the role of the expert and the court

• objective: considers both prosecution and defence propositions

• no ‘cliff-edge’ decisions with numbers

Issues with LRs

• what is the relevant population? for typicality

• there is a lack of population statistics to calculate typicality

• courts struggle to interpret statistics

  • Halton 2016 —> as many interpretations as there are jurors

R v T (2010)

• criminal division of court of appeal

• footmark evidence presented in the form of a numerical LR

• estimating population statistics

• fundamental misunderstanding/mistrust of statistics

• numerous responses by forensic experts and statisticians highlighting issues with ruling

What is current practice for LRs?

• Gold and French surveys 2011 v 2019

• verbal LR = 26% (compared to 11.4 in ‘11)

• CPS. = worryingly 23.7% use them! was 40% in 2011

What is now used in the UK to present conclusions to the court?

• tables for phoneticians with columns: support statement, score, relative probabilites

• presented to court is just the RP

• will say something like:

  • the probability that these results could be found under a different speaker hypothesis can effectively be ruled out