Eyewitness Identification: Which Results Matter for Which Decision-Makers?

View Segments Segment :

  • Citations
  • Add to My List
  • Embed
  • Link
  • Help
  • Citations
  • Add to My List
  • Embed
  • Link
  • Help
Successfully saved clip
Find all your clips in My Lists
Failed to save clip
  • Transcript
  • Transcript

    Auto-Scroll: ONOFF 
    • 00:10

      LAURA MICKES: This is what I'll talk about.I'll talk about Eyewitness Identifications--Which Results Matter for Which Decision-Makers?[Eyewitness Identification--Which ResultsMatter for Which Decision-Makers?] Now,this is something that is new.This distinction that I'm going to makeis new to the field of eyewitness identification,but it's really well understood in diagnostic medicine,so I'll talk a little bit about that at the end.But very briefly or very quickly,I'll talk about lineup procedures.

    • 00:31

      LAURA MICKES [continued]: And you've already heard it from John,so it's a bit of a review.Eyewitnesses to a crime are oftencalled upon by the police to identifya perpetrator from a lineup that looks something like that.The most common lineup procedure in the US,where most of the research has been conducted,is a simultaneous six-person photo lineup with five fillers.

    • 00:51

      LAURA MICKES [continued]: And these fillers are known to be innocent.So even if they're picked, there'sno risk of wrongfully investigatingthem or convicting them.And then there's the police suspect somewherein that lineup.Now, the police suspect, like John talked about,may be innocent or not.So in the lab we know if the suspect's innocent or not,but the police don't know if their suspect is.

    • 01:13

      LAURA MICKES [continued]: It's just their suspect.So a witness can make one of three responses.They can pick a filler.If a filler's picked, there is no riskof being wrongfully investigated or convicted, as I said.And eye witness can pick a suspect,and here the focus is typically placed on whether or notthe suspect is identified because it's

    • 01:34

      LAURA MICKES [continued]: this identification that matters in the courts of lawand for the criminal proceedings.This person may be investigated and convicted.So here are possible outcomes.If the suspect's true state is that they're guiltyand the eyewitness picks them, then that'sa correct identification.If the true state is that the suspect is innocentand the witness picks them, that's a false identification.

    • 01:57

      LAURA MICKES [continued]: The third possible response is that the witness couldpick no one, in which case if the true state isthat the suspect is in the lineup and their guiltyand the witness doesn't pick them, that's a miss,and if the innocent suspect's in the lineupand the witness doesn't pick them,then that is a correct rejection.So here we have two possible correct outcomes,a correct identification and a correct rejection,

    • 02:17

      LAURA MICKES [continued]: and two potential incorrect outcomes, a missand a false identification.OK, so in the typical lab experiment, eyewitness labexperiment, there's a study phase and a test phase.During the study phase, participantsview a video of a mock crime.There's a screenshot of it.There is some sort of delay where they do a distractortest, typically, and then memory is tested,so there's a lineup test during the test phase, where they

    • 02:39

      LAURA MICKES [continued]: pick no one or pick someone.Again, in the lab, we know if the guilty suspect'sin the lineup.The false ID rates come from the target-absent lineups.So the false ID rate is the proportion of innocent suspectsidentified from the target-absent absent lineups.[False ID Rate - proportion of innocent suspects identifiedfrom target-absent lineups] There are a couple ways to dealwith the innocent suspect, as John mentioned.You could designate an innocent suspect in the lineup,

    • 03:01

      LAURA MICKES [continued]: or more commonly, you could divideby the number of lineup members, so in this case would be six.The correct ID right comes from the target-present lineups,so the correct ID rate is the proportion of guilty suspectsidentified from target-present lineups. [Correct ID Rate -proportion of guilty suspects identified from target-presentlineups]OK, so now there are two types of decision makers interestedin eyewitness identification research.

    • 03:23

      LAURA MICKES [continued]: We've got policy makers such as police chiefs and judgesand jurors.And they need to know different resultsto help them make their decisions.So what type of information they needis largely dependent on whether or not the variable in questionis a system variable or an estimator variable,and I'm talking about variables that

    • 03:43

      LAURA MICKES [continued]: affect eyewitness accuracy.So a system variable can be controlledacross criminal cases.For example, what type of lineup formatto use, how many people to put in the lineup.And an estimator variable, these are variablesthat can't be controlled, so whether or nota weapon was used, how long the witness saw the perpetratorfor.These are examples of estimator variables.

    • 04:04

      LAURA MICKES [continued]: OK, so now two relatively new analyseshave been recently recommended to helpelucidate certain issues in eyewitness identificationresearch, and that's receiver operating characteristic,or ROC analysis, and calibration analysis.And so ROC analysis measures discriminability,which is the ability to discriminate or distinguish

    • 04:24

      LAURA MICKES [continued]: innocent from guilty suspects.An ROC plot is simply a plot of the correct ID rateand the false ID rate for every level of confidence.So here, this is the correct ID rate and false IDrate for those identifications made with 100% confidence.This is the correct ID rate and false IDrate for those identifications made with 90% confidence

    • 04:45

      LAURA MICKES [continued]: and above, and so forth.This dashed line is the line of chance performance,so that if the points fell on the line of chance performance,the eyewitness would just have no ability to distinguishinnocent from guilty suspects.So the further that those points fallfrom the line of chance, the better the discriminability.There's another point that's important to make here,

    • 05:06

      LAURA MICKES [continued]: is that discriminability is the same along that ROC curve.So regardless of which point you're talking abouton that ROC curve, whether it's the left-most conservativepoint or the right-most liberal point,discriminability is the same across.This ROC is higher than the previous ROC,so discriminability is better for this.And if you want to run a statistical test to compare

    • 05:27

      LAURA MICKES [continued]: the two curves, you just have to measure areaunder the curve for both conditions or procedures,and then do some sort of T-test.So this area under the curve would obviously be smaller.OK, now a calibration analysis measures the relationshipbetween subjective probability that an ID is correctand the objective probability that it's correct.

    • 05:48

      LAURA MICKES [continued]: Now, it's really important to notethat it's confidence at first identificationthat we're talking about.It's not later in court when confidence islikely to have been inflated.Confidence analysis is often conducted wherethey include filler picks.So to plot a calibration curve, youcompute this calibration DV.

    • 06:09

      LAURA MICKES [continued]: So it's the correct suspect IDs dividedby the correct suspect IDs plus the filler IDsfor every level of confidence.And then you plot it, and this is an exampleof perfect calibration where you have the proportion correctbased on the formula I just told you on the y-axis, and thenconfidence on the x-axis.OK, so I'm going to talk about a close relative of calibration

    • 06:30

      LAURA MICKES [continued]: analysis that we call confidence accuracy characteristicanalysis, or CAC analysis.And this involves plotting only suspect identifications,so it's the suspect identifications, again,that matter.Not the fillers, not for policymakersand judges and juries.OK, so one of the reasons we do this, besides what I just

    • 06:51

      LAURA MICKES [continued]: mentioned, is that because calibration plots can oftenunderestimate suspect ID accuracy.And that's, again, because they include filler IDs.So to us, a better DV is this.So you would take the correct suspect IDs dividedby correct suspect IDs plus the incorrect suspect IDs,and then you plot that for every level of confidence.

    • 07:14

      LAURA MICKES [continued]: OK, so I'm going to run through four experiments,and I'm going to show you ROC and CAC resultsfor each of the experiments.Two of those experiments measure system variables,so these are types of procedures, and two measureestimator variables.When the case moves to the court of law,the lineup procedure becomes an estimator variable

    • 07:37

      LAURA MICKES [continued]: because at this point in the process,the judges and jurors need to know the trustworthinessof an identification.Now they need to know results from CAC analysis.OK, so I'm also going to talk about the Palmer et al study.It is a very, very good study so I'll talk about experiment 1,where they manipulated exposure time.

    • 07:58

      LAURA MICKES [continued]: That's tough to see, but it says exposure time.So in this study, John talked about it,research assistants approached people in a park,and so one did, and said, are youwilling to take part in this experiment?And another one came out from behind a tree for either 5seconds or 90 seconds.And then later, those participants

    • 08:20

      LAURA MICKES [continued]: were tested on a six-person simultaneous lineup,where half target-present, half were target-absent,and then they collected confidence ratings.So the data from the 5-second and 90-second conditionswere used to construct the ROC plots and the CAC plots.And here's the ROC plot.Discriminability is higher in the 90-second condition,

    • 08:40

      LAURA MICKES [continued]: as shown by the higher ROC.This is no surprise.This was totally expected.So there's more time to encode the face, it should be higher.But the mere fact that the 5-second conditionyields a lower ROC doesn't necessarilymean that a high-confidence identification

    • 09:00

      LAURA MICKES [continued]: is less trustworthy than a high-confidence identificationfrom the 90-second condition.But ROC analysis won't tell us that.And it's important to think about high-confidenceidentifications because these arethe ones that matter for prosecutionsand investigations.So CAC analysis shows us the trustworthinessof an identification, and here's the plot for that.

    • 09:21

      LAURA MICKES [continued]: Here you can see that high-confidence identificationsfrom the 5-second condition was aslikely to be correct as a high-confidence identificationfrom the 90-second condition.OK, so the key point here is that despite yielding a lowerROC, high-confidence decisions madeby those in the short exposure conditionwere as trustworthy as high-confidence decision made

    • 09:45

      LAURA MICKES [continued]: in the long exposure condition.And these considerations illustratewhy ROC analysis is generally lessinformative than CAC analysis for an estimator variable.OK, in the next experiment, I measured another typeof estimator-like variable.So ID decisions can be made based on familiarity

    • 10:08

      LAURA MICKES [continued]: of the suspect's face, or it could alsoinvolve recollection of additional informationor details, such as what the suspect was wearingor the perpetrator was wearing.And because the legal system has no control over whether or notrecollection occurred, it's like an estimator variable.So in this experiment, participantswatch a video of a mock crime, engage in a distractor test,

    • 10:28

      LAURA MICKES [continued]: tested on a six-person simultaneous lineup,half target-presents, half target-absents.And I plotted the ROC curves for recollected and non-recollectedresponses, and here's the plot.ROC was significantly higher for the recollection-basedresponses than for the no-recollection-basedresponses, which just means that those who recollected detailsmore were better able to discriminate

    • 10:49

      LAURA MICKES [continued]: between innocent and guilty suspects.And consequently, you might be temptedto think that an ID accompanied by the recollection of detailsis more trustworthy than an ID that's not accompanied by that.But that's not addressed by ROC analysis,it's addressed by CAC analysis, and here's the plot for that.First of all, those who didn't recollect a lot of detailsweren't even giving high-confidence responses,

    • 11:10

      LAURA MICKES [continued]: but the moderate responses were just as trustworthyas you can see there.OK, so this suggest that participantsappreciate when their memories are not strongand appropriately adjust their confidenceto reflect that fact.And a practical implication here isthat even though overall memory performance is clearly worse

    • 11:31

      LAURA MICKES [continued]: when recollection did not occur, it'snot a relevant consideration for judges and jurorsif confidence ratings are available.So as with the exposure duration manipulation,a moderately high-confidence ID is similarly accurate,and therefore similarly trustworthywhether memory conditions were good or bad.OK, system variables.

    • 11:52

      LAURA MICKES [continued]: Because the legal system does have the opportunityto determine which lineup procedure to use,it's considered to be a system variable.So I'm going to talk about an experimentthat we did where we compared simultaneous and sequentiallineups.So I already told you about this simultaneous procedure.In a sequential procedure, suspect and fillersare presented on at a time, and there are the same three

    • 12:13

      LAURA MICKES [continued]: possible responses-- pick a suspect, pick a filler,pick no one.OK, so in this experiment, they watch a video of a mock crime,there's a delay, they're tested on a six-person lineup, eithersimultaneous or sequential, half target-present, halftarget-absent.And the question in that paper waswhich procedure yielded the better discriminability,and here's the ROC for that.

    • 12:34

      LAURA MICKES [continued]: The ROC was significantly higher for simultaneous procedurethan for the sequential procedure.And by the way, this result's replicated again and again.OK, when the case is in the court of law,judges and jurors need to know, despite the factthat a sequential lineup procedure wasused, how trustworthy is that identification?So they need to know what the CAC plot looks like,

    • 12:55

      LAURA MICKES [continued]: and here's what it looked like.At the medium- to high-confidence levels,proportion correct is the same for both procedures.So according to this experiment, they're equally trustworthy.Given that policymakers are in the positionto implement either lineup procedure,from their perspective lineup procedure is a system variable.And ROC analysis, those results, are what

    • 13:17

      LAURA MICKES [continued]: they need to pay attention to.Those are highly relevant even if the CAC curves associatedwith the simultaneous and sequential proceduresare identical.They need to use the procedure that falls on the higher ROC.OK, in the next experiment, I compared two procedures,simultaneously and a showup.Again, they're system variables for a policymaker.

    • 13:37

      LAURA MICKES [continued]: And a showup, it's just the suspect that's shown,so innocent or not.And there are two possible responses-- pick the suspector pick no one.So in this experiment, I showed a few facesto participants in an incidental learning task,so they didn't know that they'd later betested on one of those faces.And then there was a delay, and then Itested them on a six-person simultaneous lineup

    • 13:59

      LAURA MICKES [continued]: or a showup, half target-present and half target-absent,and asked them about confidence ratings,and then conducted the ROC analysis.So which one yields better discriminability?The simultaneous lineup procedure does,so it falls on a higher ROC than the showup,and this replicates work by [INAUDIBLE] and Wetmoreand their colleagues.So this is good for policymakers to know

    • 14:19

      LAURA MICKES [continued]: because if possible and if these results continue to replicate,a simultaneous lineup procedure should be used.All right, but what about when those system variables becomeestimator variables in the court of law?Is a high-confidence identificationmade on a showup as trustworthy asa high-confidence identification made on a simultaneous lineup?

    • 14:40

      LAURA MICKES [continued]: Here's what the CAC plot looks like.No.So they're not.Proportion correct for the high-confidence responsesfor the showup was not as high as thosefor the simultaneous lineup.So as a system variable, ROC analysis is relevant,so policymakers should endorse the lineup over the showupprocedure.As an estimator variable, CAC analysis is relevant,

    • 15:03

      LAURA MICKES [continued]: so judges and jurors should know the relative trustworthinessif this continues to replicate.Done.Totally out of time?Yes.OK, so we're not talking question time I'm out of?I mean, are we missing our planes?We're 3 minutes to 12, which is the end of everything.

    • 15:24

      LAURA MICKES [continued]: OK.Right.So I think I've made my point.ROC analysis measures discriminability.The result-- policymakers should listen to that, those results,and judges and jurors should hear about the resultsof CAC analysis.OK, so I don't even have time for all this apparently,but I'd be happy to talk to you about any of those things.

    • 15:45

      LAURA MICKES [continued]: And moving forward, we have a lot to do.Now I want to know exactly the CACresults from older participants would be great,after hearing Chad's talk.OK, thanks.

Eyewitness Identification: Which Results Matter for Which Decision-Makers?

View Segments Segment :


Dr. Laura Mickes discusses her research into eyewitness identification and confidence. Some types of results are more important for judges and juries, while others are more important for policymakers. She distinguishes between the two types, then shows the outcomes of her own studies.

SAGE Video Forum
Eyewitness Identification: Which Results Matter for Which Decision-Makers?

Dr. Laura Mickes discusses her research into eyewitness identification and confidence. Some types of results are more important for judges and juries, while others are more important for policymakers. She distinguishes between the two types, then shows the outcomes of her own studies.

Copy and paste the following HTML into your website

Back to Top