Open forensic science, and some general comments on the problems of legalistic thinking when discussing open science

Jason Chin, Gianni Ribeiro, and Alicia Rairden write:

The mainstream sciences are experiencing a revolution of methodology. This revolution was inspired, in part, by the realization that a surprising number of findings in the bioscientific literature could not be replicated or reproduced by independent laboratories. In response, scientific norms and practices are rapidly moving towards openness. . . .

In this article, the authors suggest that open science reforms are distinctively suited to addressing the problems faced by forensic science. Openness comports with legal and criminal justice values, helping ensure expert forensic evidence is more reliable and susceptible to rational evaluation by the trier of fact. In short, open forensic science allows parties in legal proceedings to understand and assess the strength of the case against them, resulting in fairer outcomes. Moreover, several emerging open science initiatives allow for speedier and more collaborative research.

This all sounds good, but I imagine there will be a lot of resistance. The adversary nature of the legal system seems so much different from the collaborative spirit of open science. Indeed, one of the many problems I’ve seen in trying to promote open science is that it sometimes seems to have a kind of legalistic frameworks, focusing for example on arguments about p-values rather than on the substance of what is being studied. For example, when those ovulation-and-clothing researchers criticized my criticisms of their work, they didn’t address questions of measurement and effect size which, in my opinion, were central to understanding what went wrong with their study. One problem with a legalistic attitude toward science is that it can encourage the attitude that, if you just follow the protocols, you’ll do good science. Actually, though, honesty and transparency are not enough.

Related: The purported CSI effect and the retroactive precision fallacy.

20 thoughts on “Open forensic science, and some general comments on the problems of legalistic thinking when discussing open science

  1. What a fruitful subject to explore. I would like to see Nathan Schactman and Susan Haack contribute to this and similar discussions. Sander Greenland and others as well. This is the frontier that is proving to be an epistemological/epistemic challenge.

  2. > collaborative spirit of open science

    “collaborative spirit” makes it sounds like collaboration is somehow natural to science — like it should happen automatically!

    But I think collaborative is the goal, right? There’s plenty of models of scientific discovery whereby people are not necessarily collaborating. Companies competing against each other for profit would be an example (down with capitalism!). But then even benign things — like if I was working on a crossword puzzle, I might be annoyed if you told me the answer.

    The assumption is that if we work together we’ll do better science for humanity or whatever, and that we should work for this because it hasn’t happened naturally.

    Maybe I’d go for:

    “The adversarial nature of the legal system doesn’t seem like it would support collaborative science”

    > Actually, though, honesty and transparency are not enough.

    I quickly read that. I don’t agree with the arguments in that article, though the literal sentence is likely true (all models are wrong!). The way I read your argument (ignoring dog-ate-my-research excuses), it is that because it is possible to honestly make mistakes with poorly designed experiments, honesty is not enough. You also make the argument that people might to start inferring mistakes are because of dishonesty.

    To the first, of course, no one can be expected to be all knowing about research methods. Honesty is how we double check people. I think honesty does the job there. “I believe you made an honest mistake and this result is too weak to publish” is an honest response!

    The second seems more like an argument about what we should do when bad research gets discovered. I don’t know the correct answer there, but I think your argument is incomplete. Was the intent to deceive? Or was it accidental? Do super-successful researchers have too much professional security? Do junior researchers have enough professional security? How are consequences factored in when we know them? When we don’t?

    But anyway it’s not clear to me honesty isn’t enough to solve the second thing too. So I think you are overconfident in your title!

    • Ben:

      1. I would say that all science is collaborative in the weak sense that we are collaborating with our counterparts in the past (standing on shoulders of giants etc.). In any case, I agree that not all science is directly collaborative—in business and the military, there can be a direct avoidance of contact with people who, for intellectual reasons, would otherwise be natural collaborators—but I do think that open science is inherently collaborative. And my phrase was “the collaborative spirit of open science.”

      2. You write, “no one can be expected to be all knowing about research methods. Honesty is how we double check people. I think honesty does the job there.” The problem is that there may be no knowledgeable people in the system to do that double checking! So you can have a group o of peers, who are well connected in the scientific establishment and the news media, all honestly and openly using bad methods. Sure, eventually some outsider or innovator will realize the problem—but it can take awhile before this happens, and lots of resources can be wasted in the meantime. Sometimes the problem is a flawed intellectual framework: it’s not a problem of dishonesty or lack of transparency, and, indeed, people can mistakenly believe that their own personal honesty can make them somehow immune to intellectual errors. That’s what I was seeing a few years ago in my exchanges with the ovulation-and-clothing researchers: they kept insisting on their own personal probity and couldn’t get beyond that to reassess their statistical choices. I do think honesty and transparency are helpful and in some cases even necessary; I just don’t think they’re enough.

  3. “The adversary nature of the legal system seems so much different from the collaborative spirit of open science.”
    Sure, different, but the adversarial system is in many ways so much better because there is a party *enforcing* co-operation. When I testify against someone with every incentive to prove I’m a idiot, I nonetheless have to co-operate because a failure to co-operate will mean my analysis is discarded. We don’t share our data and answer hostile questions because we like doing so, or feel co-operative, but because non-co-operation is forbidden. (I’m not saying there aren’t games that can be played… and are played, e.g. data turned over in bad photocopies or undocumented computer programs in obscure languages, but those games are nothing compared with being ignored in academia.) Now I grant that the *assessment* process in academia is far superior, as the people doing the assessments are actually qualified to do so. But from the standpoint of the *openness* of science, the legal system is vastly superior to the academic system — what is not open to the other side does not exist.

    • I also find the legal system superior in many ways to the academic/research system – at least, potentially. As you say, games are still played. Data can be requested and is sometimes produced. It is too often protected under a claim of “confidentiality” and then becomes difficult to obtain. But at least there are legal channels that can be used. And, cross-examination provides a mechanism under which people are accountable for their analysis. As we have seen, publications often have flaws (at times intentional), and getting data can be difficult, and accountability for mistakes is haphazard at best. Still, it isn’t hard to muster a critique of the legal system. The adversarial process often makes both sides over-hype their claims (e.g., look at the expert testimony in the Harvard admissions discrimination case). But I am more optimistic about improving the legal system than the publication system because there are at least rules than can be used and enforced. Research is more open in an unfortunate way – the rules are malleable. They vary across fields, time, publications, etc. in so many ways that improvement is less assured. But I suspect that if we had legal strictures for scientific research, we might regret what we asked for (I’m not sure, I just worry about unintended consequences).

      • The need to overstate results is a result of the extremely poor assessment filter — judges and juries. If other experts were judging the quality of the research, the claims would be much more modest. I think I mentioned this to you before, but just about my best experience in the expert witness game was serving as special master to a judge assessing the reliability of two opposing experts. It was astonishing how hedged their conclusions became when I was asking them questions, as opposed to when the judge was asking them questions. (It was a bench trial.)

        • Jonathan (another one): > need to overstate results is a result of the extremely poor assessment filter

          Agree, but I think Andrew is pointing to how some people opt out of paying attention to better assessment filters with I am honest so don’t be critical of me. Or a variation I have run into, my good intentions and hard work explaining statistics to the public is really important, so don’t point out any mistakes I might make.

          > astonishing how hedged their conclusions became when I was asking them questions
          I had very similar experiences with prominent academics that pharma firms hired as consultants, their hyped claims quickly gave way with a few targeted technical questions. Unfortunately in academia, some opt out using their colleagues and underlying implicit meta-physics such as its not Kosher to check priors or there is never any need to worry about multiple comparisons in a Bayesian approach.

  4. The problem is that the courts have forgotten what evidence means. It originally meant something perceived by one of the senses. Because causation cannot be seen either common sense principles are applied (the force of a car crash can break a leg) or experts are brought in to explain e.g. how Listeria monocytogenes may be identified, sequenced and traced to its source. The result of failing to put the observation first is that forensic scientists don’t test whether the bite marks really look like Fred’s, as someone has previously claimed, but rather they work backward from Fred to the bite marks and create a narrative of how they might be his though no one has ever made such a claim.

    Similarly with statistics things that have never been noticed or remarked upon by anyone are routinely summoned like so many phantoms out of data provided by the parties. Any allegation that might possibly be true triggers a discovery process which leads to the production of data and thereafter, as in Fred’s case, not a testing of something that has been observed but rather a search for bits of evidence consistent with Fred having been at it again.

    The courts should require the parties to state their theory of the case, how they intend to analyze the requested data and what it ought to show if they’re correct. Alas, having forgotten what evidence means the adversarial system instead compels the courts to turn over all the data to both sides and asks them to come up with their best arguments after having digested the data by whatever method(s) that passes the red-face test.

    • I’m not sure I get exactly what you mean. I get what you mean in the Prosecutor’s Fallacy with Fred and the bite marks, but that would seem to characterize only a tiny portion of what statistical evidence is used for. In most of my cases, the actions of the defendant are observed and the case revolves around one of two questions: (a) liability, i.e. did the plaintiff’s actions cause some harm to the plaintiff; and/or (b) damages, i.e. what would the position of the plaintiff be but-for the actions of the defendant.

      In both cases, experts use the data (admittedly whatever is at hand and relevant) to construct but-for scenarios (these are indeed generally made up, and their plausibility revolves around common sense and academic literature) and use the data to find which of the two competing but-for narratives comports well with the data; or whether neither or both does, in which case the data (and the testimony) is useless to the finder of fact, as it often turns out to be. How is this not evidence?

      • > in which case the data (and the testimony) is useless to the finder of fact, as it often turns out to be.

        I think this is common, but the experts will still argue that their model is relevant and well founded and soforth. What is the evidence that the model is meaningful? too often none. The bite marks stuff isn’t just about the prosecutors fallacy and base rates etc… it’s that these scientists were basically doing dowsing rods… they were in essence unchecked shills for the prosecution, according to the FBIs own reports…

        https://slate.com/news-and-politics/2015/04/fbis-flawed-forensics-expert-testimony-hair-analysis-bite-marks-fingerprints-arson.html

        with stuff like economic damages, where is the evidence that either the prosecution or the defense have an even approximately accurate model? what if they are both wrong, and no better than a monkey throwing darts… with investment managers it’s routine for randomly selected portfolios to outperform a large fraction of them… and yet people still pay for their expertise.

        • Sure. Advocates argue that they’re not just being advocates… And often they’ve convinced themselves. But only the finder of fact’s opinion matters. As to damages, the models are often not solidly grounded, but damages need to be determined… What’s the alternative? Agreed on the FBI.

        • What’s the alternative? You gave one yourself: finders of fact supported by 3rd party experts such as you acting as special master. Allow the jury to request analyses from the special master… Requirements that expert arguments come along with uncertainty quantification. Require the experts to provide a coded model that can be refit by the opposing party with alternative assumptions… etc

          Basically less protection for the experts as experts and more ways to examine the expert’s findings in detail by the finder of fact.

        • I was responding to your point of ungrounded models… not improvements that could be made to the process. My point is that there may be no good data to support a but-for model, and all damages require but-for models. But nonetheless, damages must be assessed somehow.

        • Well there has to be *something* supporting the model. I mean, right now my model of how much you owe me for making me write this post is $158 Million if the model for damages in a lawsuit is no better than that, then it should get the same treatment… $0 damages…

          The overall model may have nothing much to compare it to, but sub-models / assumptions can always be compared at some level. And then it’s up to the experts to assess how well put together the model is, and that’s where the special master could be valuable, since they will know how to assess a model. You can’t expect a jury to know how to debug a statistical model, but you can expect them to listen to a special master give information about what the consequences of the choices made by the modelers are and how reasonable those choices are etc.

          incidentally I didn’t mean anything by that $158 Million remark… thanks for your thoughts on this. I’ve been doing some forensic type projects recently and it’s useful to think about how to do a good job.

        • Ha. Actually, this comes up a lot. You sue me for $158MM. I claim that the true value is zero. You come up with a model that jumbles together a bunch of numbers and, sure enough, there’s $158MM on the bottom line. At this point, I am free to simply cast aspersions on your model and *hope* the judge/jury understands my critique. I might also make my own model that shows an answer of zero. Many attorneys hate method 2, on the grounds that the jury will try, if at all possible, to split the difference, and $79MM looks pretty good.

          But also remember the Daubert rules — the judge is responsible as a gatekeeper to ensure that the model supporting the $158MM is sufficiently reliable to be admitted; implicitly that is reliable if unrebutted. *That’s* where this debate would be carried out.

        • If we go all the way back to Thanatos’ original argument I’d guess he’d say that the Daubert rules are not doing their job. I would guess this because I have inside information about his ideas on the matter, having had out-of-band discussions ;-)

          Also there’s a serious conflict of interest here… Lawyers make money off both prosecuting and defending lawsuits… If they can find a way for lawyer number 1 to sue defendant on behalf of plaintiff for $158M and then jury awards $70M to plaintiff, and insurance company pays out $70M on behalf of defendant, both Lawyer 1 and Lawyer 2 make out like bandits on the billable hours, and the insurance company simply writes it into the cost of doing business. Only consumers who pay the higher prices of goods lose.

  5. People discussing here make it seem like judges in courts around the world follow the same procedure. In the country I live in, studying law, it is required to order a publically registered specialist in the topic by the judge, if the judge seems unable to assess the case due to a lack of scientific knowledge by himself. He can then refuse to weigh all other expert testimonials as valid or weightful evidence, by simply following the public experts assessment. This weighting procesure is a legal(!) option and not necessary. The country is Austria.

    Now ignoring national habits: According to the new york convention international parties can choose arbitration & the law which is taken for any dispute. This allows no further court rulings on the case by national law & the ruling becomes globally enforceable. Therefore one could choose the law for court proceedings in the world, which is deemed the most scientifically accurate. One could even adapt this law with own rules.

    This is in my humble view needed for scientific court procedures globally. Otherwise law can always skew science & makes self correcting of science harder.

  6. I’m one of the authors of this article. And, I agree with Andrew. In forensic science and beyond, transparency is not enough. I do think it’s particularly important in legal contexts, especially when the one side won’t know enough to even know what to ask for.

    There is an interesting parallel with “adversarialism” in science. I think some people used to think the “meat grinder” of peer review was more effective than it was at getting the truth out. In other words, adversarialism has failed in both law and science, and part of the reason is opaque methods.

Leave a Reply to Sameera Daniels Cancel reply

Your email address will not be published. Required fields are marked *