Dear Vera, thank you for writing this insightful piece. I am still confused about why you think Fienberg was wrong, and why you think that the polygraph can still be tested using mass screening programs. Was Fienberg wrong because he tried to provide a data analysis of something that shouldn't have been analyzed? And aren't screening tests just going to show that, since we don't know when someone is lying, we cannot tell whether the polygraph's signals indicate whether they're lying? And furthermore, that there seems to be contextual bias, such that polygraph examiners replicate the task-irrelevant information given to them during their analysis? Thanks!
Dear Maria, thank you for your thoughtful comment. The NAS polygraph report's conclusion was based on estimated hypothetical outcomes stemming from an application of Bayes' rule (see https://nap.nationalacademies.org/read/10420/chapter/2#5, Table S-1A and B). This application assumed one causal mechanism, while the programs have three possible mechanisms (see this follow-up post - https://wildetruth.substack.com/p/could-polygraphs-save-lives under "Revisiting Causality").
The programs may net harm security due to the Bayes backfire Fienberg highlighted. The mistake was to assume that we knew that, when further research is needed to assess the possible addition or interaction of all three mechanisms in the field. And to insist the polygraph can’t be scientifically validated because Fienberg couldn’t think of a way to do so with a “lie detector” test under field conditions -- when its avoidance (deterrence) and elicitation (bogus pipeline) mechanisms can be so validated.
Confirmation bias warrants further thought, but is not strictly relevant here.
Dear Vera, thank you for writing this insightful piece. I am still confused about why you think Fienberg was wrong, and why you think that the polygraph can still be tested using mass screening programs. Was Fienberg wrong because he tried to provide a data analysis of something that shouldn't have been analyzed? And aren't screening tests just going to show that, since we don't know when someone is lying, we cannot tell whether the polygraph's signals indicate whether they're lying? And furthermore, that there seems to be contextual bias, such that polygraph examiners replicate the task-irrelevant information given to them during their analysis? Thanks!
Dear Maria, thank you for your thoughtful comment. The NAS polygraph report's conclusion was based on estimated hypothetical outcomes stemming from an application of Bayes' rule (see https://nap.nationalacademies.org/read/10420/chapter/2#5, Table S-1A and B). This application assumed one causal mechanism, while the programs have three possible mechanisms (see this follow-up post - https://wildetruth.substack.com/p/could-polygraphs-save-lives under "Revisiting Causality").
The programs may net harm security due to the Bayes backfire Fienberg highlighted. The mistake was to assume that we knew that, when further research is needed to assess the possible addition or interaction of all three mechanisms in the field. And to insist the polygraph can’t be scientifically validated because Fienberg couldn’t think of a way to do so with a “lie detector” test under field conditions -- when its avoidance (deterrence) and elicitation (bogus pipeline) mechanisms can be so validated.
Confirmation bias warrants further thought, but is not strictly relevant here.