The link below is the APA's official free-access path to the executive summary, full study, and FAQs of their

*Meta-Analytic Survey of Criterion Accuracy of Validated Polygraph Techniques.* http://www.polygraph.org/section/validated-polygraph-techniques/executive-summar... For our discussion purposes, questions 23 and 24 from the FAQ document might be a good starting point...

** **

"23. What is the accuracy of the polygraph? There are many factors that appear to affect polygraph accuracy. Among these factors are: the testing

technique, number of charts, number of relevant questions, independence or non-independence of the

relevant questions, ambiguity of the relevant questions, feature selection, scoring rules, decision rules,

use of the "successive hurdles" model, examiner competency, use of a "stim test", CIT or CQT, and

other factors. The use of best practices will allow maximal accuracy.

Questions about polygraph test accuracy are measurement issues involving several dimensions of

concern, including correct decisions, errors, and inconclusive results. Bayesian methods for measuring

and estimating test accuracy, while easily understood are non-resistant to differences in base-rates or prior probabilities. Accuracy estimations based on inferential statistics will be more resistant to base-rate differences.

Although there is no single statistic for characterizing polygraph accuracy, the meta-analysis compared

the accuracy of validated polygraph techniques using the unweighted average of correct decisions, excluding inconclusive results, for confirmed deceptive and confirmed truthful cases along with the unweighted inconclusive rates for deceptive and truthful cases. Because it is impossible to conduct every possible test on every possible person, answers to questions about polygraph test accuracy are always estimates and always depend on the representativeness of the sample data. Scientific calculations of test accuracy include estimates of the range of error, in the form of 95% confidence

intervals that indicate the predicted range of test accuracy if it were possible to complete numerous

additional replications of the accuracy studies.

The 95% confidence interval for unweighted accuracy was

.798 to .940, with a mean of .869. The confidence interval for the unweighted average inconclusive rate was

.068 - .187, with a mean of .128. Event-specific diagnostic techniques, conducted around a single known or alleged problem, have accuracy levels higher than the unweighted average. Multiple issue screening techniques, involving several independent target questions in the absence of any known or alleged problem, have accuracy

levels lower that than the unweighted mean along with higher inconclusive rates.

**24. What is the reliability of the polygraph?** The 95% confidence interval for available Kappa reliability statistics was .443 to .842, with a mean of .

642. The confidence range for inter-rater decision agreement, excluding outlier results and excluding

inconclusive results, was .741 to .999, with a mean of .901. Pearson correlation coefficients produced a

confidence interval of .649 to .999 with a mean of .876. All of these are interpreted as good, if imperfect, reliability coefficients."