The Polygraph and Lie Detection (2003) / Chapter Skim
Currently Skimming:
Appendix H: Quantitative Assessment of Polygraph Test Accuracy
Appendix H: Quantitative Assessment of Polygraph Test Accuracy
Pages 340-353
The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.
From page 340... ...
reported data allowing ROC curve estimation from two points, while 17 laboratory studies provided only one estimated ROC point and a single laboratory study provided ten estimated points on its single ROC. The median sample size in a dataset, including both genuine or programmed deceptive and nondeceptive examinees, was 49, with mean 62.4 (median and mean 48 and 59.6, respectively, for laboratory datasets, and 100 and 83.1, respectively, for field datasets)
|
From page 341... ...
The quality scores for protocol documentation, data analysis, internal validity, and salience were correlated as might be anticipated. With signs adjusted so that positive correlations represent agreement in quality, correlations of salience score with protocol documentation score, data analysis score, and internal validity score were respectively 0.33,0.42, and 0.49.
|
From page 342... ...
Correlations of internal validity score with protocol documentation and data analysis scores were 0.37 and 0.66, respectively, with a correlation of 0.30 between documentation and data analysis scores. Figure H-1 plots salience against internal validity scores, with points jittered slightly left and right to avoid overlap.
|
From page 343... ...
Figure H-2 shows six theoretical ROCs from this "equivariance" binormal model, with respective A values of 0.5, 0.6, 0.7, 0.8, 0.9, 0.95, along with the reverse diagonal line corresponding to sensitivity = (1false positive rate) , which can alternately be interpreted as false positive rate = false negative rate.
|
From page 344... ...
estimates of A was much smaller, generally 0.01-0.03. Had sample sizes been large enough to allow the use of a binormal estimate in all cases, we conjecture that the median values of A reported in Chapter 5 and below would have increased by 0.02-0.03 for laboratory studies and perhaps 0.01 for field studies.
|
From page 345... ...
NOTE: Vertical lines are the length of one standard error; they extend downward from the accuracy index value for convenience of presentation. have arisen by chance from studies with particularly small sample sizes, but in fact these groups of studies have sample sizes comparable to the larger collection.
|
From page 346... ...
Although there are many more comparison question than concealed information datasets, the trend appears downward for both. In view of the low methodological quality ratings that we assigned to many of these studies, we readdressed the above points in the subset of 14 laboratory studies with internal validity scores better than 3 on a 1-5 scale (with 1 as best)
|
From page 347... ...
In view of the issue of possible research bias associated with the sponsorship of research, we attempted to classify studies by source of funding: internal studies by agencies with and without polygraph programs, studies by other groups externally funded by agencies with and without polygraph programs, and locally funded studies by academics. It was not always possible to do this without ambiguity, because funding sources were not always fully clear from the publications and because of the close connections of most researchers in the field to the polygraph
|
From page 348... ...
NOTE: Vertical lines are the length of one standard error; they extend downward from the accuracy index value for convenience of presentation. profession or one or more government polygraph agencies.
|
From page 349... ...
NOTES: Each dataset is symbolized on the plot by the salience score that committee reviewers assigned to the study from which it was derived. Boldface type indicates concealed information studies; italic type indicates comparison question studies.
|
From page 350... ...
The standard errors are a bit smaller for the field studies (which have larger sample sizes) than for the laboratory studies, with that for the first study shown in Figure H-10 artificially small due to the proximity of the estimate to the maximum.
|
From page 351... ...
Of 122 polygraph tests from criminal investigations that were examined in this particular study, there were seven inconclusive tests but no false positive or false negative errors. Due to the small number of field studies available, we use a slightly more lenient criterion in restricting to higher quality studies and consider
|
From page 352... ...
Thus, the results of the field studies, if taken literally, suggest that the average accuracy of polygraph testing in field specific-incident investigations is similar to and may be slightly higher than that found in polygraph accuracy studies using laboratory models. However, inas
|
From page 353... ...
APPENDIX H 353 much as none of these studies isolates the determination of truth from performance on the polygraph examination, they share a bias that must inflate apparent accuracy, perhaps to a significant degree. This result, in conjunction with the tendency for diagnostic test performance to degrade in field application relative to under laboratory conditions, leads us to believe that actual polygraph test performance in the field, if it could be measured against actual truth, would show a lower level of accuracy than indicated by the field or laboratory datasets we examined.
|
Key Terms
This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More
information on Chapter Skim is available.