Ok, finally I get it … and I should have gotten it earlier. You’re ignoring question 1 and focusing on question 2, while the majority of the posters are ignoring question 2 and focusing on question 1. I apologize for missing this point.
However, I can’t resist pointing out that, if you had posted some evidence to support your assertion as was requested many times, I think I would have gotten it much sooner.
It is certainly interesting and relevant whether there is a positive correlation between lying and polygraph measurments. I think that the question of whether the polygraph is a valid tool is much broader and more important.
I do not dispute that average polygraph scores are consistently better than chance.
Contrary to your assertion, there is evidence regarding this in the posted links. It took me about three minutes to find it. Some of that time was spent finding the links in the thread. It took me about two minutes to edit out the line breaks so it looks reasonable in a quote. You probably could have found it in 2-3 minutes from the original post, had you invested the effort; and you would have has the links to the tables right in front of you. From Scientific Validity of Polygraph Testing: A Research Review and Evaluation - Review and Analysis of Polygraph Field Studies:
"… data from the available field studies were analyzed to ascertain whether polygraph examination accurately differentiate deceptive suspects from nondeceptive subjects. For this analysis, the outcome frequencies for each category were converted to percentages, and average percentages within each category were calculated. A measure of predictive association (lambda, see 64,73) was also calculated, although the use of a single measure is very limited due to the wide variability in study design.
The lambda(b) index shows the proportional reduction in the probability of error in predicting one category (in this case, deception) when a second category (in this case, polygraph examination results) is known. If the information about the second category does not reduce the probability of error in predicting the first category at all, the index is zero, and one can say that there is no predictive association. On the other hand, if the index is 1.00, no error is made in predicting
one category from another, and there is complete predictive association. Essentially, lambda provides an index that translates to the percent improvement over the base rate and indicates the percent improvement in prediction when the polygraph examinations are considered versus no further information. There is almost no direct research on the percent improvement of the polygraph over other forms of investigation (cf. 207). The results of this analysis of predictive association are shown in tables 4 and 5. The average lambda~ across studies is 0.65, which means that, on the average in these field studies, the polygraph diagnosis reduced 65 percent of the error of chance prediction. The lambda(b) for individual studies ranged from 0.13 to 0.90.
To summarize, the analysis of the 10 field studies included in the analysis indicates that while polygraph examinations using CQT in criminal investigations detect deceptiveness and nondeceptiveness better than chance, there is also what in some cases might be considered a high error rate, particularly for nondeceptive subjects. The one study which tested the validity of the relevant/ irrelevant question technique (the general question test (GQT) portion of the Bersh study) also detected deceptiveness and nondeceptiveness better than chance.
Variation Among Studies
As implied in the introduction to this section, the use of a single statistic or summary number to describe the results of field tests of validity may be misleading. As shown in table 3, although the field studies of polygraph validity are similar in that almost all of them tested control question techniques in criminal investigations, they differ in operationalizations of ground truth and type of examiner decision. The result is that there is a great deal of variability in the results of studies. Correct guilty detections range from 70.6 percent in one condition of the Bersh study to 98.6 percent in a condition of the Wicklander and Hunter study. Correct innocent detections are even more variable, ranging from a low of 12.5 percent in the Barland and Raskin judicial outcome study to a high of 94.1 percent in one condition of the Bersh study. Table 5 also indicates the range of incorrect judgments and inconclusive among studies. False negatives range from 29.4 percent of the Bersh study to zero percent. False positives range from 75 percent in Barland and Raskin (22) to zero percent in two studies. Inconclusive range from zero to 25 percent. This section compares studies that used comparable operationalizations of construct and criterion validity in an attempt to discover reasons for the range of results. However, even using this method results in considerable variability."
I hoipe you realize that, although the tables are online, it’s against board rules to link directly to them
OK, question 2 is put to bed.
Let’s move on to what interests me and, I suspect, interests any others who may still be reading this thread. Care to discuss whether the polygraph is accurate enough to be useful as a tool in any employment decsisions?