Here is an excerpt from http://fas.org/sgp/othergov/polygraph/ota/conc.html to get the discussion under way.

In reading this, one can see that the reviewing entity states quite clearly that polygraph does show a better than chance ability to detect deception.  "The preponderance of research evidence does indicate that, when the control question technique is used in specific-incident criminal investigations, the polygraph detects deception at a rate better than chance.."  


Quote:

Scientific Validity of Polygraph Testing: A Research Review and Evaluation A Technical Memorandum Washington, D. C.: U.S. Congress Office of Technology Assessment OTA-TM-H-15 November 1983 Chapter 7, Section 3, Sub-Section 1 SPECIFIC SCIENTIFIC CONCLUSIONS IN POLICY CONTEXT Specific-Incident Criminal Investigations A principal use of the polygraph test is as part of an investigation (usually conducted by law enforcement or private security officers) of a specific situation in which a criminal act has been alleged to have, or in fact has, taken place. This type of case is characterized by a prior investigation that both narrows the suspect list down to a very small number, and that develops significant information about the crime itself. When the polygraph is used in this context, the application is known as a specific-issue or specific-incident criminal investigation. Results of OTA Review The application of the polygraph to specific-incident criminal investigations is the only one to be extensively researched. OTA identified 6 prior reviews of such research (summarized in ch. 3), as well as 10 field and 14 analog studies that met minimum scientific standards and were conducted using the control question technique (the most common technique used in criminal investigations; see chs. 2, 3, and 4). Still, even though meeting minimal scientific standards, many of these research studies had various methodological problems that reduce the extent to which results can be generalized. The cases and examiners were often sampled selectively rather than randomly. For field studies, the criteria for actual guilt or innocence varied and in some studies were inadequate. In addition, only some versions of the control question technique have been researched, and the effect of different types of examiners, subjects, settings, and countermeasures has not been systematically explored. Nonetheless, this research is the best available source of evidence on which to evaluate the scientific validity of the polygraph for specific-incident criminal investigations. The results (for research on the control question technique in specific-incident criminal investigations) are summarized below:    * Six prior reviews of field studies:          * average accuracy ranged from 64 to 98 percent.    * Ten individual field studies:          * correct guilty detections ranged from 70.6 to 98.6 percent and averaged 86.3 percent;          * correct innocent detections ranged from 12.5 to 94.1 percent and averaged 76 percent;          * false positive rate (innocent persons found deceptive) ranged from O to 75 percent and averaged 19.1 percent; and          * false negative rate (guilty persons found nondeceptive) ranged from O to 29.4 percent and averaged 10.2 percent.    * Fourteen individual analog studies:          * correct guilty detections ranged from 35.4 to 100 percent and averaged 63.7 percent;          * correct innocent detections ranged from 32 to 91 percent and averaged 57.9 percent;          * false positives ranged from 2 to 50.7 percent and averaged 14.1 percent; and          * false negatives ranged from O to 28.7 percent and averaged 10.4 percent. The wide variability of results from both prior research reviews and OTA?S own review of individual studies makes it impossible to determine a specific overall quantitative measure of polygraph validity. The preponderance of research evidence does indicate that, when the control question technique is used in specific-incident criminal investigations, the polygraph detects deception at a rate better than chance, but with error rates that could be considered significant. The figures presented above are strictly ranges or averages for groups of research studies. Another selection of studies would yield different results, although OTA?S selection represents the set of studies that met minimum scientific criteria. Also, some researchers exclude inconclusive results in calculating accuracy rates. OTA elected to include the inconclusive on the grounds that an inconclusive is an error in the sense that a guilty or innocent person has not been correctly identified. Exclusion of inconclusive would raise the overall accuracy rates calculated. In practice, inconclusive results may be followed by a retest or other investigations.