Objective
Scoring System – version 3
by Raymond Nelson, Mark Handler, and Donald Krapohl 
Home Poster #1  Development and validation Poster #2  Validation with mixed format field investigation polygraphs Poster #3  Additional validation with LEPET and PCSOT screening exams The OSS3 Report ** NEW ** Lafayette Instrument's OSS3 Report Other Materials 
How to read the OSS3 Report
There are six main sections in the OSS3 report, the six sections are listed below and shown highlighted in red below.
Section 1 – Reading the results In the results section you will be given the test result, and pvalue, or level of significance, for that result, along with information about the type of test technique and the decision rules used to score the test. The OSS3 algorithm regards all variants of Zone techniques as single issue examinations in which it is inconceivable that a test subject could lie to one or more test questions while being truthful to other questions regarding involvement in a single known issue. All variants of the MGQT format are regarded by the OSS3 algorithm as multiple facet examinations, in which it is theoretically conceivable that a test subject could lie to one or more questions while being truthful to other questions about various forms of involvement in a single known issue. The OSS3 algorithm does not make truthful and deceptive classifications simultaneously, and any indication of significant reactions to any test question precludes the possibility of a truthful classification for all other questions. Screening examinations, for risk assessment or compliance monitoring, are regarded by the OSS3 algorithm as mixedissues examinations in which it is conceivable that a test subject could lie to one or more test questions while being truthful to other questions about involvement in several distinct unknown issues (for which there is no known allegation). Again, the OSS3 algorithm does not make truthful and deceptive classifications simultaneously, and any indication of significant reactions to any test question precludes the possibility of a truthful classification for all other questions. By default, the OSS3 algorithm will score all variants the Zone and MGQT techniques using the OSS3/Senter algorithm which uses twostage rules (Senter and Dollins, 2003; Senter, 2003) that have been shown to provide optimal classification accuracy, including optimum sensitivity to deception, balanced accuracy, and minimal inconclusive classifications. Screening exams are scored using the OSS3/Screening algorithm which is optimized to maximize sensitivity to deception when scoring multiple distinct/mixed question targets, and uses an omnibus test of between question variance to reduce inconclusive classifications of test data that cannot be classified as deceptive. Truthful results, reported as “No Significant Reactions,” occur when the observed pvalue indicates a statistically significant difference between the observed numerical score and that expected from deceptive test subjects, using normative data obtained through bootstrap training with the confirmed single issue examinations from the development sample. Deceptive results, in which an observed pvalue indicates a statistically significant difference between the observed numerical score and that expected from truthful persons, and are reported as “Significant Reactions.” When the observed pvalue fails to meet decision alpha thresholds for truthful or deceptive classification the test result will be reported as “Inconclusive.” No opinion can be rendered regarding those results. The OSS3 algorithm regards the measurable presence or absence of reactions, and observed mathematical significance, as unequivocal according to the specified decision alpha level. Professional opinons about deception or truthfulness are made by field examiners according to professional standards of practice, training, and agency policies regarding tollerance for risk or error. The pvalue for truthful results can be thought of as the estimated proportion of deceptive persons that might produce a numerical score of equal or greater magnitude, (obtained as the mathematical mean of the combined weighted means of standardized lognormal R/C_{mean} ratios). The pvalue for truthful results may also be thought of as the probability that a test result was produced by a deceptive person, or the probability that a test result is produced by chance or error. Conversely, pvalues for deceptive classifications can be thought of as the estimated proportion of truthful persons that might produce a numerical score of equal or greater magnitude. Pvalues for deceptive results may also be though of as the probability that a test result was produced by a truthful person, or the probability that a test result was produced by chance or error. Users of the OSS3 computer scoring algorithm should remain aware that it would be a theoretically incorrect oversimplification to attempt to describe a pvalue as a probability of deception or truthfulness.
Section 2 – Looking at the details In the details section you will find four subsections: test details, spot scores, decision alpha, and the component weights. The test details consist of the names of the Examinee and Examiner, in addition to the examination date and date on which the examination was scored with the OSS3 computerized scoring algorithm. Because numerical scores are aggregated through the mathematical mean of the combined weighted mean of component scores, and not through simple cumulation, the range of scale for individual questions is identical to the range of scale for the aggregated numerical scores (grand mean). Pvalues for individual questions are therefore calculated in the same manner as the pvalue for the overall score. Classification results for individual questions are provided on the OSS3 report for multiplefacet and mixedissues examinations. For eventspecific/singleissue examinations, the individual question stimuli are assumed to be logically synonymous and are not included in the examination report. Other information in the detail session may be of interest to an examiner, auditor, or quality assurance reviewer, and include the component contribution or weighting, and the specified decision decision alpha thresholds at which deceptive and truthful classifications are permitted. Also included is information about the use of a Bonferonni correction to the potential for inflated alpha levels with multiplefacet and single issue exams when using twostage or spot scoring rules.
Section 3 – Reviewing the questions The questions section will list the actual relevant question targets from the intest phase of the examination. The OSS3 computerized scoring algorithm can score examination techniques consisting of two to four relevant questions.
Section 4 – Verifying the measurements The measurement section displays all of the physiological measurements that were obtained for the scoring algorithm. The measurements are separated into separate matrices for each test chart, with each row of each matrix containing the measurement data for a single component. The OSS3 algorithm is designed to score examinations consisting of three to five test charts, and will not score examinations for which there are fewer than three test charts. Questions themselves are aligned vertically between the matrices for the three to five test charts. Measurements that were previously marked as artifacts are not included in the mathematical analysis, and are identifiable by the text value appended to the beginning of the numeric measurement. Measurements that have been edited prior to analysis are also easily identified by the “&” within the numeric string composed of the original and edited measurements. In this way the original measurement value is retained for auditing purposes, and edited measurement values are easily identified in the OSS3 report. The OSS3 algorithm does not use measurement values that would result from descending EDA or cardio tracings. Negative measurement values are replaced with the value zero (0) at the time of measurement. Zero values in the measurement table alert the examiner or reviewer that a tracing segment was descending and uninterpretable at the time of the question stimulus.
Section 5 – Understanding the Lognormal ratios Standardized lognormal ratios are the final result of the mathematical transformation and aggregation of the Kircher measurements. Transformations begin with the calculation of R/C_{mean} ratio, using the mean comparison values within each component. In order to accommodate polygraph techniques with two to four target questions and three to five test charts, it is necessary to plan for the aggregation of data through averaging instead of cumulative totals. Because the calculation of R/C_{mean} ratios creates an asymmetrical distribution of possible data values, with lognormal distribution shape, consisting of all possible positive real numbers with a mean of one (1), averaging is not appropriate without some initial transformation of the data. Transformation of the asymmetrical R/C_{mean} ratios to their natural logarithms results in a symmetrical distribution of positive and negative real numbers with a mean of zero, which can be averaged without introducing mathematical bias to the data. Lognormal values are standardized per each component, using normative data (means and standard deviations) obtained through bootstrap resampling of the training sample. Standardized lognormal R/C_{mean} ratios are presented on the OSS3 report, and the component values are further combined through weighted averaging for each iteration of each relevant stimulus, using the component weights that are indicated in the details section of the OSS3 report. Those weighted mean values are then combined, through the mathematical mean of each of the relevant stimuli, between the three to five test charts. Mean values for the two to four relevant questions are further averaged to achieve a grand mean score that is compared to the cumulative normal distributions for confirmed truthful and confirmed deceptive cases, using normative data that was obtained through bootstrap resampling. When using twostage, spot scoring, or screening rules, mean values for each of the individual relevant stimuli are also compared to the cumulative normal distribution for truthful subjects using the specified decision alpha for deceptive classifications. In addition to the standardized lognormal R/C_{mean} ratios, the OSS3 report displays the mathematical mean of the weighted means of the lognormal R/C_{mean} ratios for each of the test stimulus targets, along with the grand mean score for the combined stimulus questions.
Section 6 – Reading the remarks The remarks section will display any remarks entered by the examiner during or after the examination. Extrapolygraphic information, that may affect the interpretation of a test, should be entered into the remarks section, including information about health, medications, mental status, noncooperation and attitudinal information might be useful to include in this field. Alternatively the remarks section may be used to display information obtained from the examinee during the posttest review.
Summary The OSS3 report summarizes all critical information used to achieve the test result, including information about the advanced user settings of the OSS3 algorithm, along with auditing information regarding the examiner's review or editing of the raw data from the automated physiological measurements.
Disclaimer:
The OSS3
developers have provided the scoring algorithm without any warranties
or assurances. All use of the OSS3 computerized polygraph scoring
algorithm in whole or part is done with the understanding that it is
the examiners responsibility, at the time of the examination, to assure
that the examinee's truthfulness or deception is the sole cause for the
presence or absence of significant reactions. Just as legal evidence is
not itself a legal finding of fact, which is the role always of judges
and juries, it is an ethical axiom in testing sciences that tests
themselves do not make decisions but provide information to the
professionals who use tests. Decisions are always made by
professionals, who are trained and authorized to use test data and
other information as a basis for professional opinion. Decisions, and
decision thresholds are sometimes necessarily influenced by
matters of administrative policy. While every effort has been made to
design and validate a scoring algorithm with maximum field utility and
maximum classification accuracy, effective use of the OSS3 algorithm
in field settings depends in large part on a properly administered
examination. It is the responsibility of each professional to become
familiar with the empirical methods and principles underlying any test
method and scoring paradigm, including its strengths and limitations,
and to use testing information according to one's level of competence,
training, and established policies. Installation
and use of the OSS3 algorithm signifies the users agreement to
indemnify the developers against any and all forms of liability
associated with its use or misuse.
