Empirical Scoring System - Multinomial
Usage Guide
Expand AllCollapse All
Multinomial tables are available for grand total scores and multiplicity-corrected subtotals for
test formats with 2, 3, and 4 RQs, with and without the vasomotor sensor. Use of the multinomial
reference tables differ slightly from the use of the original ESS reference tables. Statistical results
using the ESS-M and Bayesian analysis can be thought of as a posterior estimate of deception or
truth-telling.
Each of the below steps can be quickly and easily
accomplished using the LXSoftware ESS-M Report
Generator included in LXSoftware v. 11.8+.
Use these APA Online Reference Tables with the following Guide: Online Reference Tables
-
BEFORE CONDUCTING THE EXAM
- Locate the ESS-M reference table for the sensors and the number of RQs in the test question format
- ESS-M reference tables are available with and without the vasomotor sensor.
- ESS-M tables are the likelihood function (that provides a statistical value for the test scores).
Three of the table columns are of interest to field polygraph examiners.
- score: the grand total or subtotal ESS score
- odds: The odds of deception or odds of truth-telling are the likelihood function for a Bayesian analysis of the posterior probability or posterior odds of deception or truthtelling. The odds can also be though of as the Bayes factor or the posterior odds when the prior odds are 1 to 1.
- oddsLL05: is the .05th percentile lower-limit of the Bayesian credible interval for the posterior odds of deception or truth telling. The lower limit of the credible interval determines the numerical cutscore and tells us the level of significance for a categorical test result.
- Determine the alpha boundaries and numerical cutscores.
- Alpha is commonly set at .05
- Use the oddsLL05 column and locate the largest lower limit value that is greater than the
prior odds (prior is usually 1 to 1, unless other high quality information is available). Then
determine the cutscore by selecting the value from the score column. Locate the lower
limit odds and cutscores for deceptive and truthful scores. Use the table for a single RQ
to determine subtotal cutscores using the columns for the multiplicity-corrected odds
with 2, 3 and 4 RQs.
- Alpha is an administrative decision, often not determined by field practitioners
- Alpha = .01 for increased precision (may increase inconclusive results).
- Alpha = .10 for decreased inconclusive results (may increase error rate)
-
AFTER CONDUCTING AND SCORING THE EXAM
- Calculate the posterior probability and lower limit using the correct reference table for the number of RQs in the test format.
- Using the correct reference table, locate the row for the grand total in the score column and
determine the posterior odds using the same row in the odds column.
- Use the table for a single RQ when using subtotal scores. The odds and oddsLL05 columns
include statistical correction for both truthful and deceptive classifications.
-
Interpret the result (translate the numerical and statistical result into usable human language).
- A complete interpretation of the test result will provide sufficient information to reproduce the analytic conclusion.
- Method of analysis
- Input parameters for probabilistic inference (i.e., prior and alpha)
- An explanation of the procedural decision rules used to obtain the probabilistic and
categorical results from the grand total or subtotal scores.
- Numerical scores (grand total or subtotal) used to determine the probabilistic and
categorical result
- Bayes factor allows recalculation of the analytic results with a different prior and alpha
- Posterior odds of deception or truth-telling (or posterior probability)
- Lower limit of the Bayesian credible interval for the posterior odds of deception or truthtelling
(or posterior probability)
- Use of any statistical correction for multiplicity when using subtotal scores
- An explanation of the empirical and scientific meaning of the test result.
- Categorical test result (DI/SR or NDI/NSR)