Eye Detect Lie Detection Testing

Feel free to scroll down and find articles that apply to your questions.

You can also contact Mr. Mike Delbow of Assured Polygraph LLC to consult. Mr. Delbow is the first polygraph examiner to field study Eye Detect in Minnesota. 

   Along with polygraph, Mr. Delbow administers them on Law enforcement investigative exams, treatment monitoring, infidelity and relationship issues and equips and supports facilities that choose to administer their own Eye Detect instrument. 

Mr. Delbow has been a polygraph examiner specializing in sex related issues alone for over 15 years. He is a full member of the American Polygraph Association, National Polygraph Association and is the former yet founding Vice President of the Minnesota Polygraph Association. 

You can contact him at 612-501-0031 or email him at mdelbow@bevcomm.net for questions, cost or a consultation.

On April 11, 2008, Eye Detect had been admitted into evidence in a Daubert hearing in New Mexico. This new technology is changing the world of lie detection. 

In 2019, Eye Detect had been admitted into evidence in a China hearing in Iowa (child in need of assistance). 

In 2019, Minnesota starts using Eye Detect in contracted clinical monitoring or sex offenders to rule out false positives in polygraph. DFO Corrections

Slashing the expense, simplifying the process, reducing exam time over traditional polygraph methods. Eye Detect is putting the control back in the theosophist a parole officers’ hands as any educated individual can operate the instrument. 

This instrument is currently being used in courts, behind the wall in corrections, law enforcement screening, criminal investigations and specific issue exams all over the country. It is as simple as this: 

1. Instruct the examinee have a seat. 
2. Instruct the examinee to follow the directions on the computer. 
3. Upon completion of the exam, upload the data to the server.
4. Wait 5 minutes and receive a printed report with the determination made by Converus. Therefore, any officer or agent, or professional can administer the exam because the test proctor does not make the determination of the outcome. This maximizes standardization as the technology is rendering the decision instead of a human being thus eliminating human error and bias from a challenging or difficult examinee.

Eye Detect is a great option for someone or an agency needing to assess credibility on the go as it is very mobile and can be set up within minutes. It is a great option for facilities that don’t have examiners readily available. 

It is also a great option to rule out examinees that claim they just can’t pass the polygraph test due to anxiety. Eye Detect measures 16 different variances of one’s ocular motor movements. 

While polygraph being just as accurate measures one’s physiological responses to deception, Eye Detect also measures physiological responses but more being the cognitive load thus eliminating many of the issues polygraph examiners have with measuring traditional physiological responses due to anxiety, medication, apnea etc.

Administering an Eye Detect alongside a polygraph with the same questions can produce incredible accuracy, up to 98%, thus nearly ruling out the argument of false positive.

It has been said that Eye Detect can be used anywhere in Minnesota where a polygraph is used, and therefore, Eye Detect meets the legal definition of a “polygraph” in MN.

A very recent case in which the MN Court of Appeals stated polygraph does not meet Frye standards and the test results will not be allowed admissible in civil or criminal court, yet we still have permission to monitor our offenders with them. Eye Detect being equally valid in monitoring offenders.

There is a general definition of “polygraph” in Minnesota law 611A.26 Subd. 6 Definitions: d) “Polygraph examination” means any mechanical or electrical instrument or device of any type used or allegedly used to examine, test, or question individuals for the purpose of determining truthfulness.

Eye Detect meets the legal definition of a “polygraph” in MN. And the commissioner can require an offender to take “polygraph” as a part of their treatment. And no polygraph test results, themselves, are admissible under Frye.

Here is the ruling, although evidence of a polygraph is generally inadmissible at trial as a matter of law, Minn. Stat. § 609.3456 (2014) allows the district court to order that an offender submits to polygraph examinations as a condition of probation when the offender has received a stay of imposition or execution of sentence. 

Specifically, section 609.345 provides: (a) A court may order as an intermediate sanction

under section 609.135 and the commissioner of corrections may order as a condition of release under section 244.05 or 609.3455 that an offender under supervision for a sex offense submit to polygraph examinations to ensure compliance with the terms of probation or conditions of release.

1 Frye v. United States, 293 F. 1013, 1014 (D. C. Cir. 1923); State v. Mack, 292 N.W.2d

764, 768 (Minn. 1980).6

Volume 10 • 2016 • Number 4 (38)

John C. Kircher*

David C. Raskin

University of Utah,

Salt Lake City, UT, USA

Laboratory and Field Research

Keywords: "ocular-motor”, “ocular-motor detection test", "instrumental detection of deception"

Disclosure: The authors have financial interests in Converus Inc. (www.converus.com),

a company that has commercialized the technology described in this report. We have

disclosed those interests to the University of Utah and have in place an approved plan for managing any potential conflicts that arise from involvement in Converus.

The present paper reviews the rationale and theoretical assumptions that underlie

the ocular-motor deception test (ODT) as well as empirical evidence of its criterion-related validity. 

The research suggests that the ODT may contribute to pre-employment and periodic screening programs, particularly in government agencies concerned with law enforcement and national security.

* John.Kircher@utah.edu

** davidc.raskin@me.com

DOI: 10.1515/ep-2016-0021

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

160 JOHN C. KIRCHER, DAVID C. RASKIN

Rationale underlying the Ocular-motor Deception Test (ODT)

Cook et al. (2012) introduced a new method for detecting deception called the ocular-motor deception test (ODT). 

In contrast to the polygraph, the ODT is automated and can be completed in approximately 40 minutes. A computer presents voice-synthesized and written instructions followed by written true/false test statements concerning the examinee’s possible involvement in illicit activities. 

The instructions inform the examinee that if they do not answer quickly and accurately, they fail the test. The examinee then reads statements presented serially by the computer while a remote eye tracker recording eye movements and pupil size changes. 

The examinee presses a key on the keyboard to answer true or false. The computer processes the ocular-motor data, combines its measurements in a logistic regression equation, and classifies the individual as truthful or deceptive on the test.

The ODT uses a test format known as the Relevant Comparison Test (RCT). Originally, we developed the RCT as a new polygraph technique for use at ports of entry to screen travelers for possible trafficking of drugs and/or transporting explosives (Kircher et al., 2012). 

The RCT includes questions about two relevant issues (R1 and R2) that are intermixed with neutral questions, and it uses the difference between reactions to the two sets of relevant questions to determine if the examinee was truthful or deceptive to either of the relevant issues. 

Each relevant issue serves as a control for the other. Examinees reacting more strongly to questions concerning one of the issues are found deceptive in their answers to questions about that relevant issue.

Examinees who show little or no difference in reactions to the two sets of relevant questions are considered truthful to both issues.

The ODT is based on the assumption that lying is cognitively more demanding than telling the truth. 

A recurrent theme in the literature on deception detection techniques (Johnson, Barnhardt, & Zhu, 2005; Kircher, 1981; Raskin, 1979; Steller, 1989; Vrij, Fisher, Mann, & Leal, 2006). In contrast to truthful people, a deceptive individual must identify questions answered truthfully and questions answered deceptively. When they recognize a question as inculpatory, they must inhibit the pre-potent truthful response and do so consistently, quickly, and accurately. 

While they are performing the task, deceptive individuals may also self-monitor their performance for signs revealing their deception, by either answering too slowly or making too many mistakes. 

The recruitment of mental resources to accomplish these additional cognitive and meta-cognitive activities could account for the observed impact on pupil dilation, eye movements, response time, and error rates (Hacker et al., 2012; Kahneman, 1973; Loewenfeld, 1999; Rayner, 1998).

The pupil reacts not only to cognitive load but also to emotional stimuli. Several investigators have reported that emotional stimuli evoke pupil responses whose mag Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

LABORATORY AND FIELD RESEARCH ON THE OCULAR-MOTOR DECEPTION TEST 161

nitude depends on the intensity but not the valence of the emotional stimulus (Bradley,

Micolli, Escrig, & Lang, 2008;. Hess & Polt,1960; Hess & Polt, 1964; Steinhauer,

Boller, Zubin & Pearlman, 1983). 

Polygraph tests are based on the concept that deceptive individuals will show stronger emotional responses to test questions answered deceptively than to those answered truthfully. To the extent that emotional reactions to test questions distinguish deceptive from truthful individuals, pupil responses should reflect those differences and be diagnostic of deception. 

Consistent with this prediction, several investigators have reported that during concealed information and probable lie polygraph tests the pupil dilates more when people are deceptive than when they are truthful (e.g., Bradley & Janisse, 1979; Dionisio et al., 2001; Janisse & Bradley, 1980; Webb et al., 2009).

A reader who has difficulty reading or comprehending text shows more eye examinations, pupil enlargement, and longer reading times (Rayner, 1998; Rayner, Chace, Slattery & Ashby, 2006). 

If deceptive individuals experience greater cognitive load and difficulty processing test items than truthful individuals, we should see differences between the groups on these measures. Mock crime laboratory research on the ODT.

We have conducted a series of laboratory and field studies to determine if ocular-motor measures discriminate between truthful and deceptive individuals. 

The laboratory studies use a mock crime procedure that we modeled after laboratory research on polygraph techniques. 

Realistic mock crime experiments produce diagnostic effects on electrodermal, cardiovascular, and respiration reactions that are similar to those obtained from actual criminal suspects (Kircher, Horowitz & Raskin, 1988; Raskin & Kircher, 2014). 

In our experiments, we recruit participants from the university campus or the general community and randomly assign them to guilty and innocent treatment conditions. We instruct guilty participants to commit a mock crime, such as stealing an exam from a professor’s office, or taking $20 from a secretary’s purse, and then lie about it on the ODT. 

To simplify the research design, we have begun conducting experiments with one rather than two mock crimes. Participants in these experiments are led to believe that some guilty participants take an exam from a professor’s office, whereas others take $20 from a secretary’s purse. All guilty participants take $20 from the purse. 

Because examinees in field settings usually are highly motivated to pass the test, we offer participants a substantial monetary bonus to appear innocent of the crimes. 

The ODT consists of 16 True/False statements concerning the theft of the $20 (e.g. “I took the $20 from the secretary’s purse.”), 16 statements concerning the theft of Authenticated | davidc.raskin@me.com author’s copy Download Date | 2/26/17 8:29 PM 162 JOHN C. KIRCHER, DAVID C. RASKIN the exam (e.g. “I did not take the exam from the professor’s office.”), and 16 neutral statements (e.g. “I am seated before a computer, taking a test.”). 

In the standard format, we arrange the 48 statements so that no two statements of the same type appear in succession. The computer presents the set of 48 items 5 times with a brief break between repetitions during which the examinee answers simple arithmetic problems.

Excluding the arithmetic problems, the examinee answers a total of 80 T/F statements of each type. The correct (non-incriminating) answer to half of the items of each type is True and the correct answer to the other half is False.

Ocular-motor and Behavioral Measures. The eye trackers in our experiments measured horizontal gaze position, vertical gaze position, and one or two channels of pupil diameter depending on whether the tracker recorded data from the left and right eyes or from the right eye only. 

The trackers stored samples of each channel at either 30Hz or 60Hz. When the tracker measured left and right eyes, we computed the mean of measures from both eyes, because the two channels were highly correlated.

Mean change in pupil size for a mock crime experiment reported by Cook. (2012) is shown in Figure 1. It shows the evoked pupil response for four seconds following the onset of the statement for neutral statements, statements about the theft of the cash, and statements about the exam. 

As predicted, guilty participants reacted more strongly to statements about the cash than the exam, whereas innocent participants showed little difference between cash and exam statements. 

It is noteworthy that the mean change in pupil diameter (PD) was less than 0.1 mm.

Time in seconds:

0.0 1.0 2.0 3.0 4.0 5.0

PD Change in mm -0.12

-0.08

-0.04

0.00

0.04

0.08

0.12

Neutral

Cash

Exam

Figure 1. Mean change in pupil diameter (PD) from statement onset for guilty (left) and innocent participants (right)

To adjust for individual differences in reactivity and differences in ambient light

conditions, we transformed the pupil data for each of the five repetitions of test items to standard scores. 

Standardization also established a common metric across repetitions within individuals. From the standardized pupil response curves, we extracted.

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

LABORATORY AND FIELD RESEARCH ON THE OCULAR-MOTOR DECEPTION TEST 163

two features: the area under the response curve and the level of the response at the participant’s answer.

We derived reading measures from analyses of eye examinations on the text. 

A computer identified eye examination on the text where there was little variance in both horizontal and vertical gaze positions for a minimum duration of 100ms and a maximum duration of 1000ms (Cook et al., 2012). 

An area of interest was defined for each test item prior to the calculation of the reading measures. The area of interest began with the first character of the item and ended after the period at the end of the item. 

From the set of examinations for a test item, the computer measured three features:

• number of examinations was a simple count of the examinations in the area of interest
• first pass duration was the total duration of all examinations in the area of interest
• until an examination fell outside the area of interest, and
• re-read duration was the total duration of all examinations in the area of interest associated with leftward eye movements

We divided the three reading measures by the number of characters to adjust for differences in the length of test statements.

Behavioral measures included participants’ response times and error rates. We also measured blink rates per second for each item (item blink rate) and for the subsequent item (next item blink rate). 

Based on pioneering work by Stern, Walrath and Goldstein (1984), we expected a decrease in item blink rate when the participant was highly focused on reading an incriminating statement and an increase in blink rate when the participant encountered a less incriminating subsequent item (next item blink rate).

For a given type of measurement, such as pupil size at the time of the participant’s answer, we calculated two contrasts. 

We subtracted the pupil size for statements concerning the theft of the exam (R2) from the pupil size for statements concerning the theft of the $20 (R1); i.e. (R1 – R2). 

The other contrast was between the mean reaction to statements answered truthfully (neutral and R2 statements) and statements answered deceptively by guilty participants (R1); i.e. ( R1 – (neutral + R2) / 2). 

If the rationale underlying the ODT is correct, we should see relatively high or low scores on each of these contrasts for guilty participants and scores near zero for innocent participants.

We assessed the ability of ocular-motor measures to discriminate between guilty and innocent participants by correlating within-subject contrasts between statement types with guilt status, where guilt status was a dichotomous variable that was either 0 if the participant was innocent or 1 if the participant was guilty.

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

164 JOHN C. KIRCHER, DAVID C. RASKIN

In addition to measuring the diagnostic validity of various ocular-motor measures, we also use co-efficient alpha to assess the reliability of those measures (Chronbach, 1951). Reliability indicates the extent to which the measurements obtained from the five repetitions of the test items are consistent. 

For example, if the data from the first repetition indicated that the individual was deceptive, did the person also appear deceptive in the second, third, fourth, and fifth repetitions? Accuracy of the ODT in laboratory studies. 

Table 1 presents validity and internal consistency (reliability) co-efficient for the set of features in experiments conducted by two of our students (Patnaik, 2015; Webb, 2008). It also shows internal consistency reliability statistics for the computer-generated features.

Table 1. 

Validity and reliability coefficients for two mock crime studies of the ODT

Webb (2008) Patnaik (2015) Mean

Area under the curve R1 – R2 0.409 0.640 0.586 0.615 0.505 0.628

Area under the curve R1 – (NT+R2)/2 0.396 0.759 0.554 0.639 0.482 0.699

Level at answer R1 – R2 0.557 0.465 0.585 0.510 0.571 0.488

Level at answer R1 – (NT+R2)/2 0.548 0.527 0.634 0.575 0.593 0.551

Reading

Number of examinations R1 – R2 -0.509 0.572 -0.406 0.627 -0.460 0.600

Number of examinations R1 – (NT+R2)/2 -0.329 0.807 -0.293 0.720 -0.312 0.764

First pass duration R1 – R2 -0.549 0.582 -0.253 0.540 -0.427 0.561

First pass duration R1 – (NT+R2)/2 -0.293 0.622 -0.166 0.585 -0.238 0.604

Reread duration R1 – R2 -0.488 0.516 -0.342 0.397 -0.421 0.457

Reread duration R1 – (NT+R2)/2 -0.224 0.683 -0.115 0.407 -0.178 0.545

Behavioral

Response time R1 – R2 -0.529 0.434 -0.497 0.329 -0.513 0.382

Response time R1 – (NT+R2)/2 -0.312 0.788 -0.348 0.671 -0.330 0.730

Error rate R1 – R2 0.082 0.052 0.093 0.209 0.088 0.131

Error rate R1 – (NT+R2)/2 0.242 0.741 -0.002 0.690 0.171 0.716

Blink rate

Item blink rate R1 – R2 -0.014 0.247 -0.388 0.182 -0.275 0.215

Item blink rate R1 – (NT+R2)/2 -0.015 0.572 -0.191 0.101 -0.135 0.337

Next item blink rate R1 – R2 0.169 0.104 -0.088 0.351 0.135 0.228

Next item blink rate R1 – (NT+R2)/2 0.010 0.315 -0.105 0.381 0.075 0.348

Bolded validity co-efficient are statistically significant at p <0.05.

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

LABORATORY AND FIELD RESEARCH ON THE OCULAR-MOTOR DECEPTION TEST 165

The strong positive correlations for pupil features indicated that guilty participants showed greater increases in pupil size in response to R1 (cash) than to R2 (exam) statements. These findings are consistent with the data presented in Figure 1. 

The negative correlations for reading and response time measures indicated that guilty participants made fewer examinations and spent less time reading R1 than R2 statements. 

The effects on response time are substantial and consistent over multiple experiments and cultural groups. Based on the psychology of reading literature, we initially thought that guilty participants would experience more difficulty and spend more time on R1 than R2 statements. 

However, the data suggest that guilty participants invest more mental effort in answering those statements quickly and accurately to avoid detection (Cook et al., 2012). 

Guilty participants achieve their objective but reveal their deception. T is hypothesis explains the effects on response time and reading measures as well as the observed increases in pupil size associated with R1 statements.

Examination of mean validity co-efficient indicates that the pupil measures were more diagnostic than reading, behavioral, and blink rate measures. T e (R1 – R2) contrast for response time was almost as diagnostic as pupil size. Blink rate and error rate measures were the least predictive of guilt status.

Decision Model. To classify individuals as truthful or deceptive, we combine the scores on a subset of diagnostic measures in a mathematically optimal manner to compute the probability of deception. 

If the probability of deception exceeds 0.5, we classify the person as deceptive; if the probability is less than 0.5, we classify the person as truthful. 

Several statistical procedures have been developed to identify a subset of diagnostic measures that will represent most of the diagnostic variance in the full set of measures and work well when tested on an independent sample of cases. 

These procedures tend to select measures that are more highly correlated with guilt status and less highly correlated with each other. 

We then use logistic regression analysis to derive a unique weight for each ocular-motor measure that maximizes the separation between truthful and deceptive groups.

Much of our research has been designed to assess the effects of factors that could affect the accuracy of the ODT and to explore alternative methods for presenting test items. 

Regarding the latter objective, we have not improved on the presentation format and mock crime procedures evaluated in our first mock crime experiment (Osher, 2005), which we call the standard protocol. 

Because our attempts to improve on the standard protocol have yielded inferior results, Table 2 presents the results obtained with the standard protocol, and Table 3 presents results for non-standard protocols.

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

166 JOHN C. KIRCHER, DAVID C. RASKIN

Table 2. Percent of correct decisions under standard conditions in mock crime experiments.

Experiment Factors N nG nI Guilty Innocent Mean Validation G Validation I Mean

Osher (2005)a issues; serial format 40 20 20 85.0 85.0 85.0 85.0 70.0 77.5

Webb (2008)b sex; motivation;

difficulty 112 56 56 82.1 89.2 85.7 89.3 80.4 84.9

Patnaik (2013)a direct interrogation 48 24 24 83.3 95.8 89.6 83.3 83.3 83.3

Patnaik (2015)a distributed; pretest feedback; postresponse interval

80 40 40 82.5 90.0 86.3 80.0 90.0 85.0

Patnaik et al.

(2016)c language; culture 145 82 63 84.1 87.3 85.7 81.9 87.5 84.7

Middle East

(2016a) language; culture 112 51 61 80.4 88.5 84.5

Middle East

(2016b)d language; culture 101 52 49 75.0 85.7 80.4

Standard

Protocol 638 325 313 82.8 89.0 85.9 82.1 84.1 83.1

Validation results were obtained with the leave-one-out procedure

bT e decision model based on Patnaik et al. (2016) was used to classify participants in Webb (2008)

cT e decision model based on Webb (2008) was used to classify participants in Patnaik et al. (2016)

d T e decision model based on Middle East (2016a) was used to classify participants in Middle East

(2016b)

The results presented in Table 2 indicate that the standard protocol in mock crime

experiments yielded approximately 86% correct classifications in the original, standardization sample, and approximately 83% correct when tested on independent samples (cross-validation). On cross-validation, accuracy was slightly higher for innocent (84.1%) than guilty participants (82.1%). 

Table 3 summarizes results from non-standard conditions. The results from Osher (2005) suggest that we obtain more diagnostic information from serial presentations of individual test statements than with the simultaneous display of multiple test statements. 

Webb (2008) found that the person’s sex does not moderate the effects of deception on ocular-motor measures, whereas higher motivation to pass the test and semantic simplicity in the phrasing of test statements improves the diagnostic validity of some ocular-motor measures.

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

LABORATORY AND FIELD RESEARCH ON THE OCULAR-MOTOR DECEPTION TEST 167

Table 3. Percent of correct decisions under non-standard conditions in mock crime experiments.

Experiment Factors N nG nI Guilty Innocent Mean

Osher (2005) issues; parallel format 40 20 20 70.0 95.0 82.5

USTAR (2010) pretest questionnaire;

issues

71 47 27 59.6 77.8 68.7

NSA (2012) standardization 94 51 43 72.5 88.4 80.5

NSA (2013) validation 60 34 26 50.0 80.8 65.4

Patnaik (2013) indirect interrogation 48 24 24 58.3 79.2 68.8

Patnaik (2015) blocked 80 40 40 77.5 85.0 81.3

Non-standard protocols 393 216 180 65.3 84.5 74.9

Together, the USTAR (2010) and Patnaik (2013) studies indicated that test statements that refer directly to the matter at hand (“I did not take the $20.”) produce higher accuracies than statements that ask indirectly if the person falsified information to cover up their guilt (“I did not falsify my answers to questions about the theft of the $20.”). 

In the NSA studies, we recruited employees and tested them about minor security violations. 

The studies used a non-standard protocol because we were not permitted to provide incentives for government employees to pass the ODT, many participants were federal polygraph examiners who knew that there were essentially no consequences to failing the ODT, and we had to rely on self-report as a proxy for ground truth.

Patnaik (2015) found that the standard pseudo-random sequencing of NT, R1, and R2 statements improves the diagnostic validity of ocular-motor measures, whereas feedback about performance on a pre-ODT practice test and lengthening the interval between the answer and the presentation of the next item does not. 

Patnaik et al. (2016) found that the effects on ocular-motor measures were similar for tests administered in English or Spanish to native speakers enrolled as university students. The experiments in the Middle East required modification of the display software to present Arabic text from right to left. 

Accuracy rates on cross-validation in the Middle East were slightly lower than those obtained in the USA and Mexico, particularly for guilty participants. 

Although the differences in decision accuracy between Middle Eastern and Western participants were not statistically significant, we found it necessary to reduce the number of repetitions of test statements for measures of pupil response to achieve comparable levels of discrimination between truthful and deceptive Arabic-speaking participants as we had found for English- and Spanish-speaking participants. 

We are uncertain why it was necessary to make those changes.

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

168 JOHN C. KIRCHER, DAVID C. RASKIN

Field study of the ODT

We have recently completed a field validity study of the ODT that evaluated applicants for positions in the office of Mexico Attorney General’s, immigration, and federal police. We tested some applicants for recent use of illegal drugs (R1) compared to either corruption, arms trafficking, or affiliation with a religious terrorist organization (R2). 

Ground truth for deceptive cases were admissions by applicants during a subsequent polygraph test that they had used illegal drugs in the period covered by the statements on the ODT (n = 71). Lack of an admission is insufficient to establish conclusively that an individual was truthful on the ODT. 

To obtain data from truthful people, we created a new test for applicants for positions in immigration that asked if they had committed espionage (R1) or sabotage (R2). We assumed that all the tested individuals were truthful in their answers to both relevant issues because the base rate of deception on those issues in Mexico is very low, especially for people who had no prior government employment (n = 83).

To develop and validate a decision model with the field data, we extracted ocular motor measures from the eye tracker data. We computed validity co-efficient for the measures and used linear regression to select a subset of four measures to distinguish between the confirmed truthful and deceptive groups. 

We then used the selected variables in a 5-fold validation of a binary logistic regression model to classify cases as truthful or deceptive. 

To conduct the 5-fold validation, we divided the sample of 154 field cases into five random sub-samples such that each sub-sample consisted of approximately 20% of the deceptive cases (n=14 or 15) and 20% of the truthful cases (n=16 or 17). The first sub-sample of 14 truthful cases and 16 deceptive cases (N=30) was removed, and a decision model was created with the remaining four sub samples (N=124). 

We used that decision model to classify the holdout sample of 30 cases and recorded the percent correct for each group. The second sub-sample was then set aside (N=30), a new decision model was developed with the remaining 124 cases, and the accuracy of classifications was calculated for the second holdout sample. 

We repeated this process for the remaining three sub-samples. The results are brought together in Table 4.

Table 4. Percent of correct decisions in five decision models on independent subsamples

Fold 1 Fold 2 Fold 3 Fold 4 Fold 5 Mean Mean

N=30 N=30 N=31 N=31 N=32 N=154

truthful 75.0 87.5 88.2 88.2 100.0 87.8 86.1

deceptive 100.0 71.4 85.7 78.6 86.7 84.5

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

LABORATORY AND FIELD RESEARCH ON THE OCULAR-MOTOR DECEPTION TEST 169

Percent of correct decisions varied between the five sub-samples from 75% to 100% for truthful applicants and from 71.4% to 100% for deceptive applicants. 

Our best estimates of the performance of the model based on all 154 cases when tested on a new sample of field cases are the means for truthful (87.8%) and deceptive (84.5%)

applicants. 

Although the results obtained in the 5-fold validation using actual applicants for positions in the Mexican government suggest that the ODT may add value to a pre-employment screening program, to some degree our methods may have capitalized on chance and produced accuracy rates that are too optimistic. 

Specially, we used the entire sample to select the subset of ocular-motor measures for the decision model that was subsequently validated. This subset of measures worked well in the 5-fold validation but might be sub-optimal for a new set of confirmed field cases. 

For this reason, we recommend that the current decision model be re-evaluated with new cases from representative field settings.

Conclusions

The results of laboratory and field research indicate that the ODT yields accuracy greater than 80% on both truthful and deceptive examinees, although the accuracy rates tend to be 3% to 6% higher for truthful than for deceptive individuals. 

The accuracy rates generalize across English and Spanish languages, US and Mexican cultures, and to a lesser extent, Arabic in the Middle East. We believe the ODT to be a promising new technology that is best suited for screening applications. 

We also believe it is better suited to screening applications than specific-incident, criminal investigation, because it would be difficult to construct an RCT that contains two non-overlapping relevant issues with face validity. 

For the same reason, polygraph examiners rarely if ever use the Guilt Complex question for specific-incident polygraph examinations (Office of Technology Assessment, 1983).

Unpublished pilot research with poor readers indicated, as expected, that the ODT was ineffective. It appears that the cognitive difficulties experienced by examinees who struggle to comprehend test statements overshadow effects of deception on ocular-motor measures. 

We are exploring audio-visual and audio-only alternatives to the current text-based ODT. However, yet we have not developed a functional audio version of the test that might be effective for a broader population of individuals, including those with poor reading skills.

With specialized training and practice, polygraph tests can be defeated (Honts, 2012). To date, there have been no attempts to investigate effects of countermeasures on the ODT. 

However, because examinees are under pressure to respond quickly and accurately, attempts to implement countermeasures may be ineffective or easily. 

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

170 JOHN C. KIRCHER, DAVID C. RASKIN detected with behavioral measures, such as response time and error rates. Additional research is needed to explore these possibilities. 

Finally, research on the ODT has primarily been conducted in a single laboratory by one team of investigators. As high quality eye tracking equipment and data analysis software become available at lower costs, we are hopeful that other investigators will contribute new knowledge to this area of applied science.

References

Bradley, M.M., Micolli, L., Escrig, M.A., Lang, P.J. (2008). T e pupil as a measure

of emotional arousal and autonomic activation. Psychophysiology, 45, pp. 602–607.

Bradley, M. T., & Janisse, M. P. (1981). Accuracy demonstrations, threat, and the

detection of deception: Cardiovascular, electrodermal, and pupillary measures. 

Psychophysiology,

18, pp. 307–315.

Cook, A.E., Hacker, D.J., Webb, AK., Osher, D., Kristjansson, S., Woltz, D.J., &

Kircher, J.C. (2012). Lyin’ Eyes: Ocular-motor Measures of Reading Reveal Deception.

Journal of Experimental Psychology: Applied, 18(3), pp. 301–313.

Cronbach, L. J. (1951). Coef cient alpha and the internal structure of tests. Psychometrika,

16, pp. 297–334.

Dionisio, D. P., Granholm, E., Hillix, W. A., & Perrine, W. F. (2001). Dif erentiation

of deception using pupillary responses as an index of cognitive processing. Psychophysiology,

38, pp. 205–211.

Hacker, D.J., Kuhlman, B., & Kircher, J.C., Cook, A.E., & Woltz, D.J. (2014). Detecting

deception using ocular metrics during reading. In D.C. Raskin, C.R. Honts,

& J.C. Kircher (Eds.), Credibility assessment: Scientific research and applications. Elsevier,

pp 159–216.

Hess, E.H., Polt, J.M. (1960). Pupil size as related to interest value of visual stimuli.

Science, 132, pp. 349–350.

Hess, E.H., Polt, J.M. (1964). Pupil size in relation to mental activity during simple problem solving. Science, 143, –1190–1192.

Honts, C.R. (2012). Countermeasures and credibility assessment. In DC Raskin,

CR Honts, & JC Kircher (Eds.), Credibility assessment: Scientific research and applications.

Elsevier, pp. 131–156.

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

LABORATORY AND FIELD RESEARCH ON THE OCULAR-MOTOR DECEPTION TEST 171

Johnson, R., Jr., Barnhardt, J., & Zhu, J. (2005). Differential effects of practice on the executive processes used for truthful and deceptive responses: An event-related brain potential study. Cognitive Brain Research, 24, pp. 386–404. Kahneman, D. (1973). Attention and effort. 

Prentice-Hall: New Jersey. Kircher, J. C. (1981). Psychophysiological processes in the detection of deception. Department of Psychology, University of Utah, Salt Lake City, UT. Unpublished manuscript. Kircher, J.C., Horowitz, S.W. & Raskin, D.C. (1988). Meta-analysis of mock crime studies of the control question polygraph technique. 

Law and Human Behavior, 12, pp. 79–90. Kircher, J.C., Raskin, D.C., Honts, C.R., & Horowitz, S.W. (1994). Generalizability of statistical classifiers for the detection of deception. Psychophysiology, 31, S73. (Abstract) Loewenfeld, I. E. (1999). T e pupil: Anatomy, physiology, and clinical applications (Vol. 1). Boston: Butterworth-Heinemann.

Middle East (2016a). Ocular-motor detection of deception with Middle Eastern college students, Phase 1. Unpublished research. University of Utah, Department of Educational Psychology. Middle East (2016b). Ocular-motor detection of deception with Middle Eastern college students, Phase 2. Unpublished research. University of Utah, Department of Educational Psychology.

NSA (2012). Ocular-motor detection of deception with employees at the National Security Agency Phase 1. Unpublished research. University of Utah, Department of

Educational Psychology.

NSA (2013). Ocular-motor detection of deception with employees at the National Security Agency Phase 2. Unpublished research. University of Utah, Department of

Educational Psychology. 

Office of Technology Assessment (1983). Scientific validity of polygraph testing: A research review and evaluation. OTA-TM-H-15. Washington, DC: US Government Printing Office. Osher, D. (2005). Multimethod assessment of deception: Oculomotor movement, pupil size, and response time measures. Unpublished dissertation, University of Utah, Department of Educational Psychology.

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM

172 JOHN C. KIRCHER, DAVID C. RASKIN

Patnaik, P. (2013). Ocular-motor methods for detecting deception: Direct versus indirect

interrogation. Unpublished master’s thesis, University of Utah, Department of Educational

Psychology.

Patnaik, P. (2015). Oculomotor methods for detecting deception: Effects of practice feedback and blocking. Unpublished dissertation, University of Utah, Department of

Educational Psychology. Patnaik, P., Woltz, D.J., Hacker, D.J., Cook, A.E., Ramm, M.L., Webb, A.K., & Kircher, J.C. (2016). 

Generalizability of an ocular-motor test for deception to a Mexican population. International Journal of Applied Psychology, 6(1), pp. 1–9. Raskin, D.C. & Kircher, J.C. (2014). Validity of polygraph techniques and decision methods. 

In D.C. Raskin, C.R. Honts, & J.C. Kircher (Eds.), Credibility assessment:

Scientific research and applications. Elsevier. pp. 63–129.

Rayner, K. (1998). Eye movements in reading and information processing: 20 years of research. Psychological Bulletin, 124, pp. 372–422.

Rayner, K., Chace, K. H., Slattery, T. J., & Ashby, J. (2006). Eye movements as reflections of comprehension processes in reading. Scientific Studies of Reading, 10, pp. 241–255.

Steinhauer S. R., Boller F., Zubin J., Pearlman S. (1983). Pupillary dilation to emotional

visual stimuli revisited. Psychophysiology, 20, p. 472.

Steller, M. (1987). Psychophysiologische Aussagebeurteilung [Psychological assessment].

Hogrefe: Gottingen. Stern J.A., Walrath L.C., Goldstein, R. (1984). The endogenous eyeblink. Psychophysiology 21, pp. 22–33. USTAR (2010). Oculomotor deception detection. Unpublished research funded by the Utah Science Technology and Research Initiative.

Vrij, A., Fisher, R., Mann, S., & Leal, S. (2006). Detecting deception by manipulating cognitive load. Trends in Cognitive Sciences, 10, pp. 141–142. Webb, A.K. (2008). Effects of Motivation, and Item Difficulty on Oculomotor and Behavioral Measures of Deception. Unpublished dissertation, University of Utah, Department of Educational Psychology.

Webb, A. K, Honts, C. R., Kircher, J. C., Bernhardt, P.C., & Cook, A. E. (2009). Effectiveness of pupil diameter in a probable-lie comparison question test for deception. Legal and Criminal Psychology, 14(2), pp. 279–292.

Authenticated | davidc.raskin@me.com author’s copy

Download Date | 2/26/17 8:29 PM