project final report cover page

advertisement
BE 309
Lab Report cover
Download this form, fill in blanks and paste in summary.
Fall 2000
PROJECT FINAL REPORT COVER PAGE
GROUP NUMBER
PROJEC TITLE
M2
Design and Testing of a Biopac-Based Polygraph
DATE SUBMITTED:
12/17/00
ROLE ASSIGNMENTS
ROLE
GROUP MEMBER
FACILITATOR………………………..
Michael Kohanski
TIME & TASK KEEPER………………
Ashwin Desikan
SCRIBE………………………………..
Jenea McLaughlin
PRESENTER………………………….
Julie Bookbinder
SUMMARY OF PROJECT CONCLUSIONS
The purpose of this experiment was to monitor physiological changes in subjects to
determine lying while using the Guilty Knowledge and the Control Question Tests. It was
found that Galvanic Skin Response increased consistently by nearly 50% when subjects lied
during the Card Game and Guilty Knowledge Test. Horizontal eye movement was less reliable,
but an average change of 30% among the tests when the subjects lied. Vertical eye movement
proved to be an inconclusive parameter. Additionally, heart rate and breathing rate could not be
used to determine lying. Errors in applying Biopac filters to pulse rate data eliminated its use
from the experiment. Comparison of the Guilty Knowledge Test and the Control Question Test
showed that the Control Question Test is much less accurate in determining guilt. It was also
determined that it was not possible to design Yes/No Biopac filters that can be used uniformly
for all subjects due to large variations in subject response levels.
Objective
The Primary objective of this experiment is to create a standardized lie detector using a
series of BioPac sensors to monitor the physiological changes in GSR, breathing rate, heart rate,
pulse pressure and eye movement in different subjects. This experiment will verify if the
standard polygraph parameters (GSR, breathing rate and pulse pressure) are good physiological
indicators of lying, along with changes in heart rate and eye movement. It has been shown that
when an individual lies, his or her heart and respiratory rates, as well as pulse pressure increase
significantly due to stress brought on by feelings of guilt, fear, anger, etc. These feelings can
also cause a subject to experience anxiety, which causes the skin's conductance to increase due to
enhanced activity in the sweat glands. It has also been shown that movement in the eye
muscles, specifically the Orbicularis Oculi muscles which close eyelids, and aid passage and
drainage of tears, have been shown to move significantly under periods of duress and anxietyi.
By carefully monitoring the changes individuals experience for these parameters, an accurate
determination of guilt can be established for the individual being tested.
Specific Aims
 Determine if monitoring physiological changes in an individual is an accurate indication
of lying.
 Verify if GSR, breathing rate and pulse pressure are physiological indicators of lying
 Determine if changes in heart rate and eye movement are physiological indicators of
lying
 Determine whether Biopac filters can be set up for the designated physiological
parameters such that simple yes/no determinations can be made regarding the guilt of a
subject connected to the Biopac Polygraph.
 Determine if these filters vary person-to-person, and if so, on what parameters (i.e.
gender) these variations are based.
 To use the Biopac polygraph in conjunction with the Guilty Knowledge Test and the
Control Question Test to determine the validity of this test type in determining guilt.
Hypotheses
 Monitoring physiological changes in an individual will prove to be an accurate indication
of lying.
 GSR, breathing rate and pulse pressure (the established physiological indicatorsii) are
physiological indicators of lying.
 Changes in heart rate and eye movement are also physiological indicators of lying.
 It is not possible to design Yes/No Biopac filters that can be used uniformly for all
subjects.
 The Guilty Knowledge Test will be a better method for proving guilt than the Control
Question Test.
o More errors will be associated with the Control Question Test
Background
As stated in Detecting Lies and Deceitiii, the most common physiological measures used
to determine deceit are changes in skin conductance (Galvanic Skin Response), pulse pressure,
and breathing rate. When a subject experiences anxiety, there is a fast increase in the skin's
conductance due to increased activity in the sweat glands. In contrast, the pulse pressure has
been seen to go down as the skin conductance goes up, due to the fact that both are sympathetic
nervous system responses to emotional changes experienced by the individual. Although not a
response to the sympathetic nervous system, breathing rate has been shown to decrease when a
subject experiences a period of anxiety. As stated in Medical Physiology.iv, other physiological
responses can also be used as an indication of lying when designing a lie detector. Vertical and
horizontal movement of the Orbicularis Oculi muscles surrounding the eye has been shown to be
effective in determining guilt of an individual, as this muscle is inclined to move significantly
under periods of duress and anxiety.
To accurately determine whether an individual is telling the truth or lying, an established
method of questioning must be employed. The two most common methods of polygraph
questioning are the Control Question and the Guilty Knowledge Tests. As stated in the paper by
Honts, Kircher, & Raskin v the Control Question Test assesses credibility of an individual by
comparing the subject's physiological responses to two types of questions: relevant questions and
control questions. It is typically comprised of three relevant questions, three control questions,
and several unevaluated neutral questions. The basis for this test is guilty subjects will show the
strongest physiological responses to the relevant questions, since the control questions pose a
less immediate threat.
The Guilty Knowledge Test compares physiological responses to a series of related
questions about a specific topic with one of the questions (the “critical question”) containing
information only the investigators and a guilty individual would knowvi. If the subject
“possesses guilty knowledge,” he or she is expected to respond differently to the critical
questions, while an innocent individual will respond no differently to the critical questions than
to the non-critical questions.
Apparatus/Methods/Materials
Three tests were performed to determine the physiologic changes associated with lying.
These tests included the Card Test, the Guilty Knowledge Test (GKT), and the Control Question
Test (CQT). Each of these tests will be described below. Nine students from BE-309 lab sections
were used as test subjects, and had sensors applied to them to detect their physiologic responses
through Biopac.
Originally there were five sensors used to measure physiological changes in the test
subjects. They were GSR (Galvanic Skin Response), EOG (Electrooculogram), a respiration
transducer, ECG (Electrocardiogram), and PPG (Pulse Plethysmograph). The GSR was attached
to the non-dominant hand on the index and middle fingers to record the skin resistance across 2
fingers of the subject using 2 Ag/AgCl electrodes. The EOG measured voltage changes caused
by eye movement. There were 6 electrodes placed on the face. Two electrodes were placed on
the right and center forehand, and another electrode was placed on the left cheek. These 3
electrodes measured voltage changes due to vertical eye movement. Two more electrodes were
placed on the left and right temples, and the last electrode was placed on the left forehead. These
last 3 electrodes measured voltage changes due to horizontal eye movement. The PPG was
attached to the dominant hand on the index finger. This device is a transducer that records pulse
pressure. It consists of an infrared emitter and photo diodes, which transmit changes in blood
density caused by varying blood pressure. The respiration transducer measured chest expansion
and contraction during breathing, and the ECG measured the heart rate of the subject. These two
tests were later discontinued because they produced no useful data.
One MP30 Data Acquisition Unit was used, and the GSR, PPG, and, EOG were
connected to this unit. Analog channels 1-4 were used to collect raw data. Seven calculation
channels were set up and computed internally by Biopac. Calculation channels 1 and 2 were used
for the integrated vertical and horizontal EOG. Channels 3 and 4 were used to display the rate of
the vertical and horizontal EOG on the screen. Channel 5 and 6 were used to take the first and
second derivates respectively of the filtered PPG data. Channel 6 was displayed as the pulse
acceleration. Channel 7 was used for the smoothed raw GSR data.
A hypothetical crime was devised for lie detector test. The subject was instructed to steal
a piece of fake gold from a specified location in the BE laboratory. The subject was also told that
the thief left behind an object. The subject was tested by using questions related to this
hypothetical crime. An interview was also conducted to determine sensitive topics that could be
used for the CQT. After the subject committed the crime, he or she was connected to Biopac to
record the previously discussed physiological responses. A 5-minute baseline was recorded with
the subject at rest, followed by the Card Test, the Guilty Knowledge Test, and finally the Control
Question Test. It was initially proposed that the subjects would be visually isolated during
questioning. However, isolation of the subject proved to be too cumbersome for the purposes of
this experiment. Therefore, isolation of the subject was abandoned after the preliminary tests.
The card game was used as the initial test for the lie detector. The subject was instructed
to choose a card, and remember what it was. The examiner questioned the subject about which
card they had chosen. The subject must respond no to every card. The subject’s card was
determined based on the physiologic responses seen in Biopac. This test was mainly used to
show the subject that the lie detector works.
The GKT protocol consisted of the Examiner asking the Subject a series of yes/no
questions concerning the hypothetical crime that was the framework for this experiment. There
were four topical questions, each having five sub-questions giving alternate answer choices to
the topical question. An example of a topical question is “what was stolen?” A few examples of
sub-questions include “a knife? … a bracelet?… a gold nugget?” The subject must again answer
no to every sub-question, and was informed that answering yes would be an automatic indication
of guilt.
The Control Question Test (CQT)vii was the last test conducted during questioning.
Only yes/no responses were allowed. The subject was instructed to answer yes to all neutral
questions, and no to all control and relevant questions. “Do you live in the US?” is an example of
a neutral (N) question. The control (C) questions were designed to make the subject more
uncomfortable. For example, “have you ever stolen anything?” A relevant question was related
to the hypothetical crime such as “did you steal the gold?” The examiner reviewed all of the
questions with the Subject to ensure that the Subject understood all of the questions, and which
ones to answer yes or no too. The Examiner asked the Subject sets of questions in the order
NCR.
Results
Low Pass filters were applied as data were collected to minimize noise in order to get a cleaner
signal. These filters were based on the 4 subjects, 2 male ad 2 female, from the first week of
testing. Fast Fourier Transforms (FFT’s) were performed on the raw data of all five of the
physiological parameters, and the frequency of the data was approximated from them. The Q
value, or skew of filter was set to the Biopac recommended value of for optimal dampeningviii (Q
= 0.707). Table 1 below contains the frequencies for these ideal filters. As these filters were to
be applied directly to Biopac’s MP30 during data collection, the minimum frequency for these
filters had to be 30Hz (Biopac specification). Therefore, filters were applied by trial and error to
get the data to look as close as possible to that generated via the ideal filters. Table 1 below
contains the frequencies and Q values for these real filters (**NOTE: the words ‘real FFT’s’ used later
refers to FFT’s generated with the real filters applied**) . However, the EOG filter was not used as it was
determined that it removed too much of the signal.
Table 1: Ideal and Real Filter Values
Heart Rate Respiration Rate Pulse Rate GSR EOG
Frequency(Hz)
30
30
35
30
30
real
Q
0.00071
0.0002
0.00071 0.0009 0.13
Frequency(Hz)
0.05
0.05
0.05
3.5
3.5
ideal
Q
0.707
0.707
0.707 0.707 0.707
Figure 1 below is an example of one of the raw FFT’s generated. This is one of the EOG FFT’s.
As seen in the figure, the data is a small peak in the 1-2Hz range and the noise is seen in the
valleys between 40-80Hz. This noise is generated by the oscillation of the normal 60Hz
background noise.
Figure 1: Raw EOG FFT
Data
-46.96
noise
noise
dbV
Phase
-93.93
-140.89
-187.86
Hz
Figure 2 below is an example of one of the ideal FFT’s generated. This is the ideal FFT of the
raw EOG above. As seen in the figure below, the data is now a larger peak due to the fact that
much of the noise has been removed. The dip at around 80Hz has been completely removed and
the dip at around 40Hz has been significantly reduced. The noise could not be completely
removed as further noise reduction also resulted in significant data loss.
Figure 2: Ideal Filtered EOG FFT
Data
-60.80
noise
dbV
Phase
-121.60
-182.41
-243.21
Hz
Figure 3 below is an example of one of the real FFT’s generated. This is the real FFT of the raw
EOG above. As seen in the figure below, the data is now a larger peak due to the fact that much
of the noise has been removed. The dips at around 40Hz and 80Hz have been significantly
reduced. The noise could not be completely removed as further noise reduction also resulted in
significant data loss. This filter is not as good a filter as the ideal one because the Q value is
non-optimal for the real FFT. This allows some of the noise to remain instead of being filtered
out.
Figure 3: Filtered EOG FFT
-56.27
dbV
Phase
-112.54
-168.81
-225.08
Hz
Figure 4 below depicts the pulse acceleration and the pulse velocity for one of the test subjects.
The sharp dips in pulse velocity, which correspond to the sharp changes in acceleration, are just
1-2 seconds after the subject lied. These strong changes are not seen in the middle question,
where the subject told the truth.
Figure 4: Filtered PPG Data
st raeh fo 3
sedap s fo gnik
st raeh fo 3
45.51
Pulse Acceleration
77.7
etaR e sluP
MPB
00.0
77.7 -
55.001
Pulse Velocity
44.69
etaR e sluP
MPB
33.29
22.88
41.499
72.769
04.049
sdno ce s
35.319
Figure 5 below is the pulse acceleration data collected over the entire trial period for one of the
female subjects with the filter listed in Table 1 above. Instead of the sharp peaks expected
during lying, a sharp increase in the rate of acceleration was observed when questioning began.
Figure 5: Pulse Acceleration-Female Subject
4 diamond
king
jack
3spades
2
heart
di
queen
clubs
sa6king
mo
heart
s nds
clubs
disamo
king
ndsspades
2 clubs
3 heart3s2heart
clubs
s
Questioning
begins
knife
bracelet
gold
CD
pnugg
en
paper
hat
(before)
etsho e watch
sca
ground
rfdrawer
table
cabinet
computer
seattl
expos
e
yankee
black
orange
cham
knight
ch
ps
born
sam
borrowed
ps
insteal
US
born
st
gol
ow/out
tiad
n
len
ke
?1980?
from
gold
asking
born
fadi
frllen
knowledge
nom
ing?
in?off
Mass.
basket?
bik e?
? ofgold t
Spike due to hand
movement
minutes
Figure 6 below is the pulse acceleration data collected over the entire trial period for one of the
male subjects with the filter listed in Table 1 above. Instead of the sharp peaks expected during
lying, the subject’s pulse acceleration appeared to disappear during questioning and reappear
when he was told to relax.
Figure 6: Pulse Acceleration-Male Subject
end baseline4 diamo
57
hearts
hearts
9 nds
hearts
QJhearts
A
hearts
sp
2 diamo
Aadiamo
des
54diamo
9clubs
nds
clubs
Jnds
Kdiamo
nds
clubs
JAhearts
A
sp
nds
diamo
2adiamo
des
83Diamo
nds
8hearts
sp
nds
7 hearts
8
andes
hearts
ds
GKTknife?
questions
bracelet?
gold
CD?
pen?
nugg
paper?
hat?
sh
ewatch?
t?
osca
e?ground?
rf?
drawer?
table?
cabine
compu
seattl
t?
mets?
ter?
yanke
bl
esan
a?ck
relax
es?
diego
w/ red?
born
oran
steal?
stea
born
ingembarrased
te?
rlothtay?
ein
ke
born
gol
michig
passed
gold
dkn
in
?o
1980?
from
an?
wlege
atby
abasket?
track
frien
of the
dmae
226 .59
139 .05
51.51
mmHg/sec
Relaxed
Questioning
begins
pulse accelerat ion
Relaxed
Questioning
begins
-36 .04
-123.58
minutes
Figure 7 below is the FFT of the raw pulse velocity data from one of the subjects pre-filter.
Figure 8 below is the FFT for the raw PPG data. It is clear from the graphs that the data for the
PPG and pulse velocity lie over different ranges with only a partial overlap. The filter used for
the PPG was determined from the pulse velocity and applied to the PPG of later subjects. This
could account for the lack of strong changes after lying in the pulse acceleration for those
subjects with the filter applied directly to the PPG.
Figure 7: Raw Pulse Velocity FFT
0.0
dbV
Phase
-41.83
-83.66
-125.49
Hz
Figure 8: Raw PPG FFT
-56.93
dbV
Phase
-113.86
-170.79
-227.71
Hz
Table 2 below contains example heart and respiration data from one of the male subjects. This
data shows that there are negligible changes in both rates relative to baseline for lying and
truthful answers. There are also negligible changes in rates between lying and truthful answers.
Table 2: Example Data from a Male Subject of Heart & Respiratory Rates
Heart Rate(BPM)
Respiration Rate(BPM)
baseline lying
not lying
74.39
72.59
72.29
18.29
17.33
14.77
Figure 9 below is an example from a male subject of the heart and respiratory rates for the entire
trial period. The dashed lines correspond to questions asked. Changes in these rates do not
appear to correlate with a truthful or lying answer to a question.
Figure 9: Heart & Respiratory Rates
7 of
king
diamonds
7 of
of clubs
spadesjackqueen
of hearts
two
3 of
king
hearts
diamonds
of spades
3 of hearts
born
born
in in
1980?
6hong
weigh
feet
right
BE
kong?
tall?
vote
180
handed?
major?
see
in
lbs?
born
unbreakable?
election?
born
6infeet
weigh
in
1980?
right
hong
tall?
180?
handed?
BE
kong?
major?
votesee
in last
unbreakable
electio
129.80
55.06
BPM
Heart Rate
92.43
17.69
-19.69
23.50
9.50
BPM
Respiration Rate
16.50
2.50
-4.50
0.00
6.31
12.61
minutes
18.92
Figure 10 below is an example of the GSR data for one of the male subjects. The first dashed
line corresponds to the asking a question where a lie is the known response. The second dashed
line corresponds to a question with a known truthful response. It is clear from the figure that the
subject’s GSR increases more post-lying than post-truth.
Figure 10: GSR-Male Subject
s eda p s 8
s bul c 7
5 6. 61
8 9. 01
RSG
stl o V
2 3. 5
5 3. 0 -
2 0. 6 s dno ce s
Figure 11 below shows an example of a subjects EOG data. The top graph is the vertical EOG
and the bottom graph is the horizontal EOG. The light blue lines show a clear data area for the
horizontal EOG, where the peak to peak is different from the random movement observed
elsewhere along the data.
Figure 11: EOG data
hat
0.00
C4 - Difference (vertical)
blink
0.00
Volts
0.00
-0.00
0.00
blink
0.00
Volts
C5 - Difference (horizontal)
8.6E-004
-8.6E-004
764.00
766.00
768.00
770.00
seconds
Baseline data was collected from the 9 subjects, 5 female and 4 male. This data was collected
from the subjects while they were sitting still for 3-5 minutes. As shown in Figure 12, the female
subjects exhibited higher mean values than the male subjects consistently through the EOG
vertical, EOG horizontal, and GSR.
Figure 12: Mean Baseline Data
Baseline
Overall
Female
Male
EOG vertical
delta T (sec)
p-p (volts)
0.54 ± 0.25
0.05170 ± 0.07430
0.62 ± 0.49
0.05823 ± 0.16386
0.43 ± 0.52
0.04353 ± 0.14447
EOG horizontal
GSR
delta T (sec)
p-p (volts)
max (micro-ohms)
0.66 ± 0.44
0.01623 ± 0.02438
0.13 ± 0.34
0.68 ± 0.82
0.02207 ± 0.058713
0.17 ± 0.53
0.63 ± 1.16
0.00892 ± 0.02069
0.09 ± 0.95
Figure 13 shows the mean EOG and GSR data collected from the 9 subjects during the three
different testing methods. The card game data represents all subjects, while the GKT and CQT
data represents 6. The large confidence intervals indicate the extremely wide range of data over
the subjects. This range was caused by the individuality of each person as a unique system that
displays different physiological responses. (Graphical representation of this data is presented in
the Appendix)
Figure 13: EOG & GSR Mean Data by Test Method and Gender
Not Lying
Cards
EOG vertical
delta T (sec)
p-p (volts)
0.40 ± 0.30
0.00533 ± 0.00518
0.48 ± 0.64
0.00315 ± 0.00484
0.31 ± 0.43
0.00806 ± 0.01650
EOG vertical
0.36 ± 0.24
0.00504 ± 0.00537
0.43 ± 0.51
0.00368 ± 0.00299
0.28 ± 0.37
0.00760 ± 0.01788
EOG horizontal
GSR
delta T (sec)
p-p (volts)
max (micro-ohms)
0.34 ± 0.15 0.00249 ± 0.00254
4.80 ± 5.68
0.28 ± 0.20
0.00169 ± 0.00541
0.88 ± 0.97
0.42 ± 0.42
0.00349 ± 0.00637
9.71 ± 17.83
EOG horizontal
GSR
0.29 ± 0.14 0.00210 ± 0.00264
8.88 ± 7.00
0.25 ± 0.18
0.00200 ± 0.00691
4.49 ± 4.14
0.33 ± 0.42
0.00222 ± 0.00556
14.37 ± 26.40
EOG vertical
delta T (sec)
p-p (volts)
0.69 ± 1.40
0.00152 ± 0.00357
0.58 ± 0.93 0.00169 ± 0.27800
0.90 ± 7.29
0.00118 ± 0.00500
EOG vertical
0.83 ± 1.61
0.00164 ± 0.00461
0.74 ± 1.20
0.00185 ± 0.00356
1.00 ± 5.21
0.00122 ± 0.00943
EOG horizontal
GSR
delta T (sec)
p-p (volts)
max (micro-ohms)
0.67 ± 1.35
0.00068 ± 0.00185
1.32 ± 3.83
0.58 ± 0.93 0.00072 ± 0.00146
0.92 ± 1.94
0.85 ± 6.40 0.00059 ± 0.00205
2.12 27.71
EOG horizontal
GSR
0.81 ± 1.58 0.00139 ± 0.00343
2.64 ± 6.71
0.75 ± 1.20 0.00134 ± 0.00249
2.12 ± 4.83
0.95 ± 4.36 0.00150 ± 0.01540
3.68 ± 22.69
EOG vertical
delta T (sec)
p-p (volts)
Overall
0.44 ± 0.38
0.00112 ± 0.00338
Female
0.39 ± 0.26 0.00143 ± 0.00248
Male
0.53 ± 0.24
0.00052 ± 0.00578
Relevant
EOG vertical
Overall
0.42 ± 0.53
0.00120 ± 0.00405
Female
0.32 ± 0.26
0.00160 ± 0.00296
Male
0.62 ± 1.21
0.00041 ± 0.00461
EOG horizontal
GSR
delta T (sec)
p-p (volts)
max (micro-ohms)
0.44 ± 0.38
0.00034 ± 0.00042
4.51 ± 12.86
0.39 ± 0.26
0.00031 ± 0.00020
5.20 ± 9.63
0.53 ± 0.24
0.00039 ± 0.00357
3.14 ± 39.36
EOG horizontal
GSR
0.42 ± 0.53
0.00061 ± 0.00211
4.10 ± 17.30
0.32 ± 0.26
0.00072 ± 0.00165
5.56 ± 13.01
0.62 ± 1.21
0.00040 ± 0.00165
1.17 ± 5.05
Overall
Female
Male
Lying
Overall
Female
Male
Not Lying
GKT
Overall
Female
Male
Lying
Overall
Female
Male
Control
CQT
Equations 1 & 2 below are the equations used to quantify lying. (Data in the excel appendix):
Score = 1*(Average GSR) + 0.1*(Average vertical EOG) + 0.2*(Average horizontal EOG)
Eqn(1)
Score = 1*(Average GSR) + 0.01*(Average vertical EOG) + 0.02*(Average horizontal EOG)
Eqn(2)
Discussion
The filters designated in Table 1 above were designed to attain greater clarity of signal.
These filters are not the Yes/No Biopac filters to determine if a person is lying. All of the filters
are low pass filters as, in general, the data was found to be below about 3-4 Hz, and the majority
of the noise came from the 60Hz frequencies generated by the surrounding lab equipment and
overhead lights. The EOG filters determined from week one’s subjects functioned correctly on
week two’s subjects. However, the GSR and PPG filters did not function as expected.
Therefore, the GSR filter was eliminated, as the signal was clear enough without it. However,
the PPG filter was used since it was not evident until after data was collected that the week two
pulse accelerations were not similar to the pulse accelerations of week one’s subjects. It was
then determined that the deviation in the pulse acceleration for week two’s subjects was due to
applying an incorrect filter to the PPG data. As seen from figures 7 & 8 above, different low
pass filter cutoffs are obtained from the raw PPG data versus the raw pulse velocity data. Since
the velocity cutoff is lower than the PPG cutoff, some PPG data was lost during collection. This
accounts for the aberrant pulse accelerations observed during the second week of trials (Figures
5&6).
Heart rate and respiratory rate were found to show no significant changes (Table 2) from
baseline when lying or when not lying. Therefore, both of these measurements were not taken
during the second week of trials to save time and to lessen subject discomfort (subjects
complained about the number of sensors during week one). From this data, it appears that heart
rate, which is not an established indicator of lying, does not exhibit fast autonomic responses to
lying. This makes sense, as heart rate is a well-controlled physiological phenomenon where
sympathetic and parasympathetic systems are constantly sending signals to the heart to maintain
a tight control over heart rate. Respiratory rate is supposed to be one of the established
physiological measures of lying. The fact that it did not appear to correlate with lying here could
mean a number of things. First, respiratory rate may not actually be a good indicator of lying.
Second, the respiratory transducer used by Biopac might not be sensitive enough to pick up the
changes normally exhibited during lying. A third possibility is that the sensor was not directly
against the skin. The movement of the clothing against the sensor may have dampened the
respiratory rate changes exhibited during lying.
For the reasons discussed above, PPG, heart rate and respiratory rate are not included in
the baseline or other data. Remaining measurements are the GSR and EOG data. The wide
confidence intervals in the baseline data for these parameters support the hypothesis that it would
not be possible to design yes/no Biopac filters to apply universally. This experiment showed that
females have higher resting values than males for GSR and horizontal and vertical EOG. Males,
however, had higher lying and not lying values for GSR in both the Card Game and the GKT.
This difference between the genders illustrates one impediment in designing uniform filters.
It was also hypothesized that GSR would be a physiological indicator of lying, and it did
increase significantly during lying in the Card Game (by 45.9%, combined data) and the GKT
(by 50.0%, combined data). The CQT, however, showed different GSR trends for males and
females. It was expected that guilty individuals, who were all of the subjects in this experiment,
would have higher responses to the relevant questions. Instead, the GSR was 62.9% higher in
males for control questions compared to relevant questions. It was 6.6% greater in females for
relevant compared to control questions. The conclusions that can be drawn from these data are
as follows. The data might have been tainted because the CQT was the last test administered to
the subjects. At this point, boredom had ensued for many subjects, which skews the data
because a polygraph is only accurate if the subjects are concentrating on the questions and
nothing else. The different trends in the CQT also support the hypothesis that the GKT would be
more accurate in determining guilt than the CQT. Additionally, the strict method for giving the
CQT is to rearrange the 3-question sets so that each set is asked three times. This methodology
is meant to eliminate false positive responses from the data. Due to time constraints it was
prohibitive to perform the test this way, so instead each 3-question set was asked only once. The
mean data therefore might have been skewed by false positive responses.
The card game results showed a decrease in vertical (5.6%) and horizontal (15.7%) eye
movement when the subject was known to be lying. This is contrasted, however, by the fact that
vertical and horizontal eye movements increased in both the GKT (7.4%-vert, 51.2%-hori) and
the CQT (6.6%-vert, 44.7%-hori) when the subject was known to be lying. This is quite
surprising, as the GKT and the card game are based on similar principles, so one would expect
similar results.
The results do show that the horizontal eye movements change by a larger degree than
the vertical eye movements for all three tests and for both male and female subject types. This
change can be explained by the fact that horizontal eye movement during lying is not a random
eye movement. Rather, it is indicative of accessing the part of the brain used during lying. This
concerted eye movement is seen more in the horizontal than vertical direction.
The data obtained from the EOG tests was inconclusive in determining guilt. In the
literatureix, it was stated that eye movements are not standard physiological parameters used
during lie detection, as the subject has a tendency to become distracted, and thus move their eyes
in a manner that may be construed incorrectly. From the tests performed on the subjects, it can
be stated that vertical and horizontal eye movements measured using the EOG are inconclusive
in determining guilt.
In comparing the two main lie detection methods, the Guilty Knowledge Test and the
Control Question Test, one can see that the GKT is much more conclusive in determining guilt
than the CQT. The CQT assumes subjects respond more to relevant than control questions based
on the fact that guilty subjects would be more responsive to questions dealing with a topic they
are more familiar with than an innocent subject would. However, the data obtained show
significant deviation from this assumption, particularly between the genders. In males, the
galvanic skin response was 62.9% greater for the control questions as compared to the relevant
questions, 21.0% for the vertical EOG p-p for control as compared to relevant questions, but was
3.5% greater for relevant as compared to control questions for the horizontal EOG p-p. A similar
discrepancy was seen in females, as the vertical and horizontal EOG p-p was significantly
greater for the relevant questions as compared to the control questions, but was greater for the
control as compared to the relevant for the GSR. This inconsistent data shows that responses to
control questions are not always less than for relevant, indicating that the CQT may not be
helpful in determining guilt.
In contrast to the discrepancies seen in the CQT data, the GKT EOG data shows a clear
distinction between lying and not lying. When the subject was known to be lying, the galvanic
skin response increased by an average of 50.1%  4.5% for all of the subjects, while the vertical
EOG p-p and the horizontal EOG p-p increased by 7.4%  1.3% and 51.2%  8.9% respectively.
As stated in the literaturex, the GKT has been proved to be more effective at determining guilt
than the control question test, and based on the results obtained in this experiments, those
assertions seem to be accurate.
Given these results, is a unifying equation of lying possible? In an attempt to quantify the
physiological changes exhibited during lying, a weighted formula was derived to determine a
person’s guilt/innocence. The first version of the formula was Equation 1, listed in the Results
section. Equation 1 included GSR weighted at 100% of its value as that was found to be the
strongest indicator of lying for all subjects. Horizontal EOG was given a weighting factor of
20% since the data from this experiment was not a clear indicator of lying. Vertical EOG was
given a weighting factor of 10% as the data from this experiment was the least conclusive
indicator of lying. A lying and not lying score were compiled for each test for each subject. As
an example, the following data is based upon two of the subjects, but is representative of the
inconsistency of the entire data set. It was found with Equation 1 that Subject 1’s lying score
was 9.9% greater than his non-lying score and that Subject 2’s lying score was 235% smaller
than her non-lying score. The weighting factors were then manipulated, yielding Equation 2
from the Results section. It was found with Equation 2 that Subject 1’s lying score was 23.7%
greater than his non-lying score and that Subject 2’s lying score was still 14.6% smaller than her
non-lying score. This example is representative of the entire data set. This data seems to
indicate that a unifying equation of lying cannot be determined from this limited data set.
Further research revealed that even the expert FBI polygraph analysts do not quantify the
polygraph results. Rather, they subjectively scan the data and look for large deviations from the
changes to non-lying questions.xi
Conclusions



Increases in GSR proved to be strongest indicator of lying.
o Overall, changes in eye movement were inconclusive as physiological indicators
of lying.
 Horizontal EOG changes are a stronger physiological indicator than
vertical EOG changes
o Changes in pulse pressure cannot be assessed due to improper filter choice.
o Changes in heart rate and respiratory rate were not helpful physiological
indicators in determining deceit.
The Guilty Knowledge Test proved more accurate in determining guilt than the Control
Question Test.
GKT data was consistent in determining lying amongst the subjects.
It is not possible to design Yes/No Biopac filters that can be used uniformly for all
subjects
Variations in subject response levels make it difficult to set up uniform
thresholding filters that yield binary (Truth/Lie) results displayed in
Biopac.
Appendix - Graphs
Figure A-1: Baseline EOG p-p Data
EOG(p-p) Baseline
0.25
0.2
0.15
volts
0.1
0.05
0
Vert
Horz
-0.05
-0.1
-0.15
Male Female
Figure A-2: Baseline EOG T Data
EOG Delta T Comparison
0.8
0.7
Delta T (sec)
0.6
0.5
0.4
0.3
0.2
0.1
0
Vert
Horz
Male
Female
Figure A-3: Baseline Mean GSR Data
Baseline Mean GSR comparison
0.25
micro-ohms
0.20
0.15
0.10
0.05
0.00
1
Males Females
Figure A-4: GSR Max Data – Card Test
GSR(max)-Card Test
16.00
14.00
MicroOhms
12.00
10.00
8.00
6.00
4.00
2.00
0.00
Male
Female
Not Lying
Combined Data
Lying
Figure A-5: GSR Max Data – GKT
GSR Max - Guility Knowledge Test
4.00
3.50
Microohms
3.00
2.50
2.00
1.50
1.00
0.50
0.00
Male
Female
Not Lying
Combined
Lying
Figure A-6: GSR Max Data – CQT
GSR Max - Control Question Test
6.00
5.00
Microohms
4.00
3.00
2.00
1.00
0.00
Male
Female
Control
Relevant
Combined
Figure A-7: EOG p-p Data – Card Test
EOG(p-p)-Card Test
0.00900
0.00800
0.00700
Volts
0.00600
0.00500
0.00400
0.00300
0.00200
0.00100
0.00000
Male-(vert)
Female(vert)
All(Vert)
Male(Horz) Female(Horz)
Not Lying
All(Horz)
Lying
Figure A-8: EOG p-p Data – GKT
EOG p-p - Guility Knowledge Test
0.00200
0.00180
0.00160
0.00140
Volts
0.00120
0.00100
0.00080
0.00060
0.00040
0.00020
0.00000
Male (vert)
Female
(vert)
Overall
(vert)
Not Lying
Male (hori)
Lying
Female
(hori)
Overall
(hori)
Figure A-9: EOG p-p Data – CQT
EOG p-p - - Control Question Test
0.00180
0.00160
0.00140
Volts
0.00120
0.00100
0.00080
0.00060
0.00040
0.00020
0.00000
Male (vert)
Female
(vert)
Overall
(vert)
Control
Male (hori)
Female
(hori)
Overall
(hori)
Relevant
Figure A-10: EOG T Data – CQT
EOG Delta T - - Control Question Test
0.70000
0.60000
Seconds
0.50000
0.40000
0.30000
0.20000
0.10000
0.00000
Male (vert)
Female
(vert)
Overall
(vert)
Control
Male (hori)
Relevant
Female
(hori)
Overall
(hori)
i
http://www.meddean.luc.edu/lumen/MedEd/GrossAnatomy/dissector/muscles/ooc.htm
Virj, Aldert. Detecting Lies and Deceit. John Wiley & Sons Ltd, New York, 2000
iii
Virj, Aldert. Detecting Lies and Deceit. John Wiley & Sons Ltd, New York, 2000
iv
Guyton & Hall. Medical Physiology. Ch. 60
v
Honts, C. R., Kircher, J. C., & Raskin, D. C. (1995). Polygrapher's dilemma or psychologist's chimaera: A reply to
Furedy's logico-ethical considerations for psychophysiological practitioners and researchers. International Journal of
Psychophysiology, 20, 199-207
vi
http://www.forensicexaminers.com/polygraph.html
vii
Virj, Aldert. Detecting Lies and Deceit. John Wiley & Sons Ltd, New York, 2000
viii
Biopac Pro Manual, Ch. 6
ix
Virj, Aldert. Detecting Lies and Deceit. John Wiley & Sons Ltd, New York, 2000
x
Virj, Aldert. Detecting Lies and Deceit. John Wiley & Sons Ltd, New York, 2000
xi
http://www.howstuffworks.com/question123.htm
ii
Download