The Polygraph

1
The Polygraph

• The ideal a machine to detect lies
• No personal bias
• Reliable, objective, automatic
• Since 1890s the polygraph
• A physiological measuring device
• Measures several channels (heart rate,
  respiration, GSR, blood pressure)
• In wide use worldwide
• Popular in South Africa (insurance companies,
  recruiting)
• Eg. Pick n Pay, De Beers Marine, First National
  Bank, Kentucky Fried Chicken, SA Revenue Service

2
The polygraph
A polygraph examination underway
A paper recording of polygraph data (digital
version similar)
Actual polygraph output
3

• Technology not changed since 1900s
• Now it records digital
• Physiological measurements are very accurate
• Some sensitivity to movement, etc but can be
  compensated for
• Can record for extended periods of time
• Only measures physical variables
• Not lying/innocence!
• Lying is determined by making inferences about
  the physical measurements

4
Inferences about lying

• How do you determine lying from physiological
  data?
• No actual theory!
• Basic idea Lying will lead to increase in
  arousal
• Increase in arousal has a particular reaction
• Increase in blood pressure, heart rate,
  respiration
• Decrease in GSR
• Look for this pattern in the printout
• These variables also vary naturally, often a lot

5
How to look for a lie

• Look at all four channels
• Any one of them may tell you
• An increase will indicate an increase in arousal
  and thus a lie
• How much of an increase indicates a lie?
• Depends on each person
• Must compare within subjects
• Compare a truth situation with a lie
  situation
• Obtain baselines
• Ask subject to lie about something unrelated,
  check levels.
• Do the same for truth telling

6
An arousal increase?

• Is it true that an arousal increase goes with
  lying?
• Assumed rather than demonstrated
• Arousal increases can occur due to a number of
  situations
• Not only lying (eg. stress about the test)
• The machine cannot differentiate between these!
• A problem What do you use as your baseline?
• A neutral statement
• A harmless lie (?)

7
The relevant/Irrelevant test (RIT)

• One way of using the polygraph
• The original way
• Two types of questions asked
• Simple statements, short answers (yes/no, etc)
• Relevant questions (about the crime, etc) Did
  you take the money?
• Irrelevant questions (used for baseline/control)
  Do you live in Cape Town?
• If activity is greater in relevant questions,
  conclude the subject is lying
• BUT Relevant questions will lead to an increase
  in arousal anyway! (false positive rate is high)

8
Control Question Test (CQT)

• Most common polygraph test in use
• Compare critical questions with unrelated lies
• Critical Did you take the money?
• Unrelated Have you ever stolen anything before
  this year?
• Questions discussed before the examination
• If the critical response is greater than the
  unrelated one, conclude he was telling a lie

9
Problems with the CQT

• It is necessary for the subject to believe the
  polygraph works
• To establish the unrelated lie baseline
• stimulus test (eg. fake card trick)
• Much of the effect of the test occurs before
  you begin!
• Trick your subject
• Examiner establishes themselves in a position of
  power over the subject
• Great variability on results depending on the
  examiner
• A lot depends on the questions chosen

10
Control in the CQT

• The control questions (unrelated lies) are not
  effective controls
• They do not show that the increase in critical
  questions can only be due to lying
• The content of the critical question may greatly
  increase arousal in an innocent subject
• The unrelated lie may not lead to significant
  arousal (didnt care)
• In legal disputes, critical questions will
  probably lead to high arousal, even in innocent
  subjects

11
External information in the CQT

• The polygraph operator has several roles
• Operates the machine
• Interrogates the subject
• expected to provide the answer to the mystery
• Polygrapher often knows about the case before the
  test
• External information is used to reach a
  conclusion
• Removes the machine objectivity of the test
• Polygraph used as a tool for coercing confessions
• Should use blind examiners only

12
Beating the polygraph

• All polygraph tests work on the basis of an
  arousal comparison
• Base state vs. lying state
• You will know which questions are control
  questions and which are relevant
• Increase arousal in control state to remove the
  difference
• Confuses the examiner (strange pattern)
• How to increase arousal
• Clench leg muscles, count backwards from 100 in
  13s, think of something annoying, etc.
• Must do it without the examiner knowing
• Will prevent non-polygraph information from being
  emphasized

13
So what if the theory sucks?

• Even if lying/arousal is not related so what?
• If the machine can detect lies, theory is
  irrelevant
• We are solving a practical problem!
• Use empirical studies to measure the usefulness
  of each test
• The RIT does very badly
• Correctly identifies lies only 50 of the time
• Effective guessing the result (coin toss would
  be as good)
• Most researchers agree the RIT is useless to
  detect lies.

14
How good is the CQT?

• Attracted a lot of research
• Lab experiments and field studies
• Confused results (40 studies)
• Lies accurately detected with 78 accuracy (53 -
  90)
• Innocents accurately detected with 84 accuracy
  (70 - 90)
• Lab experiments have been criticized
• Unrealistic (low external validity)
• Perfect conditions for the machine (overestimate
  accuracy)
• Big difference between real-world lying and lab
  lying

15
Field studies of CQT accuracy

• Major problem Was a lie really told?
• Ground truth mostly not available
• Confession or external corroboration (rare)
• No clear agreement on what is acceptable to
  include
• Iacono Lykken (big critics)
• Sampling bias in confession cases
• Innocents who failed the test are omitted from
  the sample
• Guilty cases who got away with it are not
  included in the sample
• Studies are heavy with cases of successful
  identification (failures missing)

16
Field results for the CQT

• Raskin Honts (proponents of CQT)
• Guilty correct identifications average 86 (73 -
  100)
• Innocent correct identifications average 50 (30
  - 83)
• Iacono Lykken (oppose the CQT)
• Find about the same numbers
• Numbers are not very good
• Average at catching liars
• Very likely to generate false positives (horror!)

17
The polygraph and employment screening

• Difference between criminal use and employment
  use
• Employers want to know if a person is honest,
  truthful
• Event-free use of the polygraph
• Orwellian fantasy
• People will be honest if the machine can tell
  when they lie!
• In event free situations, the RIT is often used
• The CQT designed to ask about a specific thing
• RIT you can ask about anything

18
A big problem base rate

• Types of events management wants to uncover are
  very rare
• But the accuracy of the polygraph itself is low
• This leads to extremely high false positive rates
  (Bayesian probability calculation)
• A lot of people being turned down/fired
• With 2M screenings, as many as 320 000 in the US
  each year (estimate mid 1980s)
• USA now has a law preventing polygraph use in the
  workplace
• But we still use it (Yay! Yay!)