The Physionomics of Tax Evasion: Cheating Rationally or Deciding Emotionally

1
The Physionomics of Tax Evasion Cheating
Rationally or Deciding Emotionally?
Conference on Behavioral Economics and
Experiments AFSE-JEE Lyon, 23-25 May 2007

• Giorgio Coricelli (ISC, CNRS, Lyon)
• Matteus Joffily (ISC, CNRS, Lyon)
• Claude Montmarquette (CIRANO, U. of Montreal)
• Marie-Claire Villeval (GATE, CNRS, U. of Lyon,
  and IZA, Bonn)

2
Motivation

• Standard view of tax evasion cheating on taxes
  is a game
• Allingham and Sandmo 72, Yitzhaki 74)
• Ignores the impact of social interactions on
  behavior
• Multiple equilibria (Schelling 78 Akerlof
  80)
• Social multiplier effect (Glaeser, Scheinkman,
  Sacerdote 03)
• Impact of social interations on tax evasion
• Social conformity effect (Gordon 89, Myles and
  Naylor 96, Fortin, Lacroix and Villeval 07)
• Social learning effect
• Fairness effect (Spicer and Becker 80)
• Satisfaction effect (Cowell 90, Pommerehne et al.
  94)
• Literature largely ignores the role of emotions

3

• Why are emotions important in this context?
• If there is a psychic cost in evading taxes, not
  accounting for it will reduce the predictive
  power of the model
• Assumption the psychic cost can be related to 2
  types of emotions
• Internally-oriented emotions (guilt, anxiety) if
  the individual has not
• reported all his income and has been audited
  and punished
• Externally-oriented emotions (shame) if the
  cheating behavior is made
• public to the group

• Problems
• Emotions are usually either ignored or
  mentionned as an indirect explanation of behavior
  in the interpretation of data
• When do emotions arise? When people report their
  income? When they are audited?

4

• This paper An experimental measure of internally
  and externally-oriented emotions in tax evasion
• A standard tax evasion game with risk
• in which we allow for social interactions
• including a physiological measure of emotions
  (Skin Conductance Measures)
• Originality
• - A study of the role of emotions in tax evasion
• Distinct physiological measures of emotions
• when people report their income
• when they receive a feedback on audit
• Physiological measures of emotions in the
  context of social interactions
• Manipulation of the dissemination of information
  (a ceteris paribus change) to
• differenciate internally-oriented and
  externally-oriented emotions

5
Experimental design

• The game
• N8
• 4 individual income levels 50, 100, 150 and 200
• Distribution of income is not made common
  knowledge
• Random allocation of income levels
• A unique tax rate 55
• 2 individual audit probabilities
• If the reported income is among the 4 highest
  reported income 35
• If the reported income is among the 4 lowest
  reported income 65
• One decision how much income to report?

6

• Payoffs
• - No audit ?i
  (income tax on reported income)
• - Audit and no evasion ?i (income
  tax)
• - Audit and evasion ?i
  (income tax - fine)
• with tax rate applied to the actual income
• and fine tax rate evaded income

• Information
• Common knowledge tax rate, individual income,
  audit
• probabilities
• No information about the distribution of income
  in the group, the
• actual probability of audit, the number of
  cheaters

• Equilibrium prediction
• No income to be reported under risk neutrality

7

• 2 treatments
• Benchmark treatment
• the agent receives an income
• s/he reports her/his income
• s/he is audited or not
• s/he receives her/his payoff

• Picture treatment
• The agent knows that her/his picture will be
  disseminated in the
• room if an audit shows that s/he has not
  reported her/his full
• income
• Only one picture disseminated on each screen
• An audited cheater can only see his/her own
  picture

• A within-subject comparison
• 30 periods
• Alternation of sets of 5 periods with
  either the Benchmark or the
• Picture treatment

8
Physiological measure of Electrodermal Activity
(EDA)

• EDA is primarily related to the activity of the
  eccrine sweat glands located on the palmar and
  plantar surface. Those glands are predominately
  controlled by the sympathetic division of the
  autonomic nervous system (ANS). (Dawson et al,
  2000)
• EDA activity can be decomposed into a tonic
  component, EDA level, and a phasic component, EDA
  response.

Modified from Ebling et al, 1992
Modified from Dawnson and Nuechterlein, 1984
9
Skin Conductance Response (SCR)

• EDA responses measured with a constant voltage
  (0.5V) applied between two electrodes (Ag-AgCl)
  placed on the palmar surface are called Skin
  Conductance Response (SCR).
• Psychophysiological studies have shown a high
  correlation between SCR and emotional arousal
  elicited by external stimuli. (Bradley et al,
  2001)

Biopac Systems, USA
10
SCR Quantification

• SCR can be described by several parameters
  amplitude (0.2-1.0µS), latency (1-3s), rise time
  (1-3s) and half recovery time (2-10s). The most
  frequently reported parameter in
  psychophysiological studies is the amplitude.
• When a stimulus is repeated several times and an
  average amplitude of the SCR is to be calculated
  (Dawson et al, 2000)
• SCR Amplitude refers to the mean value computed
  across only those trials in which a measurable
  (nonzero) response occurred
• SCR Magnitude refers to the mean value computed
  across all stimulus presentation including those
  without a measurable response.

Dawson et al, 2000
11
Experiment EDA Recording Processing

• Signal Recording
• BIOPAC MP150W TEL100C
• Constant voltage excitation at 0.5VDC
• Ag-AgCl electrodes
• Conductible gel (Biopac GEL101)
• Analogical low-pass filtering at 35Hz
• Sampling rate at 100 Hz
• Data Reduction
• EDA signal is off-line low-pass filtered at 0.5Hz
• Detected SCR amplitudes are threshold at
  0.02µSiemens
• Stimulus-specific SCRs have onset 0.7-3.7s after
  stimulus onset

Biopac Systems, USA
12
Simultaneous EDA recording from 8 subjects
13

• Control questions
• Self-reported emotions
• Self-report of the  intensity of sentiments 
• and the  nature of sentiments 
  (positive -gt negative)
• on a 7-level Likert scale
• at the end of each of the 30 periods

• A kind of PANAS test (Positive and Negative
  Affect Schedule) with self-reported sentiments
  regarding two scenarii (audit of a subject who
  evaded tax, w/ and w/o picture dissemination)
• happiness, guilt, regret, joy, shame, sadness,
  stress, annoyance
• 7-level Likert scale
• at the end of the session

• Attitudinal post-experimental questionnaire
• Attitude towards fraud in public transportation,
  tax evasion by shop-keepers, non-respect of the
  traffic rules
• 7-level Likert scale

14

• A parsimonious test of attitudes towards risk

• From Gneezy and Charness, 2006
• Endowment of 15 points ( show-up fee of 3)
• How much of this endowment to invest in a risky
  asset?
• P(success) 0.5
• Payoff function
• If the investment is a success
• If the investment is a failure
• Choice of the amount invested at the beginning of
  the session and random draw at the end of the
  session

15

• Procedures
• 6 sessions
• Conducted at GATE, Lyon (France)
• 48 undergraduate Ss from the local engineering
  and business schools
• Computerized experiment with REGATE (Zeiliger,
  00)
• Average duration 90 minutes
• Payoffs
• Payment of the average payoffs in two randomly
  drawn periods at the end of the session
• 100 points 15 , paid in cash in a separate
  room by a person not aware of the content of the
  experiment
• 3 as a show-up fee
• Average earnings 19.33

16
Preliminary results
Proportion of reported income

• The proportion of reported income is higher
• - for the low incomes
• - in the Picture treatment for each income level

17
Amplitude of SCR and income

• Amplitude of SCR increases in income level
• Likely due to the fact that higher income Ss
  report a lower proportion of their income

18
Amplitude of SCR by evasion behavior and by
treatment
Tax evasion raises emotions The difference in
emotional intensity between evaders and
non-evaders is maximum at the time of audit and
in the picture treatment
19

• A 2-step estimation procedure to control for
  endogeneity
• First step Determinants of the reported income
  (in proportion of income)
• Random-effects tobit model

Ss with high income in points report less Ss
report more in the Picture treatment Males, less
wealthy Ss, less risk averse Ss, and students in
business report less Ss who have a poor opinion
of traffic offenders and fare dodgers report more
20

• Second step Determinants of the SCR amplitude
• Random-effects Tobit model with predicted of
  reported income

Guilt
Shame
Emotions arise with a lower reported proportion
of income, picture dissemination, risk
aversion Hesitant Ss feel more emotions An audit
in the previous period increases decision SCR,
and a punishment in the current period increases
audit SCR
21
Conclusion

• Srong results
• Tax evasion raises emotions
• Existence of a shame situation reduces cheating
• Emotions matter
• punishment increases SCR due to
  internally-oriented
• emotions (guilt, regret)
• picture dissemination increases SCR due to
  externally-
• oriented emotions (shame)
• Risk averse Ss report more but, controlling for
  tax evasion, do not feel more emotions
• Implications
• Emotions influence economic behavior
• Creating a situation of guilt and shame generates
  emotions
• Possible extensions
• Other estimations / other SCR indicators
• Reinforcement of the shame / guilt environment