Vous êtes sur la page 1sur 23

Privacy Protection

and Risk Attitudes


Alisa Frik and Alexia Gaudeul
Universit degli Studi di Trento
Georg-August-Universitt Gttingen

ESA Jerusalem 2016

2 Motivation
Currently established methods for assessing
the value of privacy:
WTP / WTA payment to avoid personal
information being revealed to others

Surveys with questionnaire on attitudes and


behavior in terms of privacy protection.

Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

3 Motivation
Those elicitation methods do not replicate
actual decisions people make.
Risk to privacy are not 0-1 decisions.
People are rarely faced with the certainty of a loss
of privacy, to be traded. Rather, they are faced with
the probability of a loss.
Not incentivized or incentives too complex to
understand.

Stated preferences differ from observed


preferences.
Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

4 Contributions
1. Original method to elicit concern for
privacy.
2. New method of elicitation of privacy
valuations, by using privacy lotteries.
We indirectly obtain implicit valuation of privacy.

3. Test of the relation between privacy and


risk attitudes.
We elicit attitudes vs. monetary risk.

We compare those with the willingness to protect


private information.
Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

Design
1. Generation of private
information
2. Lotteries and MPL

3. Controls

Privacy and risk attitudes - Frik & Gaudeul


- ESA Jerusalem

Sunday, 4 Tammuz, 5776

6 Generation of privacy concern


Name and surname + picture.
14 questions about sensitive and socially relevant
topics, on which opinions differ in the population.
Abortion
Illegal immigration
Contraception
etc

This means there is no way to answer in a way


that is agreeable to all.
But some people may enjoy getting their opinion known!
Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

7 Information revealed to others


Name and surname announced to others.

Subjects stands in front of others in the


lab.
Participants see information on their
screen as shown in the next slide:

Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

8 Information revealed to others


Seat #23:
thinks it is morally justified to abort after
discovering serious disability in the fetus, while
36 % of other participants do not agree
is in favor of chemical castration as and
appropriate penalty for rape, while 87% of
other participants did not choose this option

...
...

Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

9 Monetary lotteries
Row Option A

Option B

4 lists

You get
56 ECU

You get 55 ECU, but with probability 30 %


you lose 10 of those ECUs

You get
55 ECU

You get 55 ECU, but with probability 30 %


you lose 10 of those ECUs

You get
54 ECU

You get 55 ECU, but with probability 30 %


you lose 10 of those ECUs

You get
53 ECU

You get 55 ECU, but with probability 30 %


you lose 10 of those ECUs

You get
52 ECU

You get 55 ECU, but with probability 30 %


you lose 10 of those ECUs

You get
51 ECU

You get 55 ECU, but with probability 30 %


you lose 10 of those ECUs

You get
50 ECU

You get 55 ECU, but with probability 30 %


you lose 10 of those ECUs

You get
49 ECU

You get 55 ECU, but with probability 30 %


you lose 10 of those ECUs

You get
48 ECU

You get 55 ECU, but with probability 30 %


you lose 10 of those ECUs

10

You get
47 ECU

You get 55 ECU, but with probability 30 %


you lose 10 of those ECUs

11

You get
46 ECU

You get 55 ECU, but with probability 30 %


Privacyyou
andlose
risk10attitudes
Frik & Gaudeul - ESA Jerusalem
of those -ECUs

always 30% chance to


get info revealed
safe payoff is either 55,
65, 75 or 35 ECU.
10 ECU = 1 euro
Value of loss of 10, 30
or 50, or a gain of 30
ECU.

Note how we allow


for a full range of risk
attitudes.
Sunday, 4 Tammuz, 5776

10 Privacy lotteries
Row Option A

Option B

You get 56
ECU

You get 55 ECU, but with probability 30 %


your personal information is disclosed

You get 55
ECU

You get 55 ECU, but with probability 30 %


your personal information is disclosed

You get 54
ECU

You get 55 ECU, but with probability 30 %


your personal information is disclosed

You get 53
ECU

You get 55 ECU, but with probability 30 %


your personal information is disclosed

You get 52
ECU

You get 55 ECU, but with probability 30 %


your personal information is disclosed

You get 51
ECU

You get 55 ECU, but with probability 30 %


your personal information is disclosed

You get 50
ECU

You get 55 ECU, but with probability 30 %


your personal information is disclosed

You get 49
ECU

You get 55 ECU, but with probability 30 %


your personal information is disclosed

You get 48
ECU

You get 55 ECU, but with probability 30 %


your personal information is disclosed

10

You get 47
ECU

You get 55 ECU, but with probability 30 %


your personal information is disclosed

11

You get 46
ECU

You get 55 ECU, but with probability 30 %


Privacy
and risk
attitudes -isFrik
& Gaudeul - ESA Jerusalem
your personal
information
disclosed

Same as monetary
lotteries but
replace gain/loss
with information
disclosure.
We allow for
negative values
for privacy.
Sunday, 4 Tammuz, 5776

11 Controls
Age, gender, education, revenues, ...

Order of elicitation (monetary, privacy).


Unavoidable risk of disclosure or not.
Questionnaire-based indexes of privacy
attitudes.
WTP and WTA payment for privacy disclosure.

Trust in others, other people known, trust in


experimenter, ...
Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

Results

12

Privacy and risk attitudes - Frik & Gaudeul


- ESA Jerusalem

Sunday, 4 Tammuz, 5776

13 Data
May/June 2015

University of Trento, Italy


148 subjects

8 sessions
2x2 treatments (order of elicitation,
unavoidable risk to privacy).

Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

14 Statistics
Monetary risk attitude.
1 + =
=

Index of aversion to privacy disclosure.


= 70% + 30%
=

Warning: This is not a value for private information.

Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

15 Regression of v over r
Panel random effect GLS regressions.
1. Positive relation of v with r subjects who
are more risk averse also are more averse
to privacy disclosure.
Does not mean they have higher privacy valuation!

2. Higher index if privacy fundamentalists,


lower if know more other participants,
lower if trust strangers.
3. Robustness check: Number of safe choice in
privacy tasks rather than v.
Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

16 Implicit privacy values


1. (1 + ) = 70% + 30% so
(+)
that =
%

2. Correcting privacy values with risk attitudes,


we find that many people have negative values
for privacy.
1.

WTA 16 average, all positive.

2.

WTP 2 average, all positive.

3.

Index v: 2.5 average, 2% negative.

4.

Implicit v: 0.4 average, 53% negative.

Maybe due to subject pool / specific elicitation of


concern for privacy.
Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

Implicit valuation
of privacy
Is privacy risk like
monetary risk?

17

Privacy and risk attitudes - Frik & Gaudeul


- ESA Jerusalem

Sunday, 4 Tammuz, 5776

18 Implicit privacy valuations


1. Monetary lotteries:
1. 75 ECU and 30% chance that you will lose 50
of those ECUS.

2. 1 + 50

= 75 30% 50

2. This gives us (50), an index of risk


aversion (if the loss is 50).

3. We get r for different values of the loss,


infer from this the function ().
Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

19 Privacy valuations
Then we look at certainty equivalent for
privacy lotteries.
Indifferent between CE and 75 ECU and 30% that
your personal information will be revealed to
others.
1 + ( ) = 75 30% so we obtain
1+ )
implicit = (75
if v() is linear.
30%

This allows us to ascertain how much of the


fear of privacy loss is due to general risk
aversion and what is due to high value for
privacy.
Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

20 Perception of privacy risks


But this assumes that people reason the
same way for privacy risks as for risks
involving money!
So we examine two variations (order of
elicitation, unavoidable risk to privacy) to
(start) to see if privacy risk is evaluated in
a different way from monetary risk.

Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

21 Privacy and control


We tested for an all-or-nothing effect.

Does knowing that ones privacy is already


at risk lead one to stop valuing it?
Treatment 1: Can avoid privacy risk entirely.
Treatment 2: Told that with 20% probability
they will have to reveal private information to
others, whatever the decision.

No difference between treatments.


Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

22 Privacy and order of elicitation


We test for the effect of framing (order of
elicitation).
We find that eliciting monetary risk
attitudes first made people less sensitive
to privacy risk.
This indicates that being asked to think in
terms of money first leads people to pay
less attention to non-monetary outcomes
(privacy).
Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

23 Conclusion
Novel privacy elicitation method inspired
by experimental literature on risk
attitudes.
Positive relation between privacy and risk
valuation, while other indexes of privacy
attitudes play a limited role.
Losing control on privacy disclosure does
not lead people to lose interest in
protecting their privacy.
Privacy and risk attitudes - Frik & Gaudeul - ESA Jerusalem

Sunday, 4 Tammuz, 5776

Vous aimerez peut-être aussi