are: I had no idea, I thought it would be 50-50, Had a feeling there will be more white.
There are also a few answers reporting more “extreme beliefs” (0-20 or 30 white chips).
These responses suggest that the unknown bag was indeed perceived as ambiguous.
The Other Treatments. The same procedure outlined for the R-A treatment is adopted
in the other three treatments, with the difference being the type of uncertainty (risk or
ambiguity) characterizing the status quo and the alternative options. Thus, the payoffs
of the status quo gamble remain (10, 4) and the difference between treatments lies in the
composition of the bags on which the status quo gamble and the alternative gambles are
performed. The compositions of the bags are as follows:
– Treatment R-R (Risky - Risky): The (10, 4) gamble and the alternative gambles are
performed on two bags with known compositions of chips: one with 50 white chips
and 50 black chips and the other with 50 green chips and 50 red chips.
– Treatment R-A (Risky - Ambiguous): As explained earlier, the (10, 4) gamble is
performed on a bag with 50 white chips and 50 black chips and the alternative
gambles are performed on a bag with 100 chips, the exact composition of which is
determined using the Dow Jones index.
– Treatment A-R (Ambiguous - Risky): The (10, 4) gamble is performed on a bag with
a composition determined by the Dow Jones index and the alternative gambles are
performed on a bag with 50 white chips and 50 black chips.
– Treatment A-A (Ambiguous - Ambiguous): The (10, 4) gamble and the alternative
gambles are performed on two bags, each with an unknown proportion of black and
white chips. Both compositions are determined according to the decimals of stock
13
market indices (one using the Dow Jones and the other using the S&P 500).
15
Another small difference between treatments lies in the range of prizes chosen for the
alternative gambles, as can be seen in Table 2. In order to maintain an interior switching
point from the (10, 4) gamble to the alternatives, we adjusted the range of prizes of the
alternative set to account for ambiguity aversion.
16
Pilot sessions were used to determine
the range of prizes that would maximize the probability of observing an interior switching
point in each treatment.
3
Results
Evidence for status quo bias is found if the status quo option (the gamble (10, 4)) is chosen
more frequently when subjects own it compared to when they do not. We analyze the data
using two methods. First, we run a random effects probit regression and test whether the
(10, 4) gamble is chosen significantly more often under the status quo frame compared to
the neutral frame. Second, exploiting the within-subject design, we construct a subject
specific status quo bias index and compare the distribution of the index across treatments.
Both approaches lead to the same conclusions.
3.1
Regression analysis
A random effects probit model is estimated to test if the frame of choice has an effect
on the likelihood that the status quo option is chosen. The probability that the (10, 4)
gamble is chosen in any question (as a function of the treatment, the frame and the prizes
15
In all treatments we purposely use two distinct bags: one for the (10, 4) gamble and the other for
the alternative gambles. As a result the status quo gamble and the alternative gambles have independent
outcomes in all treatments.
16
For example the average payoffs of the alternative gambles in treatment R-A were higher than in
treatment R-R to account for ambiguity aversion.
14
of the alternative gamble) is modeled as Φ( ˜
Y ) where Φ is the CDF of the standard normal
distribution and ˜
Y is specified as follows:
˜
Y
i,k
= β
1
(T
R-R
i,k
· SQ
i,k
) + β
2
(T
R-A
i,k
· SQ
i,k
) + β
3
(T
A-R
i,k
· SQ
i,k
) + β
4
(T
A-A
i,k
· SQ
i,k
) + δX
i,k
+
i,k
where T
j
i,k
is a dummy variable that equals 1 if question i of subject k belongs to treatment
j where j ∈ {R-R, R-A, A-R, A-A} and SQ
i,k
is a dummy variable taking value 1 if
question i of subject k belongs to part 2 of the experiment, i.e., to the status quo frame.
X
i,k
is a vector of controls and
i,k
is the error term. The status quo bias is captured by
the coefficients of the interaction terms T
j
i,k
· SQ
i,k
which measure the effect of the status
quo frame on the likelihood that the gamble (10, 4) is chosen in treatment T
j
. The results
of the estimation are reported in Table 3. Specification (1) controls only for the treatment
dummies, while specification (2) also controls for the prizes of the alternative gambles.
Under both specifications we find no significant status quo effect in the “symmetric
treatments” R-R and A-A. We do find evidence of status quo bias in the “asymmetric
treatments” R-A and A-R. After controlling for the payoffs of the alternative gambles, the
status quo frame increases the likelihood that (10, 4) is chosen by 12% in the R-A treatment
and by 10% in the A-R treatment. Both results are significant at the 1% level.
3.2
The Status Quo Bias Index
The within-subject design allows us to construct a status quo bias index for each subject by
comparing his choices under the two frames. The index is constructed as follows: Starting
off with a status quo bias index of 0, if a subject chooses an alternative over the (10, 4)
gamble under the neutral frame, but prefers to keep the (10, 4) gamble after it is given
to him as an endowment, one point is added to the index. If the opposite behavior takes
place, one point is subtracted. Thus, subjects choosing the (10, 4) gamble more often when
15
Dostları ilə paylaş: |