Econ 201A, Fall 2016

GSIs: Avner Shlain and Mauricio Ulate

Problem Set 1 Suggested Solutions

1. Random choice functions: Let X be a finite Pset. A random choice Pfunction is a function
P : 2X \ [0, 1]X such that P (A)(x) 0, aA P (A)(a) = 1 and bA / P (A)(b) = 0. For
notational ease, let PA = P (A). That is, a random choice function takes a menu of options
and outputs a probability distribution over the menu, where PA (x) denotes the probability
that x is chosen from menu A.
A random choice function admits a Luce representation if there exists a set of weights
{w(x) 0 : x X} such that
(
P w(x) if x A
PA (x) = aA w(a) .
0 if x
/A

Let P denote the set of all bijections from X to {1, 2, ..., |X|}. Note that P is equivalent
to the set of all strict rankings of objects in X, where f (x) = 1 is interpreted as x being
the most preferred object under ranking f . A random choice function admits a Falmagne
representation
P if there exists a probability distribution [0, 1]P (that is, (f ) 0 and
f P (f ) = 1) such that:
 P
{f :f (x)f (a) aA} (f ) if x A
PA (x) = .
0 if x
/A

A random choice function satisfies Random Independence of Irrelevant Alternatives

if
PA (x) PB (x)
=
PA (y) PB (y)
whenever PA (x), PA (y), PB (x), PB (y) > 0

(a) Interpret the Luce and Falmagne representation.

Solution: A random choice function admits a Luce representation if we can assign

non-negative weights to each of the elements in X such that the probability that the
random choice function assigns to x out of a given set A (which is written PA (x)) is
equal to the weight ofP x (which is w(x)) divided by the sum of the weights of all ele-
ments in A (which is aA w(a)) if x A, and zero if x / A. The advantage of this
approach is that if a random choice function admits a Luce representation the probabili-
ties PA (x), which in general could be very abstract objects, take on a very simplified form.

A random choice function admits a Falmange representation if we can assign proba-

bilities (these are the (f )) to all the possible rankings of elements in X (these are the
functions f : X {1, 2, ..., |X|}) such that the probability that the random choice func-
tion assigns to x out of a given set A (which is written PA (x)) is equal to the sum of
the probabilities of all rankings that would rank x ahead of all the other elements of A
if x A, and zero if x / A.

(b) Interpret the Random IIA condition.

Page 1 of 4
Econ 201A, Fall 2016
GSIs: Avner Shlain and Mauricio Ulate
Problem Set 1 Suggested Solutions

Solution: The RIIA has a more straight forward interpretation. If both x and y have
a positive probability of being chosen out ot the sets A and B, then the probability of
choosing x relative to the probability of choosing y is the same in A and B and hence
cannot depend on the presence of different elements in A with respect to B.

(c) Prove or provide a counterexample to the following statement: Let P admit a Luce rep-
resentation. Define C(A) = {x A : PA (x) > 0}. Then C is rationalizable.

Solution: Notice that even though the definition of Luce representation only requires
the set of weights to fulfill: {w(x) 0 : x X}, if any of the w(x) were zero then
P{x} (x) will not be well defined, hence for this definition to make sense we must actually
impose w(x) > 0 x X. This implies that A X, A 6= , PA (x) > 0 x A. And
hence we have C(A) = A. In particular this implies that C is non-empty, so to show that
it is rationalizable we will show that it satisfies Houthakkers axiom. If x, y A B,
x C(A), y C(B) we actually have that x B = C(B), which implies that indeed C
satisfies Houthakkers axiom.

(d) Prove or provide a counterexample to the following statement: If P admits a Luce rep-
resentation, then it satisfies RIIA.

Solution: Let x and y have possitive probability of being chosen out of both A and
B (this implies x, y A B). Assume P admits a Luce representation, the previous
statement implies w(x), w(y) > 0. Also we can write:

P w(x) P w(x)
PA (x) aA w(a) w(x) bB w(b) PB (x)
= = = =
PA (y) P w(y) w(y) P w(y) PB (y)
aA w(a) bB w(b)

Which implies that P satisfies the RIIA condition.

(e) Suppose PA (a) > 0 whenever a A. Prove that if P satisfies Random IIA, then P
admits a Luce representation (Hint: Consider a candidate for w).

Solution: Our candidate for the weight of x will be w(x) = PX (x), that is, the proba-
bility that the random choice function assigns to x when we are considering the whole
set X as domain. Notice that by the assumption PA (a) > 0 whenever a A we know
that all weights will be positive since we are using X as the reference probability set
and this set by definition includes all elements. We will now prove that if x A then:

w(x) PX (x)
PA (x) = P =P
aA w(a) aA PX (a)

Notice that for all elements a A, by the RIIA and the fact that x, a A we know:

PA (a) PX (a) PX (a)

= PA (a) = PA (x)
PA (x) PX (x) PX (x)

Page 2 of 4
Econ 201A, Fall 2016
GSIs: Avner Shlain and Mauricio Ulate
Problem Set 1 Suggested Solutions

Adding this equations over all the elements in A we obtain:

P
X PX (a)
PA (a) = PA (x) aA
PX (x)
aA
P
But remember that aA PA (a) = 1, so we obtain:

PX (x) w(x)
PA (x) = P =P
aA PX (a) aA w(a)

On the other hand if x

/ A then by definition of a random choice function we must have
PA (x) = 0. So we have shown that P admits a Luce representation.

(f) Prove or provide a counterexample to the following statement: If P admits a Falmagne

representation, then it satisfies Random IIA.

Solution: Consider X = {L, M, N }, where random choice function admits a Falmange

representation whose probability distribution is described in the next table:

1 2 3 Probability
L M N 0
L N M 3/4
M L N 1/8
M N L 0
N L M 0
N M L 1/8

Then label A = L, M, N , B = L, M and notice:

PA (L) = 3/4

PA (M ) = 1/8
PB (L) = 3/4
PB (M ) = 1/8 + 1/8 = 1/4
So that:
PA (L) 3/4 3/4 PB (L)
= = 6 6= 3 = =
PA (M ) 1/8 1/4 PB (M )
Which means RIIA is violated.

2. Bottles of wine: Let X = {1, ..., n} be a set of bottles of wine, indexed by their prices.
Consider a consumer who always selects the second-cheapest bottle of wine from the menu.
Prove or provide counterexamples to the following statements:

Page 3 of 4
Econ 201A, Fall 2016
GSIs: Avner Shlain and Mauricio Ulate
Problem Set 1 Suggested Solutions

(a) C satisfies Sens .

Solution: C doesnt satisfy Sens , to see this let X = {1, 2, 3} with prices: pi = i i
X. Notice that 2 {1, 2} {1, 2, 3}, also 2 C({1, 2, 3}) = {2}, but 2
/ C({2, 3}) = {3}.
This is a direct violation of Sens .

(b) C satisfies Sens .

Solution: The question is open to several interpretations, which we discuss next:

i. You can think that X = {1, ..., n} being indexed by their prices means that pi =
i i X. Under this interpretation there will never by any ties, choices will only
be composed of one element and hence Sens will be vacuously satisfied.
ii. You can think that X = {1, ..., n} being indexed by their prices means that they
each have a price which could be any number, for example: p1 = 12, p2 = 8, p3 =
20, p4 = 15, etc. Under this interpretation there could be ties and in particular
there could be ties at the second cheapest wine. If ties are broken in some way,
then again choices will only be composed of one element and hence Sens will be
vacuously satisfied.
iii. Under the same interpretation as in ii. where ties are a possibility, we could also
assume that ties are not broken, so that if there is a tie at the second cheapest wine
the choice for that set would be all the bottles with the second cheapest price. In
this case Sens will still be satisfied. To see this assume that x, y C(A), A B,
and y C(B). Since the only way that different bottles of wine could be chosen
from A is if they have the same price we know px = py . We also know that x B.
Since y is the second cheapest bottle of wine in B and px = py , we know x is also
the second cheapest bottle of wine in B and hence x C(B), which means Sens
is satisfied.

Page 4 of 4