Department of Economics, Bilkent...
Transcript of Department of Economics, Bilkent...
List rationalizable choice
KEMAL YILDIZ
Department of Economics, Bilkent University
May 7, 2014
We develop and analyze a boundedly rational choice model both indeterministic and stochastic choice setups
General setup
X is an alternative set with n elements
choice sets are S ⊂ X
choice space is the collection of all choice sets; Ω
a choice function is a mapping c : Ω→ X such that for each S,c(S) ∈ S
General setup
X is an alternative set with n elements
choice sets are S ⊂ X
choice space is the collection of all choice sets; Ω
a choice function is a mapping c : Ω→ X such that for each S,c(S) ∈ S
Rational choice model
The decision maker(DM) is endowed with a preferencerelation, , over the alternative set,
and chooses the -best alternative from each choice set,
i.e. a rational DM chooses from each choice set as ifmaximizing a preference relation
LRC has a single main primitive:
a *list* is an ordering of the alternatives
for each x, y ∈ X we write x f y iff x follows y in the list
P denotes a complete asymmetric binary relation on X
LRC has a single main primitive:
a *list* is an ordering of the alternatives
for each x, y ∈ X we write x f y iff x follows y in the list
P denotes a complete asymmetric binary relation on X
LRC has a single main primitive:
a *list* is an ordering of the alternatives
for each x, y ∈ X we write x f y iff x follows y in the list
P denotes a complete asymmetric binary relation on X
For each < f , P >, we define CP,f : Ω→ X recursively as follows.
Let S = x, y, z,
Suppose x P y P z P x
z y xP P
P
For each < f , P >, we define CP,f : Ω→ X recursively as follows.
Let S = x, y, z,
Suppose x P y P z P x
z y xP P
P
C1P,f (S) = f1(S)
z y x
C1P,f (S) = f1(S)
C2P,f (S) = f1(S) ∨P f2(S)
z y xP
C1P,f (S) = f1(S)
C2P,f (S) = f1(S) ∨P f2(S)
C3P,f (S) = C2
P,f (S) ∨P f3(S)
z y xP P
C1P,f (S) = f1(S)
C2P,f (S) = f1(S) ∨P f2(S)
.
.
.CkP,f (S) = Ck−1
P,f (S) ∨P fk(S) for each k ∈ 2, ...s
Finally, let CP,f (S) = CsP,f (S)
z y xP P
C1P,f (S) = f1(S)
C2P,f (S) = f1(S) ∨P f2(S)
.
.
.CkP,f (S) = Ck−1
P,f (S) ∨P fk(S) for each k ∈ 2, ...s
Finally, let CP,f (S) = CsP,f (S)
z y xP P
List Rationality
Definition
A choice function c is list rational if there exists a list f such thatfor each S ∈ Ω, if x is the last alternative in S according to f , then
c(S) = c(c(S \ x), x)
motivation
LRC procedure offers an intuition for the choice behavior of a DMwho:
process information sequentially
are constrained by single memory
can exhibit binary choice cycles.
motivation
LRC procedure offers an intuition for the choice behavior of a DMwho:
process information sequentially
are constrained by single memory
can exhibit binary choice cycles.
motivation
LRC procedure offers an intuition for the choice behavior of a DMwho:
process information sequentially
are constrained by single memory
can exhibit binary choice cycles.
motivation
Salant (2003) argues that list rationality is the unique choiceprocedure that uses a single memory cell.
Russo & Rosen (1975) argues that LRC is followed by thesubjects to minimize the short-term memory load.
Shugan (1980) proposes a formulation based on cost ofthinking, and argues that the lowest choice cost would beachieved by first comparing among an alternative pair with lowthinking cost, and then comparing the winner with the next
Liu & Simonson (2005) argue that decision makers guided bylist rationality are more confident in their choices compared tonon-guided decision makers.
List rationality vs Choice from the lists
Rubinstein and Salant’06 analyze a new choice model whereDMs choose the alternatives out of given observable lists
We model the list as a subjective part of the choice procedure.Even if there is a given list, a DM can follow a different virtualordering.
one can start the comparison from a referencepoint,(Tversky’91)
group the similar items together, (Russo’75) or
can order them to minimize the total thinking cost.(Shugan’80)
List rationality vs Choice from the lists
Rubinstein and Salant’06 analyze a new choice model whereDMs choose the alternatives out of given observable lists
We model the list as a subjective part of the choice procedure.Even if there is a given list, a DM can follow a different virtualordering.
one can start the comparison from a referencepoint,(Tversky’91)
group the similar items together, (Russo’75) or
can order them to minimize the total thinking cost.(Shugan’80)
List rationality vs Choice from the lists
Rubinstein and Salant’06 analyze a new choice model whereDMs choose the alternatives out of given observable lists
We model the list as a subjective part of the choice procedure.Even if there is a given list, a DM can follow a different virtualordering.
one can start the comparison from a referencepoint,(Tversky’91)
group the similar items together, (Russo’75) or
can order them to minimize the total thinking cost.(Shugan’80)
List rationality vs Choice from the lists
Rubinstein and Salant’06 analyze a new choice model whereDMs choose the alternatives out of given observable lists
We model the list as a subjective part of the choice procedure.Even if there is a given list, a DM can follow a different virtualordering.
one can start the comparison from a referencepoint,(Tversky’91)
group the similar items together, (Russo’75) or
can order them to minimize the total thinking cost.(Shugan’80)
List rationality vs Choice from the lists
Rubinstein and Salant’06 analyze a new choice model whereDMs choose the alternatives out of given observable lists
We model the list as a subjective part of the choice procedure.Even if there is a given list, a DM can follow a different virtualordering.
one can start the comparison from a referencepoint,(Tversky’91)
group the similar items together, (Russo’75) or
can order them to minimize the total thinking cost.(Shugan’80)
List rationality vs Choice from the lists
Rubinstein and Salant’06 analyze a new choice model whereDMs choose the alternatives out of given observable lists
We model the list as a subjective part of the choice procedure.Even if there is a given list, a DM can follow a different virtualordering.
one can start the comparison from a referencepoint,(Tversky’91)
group the similar items together, (Russo’75) or
can order them to minimize the total thinking cost.(Shugan’80)
Rationalization by a binary game tree
Rationalization by a binary game tree
Rationalization by a game tree (Xu & Zhou’07)
Rationalization by a game tree (Xu & Zhou’07)
Q How to identify the non-observed list from the observed choicesof the DM ?
Answer: This can be achieved if we can conclude from theobserved choices of the DM that an alternative x unambiguouslyfollows another alternative y in his considerations
Random Choice
Random choice
∆(S) is the collection of probability distributions over S
a random choice rule is a mapping c, such that
∀S ∈ Ω, c(S) ∈ ∆(S)
a random binary relation is a mapping P : X ×X → [0, 1]
Random choice
∆(S) is the collection of probability distributions over S
a random choice rule is a mapping c, such that
∀S ∈ Ω, c(S) ∈ ∆(S)
a random binary relation is a mapping P : X ×X → [0, 1]
Random choice
∆(S) is the collection of probability distributions over S
a random choice rule is a mapping c, such that
∀S ∈ Ω, c(S) ∈ ∆(S)
a random binary relation is a mapping P : X ×X → [0, 1]
Random choice
∆(S) is the collection of probability distributions over S
a random choice rule is a mapping c, such that
∀S ∈ Ω, c(S) ∈ ∆(S)
a random binary relation is a mapping P : X ×X → [0, 1]
why modeling random choice?
why modeling random choice?
why modeling random choice?
Random choice accommodates:
repeated decisions of a single individual
choice data of a group of individuals
why modeling random choice?
Random choice accommodates:
repeated decisions of a single individual
choice data of a group of individuals
why modeling random choice?
Random choice accommodates:
repeated decisions of a single individual
choice data of a group of individuals
LRRC
Consider the LRRC procedure described by < f ,P >, where
z y x1/2 1/2
1/2
LRRC
LRRC which follows f , yields for each S, Pf (S) ∈ ∆(S) :
z y x1/2
1/2
⇐ z y x
Pf (S) 1/4
LRRC
LRRC which follows f , yields for each S, Pf (S) ∈ ∆(S) :
z y x1/2
1/2
⇐ z y x
Pf (S) 1/4 1/4
LRRC
LRRC which follows f , yields for each S, Pf (S) ∈ ∆(S) :
z y x
1/2
1/2
⇐ z y x
Pf (S) 1/4 1/4 1/2
Results
Plott (1973)
Stochastic Path Independence (SPI)
Thm: A choice function is rational iff path independence issatisfied.
Plott (1973)
Stochastic Path Independence (SPI)
Thm: A choice function is rational iff path independence issatisfied.
Path Independence
Path Independence: For each S, T such that T ⊂ S, if x = c(T )and y = c(S \ T ), x = c(x, y) implies x = c(S).
x yz w
=y
z w∪ x
Path Independence
Path Independence: For each S, T such that T ⊂ S, if x = c(T )and y = c(S \ T ), x = c(x, y) implies x = c(S).
x yz w
=y
z w∪ x
Path Independence
Path Independence: For each S, T such that T ⊂ S, if x = c(T )and y = c(S \ T ), x = c(x, y) implies x = c(S).
x yz w
=y
z w∪ x
⇒ x, y → x
Path Independence
Path Independence: For each S, T such that T ⊂ S, if x = c(T )and y = c(S \ T ), x = c(x, y) implies x = c(S).
x yz w
=y
z w∪ x
⇒ x, y → x
Path Independence
Path Independence: For each S, T such that T ⊂ S, if x = c(T )and y = c(S \ T ), x = c(x, y) implies x = c(S).
x yz w
=y
z w∪ x
⇒ x, y → x
Path Independence
Path Independence: For each S, T such that T ⊂ S, if x = c(T )and y = c(S \ T ), x = c(x, y) implies x = c(S).
x yz w
=y
z w∪ x
⇒ x, y → x
The Stochastic Case
Stochastic Path Independence (SPI)
Stochastic Path Independence: For each S ∈ Ω, x ∈ S andy 6∈ S,
cx(S ∪ y) = cx(S) · c(x, y)
The Stochastic Case
Stochastic Path Independence (SPI)
Stochastic Path Independence: For each S ∈ Ω, x ∈ S andy 6∈ S,
cx(S ∪ y) = cx(S) · c(x, y)
The Stochastic Case
For each x, y ∈ X, x is revealed to follow y (x Fc y) iff SPI isviolated between x and y, that is : for some S ∈ Ω s.t x ∈ S andy 6∈ S,
cx(S ∪ y) 6= c(x, y) · cx(S)
Proposition
Proposition: c is list rationalizable iff fc is acyclic.
Moreover, the followed list is identified unique up to thecompletions of transitive closure of fc.
The Stochastic Case
For each x, y ∈ X, x is revealed to follow y (x Fc y) iff SPI isviolated between x and y, that is : for some S ∈ Ω s.t x ∈ S andy 6∈ S,
cx(S ∪ y) 6= c(x, y) · cx(S)
Proposition
Proposition: c is list rationalizable iff fc is acyclic.
Moreover, the followed list is identified unique up to thecompletions of transitive closure of fc.
The Stochastic Case
For each x, y ∈ X, x is revealed to follow y (x Fc y) iff SPI isviolated between x and y, that is : for some S ∈ Ω s.t x ∈ S andy 6∈ S,
cx(S ∪ y) 6= c(x, y) · cx(S)
Proposition
Proposition: c is list rationalizable iff fc is acyclic.
Moreover, the followed list is identified unique up to thecompletions of transitive closure of fc.
The Stochastic Case
For each x, y ∈ X, x is revealed to follow y (x Fc y) iff SPI isviolated between x and y, that is : for some S ∈ Ω s.t x ∈ S andy 6∈ S,
cx(S ∪ y) 6= c(x, y) · cx(S)
Proposition
Proposition: c is list rationalizable iff fc is acyclic.
Moreover, the followed list is identified unique up to thecompletions of transitive closure of fc.
The Deterministic Case
For each distinct x, y ∈ X, x is revealed to follow y (x Fc y), if forsome S ∈ Ω, we have either
(i) x = c(S ∪ y) and [y = c(x, y) or x 6= c(S)] or
(ii) x 6= c(S ∪ y) and [x = c(x, y) and x = c(S)].
Proposition
Corollary: c is list rational iff Fc is acyclic.
Followed list is identified unique up to the completions oftransitive closure of fc.
The Deterministic Case
For each distinct x, y ∈ X, x is revealed to follow y (x Fc y), if forsome S ∈ Ω, we have either
(i) x = c(S ∪ y) and [y = c(x, y) or x 6= c(S)] or
(ii) x 6= c(S ∪ y) and [x = c(x, y) and x = c(S)].
Proposition
Corollary: c is list rational iff Fc is acyclic.
Followed list is identified unique up to the completions oftransitive closure of fc.
The Deterministic Case
For each distinct x, y ∈ X, x is revealed to follow y (x Fc y), if forsome S ∈ Ω, we have either
(i) x = c(S ∪ y) and [y = c(x, y) or x 6= c(S)] or
(ii) x 6= c(S ∪ y) and [x = c(x, y) and x = c(S)].
Proposition
Corollary: c is list rational iff Fc is acyclic.
Followed list is identified unique up to the completions oftransitive closure of fc.
Analysis of Stochastic Path Independence
Stochastic Path Independence (SPI)
SPI: For each S ∈ Ω, and y 6∈ S, cx(S ∪ y) = c(x, y) · cx(S)
a preference relation is a near linear order if the length of ∼ is atmost two
Proposition
Proposition: c satisfies SPI iff there is a near linear order on Xsuch that; for each S ∈ Ω
c(S) = c(max(S,))
Analysis of Stochastic Path Independence
Stochastic Path Independence (SPI)
SPI: For each S ∈ Ω, and y 6∈ S, cx(S ∪ y) = c(x, y) · cx(S)
a preference relation is a near linear order if the length of ∼ is atmost two
Proposition
Proposition: c satisfies SPI iff there is a near linear order on Xsuch that; for each S ∈ Ω
c(S) = c(max(S,))
Analysis of Stochastic Path Independence
Stochastic Path Independence (SPI)
SPI: For each S ∈ Ω, and y 6∈ S, cx(S ∪ y) = c(x, y) · cx(S)
a preference relation is a near linear order if the length of ∼ is atmost two
Proposition
Proposition: c satisfies SPI iff there is a near linear order on Xsuch that; for each S ∈ Ω
c(S) = c(max(S,))
X be a real linear vector space
for each S ∈ Ω, c(S) ∈ ∆(S) ⊂ X
c is continuous if for each x, y ∈ X and xnn≥1 ⊂ X,
lim xn = x ⇒ lim c(xn, y) = c(x, y)
X be a real linear vector space
for each S ∈ Ω, c(S) ∈ ∆(S) ⊂ X
c is continuous if for each x, y ∈ X and xnn≥1 ⊂ X,
lim xn = x ⇒ lim c(xn, y) = c(x, y)
X be a real linear vector space
for each S ∈ Ω, c(S) ∈ ∆(S) ⊂ X
c is continuous if for each x, y ∈ X and xnn≥1 ⊂ X,
lim xn = x ⇒ lim c(xn, y) = c(x, y)
Proposition
Proposition: If X contains the convex hull of three non-collinearpoints, then there is no continuous rcr over X which satisfies SPI
List Rationality and Two-stage choiceprocedures
List Rationality vs Shortlisting (M & M’06)
P1 is transitive
P2 is a linear order
c is shortlisting if for each S, c(S) = max(S, P1)
List Rationality vs Shortlisting (M & M’06)
P1 is transitive
P2 is a linear order
c is shortlisting if for each S, c(S) = max(S, P1)
List Rationality vs Shortlisting (M & M’06)
P1 is transitive
P2 is a linear order
c is shortlisting if for each S, c(S) = max(S, P1)
List Rationality vs Shortlisting
P1 is transitive
P2 is a linear order
c is shortlisting if for each S, c(S) = max(max(S, P1), P2)
List Rationality vs Shortlisting
P1 is transitive
P2 is a linear order
c is shortlisting if for each S, c(S) = max(max(S, P1), P2)
List Rationality vs Shortlisting
P1 is transitive
P2 is a linear order
c is shortlisting if for each S, c(S) = max(max(S, P1), P2)
equivalently;
c(S) is the solution to the problem:
maxx∈S u(x)
subject to
there is no y ∈ S s.t y P1 x
List Rationality vs Shortlisting
P1 is transitive
P2 is a linear order
c is shortlisting if for each S, c(S) = max(max(S, P1), P2)
equivalently;
c(S) is the solution to the problem:
maxx∈S u(x)
subject to
there is no y ∈ S s.t y P1 x
For a given choice function c, and for each distinct x, y ∈ X, x isrelated to y (x Rc y) if for some S ∈ Ω, we have either
(i) x = c(S ∪ y) and x 6= c(S) or
(ii) y = c(S ∪ x) and x = c(x, y).
Proposition
Proposition: A choice function c is shortlisting iff Rc is acyclic.
Note that:
x Ric y ⇒ x Fc y
x Riic y ⇒ y Fc x.
For a given choice function c, and for each distinct x, y ∈ X, x isrelated to y (x Rc y) if for some S ∈ Ω, we have either
(i) x = c(S ∪ y) and x 6= c(S) or
(ii) y = c(S ∪ x) and x = c(x, y).
Proposition
Proposition: A choice function c is shortlisting iff Rc is acyclic.
Note that:
x Ric y ⇒ x Fc y
x Riic y ⇒ y Fc x.
For a given choice function c, and for each distinct x, y ∈ X, x isrelated to y (x Rc y) if for some S ∈ Ω, we have either
(i) x = c(S ∪ y) and x 6= c(S) or
(ii) y = c(S ∪ x) and x = c(x, y).
Proposition
Proposition: A choice function c is shortlisting iff Rc is acyclic.
Note that:
x Ric y ⇒ x Fc y
x Riic y ⇒ y Fc x.
conclusion
conclusion
Choice with limited attention (MNO 2012)
This choice procedure has two primitives
an attention filter Γ and a welfare preference
For each choice set
S, c(S) = maxΓ(S), where Γ is s.t. for each choice set S
and z 6∈ Γ(S), Γ(S \ z) = Γ(S)
LRC ⊆ CLA
Choice with limited attention (MNO 2012)
This choice procedure has two primitives
an attention filter Γ and a welfare preference For each choice set
S, c(S) = maxΓ(S),
where Γ is s.t. for each choice set S
and z 6∈ Γ(S), Γ(S \ z) = Γ(S)
LRC ⊆ CLA
Choice with limited attention (MNO 2012)
This choice procedure has two primitives
an attention filter Γ and a welfare preference For each choice set
S, c(S) = maxΓ(S), where Γ is s.t. for each choice set S
and z 6∈ Γ(S), Γ(S \ z) = Γ(S)
LRC ⊆ CLA
Choice with limited attention (MNO 2012)
This choice procedure has two primitives
an attention filter Γ and a welfare preference For each choice set
S, c(S) = maxΓ(S), where Γ is s.t. for each choice set S
and z 6∈ Γ(S), Γ(S \ z) = Γ(S)
LRC ⊆ CLA
a related experimental study
Liu and Simonson (2005) compared the behavior of subjects whoare asked to make a choice from a set of ten products according totwo procedures.
people were presented all ten product offers together and areasked to indicate the one they are most interested in.
then, they are asked whether they would like to purchase theselected product.
in the second procedure, subjects are asked to make a choicefrom the set of ten products according to the LRC procedureby following an exogenously specified list.
once the last item in the list is considered, participants areasked to decide whether they want to purchase the winner.
They observe that subjects guided by LRC procedure are morelikely to purchase their selected product (%45) than those inthe simultaneous condition (%34)
a related experimental study
Liu and Simonson (2005) compared the behavior of subjects whoare asked to make a choice from a set of ten products according totwo procedures.
people were presented all ten product offers together and areasked to indicate the one they are most interested in.
then, they are asked whether they would like to purchase theselected product.
in the second procedure, subjects are asked to make a choicefrom the set of ten products according to the LRC procedureby following an exogenously specified list.
once the last item in the list is considered, participants areasked to decide whether they want to purchase the winner.
They observe that subjects guided by LRC procedure are morelikely to purchase their selected product (%45) than those inthe simultaneous condition (%34)
a related experimental study
Liu and Simonson (2005) compared the behavior of subjects whoare asked to make a choice from a set of ten products according totwo procedures.
people were presented all ten product offers together and areasked to indicate the one they are most interested in.
then, they are asked whether they would like to purchase theselected product.
in the second procedure, subjects are asked to make a choicefrom the set of ten products according to the LRC procedureby following an exogenously specified list.
once the last item in the list is considered, participants areasked to decide whether they want to purchase the winner.
They observe that subjects guided by LRC procedure are morelikely to purchase their selected product (%45) than those inthe simultaneous condition (%34)
a related experimental study
Liu and Simonson (2005) compared the behavior of subjects whoare asked to make a choice from a set of ten products according totwo procedures.
people were presented all ten product offers together and areasked to indicate the one they are most interested in.
then, they are asked whether they would like to purchase theselected product.
in the second procedure, subjects are asked to make a choicefrom the set of ten products according to the LRC procedureby following an exogenously specified list.
once the last item in the list is considered, participants areasked to decide whether they want to purchase the winner.
They observe that subjects guided by LRC procedure are morelikely to purchase their selected product (%45) than those inthe simultaneous condition (%34)
a related experimental study
Liu and Simonson (2005) compared the behavior of subjects whoare asked to make a choice from a set of ten products according totwo procedures.
people were presented all ten product offers together and areasked to indicate the one they are most interested in.
then, they are asked whether they would like to purchase theselected product.
in the second procedure, subjects are asked to make a choicefrom the set of ten products according to the LRC procedureby following an exogenously specified list.
once the last item in the list is considered, participants areasked to decide whether they want to purchase the winner.
They observe that subjects guided by LRC procedure are morelikely to purchase their selected product (%45) than those inthe simultaneous condition (%34)
a related experimental study
Liu and Simonson (2005) compared the behavior of subjects whoare asked to make a choice from a set of ten products according totwo procedures.
people were presented all ten product offers together and areasked to indicate the one they are most interested in.
then, they are asked whether they would like to purchase theselected product.
in the second procedure, subjects are asked to make a choicefrom the set of ten products according to the LRC procedureby following an exogenously specified list.
once the last item in the list is considered, participants areasked to decide whether they want to purchase the winner.
They observe that subjects guided by LRC procedure are morelikely to purchase their selected product (%45) than those inthe simultaneous condition (%34)