- people have different preferences
- how should societal decisions be taken?
- navigate conflicts of preferences
- respecting individual preferences
- Examples:
- political decisions and elections
- a group of friends wants to go for drinks: how to aggregate the differing preferences over bars
- aggregating votes of several judges in sports (boxing, figure skating etc.)
- (expert) committees
- a family deciding where to spend the summer holiday
- …
- make ethical premises explicit
- derive solutions consistent with these premises
- normative (!)
- society (
$N>2$ people) has to choose one of 2 alternatives/candidates ($x$ and$y$ ) - assumption for simplicity: everyone has a strict preference over alternatives
- majority voting:
-
$x\succeq_S y$ if at least$N/2$ people prefer$x$ over$y$ -
$y\succeq_S x$ if at least$N/2$ people prefer$y$ over$x$
-
- what normative premises underlie this social welfare function?
A social welfare function is anonymous if the names of the agents do not matter, i.e. if a permutation of preferences across agents does not change the social preference.
A social welfare function is neutral if the names of the alternatives do not matter, i.e. the social preferences are reversed if we reverse the preferences of all agents.
A social welfare function is positively responsive if the following holds: if one alternative
- didn’t I claim that social choice starts with premises and then derives solutions?
If there are two alternatives, a social welfare function satisfies anonymity, neutrality and positive responsiveness if and only if it is majority voting.
- Anonymity: only number of people preferring alternative
$x$ over$y$ matters for$\succeq_S$ . - Neutrality: if
$N/2$ people prefer$x$ over$y$ and$N/2$ prefer$y$ over$x$ , then$x≈_S y$ . - Positive responsiveness: if more than
$N/2$ people prefer$x$ over$y$ ,$x\succ_S y$ and vice versa.
- How to generalize majority voting with more than 2 alternatives?
An alternative
A group of students want to tell the teacher their preferences over exam forms (open book, closed book, online exam). How to aggregate the preferences?
best | middle | worst | |
/ | < | ||
---|---|---|---|
Student 1 | ob | oe | cb |
Student 2 | oe | cb | ob |
Student 3 | cb | ob | oe |
Which alternative is Condorcet winner?
- finite set
$X=\{x_1,x_2,…,x_K\}$ of alternatives -
$N\geq 2$ agents, each has a complete and transitive preference relation over$X$
A social preference relation is a complete and transitive preference relation on the set
A social welfare function assigns to each profile of preferences
Are the following social welfare functions desirable?
- The preferences of agent 1 are the social preferences:
$\succeq_S (\succeq_1,\succeq_2,…,\succeq_N)=\succeq_1$ - Fixed social preference relation:
$\succeq_S (\succeq_1,\succeq_2,…,\succeq_N)=x_1\succ_S x_2\succ_S x_3\succ_S…\succ_S x_K$ - Borda Count:
- turn every agent’s preference order into points: the
$k$ most preferred alternative receives$k$ points - for every alternative, sum the points it gets from all agents
- order alternatives according to points (less points are better)
- turn every agent’s preference order into points: the
- judging in sports is similar to our problem
- aggregation of several judges’ rankings
- final 2002 Olympic figure skating competition
- Slutskaya is the last skater to perform
- at that moment: 1. Kwan, 2. Hughes, 3. …
- Slutskaya is doing well but not super and ends up second
- who came first? who came third?
- table contains the ranks that the 7 judges assign to the three skaters
Kwan | Hughes | Slutskaya | |
/ | < | ||
---|---|---|---|
judge 1 | 2 | 3 | 1 |
judge 2 | 2 | 3 | 1 |
judge 3 | 1 | 2 | 3 |
judge 4 | 1 | 2 | 3 |
judge 5 | 3 | 1 | 2 |
judge 6 | 3 | 1 | 2 |
judge 7 | 3 | 1 | 2 |
Points |
If
There is no individual
Take two profiles of preferences
Let there be at least 3 alternatives in
- no social welfare function satisfies even minimal criteria
- we have to give up even some of these minimal criteria if we want to proceed!
- some ways to proceed:
- pick only one alternative: no complete social ordering necessary
- leads to similar result
- domain restriction
- we implicitly assumed that all preference profiles were possible (in the definition “social welfare function”)
- more positive results if we can rule out certain preferences
- cardinal utility
- we only looked at orderings not at intensity of preference
- assuming that there is something like intensity of preferences and this intensity is comparable across agents helps to aggregate preferences but is a questionable assumption
- pick only one alternative: no complete social ordering necessary
- imagine alternatives are ordered on the real line:
$x_1 < x_2 < … < x_K$ - assumptions:
- common ordering of alternatives
- everyone has a most preferred alternative
- of two “too high” (or “too low”) alternatives, an agent prefers the one closer to his most preferred alternative
- for simplicity: odd number
$N$ of agents
- more precisely:
- each agent
$i$ has a most preferred alternative$x^*(i)∈\{x_1,x_2,…,x_K\}$ - if
$x_k,x_m>x^*(i)$ , then$x_k \succ_i x_m$ if and only if$x_k < x_m$ - if
$x_k,x_m < x^*(i)$ , then$x_k\succ_i x_m$ if and only if$x_k > x_m$
- each agent
- if we represent preferences by utility function, this function is “single peaked”
An agent
Note: a median agent always exists.
Let preferences be single peaked and
- Consider a pairwise majority vote between
$x^*(i)$ and$x_m > x^*(i)$ .\vspace*{1.5cm}
- Consider a pairwise majority vote between
$x^*(i)$ and$x_m < x^*(i)$ .\vspace*{1.5cm}
- consider pairwise majority voting between arbitrary alternatives, i.e. say
$x_k$ is socially preferred to$x_m$ if$x_k$ wins in a majority vote over$x_k$ and$x_m$
If preferences are single peaked, pairwise majority voting induces a social welfare function.
to show: resulting preferences are complete and transitive
- As
$N$ is odd and preferences are strict, pairwise majority voting yields a strict winner between any two alternatives.\linebreak$⇒$ social preference ordering resulting from pairwise majority voting is complete and strict. - Transitivity: let
$x_m\succ_S x_k$ and$x_k\succ_S x_l$ …
\vspace*{2cm}
Reminder:
A complete preference relation
- suppose we have 2 agents and
$x\succ_1 y$ while$y\succ_2 x$ - we choose utility functions for the two agents
-
$u_1(x)=3$ ,$u_1(y)=1$ -
$u_2(x)=0$ ,$u_2(y)=1$
-
- which alternative should society prefer?
- if we assign meaning to utility, social welfare function is not invariant to strictly monotone transformations
- allows to get around Arrow’s impossibility theorem
- problem: choice of specific agent utility functions implicitly makes normative judgments beyond our criteria
- for now:
- accept some given utility functions
$u$ - let welfare depend on the utilities of the agents and be represented by a function
$W:\Re^N→\Re$ that aggregates agent utilities into “welfare”- we abuse notation and call
$W$ also “social welfare function”
- we abuse notation and call
- what are reasonable choices for
$W$ ? what normative judgments are expressed by the choice of$W$ ?
- accept some given utility functions
Alternative
An alternative
Pareto dominating alternatives are socially preferred to the alternatives they dominate if social welfare function
note: weak Pareto criterion is satisfied if
- $WRawls$ satisfies weak Pareto criterion
- $WRawls$ is anonymous
- $WRawls$ is “utility level invariant”:
- social preferences remain the same if we transform all agent’s utility using the same strictly increasing transformation
- $WRawls$ satisfies “Hammond Equity”:
- take two utility vectors
$(\bar u_1,\bar u_2,…,\bar u_N)$ and$(\hat u_1,\hat u_2,…,\hat u_N)$ and suppose$\bar u_i=\hat u_i$ for all$i$ except$j$ and$k$ - suppose further
$\bar u_j<\hat u_j<\hat u_k<\bar u_k$ - Hammond equity states that then
$W(\hat u)\geq W(\bar u)$
- take two utility vectors
A continuous social welfare function
-
$≈$ Rawlsian welfare is equivalent to weak Pareto criterion + Hammond equity
see Jehle and Reny (2011), section 6.3.1
- most commonly used welfare function (sometimes with individual weights)
- $Wut$ respects Pareto efficiency
- $Wut$ is anonymous (not true if weights are used)
- $Wut$ is “utility-difference invariant”
- social preferences are the same if we transform all agents utility using the transformation
$ψ_i(u_i)=a_i+b u_i$ with$b>0$
- social preferences are the same if we transform all agents utility using the transformation
A continuous social welfare function
see Jehle and Reny (2011), section 6.3.2
- thought experiment
- you will be one of the agents in society
- you have to decide which alternative to choose
- you do not know which agent you are going to be
- some people have argued that whatever a “fair-minded” person would choose in this hypothetical situation is a good societal decision
- Harsanyi:
- my chance of being agent
$i$ is$1/N$ - my choice should maximizes the expected utility $∑i=1^N (1/N) u_i(x)$
-
$→$ utilitarian welfare
- my chance of being agent
- Rawls:
- I do not know who I am going to be and there is no basis for assigning probabilities.
- risk aversion implies maximizing the worst case utility
-
$→$ Rawlsian welfare
- Arrow:
- Rawls makes a mistake as he assumes not risk aversion but infinite risk aversion, i.e. risk aversion does not imply maximizing worst case utility.
- so far: preferences of all players are known
- problem: aggregation
- what if everyone knows his preferences privately?
- ask for preferences
- aggregate
- additional problem: gaming the system by misreporting preferences!
- result due to Gibbard and Satterthwaite:\linebreak If there are at least three alternatives and a social welfare function is (i) Pareto efficient and (ii) creates no gaming possibilities, then it is dictatorial.
- one example for manipulability
most preferred | middle preferred | least preferred | |
/ | < | ||
---|---|---|---|
Agent 1 | x | y | z |
Agent 2 | y | x | z |
Agent 3 | y | x | z |
Points |
Could agent 1 manipulate the social preference relation by misrepresenting his own preferences? Would he want to do so?
- to discuss such topics properly:\linebreak
extend decision and game theory to incomplete information
- that’s what we will do in the coming weeks!
bibliographystyle:chicago bibliography:/home/christoph/stuff/bibliography/references.bib