I have read the Arrow impossibility theorem in Foundations of Mathematical Economics(Michael Carter). It is just too difficult to understand.
So, does Arrow'theorem mean that there is always a dictator in our society,regardless of any system of election ?
Can you recommend me some books that delve deeper into this matter ?
I thank you very much for your answer.
Arrow's Theorem doesn't say that you will always have a dictator. It says if you wanted to preserve two other "nice" properties in a voting system (I'll say later what those are) of your design, you can't have them without also having a dictator (can't have your cake and eat it too sort of thing).
Suppose there are $n$ voters and a set of alternatives, $S$, that they are voting on. Alternatives can be whether to paint the city hall green, blue or purple, whether to choose Joe, Bill, Susie, or Alice for class president etc. The only restriction here for the theorem to hold is that you have to be voting on three or more alternatives. Now suppose each of the voters $i \in \{1, \ldots, n\}$ has a personal ranking over the alternatives, P_i (e.g., $P_i = (Susie, Bill, Joe, Alice)$). Assume that these personal rankings are strict, so there are no ties in preferences (if there are ties, break them randomly and theorem works). Now $F$ is a social welfare function if it maps $\Pi_iP \to P$, where $P$ is the set of all possible rankings of the alternatives (so if there are $k$ alternatives, all permutations of $(1, \dots,, k)$). So think of this social welfare function as taking in everyone's ballots and deciding based on its input what society's ranking of the outcomes will be.
Now Arrow's theorem says we can't have all three of the following properties of our social welfare function be true:
If $aP_ib$ $\forall i$, $aF(\Pi_iP_i)b$. ($aPb$ means under ranking $P$, $a$ holds a higher position. So this condition says if everyone likes $a$ better than $b$, the ranking that the social welfare function spits out will also respect that relative ranking of the two outcomes; seems natural, no?)
If $\Pi_iP_i$ is one set of strict rankings over the alternatives by the $n$ voters, and $\Pi_iR_i$ is another, and moreover, alternatives $a$ and $b$ have the same relative ranking under $R_i$ as they do under $S_i$, then $F(\Pi_iP_i)$ and $F(\Pi_iR_i)$ have the same relative rankings of $a$ and $b$.
There is no dictator:, i.e, $F$ is not a projection map. So the social choice function does not merely spit out the preference list submitted by the $ith$ voter for some $i$.