The book culminated in what Arrow called the "General Possibility Theorem," thereafter better known as Arrow's (impossibility) theorem. The theorem shows that, absent restrictions on either individual preferences or neutrality of the constitution as to feasible alternatives, there exists no social choice rule that satisfies a set of seemingly plausible requirements. The result is a generalization and extension of the voting paradox, which shows that majority voting may fail to yield a stable outcome.
In the simplest case of the voting paradox, there are 3 candidates, A, B, and C, and 3 voters with preferences listed in decreasing order as follows.

The analysis uses ordinal rankings of individual choice to represent behavioral patterns. Cardinal measures of individual utility and, a fortiori, interpersonal comparisons of utility are avoided on grounds that such measures are unnecessary to represent behavior and depend on mutually incompatible value judgments (p. 9).
Following Abram Bergson, whose formulation of a social welfare function launched ordinalist welfare economics, Arrow avoids locating a social good as independent of individual values. Rather, social values inhere in actions from socialdecision rules (hypostatized as constitutional conditions) using individual values as input. Then 'social values' means "nothing more than social choices." (p.106)
Topics implicated along the way include game theory, the compensation principle in welfare economics, extended sympathy, Liebniz's principle of the identity of indiscernibles, logrolling, and similarity of social judgments through singlepeaked preferences, Kant’s categorical imperative, or the decision process.
Example: Three voters {1,2,3} and three states {x,y,z}. Given the three states, there are 13 logically possible orderings (allowing for ties).* Since each of the individuals may hold any of the orderings, there are 13*13*13 = 2197 possible "votes" (sets of orderings). A welldefined socialdecision rule selects the social state (or states, in case of tie) corresponding to each of these "votes." * Namely, from top to bottom for each possible ranking and with 'T's indexing ties:
x y z x y T z (x T y z repeats y x z y T x z y T x z, so is omitted, etc.) z x y z x T y x z y x T z y y z x y z T x z y x z T y x x T y T z 
The comprehensive character of commodities, the set of social states, and the set of orderings was noted by early reviewers.
The 2 properties that define any ordering of the set of objects in question (all social states here) are:
A standard indifferencecurve map for an individual has these properties and so is an ordering. Each ray from the origin ranks (conceivable) commodity bundles from least preferred on up (no ties in the ranking). Each indifference curve ranks commodity bundles as equally preferred (all ties in the ranking). 
The denotations of these 3 "ballot" options are respectively:
It is convenient for deriving implications to compact the first 2 of these options on the ballot to 1, an "at least as good as" relation, denoted R:
The above 2 properties of an ordering are then axiomatized as: connectedness: For all (the objects of choice in the set) x and y, either x R y or y R x. transitivity: For all x, y, and z, x R y and y R z imply x R z. Thus, alternation ('or') and conjunction ('and') of R relations represent both the properties of an ordering for all the objects of choice. The I and P relations are then defined as: x I y: x R y and y R x (x as good as y means x at least as good as y and vice versa). x P y: not y R x (y R x includes 1 of 2 options. Negating that option leaves only x P y, the third of the original 3 options, on the ballot.) From this, conjunction ('and') and negation ('not') of mere pairwise R relations can (also) represent all the properties of an ordering for all the objects of choice. Hence, the following shorthand. 
If voter i changes orderings, primes distinguish the first and second, say $R\_i$ compared to $R\_i$' . The same notation can apply for 2 different hypothetical orderings of the same voter.
The interest of the book is in amalgamating sets of orderings. This is accomplished through a 'constitution'.
A social ordering of a constitution is denoted R. (Context or a subscript distinguishes a voter ordering R from the same symbol for a social ordering.)
For any 2 social states x and y of a given social ordering R:
x P y is "social preference" of x over y (x is selected over y by the rule).
x I y is "social indifference" between x and y (both are ranked the same by the rule).
x R y is either "social preference" of x over y or "social indifference" between x and y (x is ranked least as good as y by the rule).
A social ordering applies to each ordering in the set of orderings (hence the "social" part and the associated amalgamation). This is so regardless of (dis)similarity between the social ordering and any or all the orderings in the set. But Arrow places the constitution in the context of ordinalist welfare economics, which attempts to aggregate different tastes in a coherent, plausible way.
The social ordering for a given set of orderings as to the set of social states is analogous to an indifferencecurve map for an individual as to the set of commodity bundles. There is no necessary interpretation from this that "society" is just a big voter. Still, the relation of a set of voter orderings to the outcome of the voting rule, whether a social ordering or not, is a focus of the book. 
Arrow (pp. 15, 2628) shows how to go from the social ordering R for a given set of orderings to a particular 'social choice' by specifying:
The social ordering R then selects the topranked social state(s) from the subset as the social choice set.
This is a generalization from consumer demand theory with perfect competition on the buyer's side. S corresponds to the set of commodity bundles on or inside the budget constraint for an individual. The consumer's top choice is at the highest indifference curve on the budget constraint. 
The social choice function is denoted C(S). Consider an environment that has just 2 social states, x and y: C(S) = C([x, y]). Suppose x is the only topranked social state. Then C([x, y]) = {x}, the social choice set. If x and y are instead tied, C([x, y]) = {x, y}. Formally (p. 15), C(S) is the set of all x in S such that, for all y in S, x R y ("x is at least as good as y").
The next section invokes the following. Let R and R' stand for social orderings of the constitution corresponding to any 2 sets of orderings. If R and R' for the same environment S map to the same social choice(s), the relation of the identical social choices for R and R' is represented as: C(S) = C'(S).
Arrow proposes conditions to constrain the social ordering(s) of the constitution (pp. 9697, 2531). The conditions, presented below, can be interpreted as general, practically necessary, or apparently reasonable.
Each voter is permitted by the constitution to rank the set of social states in any order, though with only one ordering per voter for a given set of orderings.
Arrow refers to a constitution satisfying this condition as collective rationality. It can be compared to the rationality of a voter ordering. But the prescription of collective rationality, which Arrow proposes, is distinct from the descriptive use of a voter ordering, which he deploys. Hence, his denial at the end of the book that collective rationality is "merely an illegitimate transfer from the individual to society." (p. 120) 
Condition I: Let $R\_1$, ..., $R\_n$ and $R\_1$' , ..., $R\_n$' be 2 sets of orderings in the constitution. Let S be any subset of hypothetically available social states from the set of social states. For each voter i and for each pair of x and y in S, let x $R\_i$ y if and only if x $R\_i$' y. Then the social choice functions for the 2 respective sets of orderings map to an identical social choice set: C(S) = C'(S). 
This identical mapping happens even with differences in rankings of any voter between the two sets of orderings outside that subset of social states. Consider a hypothetical “runoff vote” between say only 2 available social states. The social choice is associated with the sets of rankings for that subset, not with rankings of unavailable social states beyond the subset. Thus, that social choice for the subset is unaffected by say a change in orderings only beyond the subset. 
Condition P: For any x and y in the set of social states, if, for every voter i, x $P\_i$ y, then x P y. 
As Sen (ch. 3.4) suggests, Pareto unanimity (with universal domain) overrides any social convention selecting some social state that is otherwise voteimmune. 
An alternate statement of the theorem adds the following condition to the above:
Condition D: There is no voter i in {1, ..., n} such that for every set of orderings in the domain of the constitution and every pair of social states x and y, x $P\_i$ y implies x P y. 
Each voter has an ordering (by attribution). Yet a set of orderings used as an argument of the voting rule does not carry over to a social ordering, with a corresponding loss of social adaptivity and constitutional generality, whether descriptive or prescriptive. 
# Pareto is stronger than necessary in the proof of the theorem that follows above. Arrow (1951) uses 2 other conditions, instead of Pareto, that with (2) above imply Pareto (Arrow, 1987, p. 126):
Arrow (1951, p.26) describes social welfare here as at least not negatively related to individual preferences.
Under imposition, for every set of orderings in the domain, the social ranking is only x R y. The vote makes no difference to the outcome.
For the special case of all preferring y over x, the vote still makes no difference. If the invariant social ranking applies to only one pair of distinct social states, the constitution would violate N. In this respect, as a representation of excluding convention, N is thorough. 
The proof is in two parts (Arrow, 1963, pp. 97100). The first part considers the hypothetical case of some one voter's ordering that prevails ('is decisive') as to the social choice for some pair of social states no matter what that voter's preference for the pair, despite all other voters opposing. It is shown that, for a constitution satisfying Unrestricted Domain, Pareto and Independence, that voter's ordering would prevail for every pair of social states, no matter what the orderings of others. So, the voter would be a Dictator. Thus, Nondictatorship requires postulating that no one would so prevail for even one pair of social states.
The second part considers more generally a set of voters that would prevail for some pair of social states, despite all other voters (if any) preferring otherwise. Pareto and Unrestricted Domain for a constitution imply that such a set would at least include the entire set of voters. By Nondictatorship, the set must have at least 2 voters. Among all such sets, postulate a set such that no other set is smaller. Such a set can be constructed with Unrestricted Domain and an adaptation of the voting paradox to imply a still smaller set. This contradicts the postulate and so proves the theorem.
The set of conditions across different possible votes of values refined welfare economics and differentiated Arrow's constitution from the preArrow social welfare function. Thus, one dictator across every possible vote on social alternatives eliminates any single nonvacuous ordering as the social ordering. It also makes redundant an agent or official intent on implementing the values of others in the society through the constitution. The remaining alternative, nondictatorship, excludes a preArrow social welfare function as a consistent voting machine. The result generalizes and deepens the voting paradox to any voting rule satisfying the conditions, however complex or comprehensive.
The 1963 edition includes an additional chapter with a simpler proof of Arrow's Theorem. It also elaborates on advantages of the conditions and cites studies of Riker (1958) and Dahl (1956, pp. 3941) that as an empirical matter intransitivity of the voting mechanism may produce unsatisfactory inaction or majority opposition. These support Arrow's characterization of a constitution across possible votes (that is, collective rationality) as "an important attribute of a genuinely democratic system capable of full adaptation to varying environments." (p. 120)
The theorem might seem to have unravelled a skein of behaviorbased socialethical theory from Adam Smith and Bentham on. But Arrow himself expresses hope at the end of his Nobel prize lecture that others might take his result "as a challenge rather than as a discouraging barrier."
The large subsequent literature has included reformulation to extend, weaken, or replace the conditions and derive implications. In this respect Arrow's framework has been an instrument for generalizing voting theory and critically evaluating and broadening economic policy and social choice theory.