We consider social choice rules which select a lottery over outcomes f
or each profile of individual preferences. Agents are assumed to have
preferences over lotteries satisfying the axioms of expected utility.
We exhibit a large class of rules satisfying strategy-proofness. All t
hese rules are obtained by combining one of the following principles:
(1) start from a fixed subset of lotteries, and for each profile let o
ne fixed agent choose her preferred lottery from that subset (we call
them unilateral rules); or, (2) start from two fixed lotteries and a r
ule assigning weights to each of them depending on the coalition of ag
ents which prefer one of the two lotteries to the other; let the outco
me at each profile be the convex combination of these two given lotter
ies according to the weights which correspond to them at that profile
(these rules are called duples). All probabilistic mixtures (convex co
mbinations or integrals) of unilateral and duple rules satisfying some
additional and natural requirements are strategy-proof. Because we ar
e facing a wide class of procedures, we investigate the possibility of
designing some which are not only strategy-proof but also continuous
or even smooth in their responses to changes in preferences. Smoothnes
s requirements are not only attractive per se, but they can also be ex
pected to help in telling apart different types of rules, Notice that
unilateral rules can be very smooth, while no duple can even be contin
uous. Yet, continuity can be regained by combining a continuum of dupl
es, we provide an example of a continuous strategy-proof probabilistic
rule which is an integral of duples. However, there is a limit as to
how smooth a rule can be without resorting to unilateral schemes. We p
rove that any strategy-proof probabilistic function of class C-2 must
indeed be also a convex combination of unilateral schemes. (C) 1998 El
sevier Science B.V.