J. Aczel et Jc. Falmagne, Consistency of monomial and difference representations of functions arising from empirical phenomena, J MATH ANAL, 234(2), 1999, pp. 632-659
Choice probabilities in the behavioral sciences are often analyzed from the
standpoint of a difference representation such as P(x, x, y) = F[u(x, x) -
g(y)]. Here, x and y are real, positive vector variables, x is a positive
real variable, P(x, x, y) is the probability of choosing alternative (x, x)
over alternative y, and ii, g and F are real valued, continuous functions,
strictly increasing in all arguments. In some situations (e.g. in psychoph
ysics), the researchers are more interested in the functions u and g than i
n the function F. In such cases, they investigate the choice phenomenon by
estimating empirically the value x such that P(x, x,y)= rho, for some value
s of rho, and for many values of the variables involved in x and y. In othe
r words, they study the function xi satisfying xi(x,y; rho) - x double left
right arrow P(x, x,y) = rho. A reasonable model to consider for the functi
on xi is offered by the monomial representation
xi(x, y; rho) = Pi(i = 1)(n - 1) x(i)(-eta i(rho)) Pi(j = 1)(m) y(j)(zeta j
(rho))C(rho),
in which the eta(i)'s, the zeta(j)'s and C are functions of rho. In this pa
per we investigate the consistency of these difference and monomial represe
ntations. The main result is that, under some background conditions, if bot
h the difference and the monomial representations are assumed, then: (i) al
l functions eta(i) (1 less than or equal to i less than or equal to n - 1)
must be constant; (ii) if one of the functions zeta(j) is nonconstant, then
all of them must be of the form zeta(j)(rho) = theta(j)exp[delta F-1(rho)]
, for some constants theta(j) > 0 (1 less than or equal to j less than or e
qual to m) and delta not equal 0, where F-1 is the inverse of the function
F of the difference representation. Surprisingly, F can be chosen rather ar
bitrarily. (C) 1999 Academic Press.