* This paper is the product of a study of the causes of government expenditure funded by the Lilly Foundation. It was written while I was a member of the UCLA economics department. I would like to thank my colleagues in that department, and especially Harold Demsetz, for suggestions and comments.

^{1}The idea that the main determinant
of present expenditure is past expenditure is discussed in Sharkansky
(1967), criticized in Harlow (1968) and defended in Sharkansky
(1968). Further justifications for theories in which governments do
not behave like rational maximizers are in Buchanan and Tollison
(1972), Buchanan and Tullock (1962), Olson (1965) and Wildavsky
(1964).

^{2}The idea that government expenditure
might be affected by the elasticity of the tax base is not original
with me; see Craig and Heins (1980), Oates (1975), and references
therein. Both Craig and Heins and Oates limited themselves to the
relation between recent spending growth and roughly current measures
of elasticity, so their results show only that it takes states at
least a few years to adjust their tax levels to increases in income.

^{3}Bahl and Saunder (1965,1966
1,2),Bolton (1969),Gramlich (1969 1,2),Gramlich and Galper (1973),
Harlow (1967), Kurnow (1963), Oates (1968), O'Brien (1971), Ohls and
Wales (1972), Osman (1966,1968), Sacks and Harris (1964), Smith
(1968), Wilde (1968).

^{4}This point is discussed in Pogue and
Sgontz (1968).

^{5} It may be objected that if federal
aid really is both an effect and a cause of state revenue, then
ordinary least squares is an inappropriate procedure whether or not
federal aid is included in the regression. But in order to do a two
stage least squares or equivalent I would require an instrumental
variable--something which affects federal aid but which I know, a
priori, is unrelated to revenue except through its effect on federal
aid. It is by no means obvious that such a variable exists. In any
case, while ordinary least squares may give biased estimates for the
coefficients, the coefficients are not what I am really interested
in. If my null hypothesis is true (i.e. if the elasticity variable
does not affect current revenue) then whatever the relation between
federal aid and revenue it should not affect the results of the
regression that omits federal aid--unless federal aid is itself
related to past elasticity, a possibility I discuss. Hence the
reported t values correctly describe the improbability that, if the
null hypothesis is true, I would have gotten my results by chance.

^{6}My estimate of E, the elasticity of
all other taxes, is simply the estimate of D x E in equation 5
divided by the estimate of D