J. Angle, HOW THE GAMMA LAW OF INCOME-DISTRIBUTION APPEARS INVARIANT UNDER AGGREGATION, The Journal of mathematical sociology, 21(4), 1996, pp. 325-358
Citations number
15
Categorie Soggetti
Sociology,"Social Sciences, Mathematical Methods","Mathematical, Methods, Social Sciences","Mathematics, Miscellaneous
The Gamma Law of Income Distribution appears to be a scientific law be
cause the gamma pdf 1) fits the range of shapes seen in income distrib
utions, 2) is parsimonious, 3) appears to be scale invariant, i.e., to
show invariance under population aggregation, and 4) the gamma pdf's
shape parameter provides a convenient descriptor of the range of shape
s seen in income distributions, allowing the apparent invariance betwe
en education and the shape of the income distribution to be simply des
cribed. The Gamma Law of Income Distribution cannot, however, be a sci
entific law because it is not scale invariant. An unconditional distri
bution of income isa mixture, i.e., the weighted sum, of variously sha
ped income distributions. People at different education levels have di
fferently shaped income distributions. These distributions are well fi
tted by gamma pdfs making the corresponding unconditional distribution
a gamma shape mixture. A gamma shape mixture is not in general a gamm
a pdf. Aggregating the income distributions of population segments tog
ether can give rise to gamma shape mixtures. Thus the Gamma Law is not
scale invariant. However, under certain conditions a gamma shape mixt
ure can be hard to distinguish from GAM(alpha,lambda), the gamma pdf
whose shape parameter is alpha, the weighted average of the alpha(i)'
s, the shape parameters of the component gamma pdfs of the mixture. GA
M(alpha,lambda) has the same mean as the shape mixture. These conditi
ons allow the Gamma Law of Income Distribution to appear to be scale i
nvariant. These conditions occur in geographically defined populations
in the contemporary U.S. They are 1) the distribution of income condi
tioned on education is itself gamma distributed, 2) is invariant under
aggregation, 3) most of the population has attained an education whos
e corresponding income distribution is fitted by GAM(alpha(i),lambda)
where alpha(i) > 1, 4) there is a close relationship between the shape
of the income distribution and education, and 5) the distribution of
people over education is approximately symmetric, unimodal, and peaked
at its mode. The Gamma Law of (unconditional) Income Distribution app
ears to work because a Gamma Law of Income Conditioned on Education ex
ists.