rivista italiana di economia demografia e statistica - Sieds

34
Volume LXIII nn. 3-4 – Luglio-Dicembre 2009
In a similar way we eliminate education and demographic characteristics from
the income generating model, obtaining the predicted values netted out,
respectively, of the two groups effects. We compute the Gini indexes, ˆ and
G
Occ , Dem
G ˆ
Occ , Edu
, to obtain education and demographic characteristics contributions as
follows
1 ˆ ˆ
1
C G − G , C Gˆ
− Gˆ
. (5)
Edu
=
TOT Occ,
Dem
Dem
=
TOT Occ,
Edu
The second round estimates the contribution of factors dropped two groups at a
time from equation (2). The idea is to isolate the effect of one group starting with
the Gini indexes where only two grops are considered. From the first round we
have G ˆ Edu , Dem
, G ˆ Occ , Dem and G ˆ
Occ , Edu , the total inequality given by the groups
couples. If the aim is to obtain the occupation contribution, we can start with
G ˆ but also Occ , Dem
G ˆ
Occ ,
, so eliminating, respectively, the effect of demographic
Edu
characteristics or education. The contribution of occupation in the second round
will be the average between
C and
21
Occ
C
22
Occ
. We can estimate the contribution of
the other two groups using the same procedure.
If we continue with the logic of group elimination, the groups contribution at
the final round will be derived by the difference between, respectively, Ĝ ,
Ĝ
Edu
, Ĝ
Dem
and the Gini index computed using the equation
Yˆ
= βˆ
+ βˆ
0 4Z.
(6)
Finally the contributions at each round are averaged across all rounds to obtain
the total marginal effect of each group on total inequality.
4. Variables
The dependent variable used in the analysis is the logarithm of the households
expenditure as proxy for the income. We considered, as determinants for the
households expenditure, groups of variables related to occupation, education and
demographic characteristics. Although we include in the income generating
equation several explanatory variables, we will decompose inequality only into
Occ