CHILD LABOUR IN RURAL INDIA
75
Individual and household characteristics are represented by the age (age and
age squared) and sex (taking value 1 for a girl, 0 for a boy) of the child, the
mother's age, the level of education of the child's father and mother
(
4
respectively, father's education i and mother's education i, where i takes value
for completed high school or higher, 3 for middle school, 2 for primary, 1 for
less than primary), total number of household members (household size),
number of preschool children (siblings 0±5), and number of school-age children
1
0
siblings 6±16). We also use dummies for the religion (Hindu, Muslim etc.) of
(
the household head.
1
1
Health policies and local environmental conditions are proxied by the village
level aggregate survival rate to age 6 (village survival rate). By using this as a
regressor, we are in eect saying that parental actions (and household char-
acteristics) cause a dispersion of individual probabilities of survival around the
1
2
village-level mean. Because the mean of the village survival rates is around
0 percent, there is clearly plenty of scope for parents to improve the survival
9
chances of their own children. In view of the high degree of correlation among
survival rates to all ages, this variable is also a predictor of aggregate survival
rates to subsequent ages.
Educational policies are represented by the variable type of school, which takes
value 0 if there is no school in the village where the child lives, 1 if there is only a
primary school, 2 if there is a primary and a middle school, 3 if there is also a high
school. The presence of a school ready at hand constitutes a reduction in the
price of educational inputs. State dummies (Andhra Pradesh, Bihar etc.) allow
for other dierences of policy, as well as of climate, ethnic mix etc.
4.1. Time use
We estimated a multinomial logit model for the probability that a school-age
child will `work only' or `work and study', as against `study only' (the reference
state). Estimates are reported in table 7.
Girls are more likely to specialize fully in either work or education than to do
13
both, and are more likely to specialize in work than boys. The probability of
1
1
1
0
1
2
Where age structure is not signi®cant, we use the total number of children (siblings).
Caste (represented by the variable social) was also tried, but proved generally insigni®cant.
Ideally, we would want the information, up to the date of conception of each child, that helped
form parental expectations of these village means. Because that is not available we computed
village-level mean survival rates from individual data. If cross-village dierences have not changed
widely over time, our estimates should be a reasonable indicator of inter-village dierences in
parental expectations. Because, for some villages, there are only a few individual observations, we
also computed and tried district-level survival means to check the robustness of our results. Results
were unchanged.
1
3
A `bootstraps' explanation is provided in Cigno (1991; ch.5). Parents perceive the return to
educating girls as lower than the return to educating boys because they observe that women are less
likely to work, and thus to pro®t from their education, than men. But women work less than men
because they are less educated and, therefore, in the marital division of labour, have a comparative
advantage in specializing in housework.
Ó Blackwell Publishers Ltd 2002