116 Mohamad Fahmi
option. Moreover, students who choose to attend one of these upper
secondary schools are assumed to graduate. This assumption is to simplify
the view that an individual is able to choose a future level of education
with no uncertainty about actually completing the current level.^1 I assume
that the individual will choose alternative j that maximizes his or her
utility. The individual utility function, U is
(1) Uij = uj (Xi, εi) j = 1,...,J,
where Xi and εi are vectors of observable and unobservable variables,
respectively, and j is upper secondary school alternatives.
I assume individuals will choose an alternative, j, if it maximizes their
utility. Then the expression of a decision process of a student who chooses
to attend one particular school is
(2) Si = k iff Uik > max Uij ∀j ≠ k
From Equations 1 and 2 the probability of individual i choosing to attend
school type j may be obtained by estimating a multinomial discrete choice
model. The model may be written as
(3) Pr(Si = j) =
exp (Uij)
1 + ΣJj=1 exp(Uij)I will follow the estimation of the multinomial logit model above to not
only investigate the determinants of the school choice decision by an
individual, but also to construct selection correction variables, λ, that
are included in the binary higher education participation choice model
to detect the problem of selection bias. As in a linear regression model,
selection bias is also a problem in the logit or probit model. According to
Dubin and Rivers (1989) the treatment of the problem of selection bias in
binary choice models is more difficult computationally. The problem of
selection bias in upper secondary schools in Indonesia may appear since
school choice is not random and may be influenced by unobservable
characteristics.
Another potential factor that affects the decision is the school’s selection
criteria. According to Bedi and Garg (2000), this condition implies that
school choice may be endogenous and students with higher ability may