ABSTRACT
L f xk i= ( ) (3.1) where Lk represents the possible answer that an individual might give, depending on their level of agreement with a proposed statement, using a k-point Likert scale, and xi represents the individual’s set of influential variables. Because opinions are expressed in the form of a discrete ordered variable, it is not advisable to use a simple regression model with an ordinary least squares procedure to determine and quantify the type of relationship with the independent variables (Borooah, 2002; Greene, 2003; O’Connell, 2006), since the difference between the first two alternatives (1 and 2, for example) would be deemed the same as the difference between subsequent alternatives (3 and 4, for example) when in fact the value that each alternative takes simply indicates its sequential order. Nevertheless, using ordered logit (or probit) models it is possible to construct models that link the observed outcome to the values of determining variables (Borooah, 2002). The initial hypothesis that is implicit in an ordered model is the existence of a latent variable of opinion that is a linear function of the set of explanatory variables. Thus, analytically,
OP xi= +β ε (3.2)
where OP is the latent variable of opinion, xi are i influential variables related to the individual, β is an i-dimensional vector to be determined and ε is the random error. Thus, the link between Lk and OP can be established through the following rule:
L
P OP P OP
k =
≤ < ≤
if if if
( )
( )
γ γ γ
...
