where k is the number of reasons to choose from by the free-riders. With a normalization that
m = 0, which reflects the fact that only relative probabilities can be identified, with respect to
the base alternative m (Jones, 2000).
The dependent variable takes one of four values depending on the reason given for not being a
regular blood donor: Yi = 1 if the individual reports that “others already do it”; Yi = 2 if the
individual reports “I have an aversion to needles”; Yi = 3 if the individual reports “I have not
thought about it”; and Yi = 4 if the individual reports “I have no reason” or “I do not know”.
One reason is recorded for each free-riding respondent. Therefore, the MNLM would identify
the probability of being of a particular free-rider category relative to the reference outcome
(“others already do it”). We consider the same set of covariates (z) as in the previous freeriding
model.
The MNLM assumes independence of irrelevant alternatives (IIA). That is, if we consider the
ratio of the probability of two different reasons to free-ride k and l, IIA implies that the
relative probability depends only on the characteristics of the two reasons and not on any of
the other reasons (i.e. if a new alternative is introduced, all of the absolute probabilities will
be reduced proportionally so that the relative probabilities between k and l remain unaffected).
We test for its appropriateness using the Hausman and Small-Hsiao tests, by first estimating
the model with all of the four reasons for free-riding, and subsequently re-estimating it by
dropping one of the reasons. This is then followed by the tests for IIA (see Scott and Freese,
2001). If IIA is violated, an alternative model should be considered (such as the nested
multinomial logit or the multinomial probit model) that relax the IIA property. In addition, a
Wald test is conducted to explore whether or not combining some of the response categories
would make the model more efficient.