| Summary: | 碩士 === 國立中央大學 === 產業經濟研究所 === 85 === Qualitative response models have been used in a variety of
situations in applied econometrics, transportation choice,
public strategy, financial decision and prediction, consumers''
selection, social and behavior science, etc. for a long history.
These models are developed more widely and deeply in
econometrics in the recent years. By far the model specification
which is used most often is the multinomial logit (MNL) model.
Yet it is widely known that a potentially important drawback of
the MNL model is the Independence from irrelevant alternatives
(IIA) property which means that all qualitative alternatives are
neither similar nor substitutable. This is a very strict
constraint. In this paper we provide three computationally
convenient specification tests for the MNL model and apply Monte
Carlo method to simulate the whole process. The first test is
HM(Hausman and McFadden,1984) test, the second is SH(Small and
Hsiao, 1985) test, and the third is MTT(McFadden, Train, & Tye)
test. One more test which we called it MTTB is made for being
compared to SH and MTT tests. The basic idea for all tests is to
test the reverse implication of the IIA property. And the tests''
type except the HM test is built on likelihood ratio test
procedure. Therefore, for the sake of a consistent comparison
base we use the Nested Logit Model to be our true model. Our
results are interesting. We examine these tests by multiple
layers and by multiple viewpoints, especially on test size,
stability, and power.In general, HM test is excellent both in
size and power, MTT is the second, which power is not bad but
its size is asymptotically accepting the null hypothesis. SH
test has the worst power among the three tests, although it has
exact size. Other results includes the problem of Small-Hsiao
corrected MTT version test, the applicability of HM test, and a
new method we find to regenerate the dependent variable can
upsurge MTT test.
|
|---|
