Wald Test

[Ahmed Rashad]

Dear group,
So if am running Wald test looking for interaction terms between variable of interest and time dummies with the purpose of conducting trend analysis. The dataset is pooled dataset from several surveys at several points of time (years 2000, 2005, 2008). However, am mainly concerned with interactions terms between variable of interest and year of survey dummy over time (across several rounds of survey). The individual estimates of interactions are insignificant while main effects are significant. The Wald test gives significant result, is it ok to rely on wald test over individual estimates. Given that am intested does the Wald test should include the main effect or only interaction? Any help?
my commands are
logistic m2a i.v013 b(last).v190##timedum2 b(last).v190##timedum3 b(last).v190##timedum1,cluster(v001r)
testparm b(last).v190##timedum2 b(last).v190##timedum3 b(last).v190##timedum1

[Nick Cox]

Cross-posted at http://stats.stackexchange.com/quest...dual-estimates

Please see the FAQ for our policy on cross-posting which is that you should tell us where else you posted a question.

[Richard Williams]

The Wald test is testing the null hypotheses that all of the parameters being tested equal zero versus the alternative that at least one does not. So, a significant Wald test tells you somewhere, someplace, somehow, a parameter significantly differs from zero. It doesn't tell you which one, nor does it tell you how many significantly differ from zero. It could be one of them, it could be all of them, or it could be somewhere in between. It looks like you are testing almost every coefficient in the model. You might want to break it up into blocks, e.g. Test only the interaction terms.

[sikacity04]

please mr williams i have conducted an oparallel test but all the five test reported insignificant what should i do. am confused

[Richard Williams]

Posting your code and results may make it possible to help you. Be sure to use code tags (see point 12 of the FAQ). Also, if you are running logit models it makes no sense to run oparallel, which is for ordinal models.

[Richard Williams]

I just noticed sikacity04 was not the original poster. If this is an unrelated problem s/he should start a new thread about it and provide much more information.

[Marcos Almeida]

Also, according to the recommendations of this forum, "ahmed890" and "sikacity04" could change their "code names" into name and family name.

Please hit the "contact us" botton below, and ask for this.