Doing a Joint test in xtmixed and interpreting the difference between output from contrast and xtmixed

[Puriya Daniel Yazdanfard]

Dear STATALIST

In my data I wish to evaluate if there is any difference between a variaty of hearing thresholds on different frequencies and gender.
My question is how do I do a joint test instead of looking at variables individually? I get three different results when using Xtmixed, contrast sex#freq and contrast sex@freq.
What are the differences in these tests?
Thanks

[Joseph Coveney]

The first tests difference in z between the two sexes at ("@") each of the frequencies. ("Simple [main] effects" as the command's help file says.) Their joint test is that all of the simple effects (all six) are zero.

The second is the test of the interaction of sex and frequency. It is what you get in a factorial ANOVA table for the interaction term's F statistic and p-value.

The last is a test of the whole model, including the random effect of patient, and not only the fixed effects (as in the the first two).

[Puriya Daniel Yazdanfard]

The first tests difference in z between the two sexes at ("@") each of the frequencies. ("Simple [main] effects" as the command's help file says.) Their joint test is that all of the simple effects (all six) are zero.

The second is the test of the interaction of sex and frequency. It is what you get in a factorial ANOVA table for the interaction term's F statistic and p-value.

The last is a test of the whole model, including the random effect of patient, and not only the fixed effects (as in the the first two).

Thank you very much. This has been helpful.
However I do not quite follow "Their joint test is that all of the simple effects (all six) are zero". Does this mean that there is no difference in Z-scores between the two sexes at different frequencies? The P value is highly significant should that not indicate the opposite? That there is a difference?

EDIT:
Or perhaps what you mean is that the joint test is a test that assesses weather all the parameters have zero difference? Thus it being significant indicates that there is a difference between each frequency depending on sex.

[Joseph Coveney]

I do not quite follow "Their joint test is that all of the simple effects (all six) are zero". Does this mean that there is no difference in Z-scores between the two sexes at different frequencies?

It's like the joint test shown below. Look at what the component null hypotheses are for the joint test statistic that's shown below.


.ÿquietlyÿsysuseÿauto

.ÿquietlyÿmvregÿpriceÿmpgÿheadroomÿtrunkÿweightÿlengthÿ=ÿi.foreign

.ÿquietlyÿtestÿ[price]1.foreign

.ÿquietlyÿtestÿ[mpg]1.foreign,ÿaccumulate

.ÿquietlyÿtestÿ[headroom]1.foreign,ÿaccumulate

.ÿquietlyÿtestÿ[trunk]1.foreign,ÿaccumulate

.ÿquietlyÿtestÿ[weight]1.foreign,ÿaccumulate

.ÿtestÿ[length]1.foreign,ÿaccumulate

ÿ(ÿ1)ÿÿ[price]1.foreignÿ=ÿ0
ÿ(ÿ2)ÿÿ[mpg]1.foreignÿ=ÿ0
ÿ(ÿ3)ÿÿ[headroom]1.foreignÿ=ÿ0
ÿ(ÿ4)ÿÿ[trunk]1.foreignÿ=ÿ0
ÿ(ÿ5)ÿÿ[weight]1.foreignÿ=ÿ0
ÿ(ÿ6)ÿÿ[length]1.foreignÿ=ÿ0

ÿÿÿÿÿÿÿF(ÿÿ6,ÿÿÿÿ72)ÿ=ÿÿÿ15.64
ÿÿÿÿÿÿÿÿÿÿÿÿProbÿ>ÿFÿ=ÿÿÿÿ0.0000

.ÿexit

endÿofÿdo-file


.ÿ