manova

[statausernumber]

Dear All,
I have two subgroups, A and B, and two variables x1 and x2. In addition, I also have two more variables q1 and q1.

FIRST TEST (subgroups and variables x1, x2)
I need to test that since x1 is lower in subgroup A (and hence greater in subgroup B), then also x2 will be lower in subgroup A (and greater in subgroup B).
Can I use MANOVA for this test? How should I test it (how to write the command)?

SECOND TEST (subgroups and variables q1, q2)
In addition, I also have these two other variables, q1 and q2 and in this case I need to test that since q1 is lower in subgroup A (and greater in subgroup B), then for q2 is the obverse, that is, q2 is gretaer in subgroup A (and lower in subgroup B).
How/where can I see the change in relation (that is: that whereas in the first test above, both x1 and x2 are lower in subgroup A, here in this second test it is the obverse, with q2 beeing greater where q1 is lower and vice versa).?

If not Manova, which other test should I run?
Thank you!

[statausernumber]

Note, I forgot to clarify: variables x1 x2 and q1 q2 are all continuous

[Alex Gamma]

Dear ?
Do you have a name? Would you mind posting with it?

You need to be more specific about what you want to compare. Whether x1 and x2 are different among A and B? Jointly different or separately different? Same for q1 and q2.

Depending on the answer you could do t-tests (-ttest-), non-parametric tests (e.g. -kwallis-), ANOVA (-anova-), simple linear regression (-regress-), multivariate regression (-mvreg-), or MANOVA (-manova-). According to Tabachnik & Fidell, "MANOVA works best with highly negatively correlated DVs [dependent variables] and acceptably well with moderately correlated DVs in either direction."

From what you're saying it sounds like x1 and x2 are positively, and q1 and q2 negatively, correlated.

Alex

Tabachnik, B. G., & Fidell, L. S. (2007). Using multivariate statistics. Pearson Education, Inc.

[statausernumber]

I need to test whether x1 and x2 are jointly dfiferent : that is, that x1 being lower in groupA leads x2 being also lower in groupA
Same for q1 abnd q2: that q1 being lower in groupA leads q2 being higher in groupA.
Luis

[Joseph Coveney]

Couldn't you just use a multivariate regression and then inspect the signs of the regression coefficients, followed by your joint tests? Something like that shown below.

.version13.1

.
.setmoreoff

.tempnameCorr

.matrixdefine`Corr'=J(4,4,0.7)+0.3*I(4)

.quietlydrawnormx1x2q1q2,doublecorr(`Corr')///
>n(100)seed(`=date("2014-05-18","YMD")')clear

.generatebytesubgroup=_n>_N/2

.generatedoublexb=subgroup-0.5

.foreachvarofvarlistx?q?{
2.quietlyreplace`var'=`var'+xb/2
3.}

.quietlyreplaceq2=-q2

.
.*
.*Beginhere
.*
.mvregx1x2q1q2=i.subgroup

EquationObsParmsRMSE"R-sq"FP
----------------------------------------------------------------------
x11002.94956610.07227.6276880.0069
x21002.99792230.07818.3075990.0049
q110021.0221430.05425.6210210.0197
q21002.96616870.142216.240370.0001

------------------------------------------------------------------------------
|Coef.Std.Err.tP>|t|[95%Conf.Interval]
-------------+----------------------------------------------------------------
x1|
1.subgroup|.5245075.18991322.760.007.1476309.9013841
_cons|-.2637928.1342889-1.960.052-.5302848.0026992
-------------+----------------------------------------------------------------
x2|
1.subgroup|.5752604.19958452.880.005.1791916.9713293
_cons|-.1821782.1411275-1.290.200-.4622412.0978848
-------------+----------------------------------------------------------------
q1|
1.subgroup|.4846734.20442862.370.020.0789916.8903552
_cons|-.204594.1445528-1.420.160-.4914544.0822663
-------------+----------------------------------------------------------------
q2|
1.subgroup|-.7787193.1932337-4.030.000-1.162185-.3952532
_cons|.3491968.13663692.560.012.0780453.6203483
------------------------------------------------------------------------------

.
.//Firsttest
.assert_b[x1:1.subgroup]>0&_b[x2:1.subgroup]>0

.test[x1]1.subgroup,notest

(1)[x1]1.subgroup=0

.test[x2]1.subgroup,accumulate

(1)[x1]1.subgroup=0
(2)[x2]1.subgroup=0

F(2,98)=4.82
Prob>F=0.0101

.
.//Secondtest
.assert_b[q1:1.subgroup]>0&_b[q2:1.subgroup]<0

.test[q1]1.subgroup,notest

(1)[q1]1.subgroup=0

.test[q2]1.subgroup,accumulate

(1)[q1]1.subgroup=0
(2)[q2]1.subgroup=0

F(2,98)=8.52
Prob>F=0.0004

.
.exit

endofdo-file


.