MANOVA - how to include control variables

[Anne-Lien Linde]

Hi everyone,




For our thesis we make use of stata 16. The aim of our thesis is to find out to which extend chain affiliation may affect quality. We want to test if chain affiliation (coded 0 if non-chain, coded 1 if chain) has an impact on several quality measures (on the one side quality of life, which consists of 11 quality measures going from 1-5, and on the other side quality of care, which consists of 9 measures and are percentages). So we want to do two regressions: one to test chain affiliation (= independent variable) on quality of life (dependent variables) and one regression to test chain affiliation (= independent variable) on quality of care (= dependent variables).

We already did a MANOVA-test to determine whether there are differences between multiple independent variables, in our case these are chain vs non-chain, are significant (and they were). But now, we want to include control variables (which are dependent variables). Does anyone know how to do this?




Tank you in advance!

[Joseph Coveney]

I'm not sure what estimation command you all did to fit your MANOVA model, but maybe just put the other predictors in the model as shown in the illustration below? I use Stata's manova command, but you could try the same with whatever you used. (Begin at the "Begin here" comment. The code at the top of the do-file just creates a phony dataset for illustration.)

version 16.1

clear *

set seed `=strreverse("1545080")'
quietly set obs 250

generate int cid = _n
generate double cid_u = rnormal()

generate byte caf = mod(_n, 2)
generate double ctl = runiform() // "control variable"

forvalues i = 1/20 {
    generate double lat`i' = cid_u + rnormal()
    quietly replace lat`i' = lat`i' / sqrt(2)
}

forvalues i = 1/11 {
    generate byte qol`i' = 0
    forvalues cut = 1/4 {
        quietly replace qol`i' = qol`i' + 1 if lat`i' > invnormal(`cut' / 5)
    }
}

forvalues i = 12/20 {
    generate double qoc`i' = normal(lat`i')
}

*
* Begin here
*
manova qol* = i.caf c.ctl
// Whatever -manovatest- you've been doing
manova qoc* = i.caf c.ctl
// Ditto

// Maybe consider alternative?
sem (qol* <- QOL) (qoc* <- QOC) (QOL QOC <- caf ctl), ///
    covariance(e.QOL*e.QOC) nocnsreport nodescribe nofootnote nolog

exit

[Anne-Lien Linde]

We tried to include the control variables and this is the result of MANOVA. We are kind of confused as due to this inclusion the model is significant, but it looks like chain no longer has a significant impact. Is this how we should interpret the results? Moreover, is this what you meant by including control variables?

Number of obs = 332

W = Wilks' lambda L = Lawley-Hotelling trace
P = Pillai's trace R = Roy's largest root

Source | Statistic df F(df1, df2) = F Prob>F
------------+-------------------------------------------------------
Model |W 0.1312 49 539.0 2959.9 1.14 0.0196 a
|P 1.8068 539.0 3102.0 1.13 0.0281 a
|L 2.3049 539.0 2972.0 1.16 0.0128 a
|R 0.5286 49.0 282.0 3.04 0.0000 u
|-------------------------------------------------------
Residual | 282
------------+-------------------------------------------------------
chain |W 0.9382 1 11.0 272.0 1.63 0.0908 e
|P 0.0618 11.0 272.0 1.63 0.0908 e
|L 0.0658 11.0 272.0 1.63 0.0908 e
|R 0.0658 11.0 272.0 1.63 0.0908 e
|-------------------------------------------------------
erkendeca~n |W 0.8812 4 44.0 1042.6 0.80 0.8266 a
|P 0.1239 44.0 1100.0 0.80 0.8235 a
|L 0.1292 44.0 1082.0 0.79 0.8298 a
|R 0.0569 11.0 275.0 1.42 0.1619 u
|-------------------------------------------------------
n_prijsEe~r |W 0.2348 34 374.0 2886.5 1.14 0.0377 a
|P 1.3223 374.0 3102.0 1.13 0.0482 a
|L 1.5980 374.0 2972.0 1.15 0.0284 a
|R 0.4024 34.0 282.0 3.34 0.0000 u
|-------------------------------------------------------
n_prijsPi~e |W 0.9105 2 22.0 544.0 1.19 0.2532 e
|P 0.0909 22.0 546.0 1.18 0.2581 a
|L 0.0968 22.0 542.0 1.19 0.2485 a
|R 0.0769 11.0 273.0 1.91 0.0384 u
|-------------------------------------------------------
bezetting~n |W 0.9007 2 22.0 544.0 1.33 0.1457 e
|P 0.1018 22.0 546.0 1.33 0.1436 a
|L 0.1075 22.0 542.0 1.32 0.1478 a
|R 0.0650 11.0 273.0 1.61 0.0945 u
|-------------------------------------------------------
n_nursing~t |W 0.7768 6 66.0 1460.9 1.07 0.3310 a
|P 0.2453 66.0 1662.0 1.07 0.3232 a
|L 0.2601 66.0 1622.0 1.07 0.3390 a
|R 0.0906 11.0 277.0 2.28 0.0111 u
|-------------------------------------------------------
Residual | 282
------------+-------------------------------------------------------
Total | 331
--------------------------------------------------------------------
e = exact, a = approximate, u = upper bound on F