Simple effect tests following two-way ANCOVA

[Jan Schneider]

Hi everyone,

I am currently working on a research project in which I run a two-way ANCOVA with one covariate. Both variables of interest (a and b) are categorical (with two levels each), and the covariate d is continuous. I used the following code: anova y a##b c.d

Now my question is about the test of simple main effects. The ANCOVA shows a significant interaction of a and b and a significant main effect of b, so I want to do simple main effect tests to see whether b is significant in both levels of a or just in one of the two. How can I do this in Stata? The covariate d still needs to be included, otherwise commands like sme do not seem to work (I tried "sme b a" and it showed no significant effect of b for either level of a, and I think this is because I did not manage to include the covariate in this simple main effect test; however, I am happy to use other commands than sme if there are better ones!).

Thanks in advance!

[Joseph Coveney]

Have you considered contrast or margins?

.ÿ
.ÿversionÿ16.1

.ÿ
.ÿclearÿ*

.ÿ
.ÿsetÿseedÿ`=strreverse("1547282")'

.ÿ
.ÿquietlyÿsetÿobsÿ28

.ÿgenerateÿbyteÿAÿ=ÿmod(_n,ÿ2)

.ÿbysortÿA:ÿgenerateÿbyteÿBÿ=ÿmod(_n,ÿ2)

.ÿ
.ÿgenerateÿdoubleÿCÿ=ÿruniform(-0.5,ÿ0.5)

.ÿ
.ÿgenerateÿdoubleÿoutÿ=ÿrnormal()

.ÿ
.ÿ*
.ÿ*ÿBeginÿhere
.ÿ*
.ÿanovaÿoutÿi.A##i.Bÿc.C

ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿ=ÿÿÿÿÿÿÿÿÿ28ÿÿÿÿR-squaredÿÿÿÿÿ=ÿÿ0.2054
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿRootÿMSEÿÿÿÿÿÿ=ÿÿÿÿ1.03243ÿÿÿÿAdjÿR-squaredÿ=ÿÿ0.0672

ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿSourceÿ|ÿPartialÿSSÿÿÿÿÿÿÿÿÿdfÿÿÿÿÿÿÿÿÿMSÿÿÿÿÿÿÿÿFÿÿÿÿProb>F
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-----------+----------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿModelÿ|ÿÿ6.3368733ÿÿÿÿÿÿÿÿÿÿ4ÿÿÿ1.5842183ÿÿÿÿÿÿ1.49ÿÿ0.2390
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿAÿ|ÿÿ.00322125ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿ.00322125ÿÿÿÿÿÿ0.00ÿÿ0.9566
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿBÿ|ÿÿ4.0071408ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿ4.0071408ÿÿÿÿÿÿ3.76ÿÿ0.0649
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿA#Bÿ|ÿÿÿ.0125271ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿ.0125271ÿÿÿÿÿÿ0.01ÿÿ0.9146
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿCÿ|ÿÿ1.4485264ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿ1.4485264ÿÿÿÿÿÿ1.36ÿÿ0.2557
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿResidualÿ|ÿÿ24.516175ÿÿÿÿÿÿÿÿÿ23ÿÿÿ1.0659207ÿÿ
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿ-----------+----------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿTotalÿ|ÿÿ30.853048ÿÿÿÿÿÿÿÿÿ27ÿÿÿ1.1427055ÿÿ

.ÿ
.ÿcontrastÿr.B@A,ÿnoeffects

Contrastsÿofÿmarginalÿlinearÿpredictions

Marginsÿÿÿÿÿÿ:ÿasbalanced

------------------------------------------------
ÿÿÿÿÿÿÿÿÿÿÿÿÿ|ÿÿÿÿÿÿÿÿÿdfÿÿÿÿÿÿÿÿÿÿÿFÿÿÿÿÿÿÿÿP>F
-------------+----------------------------------
ÿÿÿÿÿÿÿÿÿB@Aÿ|
ÿ(1ÿvsÿ0)ÿ0ÿÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿÿÿÿÿ2.04ÿÿÿÿÿ0.1669
ÿ(1ÿvsÿ0)ÿ1ÿÿ|ÿÿÿÿÿÿÿÿÿÿ1ÿÿÿÿÿÿÿÿ1.69ÿÿÿÿÿ0.2063
ÿÿÿÿÿÿJointÿÿ|ÿÿÿÿÿÿÿÿÿÿ2ÿÿÿÿÿÿÿÿ1.88ÿÿÿÿÿ0.1750
ÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿDenominatorÿ|ÿÿÿÿÿÿÿÿÿ23
------------------------------------------------

.ÿ
.ÿquietlyÿmarginsÿA#B,ÿpost

.ÿtestÿ0.A#1.Bÿ=ÿ0.A#0.B

ÿ(ÿ1)ÿÿ-ÿ0bn.A#0bn.Bÿ+ÿ0bn.A#1.Bÿ=ÿ0

ÿÿÿÿÿÿÿF(ÿÿ1,ÿÿÿÿ23)ÿ=ÿÿÿÿ2.04
ÿÿÿÿÿÿÿÿÿÿÿÿProbÿ>ÿFÿ=ÿÿÿÿ0.1669

.ÿtestÿ1.A#1.Bÿ=ÿ1.A#0.B

ÿ(ÿ1)ÿÿ-ÿ1.A#0bn.Bÿ+ÿ1.A#1.Bÿ=ÿ0

ÿÿÿÿÿÿÿF(ÿÿ1,ÿÿÿÿ23)ÿ=ÿÿÿÿ1.69
ÿÿÿÿÿÿÿÿÿÿÿÿProbÿ>ÿFÿ=ÿÿÿÿ0.2063

.ÿ
.ÿexit

endÿofÿdo-file


.

[Jan Schneider]

Thanks for the reply! I tried both options and still the simple main effects of B remain insignificant. However, I realized that in your example from above, the p values of B are higher than the p value in the ANOVA table (p = 0.1669 and p = 0.2063 vs. p = 0.0649), like in my case. I also observed that the sme command I originally used yields the same F value as both contrast and margins command, so these three tests seem to tell me the same about the effect of B in both levels of A.
But I am still wondering whether these outputs are misleading because these commands exclude the covariate (which was significant at p < .001 in the ANOVA) or whether they are not (and already include the covariate) and I am just wrong in expecting to observe one or two significant simple main effects of B because of a significant main effect of B in the ANOVA (if the latter is true, could you maybe tell me how this is possible?).