overlay two mixed-model graphs

[Xiaochen Zhang]

I'm using Stata 14.0 in windows. I was curious if there is a way to overlay two mixed model graphs.
I have data in 6 months on physical activity and temperature. I used mixed model to get the trend of physical activity during the 6 months.

xtmixed time_exercise i.week ||study_id:,
margins i.week
marginsplot, noci

I did the same to see the trend in temperature during the 6 months

xtmixed wk_temperature i.week ||study_id:,
margins i.week
marginsplot, noci

So I got two graphs : one is physical activity (y axis) vs. week (x axis), the other is temperature (y axis) vs. week (x axis). I'd like to know how can I overlay these two graphs together.
I have seen some graphs in the literature combined these types of graphs (one plot, one x axis, one y axis. and using two y axis legend to explain). But haven't figured out how to do it in Stata.

Thank you
Xiaochen

[Joseph Coveney]

You could use predict after fitting each model and plot the linear predictions for the fixed effects on one graph using graph twoway. Something like

xtmixed time_exercise i.week ||study_id:
predict double exercise_hat, xb
xtmixed wk_temperature i.week ||study_id:
predict double temperature_hat, xb
graph twoway line exercise_hat week || line temperature_hat week

But, if the two variables are related in some manner, why not just model them together in a single model? Below, I've gone with your assertion that temperature is an outcome (you modeled it as one) and shown how you can get them both into a single marginsplot. It would seem, though, that if it's ambient temperature, then it would be a predictor and not a second outcome.

.ÿversionÿ14.1

.ÿ
.ÿclearÿ*

.ÿsetÿmoreÿoff

.ÿsetÿseedÿ`=date("2015-12-05",ÿ"YMD")'

.ÿ
.ÿquietlyÿdrawnormÿu1ÿu2,ÿmeans(60ÿ20)ÿsd(1ÿ2)ÿcorr(1ÿ0.5ÿ\ÿ0.5ÿ1)ÿn(250)

.ÿ
.ÿgenerateÿintÿstudyÿ=ÿ_n

.ÿ
.ÿforvaluesÿweekÿ=ÿ1/26ÿ{
ÿÿ2.ÿÿÿÿÿÿÿÿÿgenerateÿdoubleÿexercise`week'ÿ=ÿu1ÿ+ÿrnormal()
ÿÿ3.ÿÿÿÿÿÿÿÿÿgenerateÿdoubleÿtemperature`week'ÿ=ÿu2ÿ+ÿ2ÿ*ÿrnormal()
ÿÿ4.ÿ}

.ÿquietlyÿreshapeÿlongÿexerciseÿtemperature,ÿi(study)ÿj(week)

.ÿ
.ÿ*
.ÿ*ÿBeginÿhere
.ÿ*
.ÿrenameÿexerciseÿoutcome1

.ÿrenameÿtemperatureÿoutcome2ÿ//ÿAreÿyouÿsure?

.ÿquietlyÿreshapeÿlongÿoutcome,ÿi(studyÿweek)ÿj(type)

.ÿ
.ÿquietlyÿtabulateÿtype,ÿgenerate(type)

.ÿ
.ÿmixedÿoutcomeÿi.week##i.typeÿ||ÿstudy:ÿtype1ÿtype2,ÿnoconstantÿcovariance(unstructured)ÿ///
>ÿÿÿÿÿÿÿÿÿremlÿresiduals(independent,ÿby(type))ÿnolrtestÿnofetableÿnolog

Mixed-effectsÿREMLÿregressionÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿobsÿÿÿÿÿ=ÿÿÿÿÿ13,000
Groupÿvariable:ÿstudyÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿNumberÿofÿgroupsÿÿ=ÿÿÿÿÿÿÿÿ250

ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿObsÿperÿgroup:
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿminÿ=ÿÿÿÿÿÿÿÿÿ52
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿavgÿ=ÿÿÿÿÿÿÿ52.0
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿmaxÿ=ÿÿÿÿÿÿÿÿÿ52

ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿWaldÿchi2(51)ÿÿÿÿÿ=ÿÿ101885.73
Logÿrestricted-likelihoodÿ=ÿ-23801.742ÿÿÿÿÿÿÿÿÿÿProbÿ>ÿchi2ÿÿÿÿÿÿÿ=ÿÿÿÿÿ0.0000

------------------------------------------------------------------------------
ÿÿRandom-effectsÿParametersÿÿ|ÿÿÿEstimateÿÿÿStd.ÿErr.ÿÿÿÿÿ[95%ÿConf.ÿInterval]
-----------------------------+------------------------------------------------
study:ÿUnstructuredÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(type1)ÿ|ÿÿÿ.7897238ÿÿÿ.0742181ÿÿÿÿÿÿ.6568701ÿÿÿÿ.9494475
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿvar(type2)ÿ|ÿÿÿ4.406638ÿÿÿ.4088276ÿÿÿÿÿÿÿ3.67398ÿÿÿÿ5.285401
ÿÿÿÿÿÿÿÿÿÿÿÿcov(type1,type2)ÿ|ÿÿÿ.7448146ÿÿÿ.1319014ÿÿÿÿÿÿ.4862926ÿÿÿÿ1.003337
-----------------------------+------------------------------------------------
Residual:ÿIndependent,ÿÿÿÿÿÿÿ|
ÿÿÿÿbyÿtypeÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ|
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ1:ÿvar(e)ÿ|ÿÿÿ.9974162ÿÿÿ.0178781ÿÿÿÿÿÿ.9629841ÿÿÿÿ1.033079
ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿ2:ÿvar(e)ÿ|ÿÿÿ4.028403ÿÿÿ.0722068ÿÿÿÿÿÿ3.889338ÿÿÿÿ4.172441
------------------------------------------------------------------------------

.ÿ
.ÿquietlyÿmarginsÿi.week#i.type

.ÿmarginsplotÿ,ÿnociÿ///
>ÿÿÿÿÿÿÿÿÿplot1opts(lcolor(black)ÿmsymbol(none))ÿ///
>ÿÿÿÿÿÿÿÿÿplot2opts(lcolor(black)ÿmsymbol(none))ÿ///
>ÿÿÿÿÿÿÿÿÿtext(55ÿ13ÿ"DurationÿofÿExerciseÿ(Minutes)")ÿ///
>ÿÿÿÿÿÿÿÿÿtext(15ÿ13ÿ"AmbientÿTemperatureÿ(°C)")ÿ///
>ÿÿÿÿÿÿÿÿÿylabel(0(25)75,ÿangle(horizontal)ÿnogrid)ÿlegend(off)ÿ///
>ÿÿÿÿÿÿÿÿÿytitle(LinearÿPredictions)ÿxtitle(Week)ÿtitle("")

ÿÿVariablesÿthatÿuniquelyÿidentifyÿmargins:ÿweekÿtype

.ÿquietlyÿgraphÿexportÿZhang.png

.ÿ
.ÿexit

endÿofÿdo-file


.

[Xiaochen Zhang]

Hi Joseph,

Your second solution is brilliant! That works for me. I never thought of put two outcomes together.
Thank you

Xiaochen