Two-way graph of longitudinal data

[May Blake]

I have the following data below with pre-treatment data (denoted with the negative times) and post-treatment data (denoted with the positive times). I would like to graph this two-way to create a %predicted value of MT which are already percent values. I am trying to graph it similar to figure 2 in this publication (https://pubmed.ncbi.nlm.nih.gov/27816444/). However, I get a straight line in my graph when I use the codes below.

mixed MT c.time || id:, covariance(unstructured) nolrtest nolog
quietly margins , at(time = (-5(1)5))
marginsplot , title("") plotopts(msymbol(none) lcolor(black)) ytitle(BP (%)) ylabel( , angle(horizontal) nogrid) xtitle(months) level(50) recastci(rline) ciopts(lcolor(black) lpattern(dash)) xline(0)
end

* Example generated by -dataex-. To install: ssc install dataex
clear
input byte(id time) int MT
 1  -5 110
 1  -2 116
 1   1 189
 2  -3 135
 2  -1 170
 2   1 106
 2   2 155
 8  -4 149
 8  -1 169
 8   1 160
11 -10 170
11  -1 180
11   1 144
14  -5 174
14  -1 149
14   1 145
14   2 133
14   5 100
end

[Maarten Buis]

Your model assumes that there is a straight line. Margins and marginsplot just summarize the model, so it should also result in a straight line. The solution is to change the model to allow for non-linearity, See e.g. this Stata tip: http://maartenbuis.nl/publications/leaps.html

[May Blake]

Hi Maarten, Thank you. This is exactly what I needed. I can use this with my mixed methods regression model. My only question is how I can structure the data to conduct this analysis. For the nlsw88 data use in your example, the data only has one observation per id. In my dataset, participants have multiple time and MT data depending on the pre-intervention enrollment and post-intervention measures. Do I have to reshape the data to be wide for those variables and if I do, how I do I add the time-varying variables since they will now be separated. I started using the code below but the I realized I may have it wrong since I have multiple entries.

gen aft=time>0
mixed MT c.time##c.aft || id:, covariance(unstructured) nolrtest nolog
lincom c.time#c.aft + aft
quietly margins, at(race=1 graft=1 smoke=2), over(time)
marginsplot , title("") plotopts(msymbol(none) lcolor(black)) ytitle(MT (%)) ylabel( , angle(horizontal) nogrid) xtitle(months) level(50) recastci(rline) ciopts(lcolor(black) lpattern(dash)) xline(0)