Overlaying three itsa graphs

[Claire MacDonald]

I am using Ariel Linden's excellent "itsa" package in Stata 17 for an interrupted time series analysis.

Linden, A. (2015). "Conducting Interrupted Time-series Analysis for Single- and Multiple-group Comparisons." The Stata Journal;15(2): 480-500.


I would like to be able to overlay three single group model graphs into one graph for illustrative purposes, rather than using the multigroup analysis as this only allows a comparison of two groups (as far as I have been able to ascertain). I have been able to use the overlay command with simple time series line graphs, but not with the segmented regression graphs which are generated from the itsa command.

I have put together an example dataset for this question:

* Example generated by -dataex-. For more info, type help dataex
clear
input float mdate long outcome float value
696 1  7
697 1  6
698 1  7
699 1  8
700 1  9
701 1 11
702 1 14
703 1  7
704 1  6
705 1  6
706 1  8
707 1  7
696 2 45
697 2 46
698 2 44
699 2 42
700 2 40
701 2 37
702 2 36
703 2 45
704 2 45
705 2 44
706 2 44
707 2 46
696 3 20
697 3 20
698 3 21
699 3 21
700 3 20
701 3 21
702 3 21
703 3 20
704 3 21
705 3 21
706 3 20
707 3 20
end
format %tm mdate
label values outcome outcomelbl
label def outcomelbl 1 "var1", modify
label def outcomelbl 2 "var2", modify
label def outcomelbl 3 "var3", modify

Here is the simplified code for the three single group models, which each produce a graph as well as the model parameter statistics:

 
itsa value, single treat(1) trperiod (2018m4; 2018m8) replace figure

itsa value, single treat(2) trperiod (2018m4; 2018m8) replace figure

itsa value, single treat(3) trperiod (2018m4; 2018m8) replace figure

For reference, and to demonstrate the difference, here is some code for a multi-group analysis comparing var1 as the control with var2 as the comparison group:

 
itsa value, treat (1) trperiod (2018m4; 2018m8) contid (2) replace figure

I have searched Statalist, Stata help and several instructional videos so am hoping someone here may be familiar with Linden's package and able to help.

Many thanks.