How to analyze quasi -experiment with non equivalent control group analysis using Stata?

[Mulusew Gerbaba]

I am about to analyze a quasi-experimental study with non-equivalent control group (2 control and 1 treatment group) design; with multiple pretests and post-tests( four measurements before interventions and four measurements after the interventions in all three groups).

The data points are not equally spaced but are uniform in all participating groups both before and after the interventions ( 6hrs, 12hrs, 24hrs and 48hrs after the surgery). However, the samples are separates (individuals in the pre-intervention sample are not the same as individuals in the post-intervention sample. This is also true for my 2 control groups. Which analysis method would you suggest me?

Thanks in advance,

[Clyde Schechter]

So this sounds like a repeated-measures design with measurements nested within patients. So you would probably end up using one of the -xt- commands or one of the -me- commands for this. Since I assume you want to contrast the controls with the treatment group at some or all time points there will be treatment # time interaction terms in your model. But without any description of the outcome variable(s) and the research hypothesis(es) it is not possible to be more specific than that.

[Mulusew Gerbaba]

Thank you so much for your reply. The outcome variables were pain intensity and interference measured on Numeric Rating Scale of 0_10. We hypothesized that the intervention group will presume lower mean pain intensity and interference at all measurement points compared to the other 2 controls, after individual patient education plus healthcare professionals training. No measurement (outcome) exist at the health professional level only at patient level.

[Clyde Schechter]

So, assuming that your 0-10 rating scale distributions don't just bunch up at one end of the scale or exhibit some a bizarre distribution, it is probably simplest to treat these outcomes as if they were continuous variables. Your dataset will need to include a patient identifier, a treatment group variable (presumably 1 for treatment 0 for control, unless you want to distinguish the two control groups in your analysis, in which case some 3-level coding is needed), and another variable encoding time. (Since it is probably going to be helpful to use -margins- after you run the regression, you probably want to encode time in such a way that all values come out as non-negative integers. So maybe 0 for pre-op, 1 for 6hrs post-op, etc. Or perhaps treat time as a continuous variable and just coded as 0, 6, 12, 24, 48. Since this is not randomized trial, presumably you will want to adjust for differences between treatment and control groups on other relevant variable such as age, sex, and any other relevant information you have that might affect the outcome variables and might differ between the groups. (I'm not sure what variables those might be, but presumably you and your collaborators do.)

Ultimately it sounds like your analyses will look like:

mixed outcome_variable i.treatment_group##i.time covariates || patient_id: // IF TIME IS DISCRETE

mixed outcome_variable i.treatment_group##c.time covariates || patient_id: // IF TIME IS CONTINUOUS & LINEARLY RELATED TO OUTCOME

You will probably want to follow those up with -margins- commands like:

// IF TIME IS DISCRETE
margins treatment_group#time
margins time, dydx(treatment_group)

// IF TIME IS CONTINUOUS
margins treatment_group, at(time = (0 6 12 24 48))
margins, dydx(treatment_group) at(time = (0 6 12 24 48))

If you are not familiar with the i. and c. and ## operators, read -help fvvarlist- to learn about factor variable notation. If you are not familiar with the -margins- command, I think the best introduction is Richard Williams' http://www.stata-journal.com/sjpdf.h...iclenum=st0260, which is very clearly written and covers the basics (including the applications suggested above). After that, you might want to read the -margins- chapter in Stata's PDF documentation to learn about the fancier things you can also do with it.

Good luck.

[Mulusew Gerbaba]

I thank you very much, this is more than enough!!!