How do you find the Balance of Treatment in Randomized Experiment Data?

[Ng Luke]

Hi,

I am currently doing an assignment on randomised experiments and I am facing some difficulties with it. I have a binary variable, 0 for control, 1 for group A and 2 for group B. Both A and B are given different treatments. I have the basic characteristics for these people including age, income levels and household size. I would like to ask is how am I able to do a wald test to show that at baseline, the means of the 3 groups, control, A and B, are the same on Stata? I have tried finding solutions for these online but I couldn't get any information.

Also, if I would like to test the heterogeneous treatment effect on group B based on age, how do I go about doing so on Stata? I am using Stata 12.

Thanks so much and I really appreciate the help that you guys are providing Have a wonderful day ahead.

[Marcos Almeida]

Hello Ng,

To start, your group variable is not binary, since it has 3 values (0, 1 and 2). Now trying to provide a suggestion for your query, and considering your measure is a continuous variable, you could turn its baseline value into the outcome and the group variable as the predictor.

. regress baseline_value i.group

An ANOVA could also do the trick as well.

Best,

Marcos

[Ng Luke]

Hi Marcos,

Thanks so much for replying I hope you do not mind me asking. The thing is I have 3 groups of participants which are labelled under the variable, GROUP. It contains either (0,1,2) to represent the different groups. I have a bunch of control variables as listed above, which include age, income etc. And the outcome is to determine whether they choose to do something which is classified under the binary variable, Create. I need to do up a logistic regression for this. Could you advise?

Thanks so very much for all the help.

[Marcos Almeida]

Hello Ng,

Since this is a different question and considering you take as "answered" your original query, you'd better start a new thread,

That said, in spite of the paucity of details about your outcome, it seems you need to perform a logistical regression. However, may "time to event" be of concern, a survival analysis is something to think about.

In the new thread, please give enough details about your study design.

Best,

Marcos