If I want to measure the effects of a quasi-natural experiment on a group of cases (10 cases), and I want to argue that the effects of the intervention were more robust in one group of cases (5 cases) over the other group (5 cases) (based on a dichotomized institutional difference), can I run the regression results in two separate models (one model consisting of one group of cases and another model consisting of the other group of cases) and compare the results between the two groups (models) in order to show that the effects of the intervention were more robust on one group over the other?
Your question was how to compare two separate regression results for examining the effect of intervention. If you run two separate regression models, you will get two separate results and show that estimate was changed by the intervention. It will be difficult to test the statistical proportion different between pre and post intervention. My suggestion is that you can combine the two groups of data and create a variable study group (1=post intervention and 0=pre intervention) , then include this variable in your model. You will get the coefficient for study group variable together with p-value and confident interval. This coefficient indicates that what proportion was decreased by the intervention and interpret that the effect of intervention were more robust if you see p
Terms for consideration
Quasi-experiment contains a control group and an experimental group the elements of which are not randomly selected (inferential analysis will be inappropriate_
Longitudinal panel design collects data at different points in time from the same participants called panel.
Eltion, I am a bit confused with the "givens". As a quasi-experiment, two groups are indispensable- a control group and an experimental group.
As a panel design, is it longitudinal (maybe time-series)? (where regression is possible)
Are you comparing means between two groups?
Will you have at least points from x and y to warrant a regression?
Please clarify. Ed
Hello Ed! You're right I wasn't clear enough or accurate in my description. And you are right, after thinking about this, it does not make sense to separate the cases into two groups--doing so would defeat the purpose of the intervention.
I am actually using time-series data for which I have measurements for both X and Y, for before and after the intervention. But I am not interested in simply comparing means. I want to run an OLS model to show that the effects of the intervention had a stronger impact on one group of cases over another group. Although, as you said, considering all the cases together is indispensable, what I am struggling with is how to show that the intervention was for effective in one group over another. That may sound simple enough but also consider this. I am making two comparisons--the first comparison compares the effects of the intervention in a pre versus post format, while the second comparison is concerned with the effects of the intervention only in the post period where I want to show that the intervention had a stronger effect in one group over the other group.
Regarding the first comparisons, a simple dummy variable of pre and post would suffice. But its the second comparison, which I believe is influenced by my independent variable which I am finding hard to incorporating into my model and interpret correctly.
Regarding the second comparison, you may run the separate model with independent variables and then see that what coefficient of variable are change from before intervention group to after intervention. If x1 variable changes at after intervention, the effect of X1 variable on disease was changed by the intervention.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS