hi,
this looks lime multiple regression to me, because your DV is continuous, unless you only have a few percentage values lime 0, 0.33 0.66 1. What is the exact structure of your data?
Hi,
I have a simple test where I want to measure the effect of a couple of variables on an outcome.
Independent:
Var_1: Dichotomous, particular action applied Y/N
Var_2: Continuous, age of the subject 0,1,..5
Dependent:
A Ratio. (0-1)
I ideally want to compare if var_1 provides a particular increase in the Ratio before or after var_1 action was undertaken on a particular subject and if that increase is related to var_2. I can convert ratio to dicotomous (e.g. ratio_after/ratio_before - 1) by looking at increase or decrease in pre/post var_1 scenario.
Any guidance on how to proceed? My initial hunch was that a Chi-sq test would suffice.
I have around 200 samples total, each with var_1 before/after and different var_2.
hi,
this looks lime multiple regression to me, because your DV is continuous, unless you only have a few percentage values lime 0, 0.33 0.66 1. What is the exact structure of your data?
Thanks for your reply.
While the underlying data is more complex, I will end up with a table like below.
Code:CODE Date Action Ratio AA 1-May N 0.21 AA 1-Jun N 0.19 AA 1-Jul N 0.22 AA 1-Aug Y 0.19 AA 1-Sep Y 0.23 AA 1-Oct Y 0.26 BB 1-May N 0.15 BB 1-Jun N 0.14 BB 1-Jul N 0.13 BB 1-Aug N 0.14 BB 1-Sep Y 0.15 BB 1-Oct Y 0.16 ...
I have a feeling this is a little more complicated than would be assumed.
-It looks more like repeated measures. So AA is a subject and you have three measure before and then after an event?
-Plus just looking at differences can be mis-leading. My example is if some subjects had
high or low initial values, then they may have more room to move and it is not just the intervention you need to control for.
-Everybody got the intervention between the 3rd and 4th period?
-Also, ratio outcome could mean Poisson regression or maybe beta regression.
Can you provide more study context details?
Stop cowardice, ban guns!
hi,
could this be a mixed model? Was there one intervention only or did they have an intervention each month where you have yes?
Thanks for the pointers guys. Really appreciate it.
Some context. The subject here are accounts/stores. Each account has max sales potential. The
Ratio = Achieved Sales/Total Potential Sales
where
Total Potential Sales = Max Sales Potential - Achieved Sales in the Prior Months.
The Ratio is then summarized for a Fiscal Month/Quarter/Year.
Sales itself is driven by arrival or orders and customers.
Now the action part. Assume it is applied once - not every month. Once applied, the account stays in that state.
I am trying to assess the impact of the 'Action' on Ratio. Unfortunately, I do not have accounts where action was not applied. So a simpler A/B test is not an option. A same account may have a state where action was not applied but has now been applied.
As hlsmith noted, each account can start from a different base line and has the action applied at a different time. So _lift_ between the accounts will be different. That said, I am inclined to ignore effects from this time difference.
Hope I am making some sense.
hi,
just a quick idea - you could summarize the data for each location by taking the mean and the range before and after the action and analyse by a simple two-sample t-test for both range and mean.This could give you a general idea of how useful the action is over the total population of stores.
regards
Tweet |