We have two treatments (A or B), both are independent, we want to compare their mean effects for the speed for recovery in days, However each of the treatments A or B was not prescribed as a single agent, and each one: A or B was confounded with other treatments; e.g. it could be confounded as:
A alone B alone NO confounder
A + C B + C
A + CD B + CD
A + E B + E
And so on, given that we have limited combinations for limited confounders i.e about 5 confounders including their combinations.
You can calculate ANOVA for all four conditions with the response variable recovery in days
You may want to look at using survival analysis and Cox regression (often referred to as time to event analysis and proportional hazards regression, respectively). [Note that 'survival' refers to the time free of the outcome. This could be death, but could equally be recovery, as in your case.]
Cox regression will tell you what the propensity of recovery of group A over group B is (the 'hazard ratio'), both in crude analysis and adjusting for the confounders. An advantage of Cox regression is that you can have censored data (i.e. where the outcome is not observed for all subjects - some subjects may drop out before they recover, or might never recover) and different follow up times for different subjects. As with all models, you will need to check that various assumptions are met.
I don't use SPSS myself but there's a help file on Cox regression here: http://www.ibm.com/support/knowledgecenter/SSLVMB_20.0.0/com.ibm.spss.statistics.help/idh_coxr.htm.
See also this introduction: https://www.cscu.cornell.edu/news/statnews/stnews78.pdf (and you can find much more on-line and in textbooks).
Multiple regression should work; the whole purpose of it is to figure out how much variance in the dependent variable can be uniquely ascribed to a given independent variable when other independent variables have also been included in the model.
If you have:
A alone B alone
A + C B + C
A + CD B + CD
A + E B + E
You can perform ANOVA, after that, orthogonal contrast by using the t test to check if the contrast is or not zero!!!:
So, you can obtain different responses....
Alysson has the approach you want, I think. You would conduct ANOVA with all the treatments as separate treatments, and then use a contrast to compare whichever group of treatments to any other group of treatments you want.
This is easy in Sas and relatively easy in R. Dunno about spss.
There is an example at the link below, particularly the example with multcomp.
http://rcompanion.org/rcompanion/h_01.html
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS