Discussion: View Thread

Dear All,

I have a 2 level study design with level 1 having about 1200 individuals equally distributed in each of the 4 level 2 units, with a possible chance of 15% attrition over the entire study period. 

Level2 (say clinics):       A                          B                     C                                D

Level1 (say patients): 1,2,...300              1,2,...300          1,2,...30                       1,2,...300

 
1) What is the best way to calculate power for a 2 level design (such as above) when one is expecting a certain amount of attrition (15% in my case)?  
I read Snijder's Power & Sample Size in Multilevel models but  I cant find any specific mention of the use of prior knowledge of % of attrition.. 

2) Also, for such a study while building multivariate hierarchical models how can one estimate the maximum number of allowable predictors one can include to achieve specific sample size. 

Peduzi et al(1996) , have done an extensive simulation study on events per variable in case of logistic regression  to calculate minimum sample size required but its not in multilevel scenario 

Looking forward for your guidance.

Best Regards,
Tasneem
-------------------------------------------
[Tasneem] [Zaihra]
[Post Doctoral Fellow]
[McGill University]
-------------------------------------------

Accounting for attrition is easy, just increase your numbers by 15% after your finished.

Multilevel designs are difficult to deal with and I would need to spend time finding the papers on it;
I believe there was a Statistics in Medicine paper on that subject in the last year or so, but would
recommend using the Computerized Index to Statistics to see what has been done recently.

Ray

-------------------------------------------
Raymond Hoffmann
Professor, Biostatistics in Pediatrics
Medical College of Wisconsin
-------------------------------------------