I'm working on a dataset of variables and I would like to apply a multiple imputation on a subset of variables of great interest. I'm working on STATA. I also know what's the prinicipal of the MI method : checking the MAR assumption, percentage of missing data, etc... But my main question is : What about the sample size? I read many articles (e.g Ula Nur for example) which describe the method perfectly on big datasets with around 15000 observations. In my study, I just have 300 hundred patients but more than 20% of missing data for many variables (this is why I think about imputation). So, can I apply this method on my little dataset ?
Hi Alix,....I think I might be able to assist you in some ways and since you just have 20% 'holes' in your dataset. I encountered a similar challenge before and I went online, joined Statalist (forun for all Stata users) and got my solution. The guys at the forum assisted me and I have compiled a dofile for you to go through. First read the brief from UCLA and thereafter use the codes listed towards the end of the dofile.
Let me know if this helps..
Ngozi
Thank you so much Ngozi ADELEYE for your answer and for sharing your program. I just take a look on it and it's interesting. Moreover the command mvpattern that discribes patterns of missing data is awesome. First I need to find a way to install this package quickly because my STATA software installation program have a problem to install "mvpattern", but whatever, I will fix that. Indeed, I didn't complete my dataset management but I know I have approximatively 50% of missing data. I'll keep in touch with you to tell you if I can run this work.
Thank you Alvin Vista for your indication. If I can process to MI on my 300 observation it's cool.
Hello,
I thik you have to apply the monte carlo simulation for this kind of study.
Hi,
Assuming that you are clear about the rest of the MI procedure and analyses methods, I suggest to use a large number of MI datasets, your pooled estimates should be based on more than 20 imputed datasets as suggested in the literature.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis