I hesitate whenever it is a sufficiently large group. I will be grateful for your help.
The question which statiscal method you can use depends on a lot of things, not only the number of cases. Since you didn't mention anything else here, look at these articles on Bitesizebio:
http://bitesizebio.com/19298/a-basic-guide-to-stats-comparing-two-sets-of-data/
http://bitesizebio.com/19315/lets-talk-about-stats-comparing-multiple-datasets/
It is not exactly what I meant , probably the question was wrongly formulated. I would like to apply just the decision tree method for my studies. Is there any limitation or threshold for number of investigated cases?
Literally speaking, you can make a decision tree from an N of 2, as long as you can base the separation of them on some simple yes or no basis (is this evidence of cancer, yes or no, for example).
A decision tree has to have some unbiased basis for node splitting, and the data needs to be such that it can ultimately be resolved down to a single optimal outcome (by whatever criteria you are defining optimal).
SAS has a nice white paper on using decision tress - http://support.sas.com/publishing/pubcat/chaps/57587.pdf
Without knowing anything more about your data or study, its hard to say if this is a good strategy or not. But literally speaking, sure, 45 cases is not a limitation.
I think Prof. Glasziou has published on this topic of decision trees
Just as Michael wrote, there is no "technical" limitation and it depends on your data if it makes sence.
Decision tree implementations usually use the number of cases in an end node as one of several stopping criteria. So they have some default values for the number of cases in the end nodes, and/or for the cases in a node to do a further split. For example in rpart (a CART- implementation in R) the default for the "minsplit" parameter is 20, so a node won't be split if there are less than 20 cases in it.
You can grow a large tree if you set the appropriate parameters to low values. You can even choose 1, which will probably result in an overfitted tree that should be pruned back using cross validation in the next step.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics