hai, as i know  you have to find the related study which study the same parameters. then you have to minus their  end result  between treatment and control of that study, then you put the difference in mean diff colum in PS calculation then you decide the SD then it will calculate for you. you have to do that for other parameter to get enaugh sample size.then the calculated sample size should add 20% of drop up possibility.

There is too little information to answer your question. 

First off: if you have three groups, it is important to know what hypothesis connects them. For example, you might be studying lung function and your groups might be

a - never smoked

b - currently smoker

c - former smoker

In a case like this, you can imagine that there are two hypotheses:

1. There is an effect of having ever smoked (b+c versus a) 

2. There is an additional effect of bring a current smoker (b versus a and c)

So the hypothesis, not the number of groups, determines the statistical approach.

The second thing we need to know is what you mean by categorical data. Are you referring to categorical (nominal) scales, in which the data are in unordered categories, or ordinal data, in which the data form ordered categories. Or dichotomous (yes/no, true/false type data)?

Nur Hasana is wrong about basing sample size on a previously published effect size. Sample size must be based on the minimum clinically significant difference, which is the smallest effect size that is of real-life importance. Basing effect size on previously published work means that you do not have a proper chance of detecting effect sizes smaller than that which are still of importance. 

I would recommend that you read up a bit on sample size. It requires a lot of your input in terms of specifying the hypotheses being tested and the minimum effect size to be detected. After that, the stats are pretty simple!

Could you precise which are the "categorical data" used in your study?

It is possible that the solution is very simple...