I am analyzing a large of dataset (n>100) of incident rates, with the aim of forming a normal distribution. Then I will know if a future incident rate (x%) is either close to a historical mean or not, and can score/rate it accordingly with an already created formula.
The data is positively-skewed, as most data points cluster around or near zero percent. I HAVE to transform this data into a normal distribution, correct? What is the preferred method when dealing with percentages (these will always be between 0 and 100%)? Are there alternative non-normalizing methods I can use to reach my desired output?
Anyway, let's say I've transformed the data and it follows a normal distribution. Now I can find the mean and std dev, then plot these in Excel using z-scores. Then I should be able to determine if incident rate x% is in the top 10%, top 20% of values, and score it with my formula accordingly.
Any problems with this method?