Stats - BiModal Distribution analysis

flop

Hi all,

I have a dataset which has 3 modes, which i can see when i plot the data as a histogram.

I am wondering what is the general approach to fitting Bimodal data sets to Distributions curves?

Should I a) break the analysis in 3 sections are analyze each mode as a uni modal distribution? or do i need to analyze the data as a whole and fit to a bimodal distribution as a whole?

My stats knowledge is at undergrad level so doesn't cover bimodal datasets.

any replies appreciated.

Thanks
Jonathan

Fremen

You may want to look into using a mixture model. The idea is that you take a collection of positive weights that sum to one, and associate a univariate distribution with each weight. This gives a collection (w_i, P_i), where w_i is a weight and P_i is a distribution. You generate samples from p_i with probability w_i.

This gives you with a probability distribution whose density is multimodal. Here's a wikipedia example:



You can read more here:
http://en.wikipedia.org/wiki/Mixture_density
http://en.wikipedia.org/wiki/Mixture_model