# Finding two populations in a set of data

I have a set of data (metallicity of globular clusters) and wish to determine the presence of two sub-populations. I know there are two populations but I am unsure on how to split the two.

I have plotted a histogram of the data (metallicity vs number). Is there any method aside from sight, to determine the two populations?

FeH     Frequency
0       4
-0.2    6
-0.4    14
-0.6    15
-0.8    6
-1      15
-1.2    22
-1.4    23
-1.6    19
-1.8    15
-2      8
-2.2    5

• If there are two modes, it is common to infer that there are two populations, and to hope that at least one will be named after you. I do not see two modes here. Apr 21, 2012 at 5:14
• Thanks André. I have update my question slightly and redone the histogram to show the total number agains the value. Perhaps this is a better place to start from?
– Carl
Apr 21, 2012 at 5:57
• Congraulations! Which subpopulation should be named Carl? The population is clearly bimodal. Apr 21, 2012 at 6:09
• There are. I have forgotten, it has been many years since I have done statistical consulting. That's why I included the term bimodal, for searching. Post a similar question on the stats version, bottom of page, and you will probably get a few references. Apr 21, 2012 at 6:45
• I've seen, -also some years ago, and do not remember the details- a concept and a software called "emmix" (try google) which should be able to separate mixtures of normal distributed data into the most likely subgroups. It seems to me, that this could be the required systematic approach for your question (but might require high level understanding of the required math, don't know). Apr 21, 2012 at 8:59

For anyone else interested, I used Gaussian Mixture Modeling (GMM) algorithm to determine the means of the two populations and separate them.

Details of the techniques used are explained in the paper linked on this page: http://www.astro.lsa.umich.edu/~ognedin/gmm/