$\chi^2$ test and sparsely populated bins Which article should I reference as a source of the following claim?:
The $\chi^2$ test is not recommended when too many sparsely populated bins exist in a given histogram.
Please rephrase the above sentence if it is not clear.
This article (page 328) by W. Cochran states something along that lines, but it's pretty old (1952) and it's not at all conclusive.
 A: The recommendation that all the expected values meet or exceed 5 for a chi-squared test is very old and very well-known.  You may not have to substantiate it with a citation.  It is also somewhat arbitrary, and this is also pretty well known.  There won't be a 'bright line' anywhere separating conditions under which the test works just fine vs. breaks down--there can't be.  Again, any research statistician will know this or immediately intuit it.  As a result, I doubt there will be much research substantiating the 5 rule of thumb.  If you wanted to cite something anyway, the Cochran paper may be sufficient.  
If you want to pursue this for your own edification, you may want to read Campbell (2007) and the various papers that cite it.  This is not quite the same context as yours, but the underlying issue is the same.  
A: Here's a screen capture of the first half of the relevant page from Bhattacharya and Johnson, Statistical Concepts and Methods, in response to a request in comments to @gung's answer:

