Annals of Family Medicine -- Comments for Reed, 2 (3) 201-203

QUICK SEARCH:

	Author:		Keyword(s):
Year:		Vol:		Page:

TRACK to:

Original Research:
James F. Reed, III
Adjusted Chi-Square Statistics: Application to Clustered Binary Data in Primary Care
Ann Fam Med 2004; 2: 201-203 [Abstract] [Full text] [PDF]

TRACK: Submit a comment to this article

Electronic letters published:

Analysis of clustered data.: James W. Mold (5 June 2004)

Analysis of clustered data. 5 June 2004

James W. Mold,
Oklahoma City, OK
Professor, University of Oklahoma HSC
Send response to journal:
Re: Analysis of clustered data.

The papers by Reed and Killip, Mahfoud, and Pearce were excellent and timely particularly for those of us involved in research that involves the analysis of information about the patients of clinicians in practices belonging to networks. The analytic problem, stated most succinctly in the first sentence of the Killip abstract, occurs when “subjects are randomized at a group level but analyzed at an individual level.” I was initially confused by the chi squared examples discussed by Reed because the data seemed to be presented as group data. However, I eventually realized that the chi squared tests used to analyze the data give credit for the total number of individual subjects and therefore constitute analysis at the individual level. I wonder whether the problem could be solved in this case by converting the group proportions to decimal fractions, and then analyzing the data using independent t-tests, recognizing but ignoring the violation of assumptions caused by the dichotomous nature of the individual data points.
The discussion of the advantages and disadvantages of clustered designs and the explanations of the calculations used to determine the ICC, ESS, and DE in the Killip article were especially helpful. However, I’m still not sure that I can calculate the variances between and within clusters for the examples provided by Reed. The calculations must become even more complex when the analyses involve more sophisticated statistical tests such as regression or when there are clusters within clusters (e.g. patients within clinicians within practices). It looks like there will jobs available for biostatisticians for the foreseeable future.
Competing interests: None declared