Sample size determination in health studies : a practical manual
S. K. Lwanga and S. Lemeshow
Abstract
Presents the practical and statistical information needed to help investigators decide how large a sample to select from a population targeted for a health study or survey. Desgned to perform a cookbook function, the book uses explanatory text and abundant tabular calculations to vastly simplify the task of determining the minimum sample size needed to obtain statistically valid results. The objective is to assist those investigators, undertaking health studies at local or district level, who lack detailed knowledge of statistical methodology. Acknowledging that the size of a sample will depend on the aims, nature, and scope of the study, the first part of the book provides a practical framework for working through the steps of sample size determination once a proposed study and its objectives have been clearly defined. In six chapters, readers are introduced to a variety of situations in which minimum sample size must be determined, including studies of population proportion, odds ratio, relative risk, and disease incidence. Each situation is first defined in terms of the information required and then illustrated by a hypothetical example of a study objective and the questions that will need to be answered in order to determine the appropriate sample size. The solution to each problem is clearly stated, together with relevant explanatory notes. The second part of the book features more than 50 pages of tables that enable the reader to determine the sample size required, under simple random sampling, in a given type of study without recourse to complicated calculations. Each of the illustrative examples featured in the first part of the book includes a reference to one of these tables, thus showing investigators how to move from the objectives and design of a study to the rapid calculation of an appropriate sample size. Examples and tables were selected as representing many of the approaches most likely to be adopted in health studies.
Cohen, Jacob. 1988. Statistical Power Analysis for the Behavioral Sciences. Lawrence Erlbaum Associates, Hillsdale, New Jersey. 567 pages.
In terms of programs which have "predigested" statistical power calculus, there is the software "PASS", but using that limits you in the number of different statistical tests they implemented in the software.
I tend to favor simulations in R, as it is easier to tailor things for you specific situation.
I suggest you the book of Julious S. "Sample Sizes for Clinical Trials" (see http://www.amazon.com/Sample-Clinical-Trials-Steven-Julious/dp/1584887397). There is also an available APP for IPAD (SampSize) based on Julious book (but I prefer to reproduce the estimations directly in R). I hope to be useful.
It depends on the type of experiment you are planning to to, most of the good books in statistics have at the end of the chapter some topics for the determination of the sample size. For the sample size determination you will need: a measure of the variance of the response variable, (which you can have fron previous research, or if not you can go ahead a run part of your research, for example with the control treatment); the magnitude of the difference that you expect could be important to be declared as significant; the formula for thr mean square error (MSE). In addition, you need to fix alpha to a convenient value, say for example 0.01 or 0.05; and you have to consider also beta, say for example 0.10 or 0.20.
There are several PDF files available in the www where you can find good information for the calaculation of the sample size, for any experimental design, based on the power. I had a class with Dr. F Martin, in design of experiments, whose notes covered a very nice subroutine for that purpose which could be adapted for any experimental design, by using the SAS statistical functions.
Sample size determination is a problem of common sence,. What happens if your calculations give you a sample size which make the experiment to be impossible of being carry out? . You know that we worry of the sample size, becasuse we want the standar errors for the differences between the means to be small. But we also know, that we can have small MSE, if we care about the experimental technique.
Here is a good material, on sample size determination.
PS on Vanderbilt site, G*power, and much more free software. SAS has it also. R has it. PASS has it. I wrote demo for my PhD students 15 years ago, and several of them made it even better. field is covered, plus very small percent of statisticians will ever need anything sofisticated out of the blue, so take your time.
Not really a cook book, but I would take a look at "Statistical Power Analysis for the Social and Behavioral Sciences" by Xiaofeng Steven Liu. It has examples or R, SPSS and SAS code for most of the calculations and covers a range of designs that aren't easily addressed in other software or books. It is also (in my opinion) clearly written.
[I reviewed the book for the publisher - and endorsed it; but I wouldn't have endorsed it if I didn't think it was rather good]
I found an article written by Ronán Conroy quite informative and having practical examples on sample size calculations.The title of the article is Sample size: A rough guide.