Hi friends,
I hope the summary in the subject line already makes my problem a little bit clear. I have (or rather in the future will have) two paired groups of observations. Within each observation there are two (paired) binary patterns, each consisting of a number (probably 5) dichotome outcomes (0/1). Or putting it another way: there are two (paired) series of 5 characteristics, each with dichotome results. The two patterns stem from different sources, but represent the same characteristics. In the ideal case they should be equal within an observation, but that is what actually has to be checked. How do I test, compare those 5 dichotome variables at once and how do I test whether they as a series (not individually) are equal or not (thus not unequal or not)?
I was thinking about making a binary number out of those separate (5) dichtome results and to do the testing on that number. But I fear such a combination of results in a number, though unique and containing all the info, does not show variations in the original qualities clearly and comparably enough. Another possibility would be to create some kind of compound index, e.g. the sum of all outcomes '1', but that would even discard detailed info in the original data.
After having established how to represent the two patterns, how to compare them (which statistical test), I have to do an initial (paired) sample size analysis. I assume I would have to set beta to .05 for that purpose and alpha to say .20. However, this all depends on the analysis methods to be choosen.
A little more elaboration on the data that I will have: one series of binary outcomes consists of (5) computer algorithm determined decisions on (5) different (given) aspects; the other series consists of corresponding decisions made by human experts. The decisions are being made in different situations, different subjects actually who are the observations, where the outcomes should vary largely and represent all possible situations for each characteristic. Human and machine made decisions should be compared, the human decisions being the reference.
Regards - Jim.
--
. . . . . . . . . . . . . . . .
Jim Groeneveld, MSc.
Biostatistician
Science Team
Vitatron B.V.
Meander 1051
6825 MJ Arnhem
Tel: +31/0 26 376 7365
Fax: +31/0 26 376 7305
[log in to unmask]
www.vitatron.com
[common disclaimer]
|