help on selecting a proper statistical test - (Jun/12/2006 )

Hi, I'm looking at a set of genes under 3 different treatment A, B and C. The purpose of the study is to find out genes that expressed differently in C from the AVERAGE of A and B. So it's different from the regular ANOVA test. Does anyone have suggestions on what test to do or what kind of manipulations are needed on the data set? Thanks.

QUOTE (elad @ Jun 13 2006, 12:36 AM)
Hi, I'm looking at a set of genes under 3 different treatment A, B and C. The purpose of the study is to find out genes that expressed differently in C from the AVERAGE of A and B. So it's different from the regular ANOVA test. Does anyone have suggestions on what test to do or what kind of manipulations are needed on the data set? Thanks.

Hi,

there are several things one should know before statistics: 1. what kind of data (frequencies, measurements, binary)? 2. Are data normal distributed and variances homogene (in case of measurements)? Check with box plots and tests (e.g. Levene, KS-test)
In the easiest case (data normal are distributed and variances are homogene, or they are after transformation) you can perform one way ANOVA to check if there are differences at all. Then you can make multiple comparisons. In your case a priori contrasts seem useful as you have already a hypothesis how the results may be i.e. c>or< (a+/2. These contrasts are also called orthogonal or linear contrasts and in many stat software implemented e.g. in SPSS under GLM oneway, contrasts or SAS (e.g. proc multtest). But I don't know if SPSS performs more complex contrasts, then SAS or R is better.
In case of other data its more difficult ;-)

-hobglobin-

Hi,

there are several things one should know before statistics: 1. what kind of data (frequencies, measurements, binary)? 2. Are data normal distributed and variances homogene (in case of measurements)? Check with box plots and tests (e.g. Levene, KS-test)
In the easiest case (data normal are distributed and variances are homogene, or they are after transformation) you can perform one way ANOVA to check if there are differences at all. Then you can make multiple comparisons. In your case a priori contrasts seem useful as you have already a hypothesis how the results may be i.e. c>or< (A+/2. These contrasts are also called orthogonal or linear contrasts and in many stat software implemented e.g. in SPSS under GLM oneway, contrasts or SAS (e.g. proc multtest). But I don't know if SPSS performs more complex contrasts, then SAS or R is better.
In case of other data its more difficult ;-)
[/quote]

Sorry the smiley is not disappearing, should be a "B" with brake.

-hobglobin-