in g power - sample size for F test family, ANOVA being the stats. test , under determination of effect size - sd with in each group how to calculate for 4 groups ?, my point is we have the means for all four groups but how come the standard deviation with each group can be calculated - a single value which has to be entered ....
The real problem with ANOVA is that it starts with a vague question, that some kind of difference exists between the means. Its vagueness suggests a lack of a good hypothesis based on theory. The use of post-hoc tests increases the problem rather than solving it. Now the investigator is using a p-value from a vague (and often theoretically meaningless) test with multiple degrees of freedom to justify carrying out individual comparisons and to use these as the basis for theory.
The whole logic of hypothesis testing is turned upside down by the sequence of ANOVA-post-hoc-tests.
A far better hypothesis has one or more tests, each with a single degree of freedom, derived a priori from theory.
So I'd advise you to be bold: hypothesise now, rather than being forced into post-hoc hypothesising driven by whatever pairwise comparisons are significant.
Are you attempting to determine the effect size observed within your data or to determine the power based on the current N of your study?
For the first there are many ways to determine effect size from your data through various statistical packages.
actually i want to estimate sample size for a comparative study of four groups with continuous data -- tried g power software --- used f test family, anova with fixed effects, omnibus, one way under statistical set, now my query is regarding determination of effect size - when we go to enter parameters to estimate effect size .... means are fine for each group and N but how to compute SD within each group .... kindly suggest ...
Is really the ANOVA answering your research question? Often, the research question is for differences between groups, what is adressed by a series of t-tests (call them post-hoc if you like). If so, the interesting effect is a mean difference, the degrees of freedom for the test is the total sample size minus the number of groups, and the power calulation should consider adjustments for multiple correction, if you want to control the family-wise error-rate (rather than the test-wise error rate).
definitely post hoc will be yielding my final results ... but for finalizing sample size for my research via g power under test family t test doesnt fit as per my knowledge so ANOVA i took ....... My main query is that while using the determining effect size for this - SD within each group dialog box - what to put in ....
If you are only concerned with the effect size you can input your observed effect size from previous experiments in your or other labs. It sounds like you already have data from at least one experiment so you should be able to use that effect size as a starting point at the very least.
However, as Jochen mentioned you should also consider the impact that a series of t-tests has on error rates.
i have pilot study results wth me ...... mean and sd for all four groups .... the question still remains the same EFFECT SIZE determination ............ "at least one experiment so you should be able to use that effect size as a starting point at the very least." -- whoch effect size you are talking about i cant understand ...
You should make your knowledge fit your problem, not the solution fit your knowledge...
You might consult a local statistician to help you.
The real problem with ANOVA is that it starts with a vague question, that some kind of difference exists between the means. Its vagueness suggests a lack of a good hypothesis based on theory. The use of post-hoc tests increases the problem rather than solving it. Now the investigator is using a p-value from a vague (and often theoretically meaningless) test with multiple degrees of freedom to justify carrying out individual comparisons and to use these as the basis for theory.
The whole logic of hypothesis testing is turned upside down by the sequence of ANOVA-post-hoc-tests.
A far better hypothesis has one or more tests, each with a single degree of freedom, derived a priori from theory.
So I'd advise you to be bold: hypothesise now, rather than being forced into post-hoc hypothesising driven by whatever pairwise comparisons are significant.
Did you get your answer? I am facing the exact same problem? How to get one SD representing 3 groups (in my case).
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size