If you did the discriminant analysis already, with inferential tests (parametric or non-parametric, just something to give you a p value or similar to get you beyond fixed-effects), and if you detected an effect then the answer is you detected an effect. Work backwards (people sometimes hate this). G*Power should just spit a number out at you for this. Make a conservative estimate with f at a small effect size (e.g., 0.1) to find out the minimum N to detect a signal. Use G*Power with MANOVA (just as you did). There are two answers for your particular problem. In part because multivariate techniques have been argued to 1) stay stable across varying sample sizes 1 and 2) multivariate approaches boost power as compared to univariate 2. The short answer is - it's not very well established. Power computations for multivariate-especially those based on components analysis (eigen/singular value decomposition) or maximum likelihood estimation-are heavily argued.Ī number of arguments boil down to anecdotal evidence, rules of thumb (see factor analyses - many rules of thumb such as Guilford, 1954, Cattell, 1978, Gorusch, 1983), and empirical examples. Here is my arsenal of responses that should help you. I've had to make this argument before for a publication. There are several tests that fit what I'm doing, but I'm lacking confidence in selection. Now I'm stuck on the Statistical test menu. It might come down to the fact that GPower is an inappropriate tool in this case, but I don't know of another for multivariate. So now I've used every test possible (it feels) and I have no idea where to start calculating the number of participants in G Power. The problem is, with discriminant analysis, I am doing a MANOVA, then I calculate the R 2 and T 2 values, and then the univariate F. I am trying to use G*Power to determine appropriate sample size as I am required to use a tool by my committee. I am doing a discriminant analysis and need to justify my sample size. So now I'm left with no experience, no help, and Chapter 3 of my dissertation staring at me laughing. My problem is summed up with the fact that all my stats classes are theory and practice is suppose to come from your department, but my department only does qualitative- except for me. I'm looking for some help on trying to figure out what I need to use and maybe you all have a better tool than what I know. 10 ) is that does not tell us the importance of the effect, but we can measure the size of the effect in a standardized way.I'm kind of out on my own trying to review 3 year old concepts and use it in a way that I never have. The problem with the significance (whether is. This is the opposite of the probability that a given test will not find an effect assuming that one exists in the population, which, as we have seen, is the β-level (i. “The power of a test is the probability that a given test will find an effect assuming that one exists in the population. This means that if we took 100 samples (in which the effect exists) we will fail to detect the effect in 20 of those samples. The most common acceptable probability of this error is. The opposite (or false negative) is when we believe that there is no effect where in reality there is. Type I and Type II Errors A Type I error (or false positive) is when we believe that there is a genuine effect when it is not. There are many tools and tables to calculate the effect size. 57) Effect is very important because in addition to our test being significant, we can test "how significant' is the effect. Many measures of effect size have been proposed, the most common of which are Cohen's d, Pearson's correlation coefficient r and the odds ratio" (Field, 2009, p. The fact that the measure is standardized just means that we can compare effect sizes across different studies that have measured different variables. About effect size: An effect size is simply an objective and (usually) standardized measure of the magnitude of observed effect. Also, the specific tests to be performed play a role in this calculation (For example factor analysis). The size, the power, and the effect are intimately related. Calculating Sample Size Common Scenario on Proposals on URM (Pre QRM) or Statistic Classes: “I am conducting a correlational design and my chosen sample size is 25 subject” (no explanations provided) My typical answer: The sample size is something that we cannot just arbitrarily select, but must calculated based on our type of tests, the expected power, and the expected effect.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |