When conducting Pearson correlation analysis for variables with severely skewed distributions, do using Fisher r-to-z transformation and bootstrapping have similar outcomes (answers)?
May not be the same. Bootstrap method again depends on central limit theorem, so normality may be assumed in calculation of the correlation, whereas the fisher z transformation may give different results. check it with your data, so that the difference may be observed and test the significance also.
Thanks Dr Naveen, I have already checked it with my data. The sample size and the magnitude of r played a role in the outcomes. Bootstrapping may be better and more robust than fisher z transformation. The problem is "when Bootstrapping results (based on C.I) contradict the empirical reality " while Fisher’s supports it. I think it is not acceptable to generalize from just one or two examples, thats why I asked this question.
Please visit:
Class-Specific Correlations of Gene Expressions: Identification and Their Effects on Clustering Analyses
Comparing correlations: independent and dependent (overlapping or non-overlapping)
Best:
George
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2495058/
https://seriousstats.wordpress.com/2012/02/05/comparing-correlations/
Your results may or may not be the same as the two procedures depend on the original distribution of your data. Using z, the process depends on the law of large numbers (Central Limits Theorem) to achieve a normal sampling distribution. In bootstrapping, you are resampling from YOUR distribution to create a sampling distribution to test your parameter. Since bootstrapping is based on the distribution of your sample, it may be better if your sample is stable.
Yes. Bootstrap CI sometimes give wider range as expected. please check it with simulation study by taking different sample sizes. it may give you an Idea. I suggest you to use bootstrap method (robust).
- Badges
- Science topic
- Similar topics
- Psychology