Wow so many replies I can hardly keep up. Thanks for the great help all!
Hey everyone!
I had a few questions regarding certain statistical methods I am researching. I've found a peer-reviewed journal article dealing with the menstrual cycle and it's affects on the renal-adrenal and hemodynamic responses during orthostasis in POTS patients. The link to the article can be found here:
http://hyper.ahajournals.org/content...51787.DC1.html
And I've also attached a copy of the article in this post.
My question is regarding the specific tests used for the statistical analysis of data. The methodology includes two groups, POTS patients and a control. Both groups are females and both were tested during two phases of the menstrual cycle (the mid-luteal phase and the early-follicular phase).
The specific tests used were the Man-Whitney rank-sum test, the Wilcoxon Signed rank test as well as a Holm-Sidak method for post-hoc comparisons.
I would like to know the specific reasons why these test were used if anyone would be able to clarify the differences between the tests and their function in statistical analysis.
Thank you all!
(And I hope I was descriptive enough).
Chris
Wow so many replies I can hardly keep up. Thanks for the great help all!
Don't go bananas! These are nonparametric tests used when data aren't normally distributed. The first is the analog to the two sample a ttest. So comparing a continuous variable between two groups. Second test is analog to one sample ttest. So comparing a continuous variable to a pres -specified number - typically "0".
Nobody gets paid to help you, so you get what you paid for, thus responses may be slower than if you consulted a paid professional.
Bananas (11-09-2015)
Hey, bananas - I wanted to make sure you knew we cared. I opened the article and appears the "Holm-Sidak method" was most likely used after the ANOVA procedure. You see the ANOVA will just tell you there is a difference. You then have to go back and perform pairwise comparisons between values to find the differences. Yeah, you could just run them like normal, but to honest to your a priori established level of significance (i.e., 0.05), you have to correct this level of alpha when making multiple comparisons, because by chance you may find a statistically significant result because of all of the comparisons. So you are controlling for your potential for an increased false discovery rate. So if I said on average my heart rate was higher than yours, well cool we can test that. However, if I keep walking around your hospital, I will eventually find someone with a resting heart rate higher than mine and this could be because of chance and not my hypothesis.
Bananas (11-09-2015)
hlsmith,
You actually explained things in a way to someone that doesn't understand statistics in a way that is understandable. Thank you for your help... honestly! Question though, compared to the Holm-Sidak method, when would we use the Scheffe method? This was what we were taught in univariate analysis but I know they are both pairwise tests performed after we have found a significant difference to compare the groups.
Thanks for the compliment - its probably because that I am totally applied statistics and have pretty much no theoretical skills.
There are dozens of corrections. Ideally you select one a priori, so you don't go fishing for significance. They all are about the same. You are probably fine as long as you select one that is not too esoteric. I can't recall the differences between those two you listed, but they are fairly common corrections. I was thinking the sidak-holm ranks the p-values and has a graduated level of cut-off, getting harder and harder for bigger p-values to be significant, which help show the smallest p-value is significant. Though, I could be mis-remembering. I would imagine there are papers or webpages that contrast them.
The method you see the most often based on its simplicity is the Bonferroni correction. With the Bonferroni you just divide your level of significance (alpha, which is typically 0.05), by the total number of possible pairwise comparisons. So if you had three groups you would do the following:
0.05/3 = 0.017; your new alpha cut-off or
you could just multiply your p-value by 3 and it would get the same result. The Bonferroni is known for being a little over cautious (AKA, harder to show a significant difference).
We can't answer this question for you, since it can be field and preference based. Though, I would say just do a little reading and look at their formulae. I am sure some have perks here and there.
Bananas (11-09-2015)
That's great information for me. Thank you so much! I do believe Holm-Sidak uses a ranking system like you said and then you would perform the analysis until you're unable to detect a difference between the groups. I believe I've found enough information to answer my question. I'll be discussing some other questions with my professor tomorrow. Thanks again for all of your help hlsmith. You're great for shedding light into the darkness!
Tweet |