Skip to 0 minutes and 15 secondsSo this is back to some of the things I had on the larger slide that had all that text on there that I wanted to provide you with some more detail. And there're like just a couple of things that you have to consider when you're picking these tests. So each group has a little bit of caveats with it. Now I already mentioned with tests they're meant to evaluate interval/ratio data are known as parametric tests, and they need that normal distribution. And statisticians as if just arbitrarily if you asked me, decided that it's at least 30 in each group.
Skip to 0 minutes and 46 secondsSo if they don't say that it's normally distributed or if they don't use a normally distributed assessment test to determine that it is. If you don't have more than 30 in each group, you're they're violating the use of a parametric test and should have chosen a non-parametric test. Now T test, as you can see they're primarily, T tests, ANOVA, And ANCOVA is another one that are often used. And we talked I talked about groups, two groups or two or more groups T test is the only one that has a limitation, it doesn't have use groups that uses comparisons. So you can only make one interval/ratio comparison in a study, otherwise, the overall study error rate becomes too large.
Skip to 1 minute and 26 secondsSo if you could think about all the articles that you've probably read, how many times did you have just one set of interval/ratio data? So that rarely happens. So T test is really not a good test to use in studies because it has a higher chance of that alpha error which remember is saying that something works when it doesn't. So using T test should be very limited in clinical trials. Now nonparametric tests, and I'll give you the names of the most common ones are known as Distribution-free, so you don't have to worry about the sample sizes. And that I already mentioned to you nonparametric tests which are used on ordinal data can analyze both interval/ratio and ordinal.
Skip to 2 minutes and 5 secondsAnd then, I don't know why they call them both non-parametric, but they're separate tests then that are also called nonparametric. They can handle nominal data types. So you can see I put some other little things in this caveat, but you'll see this is a snippet of a bigger table that I'll provide you. But you can see that we look at the type of experiment, the groups, remember I mentioned, it's independent or dependent for a type of experiment, so you divide them between those. The groups two or greater than two.
Skip to 2 minutes and 32 secondsBut you can see that I have some things in red that you also have to consider, Remember I mentioned to use interval/ratio statistics which are their parametric, you have to have at least 30 in each group. So that's why I gave you the red number 30 there to remind you of that. And for using T test, remember I said it's one comparison that they can do. Now there's also a fudge factor called Bonferroni rule that can get you up to six comparisons in a study, so that's why you'll see in our big table that we want to look to see if we have more than six comparisons.
Skip to 3 minutes and 4 secondsAnd then for ordinal, I listed, remember I told you that it's perfectly acceptable to use non-parametric tests on interval/ratio data, so they're listed there as well. The main statistical tests that are used are mann-whitney U, and then the exact same test it must been two different people or groups of people, so they had to name it after each other. One is Wilcoxon rank-sum and then the one for two mo groups is Kruskal- Wallis. And then you can see for nominal Fisher's exact is used for two groups, and it's more it's called an exact test, because it's exact whereas chi-square is estimation.
Skip to 3 minutes and 42 secondsSo Fisher's can use very small sample sizes, Chi-square cannot handle under 20 and it's a little iffy between 20 and 50. But Fisher's cannot handle more than two groups. Now you can see we also have our dependent tests on the other side which all those rules apply, except for chi-square has the size limitation whereas its dependent partner Cochrane Q does not. Now, I didn't do this but you can see, that under the ordinal you have Wilcoxon signed-rank.
Skip to 4 minutes and 9 secondsAnd you just said: Wait a minute.You just said that was for independent? Well, actually that's Wilcoxon rank-sum. So it wasn't me that did that. So just remember that since I always consider independent first, it's Wilcoxon rank-sum, R comes before S, in the signed rank. So just remember Wilcoxon rank-sum is for independent and Wilcoxon signed-rank is for dependent studies. And you'll notice that most studies are independent parallels, we're probably going to stay on that left-hand side, we're gonna rarely go to the right-hand side of this chart. But sometimes you can have both independent and dependent data within a study.
Skip to 4 minutes and 45 secondsBecause you can have it where you have two parallel groups, where you're comparing the endpoints to each one, as in the diabetes example and power that I used. But they also some types compare baseline to an endpoint within each group. So sometimes you may have to have both independent and dependent statistical tests within a study, so you're not just always picking for one piece of the buffet here. All right. So here's the four statistical test table that you'll be looking at and you can see that if you decided it's independent, this is the piece of the chart that you're going to be looking at. And if it's dependent, you're gonna pick from this group.
Skip to 5 minutes and 24 secondsBut again, remember, you may have some independent and some dependent tests within your studies. You're gonna have to differentiate and determine where it goes with each one. And then associations we're going to talk about that in a subsequent session. Okay. So now we've arrived. Here's this big table that I talked about at the beginning. So let's look and see about how we potentially fill this out. So here's an example that I'd like you to go through and if you want to just not stop this and just go along with my answers, and then go ahead and maybe fill it out with the next one that will be fine. But this was a superiority study that was double-blind parallel.
Skip to 6 minutes and 5 secondsYou may not get that in each one of them, but you can pretty much tell whether you're looking at before and after in each group or you looking at comparison each comparison to each group. This one is looking at ondansetron versus metoclopramide for nausea and vomiting and pregnancy. The primary outcome and they should distinguish what that is somewhere in this study. There're two of them was well-being on a ten-point scale and numeric rating of the VNRS scale. So as I mentioned, seems that's the primary outcome they should have two power calculations for those. And then their secondary outcomes as you can see what they were listed. And then I provided you with the results.
Skip to 6 minutes and 43 secondsSo what you're going to do is you need to go look for all of the data points. So you're going to look on to look in three different places for those. You're going to want to look in the baseline table, you're going to want to look in the outcomes tables, and you're going to want to look in the adverse reaction tables. To be able to locate all the different data points to ensure that there are statistical tests to cover each one of those.
Statistical Tests: Dependent and Independent
Prof. Mary Ferrill clarifies the differences between the parametric and nonparametric tests, which are also called distribution-free tests.
T tests are very limited in clinical trials. In addition, we can learn the classification of the statistical tests by the table showed in this video.
Also, we should notice that Wilcoxon rank-sum is for independent, and Wilcoxon signed-rank is for dependent studies.
Finally, we have to look for all of the data points in the example.
We have to look in three different places: the baseline table, the outcomes tables, and the adverse reaction tables. Thus, we are able to locate all the different data points to ensure that there are statistical tests to cover each one of those.
Please share your thoughts about this video below.