R demo | Friedman Test | How to Conduct, Visualise and Interpret

[@eddieNyakan]

You make STAT seems easy... Great presentation

[@yuzaR-Data-Science]

Glad you think so, Eddy!

[@SUNILYADAV-tv5ze]

Nice explanation Sir 👍

[@yuzaR-Data-Science]

Thanks and welcome, Sunil!

[@Lina-hw8gu]

Hi! thank you for this video! I was wondering if you could give me advice on how I should report my results. My research looks at Likert-Scale responses (ordinal values 1-5) on the perceived effect of 4 different compositions (/treatment group) of music. How can I find out which composition of music had the "biggest" perceived effect? I tried comparing median ranks to post-hoc results, however, many pairs did not demonstrate statistical significance though they had big difference in median ranks. Not sure if my question is clear but your advice would be very much appreciated

[@yuzaR-Data-Science]

:) the question isn't really clear, but I'll suggest, you have not enough observations per group. 4? with not much of a data you can see big differences, but still no significance, because the test isn't confident enough

[@John-xi2im]

its not known to everyone that universities create their own packages and store them in their own labs, and to use certain package cited in a journal, phd students from different universities connect with each other and share those package infos, thereby skipping the sharing process with general audience. Hence it is not easy to install all the libraries for general audience (like ggstatsplot could not get installed on my pc). Hence python is the preferred open source language because whatever is written in python, is available through pip and is always installed, unlike R.

[@yuzaR-Data-Science]

it's totally fine to use python instead of R! And I think it's a matter of choice. I tried python several times, but always got back to R, because I had similar problems in installation of packages ;) The installation of R packages always went smoothly for me. So, if you wish, I can try to help to install ggstatsplot if you provide your error message ;) Cheers and thanks for the feedback!

[@hansmeiser6078]

What is the difference between Friedmann and Kruskal-Wallis-Test, and when to use which? I know there is paired/ordered and there is sampled data involved. In R we can generate both types of datasets.

[@hansmeiser6078]

Update: I made a spread-sheet with all tests. Seems to me, the data-source itself dictates ordered vs. sampled approach.

[@yuzaR-Data-Science]

it's important to understand the concept of repeated measures. this determines what test do we take. how the data looks is only relevant for the software we use. but understanding the difference between dependent (paired/repeated measures) and not dependent data is key

[@yuzaR-Data-Science]

again, there is no ordered or samples approach. the data needs to be ordered when it is dependent. this is much clearer if we had an "ID" columns, where software would know that some of the points came from the same individuals. mixed-effects models use it perfectly. but before you confuse yourself, I would rekommend to understand those simple tests and learn a bit of stats theory, like types of data - dependent vs independent

[@hansmeiser6078]

@@yuzaR-Data-Science I did so with this: kzbin.info/www/bejne/gmPTk35qdretf5o Dependent looks like panel-data-analysis. But it's still a little bit difficult, to choose the right one between both. For example, theoretically I could use both in the same case. Compare rmse of multiple groups (undependent) AND compare rmseRank of multiple groups, repeated in dedicated time-intervals (dependent). Which test would you *trust more,* perhaps in regard of there output? Is there a specific paradigm involved?

[@yuzaR-Data-Science]

I really recommend you to read a bit more on dependent vs. not-dependent observations. The hypothesis and type of data decide then what test is right. Never trust one test over the other if you are not sure, which you should take.

[@lananss]

Hello, I have a problem in ggbetweenstats, when I get the graph the bars that exemplify p-value do not appear, and the test says p=5.03e-03. can you help me ?

[@yuzaR-Data-Science]

Hey Lana, sure, that can happen, because pairwise tests are corrected for multiple comparisons, so, they may be higher than 0.05 threshold. However, you can force the ggbetweenstats to show all the pairwise comparisons and tell it not to adjust for multiple comparisons (I do not recommend than however). For that simply type ?ggbetweenstats in the console of R studio and check out the arguments. cheers

[@SUNILYADAV-tv5ze]

Please make video for Bayesian and survival. Which is my interest area. Thanks for this video

[@yuzaR-Data-Science]

I already have two video on survival analysis for the very beginners... but did it over one year ago, so the quality of video and sound is worse than recent videos. Could be useful however. Bayesian analysis is part of parametric methods, like paired t-test, or repeated measures anova. In the future I'll try to cover more bayesian stats for sure, because I am also interested in it

[@SUNILYADAV-tv5ze]

Thank you so much Sir . I have seen that lectures those are very nice.