Thanks for this discussion. I have intuitively hesitated to treat 0.05 as a strict threshold but find the plethora of text books treating it as just that sometimes overwhelming (it can be hard to convince others of something that is widely stated). It is refreshing to hear the thoughts of others, such as yourself, on the topic. Much appreciated!
Thank you, Yury, for your educational video. In your other tutorials, you said don't use the Kruskal-Wallis test when the samples are big (n>30). Does this mean when all the combined samples are bigger than 30 or the sample number of each intervention(treatment group) is > 30? 2. You said you don't have to plot counts or proportions using box plot; use Bar Plots instead. What if you present the median counts with IQR because the data is non-parametric? Can you use bar plots and add the IQR or MAD? Thank you.
I am sorry for confusion. You can of coarse use KW for larger samples! Does not matter how large. The issue is in the normality of data. Big samples are often normally distributed, but the normality tests show significant deviation from normality. So, please, always plot density, QQ of the data and even better of the residuals. I use counts (categories) with bar plots, continuous data (e.g. 2.3) or numeric (3 or 19) are best plotted without bar plots, just my opinion. Yes, mean + CIs or Medians + IQR are perfect. Then, the bar is not needed, but some colleagues still like to put them for more dramatic visual expirience :)
Thanks for this discussion. I have intuitively hesitated to treat 0.05 as a strict threshold but find the plethora of text books treating it as just that sometimes overwhelming (it can be hard to convince others of something that is widely stated). It is refreshing to hear the thoughts of others, such as yourself, on the topic. Much appreciated!
Yeah, I fight with 0.05 often, because it is dogmatically used. Thanks for your feedback and thanks for watching!
🙏👍💪😎
thanks a lot!
Thank god, u update!
More to come!
Thank you, Yury, for your educational video.
In your other tutorials, you said don't use the Kruskal-Wallis test when the samples are big (n>30).
Does this mean when all the combined samples are bigger than 30 or the sample number of each intervention(treatment group) is > 30?
2. You said you don't have to plot counts or proportions using box plot; use Bar Plots instead. What if you present the median counts with IQR because the data is non-parametric? Can you use bar plots and add the IQR or MAD?
Thank you.
I am sorry for confusion. You can of coarse use KW for larger samples! Does not matter how large. The issue is in the normality of data. Big samples are often normally distributed, but the normality tests show significant deviation from normality. So, please, always plot density, QQ of the data and even better of the residuals. I use counts (categories) with bar plots, continuous data (e.g. 2.3) or numeric (3 or 19) are best plotted without bar plots, just my opinion. Yes, mean + CIs or Medians + IQR are perfect. Then, the bar is not needed, but some colleagues still like to put them for more dramatic visual expirience :)
@@yuzaR-Data-Science Thanks for the clarification
you are welcome! @@ibrahimlawan9663
P-value is the probability of Null Hypothesis being true.
That's one of the intuitive but not entirely correct definitions. P-value is the probability of data, not (any) hypothesis. Cheers
From now I will stop worrying about how the car works 🤪😉