Levels of variation and intraclass correlation
HTML-код
- Опубликовано: 22 авг 2024
- When working with multiple data sets, it is important to analyze the level at which individual observations vary. This can be done by calculating intraclass correlation (ICC) to quantify the variances. Using profitability data as an example, there are three levels of variation that can affect a data set: year-to-year variation within companies, variation between companies, and variation between industries.
In a small data set, graphical analysis can be used to identify patterns, while in larger data sets with manageable numbers of clusters, box plots can be used to understand between and within variances. The ICC(1) is calculated as the variance between groups divided by the total variance, helping to determine how much of the variation in the data is attributed to the groups and how much is within the groups. ICC(1) values close to 0 indicate no meaningful clustering, and values close to 1 indicate no variance within clusters. In cases where ICC(1) is around 0.5, multi-level modeling may be needed to account for the levels in the analysis.
Link to the slides: osf.io/fjumc
Very clear explanation, Mr. Rönkkö. Please continue making videos, you're good at this.
Thank you so much for your clarification, I am being helped organizing the concepts again by your explanation!!
You are welcome!
Fantastic video. Very good examples and overall a clear presentation style. Thank you!
You are welcome!
I am struggling in understanding multilevel modeling used in the paper and your videos really help me out with good example and clear explanation. Thanks for the amazing video
Great to hear!
Like the others here I appreciate the clarity. Thanks for making this video.
You are welcome!
Thank you so much for your explanation, sir. This video was very helpful!
You are welcome
Brilliant! You were very clear. Where can I find out about the other types of ICC?
Great video! Many thanks
You are welcome.
Great explanation, thank you for posting
You are welcome
would it be correct to say this? "multilevel models deal with separate distributions within larger, more general distributions, to the nth number of levels" thank you for these videos, they are finally helping me understand multilevel modelling!!
Yes, you could say that.
You are amazing! Thank you
You're so welcome!
Hi, great video, finally I got it!
But I think there may be a mistake in the video. Shouldn't the means on the plot at 7:25 be all at the same level?
@@mronkko I see now. Thanks for the explanation.
So, an ICC of 0.2 (for example) indicates most of the variation is likely coming from the variance within groups?
Correct.
Hi Mikko at @ 7:31 you say there is not variance between groups. Do you mean there is total variance as at at ICC1 = 1 it says no variance on the slide.
Mikko Rönkkö ah ok that help clear it up a bit thanks
Thank you Mikko, It is really clear. Could you explain a little bit about the relationship between ICC and repeatability?
Define repeatability.
@@mronkko almost one year no answer, I'm going to answer anyway. Repeatability means how many probability of occurance of an observed value. Correct or no correct, I have no idea as I'm still learning. Now, how about ICC sir?
@@KangJangkrik ICC1 is defined at about 6:30
Thank you for a nice video! When I calculate the variance of the 5 group means I dont get 0.00033, I get: var(c(0.220,0.183,0.222,0.220,0.236)) = 0.00039. How is the 0.00033 calculated?
I guess I found out, since each group has 5 observations; var(rep(c(0.220,0.183,0.222,0.220,0.236),5)) gives the right answer.
@@christoffer1769 Right. It is not the variance of groups but variance attributed to the groups in the 25 observations.
@@mronkko I see, thank you :)
Hi Mikko, at 6:04 how are you calculating the within variance from the group mean centred data, I can't seem to work out how you get that value?
It is just the variance of the group mean centred data. In R, you would do
var(c(-0.022, 0.025, 0.014, -0.041, 0.024, 0.006, 0.033, -0.022, -0.043, 0.027, -0.007, -0.002, 0.033, -0.009, -0.014, 0.038, 0.004, 0.037, -0.052, -0.026, -0.013, 0.016, 0.002, 0.035, -0.041))
If you want to do it by hand, square all the values, sum the squares, and then divide by 24 (N-1).
Should not the intraclass correlation be (within groups) and not between groups?
I am not sure if I understand your comment. ICC measures the similarity of observations within groups compared to the overall variation. If ICC is high, there is very little within group variation and most of the variation is between group. So you cannot really define ICC without referring to both levels of variation. I hope this helps.
@@mronkko This is a helpful explanation to something I have misunderstood for a while. Thank you.