Short answer: in my opinion, yes you can still use Cohen's f^2. Longer answer: it depends on how conservative you are with your use of statistical techniques. Traditionally, linear regression is for only continuous variables from which you can derive R^2 and Cohen's f^2. Yet, in practice, it is commonplace to see ordinal variables (Likert, etc.) used as if they were continuous variables if the ordinal scale is large enough to allow for higher amounts response variation (I teach my students no less than 5 ordinal categories). Using ordinal variables as continuous then allows you to use them for techniques like linear regression and therefore you can use them to derive Cohen's f^2. Just be mindful that people reading your work might disagree with your application if they do not believe ordinal variables should be used as continuous variables.
Thanks. Glad you found the video useful. Power is not truly 1 as that would mean you have 100% power. If there is one thing about statistics, it is that we are not 100% always sure of anything. Having said that, it is possible to be very sure of a thing. So what is likely happening in your case is that R, or whatever program you're using, has a power of 0.9999 or similarly high value and is rounding up to 1. So depending on the level of detail required for your report, I would say something like having an extremely high power value, or having a power value greater than 0.99.
You have a power of 1? That is very high. The basic definition of power is the probability your analysis will be able to find a statistically significant different IF there is an actual difference that can be found. Your power of 1 equals a 100% chance, but like all things statistics, we're never 100% certain, so it's more like 99.9+%.
Sir I have only one model in which there is one d.v and two predictors so since there’s only one R2 value so effect size of both variables will be same?
Like r^2 (and using r^2 for the calculation), the f^2 regression effect size value is for the entire regression model--the effect of all predictor variables together on the DV. So it does not matter how many predictor variables you have. However, predictor variables that do a better job in the model should increase the f^2 value.
Hi sir. thankyou for this video. I am currently doing a multiple linear regression analysis (social science field) ang got an effect size of 0.61. I am confused because I havent found a source to indicate what a good effect size for multiple linear regression and luckily came upon your video regarding the general rules of thumb. Would it be ok to ask for that source? Thanks!
Thanks for enjoying my video. If you have an f^2 of 0.61, that's very large. Great question asking about source information--citations are very important. A commonly referenced paper is: Selya, A. S., Rose, J. S., Dierker, L. C., Hedeker, D., & Mermelstein, R. J. (2012). A Practical Guide to Calculating Cohen's f(2), a Measure of Local Effect Size, from PROC MIXED. Frontiers in psychology, 3, 111. doi.org/10.3389/fpsyg.2012.00111 If you want to go further back, the paper above cites Cohen's original work: Cohen, J. 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd Edition. Routledge Check around page 410 for f^2 effect sizes, which you might be able to access here: www.utstat.toronto.edu/~brunner/oldclass/378f16/readings/CohenPower.pdf
Excellent. You make hard stuff accessible.
Thanks very much, just what I needed to quickly wrap my head around this and practically apply it.
Many thanks. Glad it was helpful!
Great video! Thanks a lot.
Thank you for the video! I was wondering how I am able to report the Effect Size in my thesis as I am struggling with the structure.
You are most welcome! Glad I could help and good luck with your thesis.
@@ShawnJanzen Thank you. I was just wondering how I am able to report this in my results (multiple regression)? Do I just present the effect size?
Thank you, this is well explained.
Many thanks for the kind words.
sir, my independent and dependent variables both are based on seven point-type Likert scale..can I still report cohen f2 in my study?
Short answer: in my opinion, yes you can still use Cohen's f^2.
Longer answer: it depends on how conservative you are with your use of statistical techniques. Traditionally, linear regression is for only continuous variables from which you can derive R^2 and Cohen's f^2. Yet, in practice, it is commonplace to see ordinal variables (Likert, etc.) used as if they were continuous variables if the ordinal scale is large enough to allow for higher amounts response variation (I teach my students no less than 5 ordinal categories). Using ordinal variables as continuous then allows you to use them for techniques like linear regression and therefore you can use them to derive Cohen's f^2. Just be mindful that people reading your work might disagree with your application if they do not believe ordinal variables should be used as continuous variables.
Hi, a quick question - my power is showing as 1. How would be best to report this? Thanks for the informative video!
Thanks. Glad you found the video useful. Power is not truly 1 as that would mean you have 100% power. If there is one thing about statistics, it is that we are not 100% always sure of anything. Having said that, it is possible to be very sure of a thing. So what is likely happening in your case is that R, or whatever program you're using, has a power of 0.9999 or similarly high value and is rounding up to 1. So depending on the level of detail required for your report, I would say something like having an extremely high power value, or having a power value greater than 0.99.
@@ShawnJanzen OK, thanks so much for clearing that up!
Sir, what is the meaning of a multiple linear regression model hiving an extremely higher power (i. e., 1)? What does it say about the relationship?
You have a power of 1? That is very high. The basic definition of power is the probability your analysis will be able to find a statistically significant different IF there is an actual difference that can be found. Your power of 1 equals a 100% chance, but like all things statistics, we're never 100% certain, so it's more like 99.9+%.
@@ShawnJanzen Sir, thank you very much for your clear explanation of the output in my research.
I feel like I need to learn all my stats from now on from a dude in blacked-out glasses and a ZZ Top style beard haha that's awesome!
Sir I have only one model in which there is one d.v and two predictors so since there’s only one R2 value so effect size of both variables will be same?
Like r^2 (and using r^2 for the calculation), the f^2 regression effect size value is for the entire regression model--the effect of all predictor variables together on the DV. So it does not matter how many predictor variables you have. However, predictor variables that do a better job in the model should increase the f^2 value.
Hi sir. thankyou for this video. I am currently doing a multiple linear regression analysis (social science field) ang got an effect size of 0.61.
I am confused because I havent found a source to indicate what a good effect size for multiple linear regression and luckily came upon your video regarding the general rules of thumb. Would it be ok to ask for that source? Thanks!
Thanks for enjoying my video. If you have an f^2 of 0.61, that's very large.
Great question asking about source information--citations are very important. A commonly referenced paper is:
Selya, A. S., Rose, J. S., Dierker, L. C., Hedeker, D., & Mermelstein, R. J. (2012). A Practical Guide to Calculating Cohen's f(2), a Measure of Local Effect Size, from PROC MIXED. Frontiers in psychology, 3, 111. doi.org/10.3389/fpsyg.2012.00111
If you want to go further back, the paper above cites Cohen's original work:
Cohen, J. 1988. Statistical Power Analysis for the Behavioral Sciences, 2nd Edition. Routledge
Check around page 410 for f^2 effect sizes, which you might be able to access here: www.utstat.toronto.edu/~brunner/oldclass/378f16/readings/CohenPower.pdf
@@ShawnJanzen thankyou so much 💖