This is a nice video explaining clearly what a suppression effect is. Well worth 7 minutes! Thank you.
@mronkko
2 жыл бұрын
I am happy that you found it useful.
@dizzyboxnine2656
Жыл бұрын
nice video :)
@mronkko
Жыл бұрын
you are welcome!
@kiethuynh5882
2 жыл бұрын
Great explanation!
@mronkko
2 жыл бұрын
Glad it was helpful!
@Marunius
3 жыл бұрын
Beautiful analysis, keep it up bro!
@mronkko
3 жыл бұрын
Glad you liked it!
@user-rm2hk2ry1s
2 жыл бұрын
Thank you!That was really helpful! Can you give me an example of how to explain it though? Like what does it say about the relation in study?
@mronkko
2 жыл бұрын
I think that I give an example in the presentation: Doctor age, tenure, and patient satisfaction. More generally, a correlation that has a different sign than regression coefficient indicates that there is a third variable causing a spurious association. You just need to look at the correlation table to figure out which variable (or variables) that is.
@laxmanpokhrel5153
3 жыл бұрын
Your entire video list is incredible. I have a question why the mean and SD value of gender is calculated? Should it not be mode or variance value?
@mronkko
3 жыл бұрын
Mode does not tell us about the dispersion of the variable, so I do not think that it can substitute for SD or variance. We use SD instead of variance because SD is on the original metric and easier to understand than the variance, which is on the squared metric. Because gender is a binary variable (has a Bernoulli distribution), the SD is completely determined by the mean. But if you need variance, it is easy enough to just take as square of the SD. The SD is given by sqrt(0.38(1-0.38)) = 0.49
@manimaydev6125
2 жыл бұрын
Very well explained. Just one doubt, should we explain the effect of Age using correlation (positive) or regression coeff. (negative)? or should we drop Age (as Tenure is already present)?
@mronkko
2 жыл бұрын
That depends on the research question. In this case, age should definitely be kept in the model.
@janedoe-ud8ii
2 жыл бұрын
Should supression variables be removed from the model? I have seen many researchers said they should when i google supressor variable? Should suppressor variables always be included in the model? Why are they saying they should be removed?
@mronkko
2 жыл бұрын
No, for two reasons. First, models are built based on theory not based on empirical results. That is, you choose the variables before running the model. (At least if you are testing a theory. For predictive modeling where the interest is on calculating likely future outcomes things are a bit different.) Second, "no suppression" is not a regression assumption. It is a result. As for "Why are they saying they should be removed", I do not know who "they" are so I cannot speculate on their motivation. But more generally, if I find a recommendation that is not justified, to me it just counts as a unsubstantiated opinion. And we should not base our decisions on opinions but on methodological justification. This probably falls into the same category of recommendations that "multicollinearity is a problem that should be fixed" does. This video might be helpful kzitem.info/news/bejne/rXqc2HmbqJyYhKQ
Пікірлер: 17