For eg. I have obtained three coefficient values from my OLS regression 0.85, 0.75, 0.63. Now I have to rank them in order of their magnitude. So is there any statistical technique for doing it, or we can simply rank them on the normal value basis? Please suggests.
Dear Tanuj Mathur,
You have already three values (and you know its magnitude). So, for the purpose of ranking them in order of their magnitude, statistical technique has not much role to play here.
If you want to assess the sensitivity/influence of individual variables, the following links will be useful to you.
https://statistics.laerd.com/stata-tutorials/multiple-regression-using-stata.php
http://www.ats.ucla.edu/stat/stata/dae/logit.htm
I hope this helps.
All the best !
Ranking in order of magnitude isn't a statistical question; the biggest coefficient is the biggest and the smallest coefficient is the smallest. There is a question of whether you would want to use un-scaled coefficients (which are not meaningfully comparable if they are on different scales), or scaled coefficients, or d effect sizes (i.e., the coefficient divided by its standard error). But in any of those cases, ranking from largest to smallest is straightforward.
Hi Tanuj,
to understand it right: we are talking about three regression coefficients (0.85, 0.75 and 0.63) of three different predictor variables in your OLS multiple regression model? And you want to rank them to show which variable of these three has the greatest impact on the criterion variable and so on?
Then I would suggest to use the standardized regression coefficients (beta coefficients) to rank them, because they are not influenced by the metric on which the variables are measured. Furthermore, you have to consider the method of entry you used (sequential and so on), if you want to talk about the relative influence of the variables.
Best,
Brian
Dear Tanuj Mathur,
You have already three values (and you know its magnitude). So, for the purpose of ranking them in order of their magnitude, statistical technique has not much role to play here.
If you want to assess the sensitivity/influence of individual variables, the following links will be useful to you.
https://statistics.laerd.com/stata-tutorials/multiple-regression-using-stata.php
http://www.ats.ucla.edu/stat/stata/dae/logit.htm
I hope this helps.
All the best !
What is mentioned above is correct in that you simply rank them by their size. However, you can only do that with a) standardized regression coefficients or b) if your coefficients are on the same scale. In your example, if .85 was for an IV like weight and .75 was for height, they are not on the same scale and could not be compared. If you used the standardized output (which most stats programs give), then you can say weight was more important than height. Or, if the three variables were on the same scale, then I think using unstandardized would be permissible (although in a lot of cases, measures are on different scales). The other advantage of standardized is that they can be compared across studies with different variables. Best of luck
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques