# Assign weights to indepedent variables in multiple regression

#### IoannaChr

##### New Member
I have performed my multiple regression using 6 explanatory variables and I would like to find the weights (as %) that should be assigned to each one of them based on their importance to my depedent variable (y). How do I do that?

#### obh

##### Active Member
Do you mean what IV influences more the DV?
You may look at the standardized coefficient.

#### IoannaChr

##### New Member
yes and then to actually assign a weight to it for example lets say that the first indepedent variable should have 10% weight etc.

#### obh

##### Active Member
So you may try the standardized coefficients.

#### IoannaChr

##### New Member
thank you! Since I am doing it in excel do you know if it is possible to do it there? If not where would you recommend me to do it?

#### obh

##### Active Member
Hi Loanna,

I checked in excel and you don't get it directly in the results.
But you may just standardize your data before running the regression.

You may use the following calculator, it will give you also the standardized coefficients.
And also the R code in the output, if you prefer to run it in R.
But in are you will need also to scale it. just add scale for the DV and IVs.

for example:
lm(formula = scale(Y) ~ scale(X1) + scale(X2), data = df1)

http://www.statskingdom.com/410multi_linear_regression.html

#### obh

##### Active Member
Ps standardized coefficient is one option, but you may decide to choose another way.
For example, if x1 varies in your data between 10 to 12, you may decide that your data for this parameter doesn't represent the full scale,
and decide to take the range of 5 to 20.

So just don't work automatically ...

#### hlsmith

##### Less is more. Stay pure. Stay poor.
I wouldn't try to create weights, I would just use their coefficients. Cuz, where you can get in trouble is X1 has coefficient 1.3 and SE 0.5 and X2 has coefficient 1.4 and SE 0.65, which is better? Also, I believe standardization gets tricky when your independent variables include continuous and categorical variables, how are you going to standardize those in a uniform manner? Plus, if you are creating weights, I am guessing you will try to generalize them to new data, which can be misleading too, if you have not tuned them based on holdout data. I usually just report coefficients with SEs along with partial R^2 with confidence intervals.

#### obh

##### Active Member
I wouldn't try to create weights, I would just use their coefficients. Cuz, where you can get in trouble is X1 has coefficient 1.3 and SE 0.5 and X2 has coefficient 1.4 and SE 0.65, which is better? Also, I believe standardization gets tricky when your independent variables include continuous and categorical variables, how are you going to standardize those in a uniform manner? Plus, if you are creating weights, I am guessing you will try to generalize them to new data, which can be misleading too, if you have not tuned them based on holdout data. I usually just report coefficients with SEs along with partial R^2 with confidence intervals.
Hi Hl,

Clearly there is no one perfect solution, and if you don't need weights, there is no need to use the standardized coefficients.
Under the assumption that for some reason you need weights, then the standardized coefficients method is one valid option.
But no method replace the common sense ...

#### IoannaChr

##### New Member
I wouldn't try to create weights, I would just use their coefficients. Cuz, where you can get in trouble is X1 has coefficient 1.3 and SE 0.5 and X2 has coefficient 1.4 and SE 0.65, which is better? Also, I believe standardization gets tricky when your independent variables include continuous and categorical variables, how are you going to standardize those in a uniform manner? Plus, if you are creating weights, I am guessing you will try to generalize them to new data, which can be misleading too, if you have not tuned them based on holdout data. I usually just report coefficients with SEs along with partial R^2 with confidence intervals.
Yes I see. But indeed as obh says, I need the weights for my calculations so I cannot skip them. My goal is to find the appropriate weights for each factor actually but I should do it based on the regression analysis that's why I am working like this