Magnitude of Coefficient Values and its interpretation



This is regarding the inference from estimate values in regression.

The general inference of coefficients is the amount of variation on dependent variable with unit change in the predictor variable.

But what i am looking for?

Do the magnitude of estimate values/coefficients has an implicit meaning other than variation it creates?

Ex: In my model, certain significant variables estimate values are 0.007,0.002…


If you have a significant variable with very low coefficient value, it just means that its scale is some orders of magnitude bigger than the target variable. If you normalize the data, coefficient values will be more meaningful in terms of importance evaluation.


So, the normalization should be done in data preparation itself (Reduce the predictor variables to DV’s scale)


Yes, the features should be scaled in the preprocessing step. Also note that this won’t affect the model predictions, only the coefficients will be rescaled.


I have a doubt in scaling…this was on my mind frm long time and also askd in interview…

Is it possible to fix a range of scaling?

For example : 0 to 1, 0 to 5 , 0 to 10…

Frm my understanding we apply scaling based on transformations using log, sqrt… then using formulas like z-standardization and min max approach…

pls clarify this…


@ssvbalan The multiple ranges you mention are just a real value multiplication of 0-1 range.

For example, if you want a range of 0-5, you would

  • convert the data into range 0-1
  • multiply all the values by 5


@ssvbalan, to bound the data within a given range, just apply the min max scaler and then multiply by the upper bound desired, as mentioned by @jalFaizy. You don’t need to apply any other kind of transformations if all you want is to get the data within a range of values.


Thanks for the input…