Why normalization of variable is needed?

[Prince Rahman]

Hello,
I'm new in stata. My question is, why should we normalize data/variables? And what issue might occur if we don't normalize the data and run into the program. Example of such variables is GDP, tax revenue, etc.

Thanks in advance for your help

[Carlo Lazzaro]

Prince:
welcome to this forum.
If you refer to linear regression, please note that normality is a (weak) requirement for residual distribution only.

[Nick Cox]

Normalization to machine learning people can mean some kind of standardization or scaling such as

(value - min) / (max - min)

or

(value - mean) / SD

It can mean bringing a value closer to a normal distribution, which is sometimes impossible and often difficult, even roughly.

[Prince Rahman]

Prince:
welcome to this forum.
If you refer to linear regression, please note that normality is a (weak) requirement for residual distribution only.

Thank you sir, but if i deal with panel data regression, is it necessary then?

[Prince Rahman]

Normalization to machine learning people can mean some kind of standardization or scaling such as

(value - min) / (max - min)

or

(value - mean) / SD

It can mean bringing a value closer to a normal distribution, which is sometimes impossible and often difficult, even roughly.

Thank you sir, I tried to normalize some macro economic factors. But i failed, then I used the variable as it is. Can you please tell me, if winsorisation is useful or not?

[Carlo Lazzaro]

Prince:
what above holds for panel data regression, too.

[Nick Cox]

It's hard to follow #5 as despite the explanation that "normalization" has different meanings, you have not indicated which one you follow.

But the aim of Winsorizing is not to create a normal distribution if that is your goal.

[Maxence Morlet]

Thank you sir, I tried to normalize some macro economic factors. But i failed, then I used the variable as it is. Can you please tell me, if winsorisation is useful or not?

Prince Rahman
Check out this great post by Dr. Cox on winsorization: https://www.statalist.org/forums/for...orize-the-data

[Jared Greathouse]

I like to think I'm a panel data researcher. The only time I've had to do normalization of any kind is when I wrote a Stata command implementing the basic code for this monster of a paper. In essence, you do PCA followed by LASSO/Ridge/OLS regression to make a counterfactual, and before you do PCA (a form of singular value decomposition), you normalize your predictors from -1 to 1.

But that's the only time, that's it.

[Prince Rahman]

Prince:
what above holds for panel data regression, too.

Thank you sir for your advice

[Prince Rahman]

It's hard to follow #5 as despite the explanation that "normalization" has different meanings, you have not indicated which one you follow.

But the aim of Winsorizing is not to create a normal distribution if that is your goal.

Thank you sir. i got this now

[Prince Rahman]

Prince Rahman
Check out this great post by Dr. Cox on winsorization: https://www.statalist.org/forums/for...orize-the-data

Thank you so much for this valuable resource

[Prince Rahman]

I like to think I'm a panel data researcher. The only time I've had to do normalization of any kind is when I wrote a Stata command implementing the basic code for this monster of a paper. In essence, you do PCA followed by LASSO/Ridge/OLS regression to make a counterfactual, and before you do PCA (a form of singular value decomposition), you normalize your predictors from -1 to 1.

But that's the only time, that's it.

Thank you sir for your valuable advice

[Carlo Lazzaro]

Prince:
please call me Carlo, just like all on (and many more off) this list do. Thanks.