Limitation of linear regression model application

linear_regression
r
predictive_model
regression
statistics

#1

Hi, all recently working on linear regression model , i like it too much but there was one doubt what are the limitation of this model and if there is limitation then how to identify by seeing the stats parameter like t or p value or coefficient .

well i am very bad in story mode and too lazy person so i would like to proceed in points mode :smile:

> if identifying linear regression model validation limitaion then tell me on following perspective:

  • R studio -> is there any command argument to check linear regression before applying on data set

    statsistics mean is there any parameter to check like t value, p value anything


#2

Hi
before to answer you should specify which type of linear regression you want to work with. As you mentioned t test you are speaking about parametric, is it the only type you want to work with ? No cox or quantile
If you use t-test then you should have normal distribution or the results will not be as good.
But the golden rule, if you should have a linear relation in any case between your predictors and the response, so back to residual.
Alain


#3

well sir my question was posted in general , sometime i feel without using libraries and using proper data visualization also we can pdo prediction .
like considering the parametrs of statistics.


#4

You may want to think about what you’re doing before you model. There are always assumptions you should check. For linear regression in R, there are commands to check assumptions, but first you need to know what the assumptions are.

A good place to start might be https://en.wikipedia.org/wiki/Linear_regression. Especially look at the assumptions heading.

Of course, there are times when you can (almost) safely ignore some assumptions, which is why it’s important to understand why an individual model works.


#5

Hi Xtremcurius,

As mentioned you can test your assumptions using the statistics, the game could be tedious for example if you have high kurtosis on a data set with out visualisation you will start to calculate the mean, range, Inter quality range to get a first feeling then the mad and to calculate the outliers and see if some really totally out of range. To decide to remove some perhaps and restart to a better understanding

You do a boxplot and one histogram and then you have all and I believe with better grasp on data set but some people good not be agree on this point in this forum.

Hope the help.

Alain


#6

yes man you were right sometime old school is strong but irritating .