0
$\begingroup$

I am running a regression model, and I need to delete outliers. However, when I ran a boxplot, it asked me to delete at least 100 datas (I only have 700 datas in total). Both y and x variables are right-skewed distribution.

boxplot: (y is total_cast)

boxplot(data)

enter image description here

I did not delete anything and tried to run the model first. Only 2 variables' p-values are strong significant with three stars, another 3 variables have weak significant with only one star. R-squared is only 0.5, which is not good enough.

so I checked the plot:

plot(cost_model, which = 2)

enter image description here

Does this mean I only have to delete 3 outliers in this data? From row 45, row 107 and row 147? I do not understand what these number means and how to delete them.

Improve this question
$\endgroup$
5
  • 1
    $\begingroup$ Why do you need to delete the outliers? // Why is $R^2\approx 0$ not good enough, because a $50\%$ is an $\text{F}$ grade in school? $\endgroup$
    – Dave
    Commented Dec 14, 2020 at 12:03
  • $\begingroup$ @Dave Hi! The professor wants to know how we deal with outliers, and since the R-squared is not high (close to 1), I want to delete the outliers to see if the model can be more ideal. $\endgroup$
    – Ching
    Commented Dec 14, 2020 at 12:21
  • $\begingroup$ Why do you think your $R^2$ value is low? // It sounds like you have just one predictor (independent) variable. Please post a scatter plot of your data (assuming nothing is proprietary or otherwise confidential). $\endgroup$
    – Dave
    Commented Dec 14, 2020 at 12:22
  • $\begingroup$ @Dave from the data I analyzed before, normally the models have at least 0.7 in R-squared. That's why I think 0.5 is a bit low. $\endgroup$
    – Ching
    Commented Dec 14, 2020 at 12:25
  • 3
    $\begingroup$ "I need to delete outliers" No, you don't. If Total_cost is your dependent variable, you most likely should transform it or use a GLM. $\endgroup$
    – Roland
    Commented Dec 14, 2020 at 15:41

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.