Linear regression - case wise diagnostics - how to remove outliers?

Comments

• EJ

  July 2021

  Hi Laura,

  This is probably not a mistake. When you filter out 48, this reduces the variance, and this may cause another outlier to be identified. I am not sure this is problematic -- there are a number of ways in which you can detect outliers, and it seems prudent to try a number of different ones and confirm that your general conclusion is robust.

  Cheers,

  E.J.
• Sameha

  September 2021

  Hi EJ,

  Apologize if I am supposed to open a new topic, I was not sure if need to open a new discussion or ask in this dicussion page since it is on the same topic.

  
  

  Here is my question, for detecting outliers before performing multiple regression, when I use boxplot in the descriptive analysis, 2 of my independent variables have few outliers. However, when I use z-score (following the below video at minute 8:05), the z-score minimum and maximum are within the range (this is without removing outliers detected by Boxplot.

  https://teachpsychscience.org/index.php/video-how-to-do-simple-linear-regression-in-jas/709/research-methods/

  
  

  Is this expected since each method works differently (Boxplot based on median, z-score based on mean and SD)?

  Since regresstion is sensitive to outliers, would it be valid to use z-score only as detector of outliers without using boxplot?

  
  

  Best Regards,

  Sameha
• EJ

  September 2021

  Yes, I'd go with the z-scores.

  E.J.
• EJ

  September 2021

  ...although you could remove the outliers as a robustness check.
• Sameha

  September 2021

  Hi EJ,

  
  

  Thank you very much. Your answers have been helpful for my studies.

  It's appreciated!

  
  

  Best Regards,

  Sameha