When to test

This topic contains 2 replies, has 2 voices, and was last updated by  datasociety@mawapps.com 6 months, 1 week ago.

Viewing 3 posts - 1 through 3 (of 3 total)
  • Author
    Posts
  • #19517

    datasociety@mawapps.com
    Participant

    My Achievements

    Concept review passed! IconExercise set done! IconSection completed! IconAnswered a post IconPost published! IconFeedback given! IconExercise done! IconConcept review passed! IconQuiz completed IconSetup completed! Icon

    Hello Merav,

    If I find a good correlation (0.75) between X and Y, do I still need to test for autocorrelation in Y residuals? Do I need to also investigate outliers in this situation?

    Thank you!

    #19521

    Dmitri Adler
    Participant

    My Achievements

    Exercise set done! IconExercise set done! IconAnswered a post IconExercise done! IconQuiz completed IconSetup completed! Icon

    Hello,

    Yes, even if you have a good correlation, you should still test for autocorrelation and outliers. You want to make sure that you either didn’t find a spurious [false] correlation or that your correlation wouldn’t be even stronger without outliers or without autocorrelation. I always perform these 2 checks when I run regression analyses (as well as a few of the other ones you saw in the course).

    From what I’ve seen in practice, assuming that good results are necessarily right is a frequent pitfall of even experienced data scientists.

    I hope this helps!

    -Dmitri

    #19524

    datasociety@mawapps.com
    Participant

    My Achievements

    Concept review passed! IconExercise set done! IconSection completed! IconAnswered a post IconPost published! IconFeedback given! IconExercise done! IconConcept review passed! IconQuiz completed IconSetup completed! Icon

    Thanks, Dmitri,

    Yes, I’ve given a Simple Linear Regression example to a couple of Data Analysts, and they seem to skip the autocorrelation test, in fact, they seem to depend on X and Y correlation R-Squared, and charts only.

    A professor told me that I should find out why there is autocorrelation in the residuals. This makes sense but I am not really sure how to go about it given a domain I am not really familiar with.

    Mark

Viewing 3 posts - 1 through 3 (of 3 total)

You must be logged in to reply to this topic.