I'm aiming to run a linear regression for some data so I'm testing assumptions. I have a few scatterplots that I'm finding it hard to read. It looks somewhat linear but I can't tell whether they violate the assumption of linearity or not. Dependent variable is on the Y axis in all examples.

enter image description here


The relationship between your variables seems to be very weak. There is possibly a linear relationship in the first plot but I'm very unsure about the other two.

It would he really good if you could give us some more context. What are the x values representing? What's the response? What values does it take (I just see 0s on the side)?

| cite | improve this answer | |
  • $\begingroup$ Hi there, thanks for your response - the Y axis DV is an interval variable (the values are cut out, sorry about that, but e.g. 1.00, 2.00, 3.00 going up the Y axis). The X axes show other demographic and questionnaire variables (in the middle graph that is gender, where 1 = female, 2 = male, 3 = prefer not to say, etc.). $\endgroup$ – Efflim Jun 11 at 19:02
  • 1
    $\begingroup$ What actually is the DV? Can you explain it in words? Also since gender doesn't have an order (I.e. male>female>other doesnt make sense) then Y will not have a linear relationship with 1,2,3. This particular variable should be treated as a factor on your model rather than a continuous variable $\endgroup$ – jcken Jun 11 at 19:07
  • $\begingroup$ The DV is scores on a measure of wellbeing. I'm assessing gender and the other variables in this way only to see if they satisfy the assumptions for linear regression, which include that the relationship between IVs and DV is linear - so I see why they aren't producing a linear relationship, but not how to satisfy the linearity assumption in this case, if that makes sense? $\endgroup$ – Efflim Jun 11 at 20:00
  • $\begingroup$ I think a linear regression isn't the most appropriate thing here. If your DV is on a likert scale then you should try an Ordinal Logistic Regression. If it is an unbounded integer then try a Poisson regression $\endgroup$ – jcken Jun 11 at 20:02
  • $\begingroup$ Thanks for your response - it is not a Likert scale, and I have seen many linear regression analyses with this DV (interval), so I would hope it would be appropriate here? $\endgroup$ – Efflim Jun 11 at 20:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.