I am currently trying to undertake a linear regression analysis in SPSS and I have come across some difficulties. My data does not show linearity and therefore is violated. I have transformed both predictor and outcome variables numerous ways but this does not change the outcome of linearity. I have read the next step could be to do a nonlinear regression analysis. Is this correct? If so, how it this performed in SPSS? Anyone know of any good websites that explain this?

  • 3
    $\begingroup$ Can you clarify what you mean by "My data does not show linearity"? Depending on what you mean, it may (or may not) be a violation of the assumptions of your model. $\endgroup$
    – Macro
    Sep 25 '13 at 14:25
  • 2
    $\begingroup$ SPSS makes this much harder to do than R or Stata, but a regression spline is one of the best general purpose solutions. As a side question what made you pay $ for statistical software? $\endgroup$ Sep 25 '13 at 15:05
  • 4
    $\begingroup$ @FrankHarrell As much as I dislike the business strategy of SPSS, I have to frankly say that it's relatively learner-friendly: interface is simple and functions are well categorized. Plus some lecturers/teachers specify the software to go with their class, it does not have to be Lauren's own decision to pick up SPSS. $\endgroup$ Sep 25 '13 at 15:30
  • 1
    $\begingroup$ I would hazard to say that it is learner-friendly only because it is restrictive and doesn't implement best statistical practice. If it doesn't make regression splines easy then watch out. $\endgroup$ Sep 25 '13 at 21:21
  • $\begingroup$ There's no cut-and dried answer. If you're serious about learning to handle a variety of situations like this, you'll want to get familiar with the literature on transformations, exploratory data analysis, and ways to use linear regression flexibly (e.g., making use of partial plots; squared or interaction terms; etc.). Eventually you may want to look into more advanced topics such as regression splines. $\endgroup$
    – rolando2
    Sep 28 '13 at 14:50

Depending on what you mean by linear (as asked by @Macro), you could do a polynomial regression. I'm not familiar with SPSS, but you could create a variable called x2 that is simply x squared, then include that term in a multiple regression model.

A quick search showed that in SPSS you don't even have to create an additional variable, but could check an option to have SPSS model a quadratic term: http://www.ucdenver.edu/academics/colleges/nursing/Documents/PDF/NonlinearExampleHowTo.pdf


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.