0

I'm running an OLS and was wondering if the 'Estimate' in my SPSS output is the same as the beta coefficient in a linear regression?

Are there specific assumptions required to run an OLS? I have age, gender and income as my predictor variables and quality perception as dependent variable (measured on a Likert scale). Is OLS appropriate here?

Thank for your help, I really appreciate it!!

Mary
  • 13
  • 3
  • Hi Mary please look at this link for an example of linear regression in SPSS (https://stats.idre.ucla.edu/spss/output/regression-analysis/). To be absolutely correct, you would need to run a censored regression model. But if this is for a preliminary statistics class, then I'm guessing the teacher would not mind as much. To be safe you can run the censored model (example here: https://stats.idre.ucla.edu/stata/dae/tobit-analysis/) and present both results. The issue with running OLS is that outside of the observed levels (1-5 etc.) the prediction equation does not make sense. Hope that helps. – Prometheus May 10 '19 at 21:22
  • The "Estimate" column does in fact show estimated beta coefficients. Your dependent variable is an ordinal variable, not interval/continuous. OLS assumptions (independent, identically distributed errors) are violated with this type of variable, and ordered logistic regression is a good candidate here. – AlexK May 10 '19 at 21:33
  • 1
    This is probably the most widely discussed topic on our site: please see https://stats.stackexchange.com/search?q=regression+assumption*+score%3A5. Although I severely limited this search, it still turns up over a thousand promising hits, so consider refining the search to meet your specific needs. – whuber May 10 '19 at 21:39
  • @AlexK I disagree somewhat here. While it is true that some fields abhor the use of Anova or linear regression in the case of Likert scale dependent variables, you will find others (eg. psychology, also I am not in psych) that accept their use. – Prometheus May 10 '19 at 22:19
  • Thank you to everyone! However, if i run a linear regression with my data, i need to check for normality of residuals and heteroskedasticity. But the residual plot only shows vertical lines...what does that mean? does it even make sense to test these assumptions regarding the nature of my data?! – Mary May 11 '19 at 05:32

1 Answers1

0

You should look at the Gauss Markov Assumptions which underlay regression. Things to check for include heteroscedasticty, non-linearity, non-normality, and multicolinearity (these are not the only assumptions, but the ones you can generally catch). A place to start is Regression Diagnostics by John Fox. That is a very old monograph now but worth reviewing as a starting point.

user54285
  • 751
  • 4
  • 9