Hi,
I am doing a final year dissertation project and wondered if I could get some help understanding my results?
I have 2 predictor variables on one outcome variable and did a linear regression with one of the IVs and the DV to test my hypothesis and then added the other IV to the model to see the additional variance.
I did a spearmans correlation and found that the IV and DV have the effect size: 0.17 (which is positive?)
But the linear regression showed (B= .037, p>.05 )- which is positively related?
And once I added in the other variable the outcome showed (B= -.032, p>0.05).
I don't know what to report this as, if it is positively related or negatively related as none of them are significant but the correlation is positvive?
Many Thanks!!
Hello Emma,
Zero-order correlations (e.g., the correlation of a single IV with the DV) do not necessarily reflect what will happen by way of the sign of regression coefficients when you construct a model with two or more IVs. Among the usual suspects for reasons one can include: (a) multicollinearity; (b) suppressor effects; or (c) noise/error.
If your estimated regression coefficient of .03 is not statistically distinguishable from zero (as implied by your p > .05 summary), then you have to behave as if zero is a plausible value, and there is not a systematic linear relationship, regardless of the sign.
First, look at the zero-order correlations...are they distinguishable from zero (again, regardless of sign)? Is the overall regression model significant (e.g., is R-squared significantly different from zero)? If not, you can make the usual boiler-plate recommendations ("further research is needed," "larger sample sizes might be helpful", etc.), but the long and the short of it is that, based on your sample, your model appears not to account for a lot of variation in the DV your're investigating.
Good luck with your work!
Spearman correlation works with zero order correlation, which means the correlation between the 2 variables regardless of the effect of any other variable. in MLR, the first variable is entered by the statistical program with its zero order correlation value, but the 2nd is entered automatically in the partial correlation. which means part of its correlation [which is shared with the other IV with DV] is removed.
Also, the result will depend on which method of SPSS you used in regression. was it Simultaneous entry or Hierarchical entry?
Oh WOW. This must be stepwise regression in SPSS. I would never trust that with my dissertation research. See the attached for details. Good Luck, David Booth
- Badges
- Science topic
- Similar topics
- Medical Science Nursing
- Dissertations