18th May 2012 at 2:02 pm #2400Raivo JutupliiatsMember
Hello, i’m a Bachelor’s student, writing my thesis and i’m in need of some advice.
I research factors that affect e-shopping (buying from internet shops). I want to do a correlation analasys between dependent and independent variables. My dependent variable has scale data (number of purchases) and my independent variables are ordinal data (Likert 5 scale questions). The independent variables are not normally distributed nor are they linear (not 100% shure about unlinearity). I use SPSS and have done correlation tables using Spearman, Kendall Tau and Pearson’s coefficient, but i dont know what results i can trust. Ive read that Spearman is the most suitable considering my ordinal (Likert 5scale) data and lack of normal distribution and linearity.
Aditional info: My sample size is 557, the correlation test showed statistically significant relations with logical direction ( + ; – ), but most of the results were very weak, under 0,3. Only a few variables had reasonable correlation with the dependent variable.
So my questions are:
2) wich correlation coefficient to use and why ?
3) Can i still use results in my thesis, that are statistically significant and logical, but the correlation coefficients value is just under 0,3 ?12th June 2012 at 12:59 pm #240329th June 2012 at 7:33 pm #2402Vanessa ArcillaMember
In using correlational analyses, you’re more of checking the association of your variables, not entirely that one factor (IV) influences or affects the other (DV). Weak correlations may mean that you may have to be cautious in reporting your results, especially for a big sample size that you have.
This site might be helpful. http://www.socialresearchmethods.net30th June 2012 at 10:15 am #2401Ingo RohlfingMember
The proper correlation depends on the measurement level of both variables. In your analysis, Spearman’s rho is the proper coefficient (Thomas Black has a nice table on page 161 in his SAGE book Understanding Social Research).
In general, it holds that the larger the sample size, the smaller the correlation coefficient needs to be in order to be significant. When doing a correlation analysis, you usually want a high and significant coefficient. But it is up to you to decide what is high in light of your research and standards of your field of research. Recall that the squared correlation coefficient is the variance of the dependent variable that is captured by the independent variable (presuming that you can determine what is independent and dependent in your analysis). If your correlation is 0.30, this means that the R squared is 0.09.
- You must be logged in to reply to this topic.