1
$\begingroup$

I have a question regarding Ordinal data. I have measured 'Attitude towards a product' using Likert scale and I have measured 'Purchase Intentions to buy the product' also using Likert scale. Now I have to check "if Attitude influences Purchase Intention". Given both the response as well as independent variables are Ordinal, which is the best test to analyze my hypothesis? Thanks! If the question has already been answered, it would be great if you could point me out in the direction.

$\endgroup$
  • $\begingroup$ Your data are either ordinal or on a Likert scale. Those are not the same thing. Look up what a Likert scale is. $\endgroup$ – John Jul 31 '15 at 9:22
  • $\begingroup$ Thanks for your reply! I read in a post (bit.ly/1De562B) that when you rank your Likert scale, it can be treated as ordinal. I should have mentioned that I would rank the scale perhaps. That is why I framed my question in that manner. But the crux of the problem is how do I find if Attitude (on a Likert scale) influences Purchase Intentions (on a Likert scale). Thanks again and pardon my ignorance. $\endgroup$ – user83783 Jul 31 '15 at 11:30
  • 1
    $\begingroup$ The link you sent me, and subsequent sublinks, especially the most endorsed one are terribly problematic. See en.wikipedia.org/wiki/Likert_scale. Consider that when you sum many ordinal scales together that the resulting ordinal problems might get averaged out or minimized so much that they don't matter very much. But all I'm really asking is that you word your question in recognition of the difference between a Likert scale that collects many ordinal value together and rating scale which considers only one. Also, the range of the scale matters here. $\endgroup$ – John Aug 2 '15 at 0:39
  • 1
    $\begingroup$ For problems like these you are very unlikely to reach an incorrect conclusion If you ignore the ordinal/interval distinction and use Pearson's r. Note that the distinction between ordinal and interval was developed long after the Pearson correlation. Pearson's r makes no measurement assumptions per se but the interpretation of it, of course, depends on measurement. However, in the real world, r works well for very non-interval data. $\endgroup$ – David Lane Feb 7 '17 at 1:09
1
$\begingroup$

If one of your variables is ordinal, you may want to consider using a Spearman Rank correlation. Based on your description I assume that there were no experimental manipulations in your study. Thus, I think it would be best to be very careful in making conclusions concerning causality ("influences"), as you cannot rule out that there is, for instance, a third variable that is responsible for the association you observed (cf. moderators/mediators).

$\endgroup$

Your Answer

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