1

Could you recommend me a book or scientific article that explains the interpretation of the results of a Spearman correlation analysis (SC)? For example, I need a reference in the literature that says:

SC = 0.5 is weak
SC < 0.7 is average
SC > 0.8 is strong
SC > 0.9 is very strong

I need it for a research paper, then to be something more "serious" than a web site.

[UPDATED after whuber´s help]: To be more specific, I'm using SC to evaluate an Information Retrieval System. I need to "correlate" the Rank result of my system against a rank calculated by some specialist.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Igor Veloso
  • 121
  • 5
  • 2
    There is no such thing in general. Possibly within a narrow subset of literature on a particular kind of data for a particular subject there will exist such guidance. But when you think about it, you should conclude this isn't an interpretation at all: it's just providing qualitative names to an otherwise quantitative result. That seems like a step backwards to me. An interpretation would relate the correlation, in the context of its standard error, to some question of interest in your research. – whuber Feb 25 '16 at 22:24
  • 1
    In many areas of social studies correlation of 0.5 would be considered as actually quite strong and 0.7 as pretty marvelous... Another counterexample: https://stats.stackexchange.com/questions/132536/how-to-choose-a-confidence-level/132538#132538 If depends on your field and on your data if some correlation can be considered as strong, or weak, there is no "objectively strong" values for it. – Tim Feb 26 '16 at 09:10
  • I agree with the previous comments... I would look in Cohen, J. 1988, Statistical power analysis for the behavioral sciences, 2nd ed. This is a popular book, but unfortunately I don't have it in hand. Cohen also explains that there is no absolute interpretation for the magnitude of effect sizes, but he gives "rules of thumb" interpretations. I think for *r*, he goes with 0.1 = small, 0.3 = medium, and 0.5 = large. I presume he would apply the same to *rho*... Better would be to find published results in your field that you can compare your values of *rho*. – Sal Mangiafico Aug 26 '17 at 21:15

0 Answers0