Today, I’ve updated the Tutorial on Correlation Coefficients in order to add a new section on correlation strength scales. I feel this is granted.
In a 7/16/2010 Search Engine Watch post, a search marketer reported an r value of 0.67 and stated:
“If 1 is perfectly correlative, then 0.67 is certainly a strong correlative relationship and a figure of some interest, when we consider there are a couple hundred factors that reportedly contribute to rank.” (http://searchenginewatch.com/3641002).
This raises the question on how to characterize correlation strengths. Several attempts have been made at classifying r values as ‘weak’, ‘poor’, ‘moderate’, ‘strong’, or ‘very strong’ using scales of correlations.
The problem with these scales is that their boundaries are often defined using subjective arguments, not to mention that not all researchers agree with using such boundaries or scales at all.
Feel free to read the updated version now.
BTW, during the updating process I found an involuntary missing “not” in one line of example 3 in the tutorial. It should read as follows: “The difference between r1 and r2 is not significant at the 95% confidence level.” This should be obvious from reading the null hypothesis. My mistake, nevertheless.
Another correction made was in the Olkin-Pratt formula for bias. In this case, there is a missing parenthesis. Instead of 2n – 3, this should be written as 2(n -3). The parenthesis was originally included in the Excel program used to draw the graphs. So the graphs were not affected.
In the future and if I can find the time, I may add an exercise section applied to IR and search engines.