Pearson product Correlation

[Becky88]

If I am looking for a relationship between several different variables (personality characteristic, attachment, eating pathology) does it matter that they have all been measured on different scales?

For example - some give a score out of 20, some of the sub scales give scores out of 5 and so on. Is there anything I need to do before conducting the analysis?

 

[Miner]

This sounds like it is ordinal data. You should try a rank correlation instead.

 

[Becky88]

Hi there,

Thanks very much for your response. I wonder if I can pick your brain further.

I am actually extending a study which looks at the relationship between attachment and eating pathology, and later will be looking at anxiety and depression as mediators.

The study I am extending has used Pearson product correlation to look at the relationship between attachment and eating pathology, so I would rather use the same method.

Would you not recommend this?

 

[noetsi]

It is common to use pearson with all data because the people doing so were not aware of the limitations of this. But Pearson's assumes interval data and thus will generate errors when you use it with ordinal data. Noting that someone used an incorrect method, and why, and then using a better method is done in journals. You could compare the results you get in spearman's (or polychoric) correlations with the previous results using pearson's (using the original data) to demonstrate the issue. Then use your own data if it is different.

You can use the previus method. It simply won't be entirely accurate. How inaccurate it is is hard to say.