BioForum -> Bioinformatics and Biostatistics -> Spearman Rank vs Pearson Correlation Coefficient

### watsonkatherine

Posted 09 August 2013 - 10:24 AM

I have observational data and want to do a correlation test to determine the relationship.

For one variable I have a hygiene scoring of dairy cattle of various farms, so this data is obviously non parametric and ordinal since I ranked the cows from 1-4

For the other variable I have TPC (total plate count x1000 CFU/mL) of the bulk milk tank for that farm during the period I scored the cattle. So this data is parametric and interval or ratio.

Question - do I still use the Spearman Rank Correlation Coefficient to analyze my data? I'm just confused since one data set is non parametric and the other is parametric.

### hobglobin

Posted 09 August 2013 - 11:02 AM

what you should look at is the distribution of the data i.e. if they have e.g. a normal distribution or not. As at least the TPC are count data it would be more a Poisson distribution which would need e.g. a non-parametric test method (or any other test that can deal with this distribution). So I'd go with Spearman's Rank Correlation Coefficient as it's a non-parametric method also for discrete variables whereas Pearson's correlation coeff. needs at least one normal distributed variable and tests only a linear correlation if my memory is right.

### watsonkatherine

Posted 12 August 2013 - 11:48 AM

Thanks for your response - I went ahead and did a Spearman Rank Correlation Coefficient test instead of a Pearson in order to decrease my type I error, so I treated all the data sets as non parametric.

### Mikhail Shugay

Posted 15 August 2013 - 11:16 AM

You can also use Kruskall-Wallis test (it is non-parametric 1-way ANOVA) if you have grouped data by hygiene scoring

©1999-2013 Protocol Online, All rights reserved.