Correlation: Ranks. |
|
For correlation analyses, the first step is to generate a scatterplot to get a sense of the data and the differences present.
There are a few measures of correlation that are used for categorical variables; here, we describe one that is commonly used.
There are a few measures of correlation that are used for categorical variables; here, we describe one that is commonly used.
Spearman's correlation coefficient (Rho)
This metric describes the strength of the relationship between 2 sets of data; this correlation does not evaluate agreement between both datasets. Note that this metric is generally not recommended over other, more robust metrics like kappa scores.
Assumptions:
Pros:
How to use:
When analyzing in R, the "cor.test" function in base R can generate Spearman's correlation coefficient by specifying method = 'spearman.'
Spearman's correlation coefficient is a number between -1 and 1, typically associated with a p-value. Values closer to -1 indicate a strong negative relationship, while those closer to 1 indicate a strong positive relationship. Values closer 0 indicate no or weak relationship. Cutoffs for what is deemed "acceptable" vary by researchers and disciplines. Values above 0.6 are typically considered a "strong" relationship, while those above 0.8 are "very strong".
Assumptions:
- The data are ordinal
Pros:
- Extreme outliers are not likely to provide undue influence
- Can use with non-normal data
- Can describe consistency between observers
- Not a measurement of agreement, instead implies there’s a relationship or association; sets of observations may be highly correlated and consistent between individuals yet have low agreement
- Insensitive for detecting systematic differences (e.g. consistent overestimation) in values between observers because it does not incorporate slope or intercept differences like regression analysis does
- The p-value associated with the correlation can indicate a statistically significant, yet weak, relationship and therefore not support strong consistency between observers
How to use:
When analyzing in R, the "cor.test" function in base R can generate Spearman's correlation coefficient by specifying method = 'spearman.'
Spearman's correlation coefficient is a number between -1 and 1, typically associated with a p-value. Values closer to -1 indicate a strong negative relationship, while those closer to 1 indicate a strong positive relationship. Values closer 0 indicate no or weak relationship. Cutoffs for what is deemed "acceptable" vary by researchers and disciplines. Values above 0.6 are typically considered a "strong" relationship, while those above 0.8 are "very strong".
