correlation with ordinal variables
correlation with ordinal variables
Hi, I am trying to run correlations with ordinal variables but I am not able to do this for some reason? It is possible to run correlations with ordinals isn't it? If so, how can I do this in jamovi?
Re: correlation with ordinal variables
Again, I would suggest setting the variables to "continuous" and running a non-parametric rank-based correlation (which is a type of ordinal correlation). Options are Spearman's Rho and Kendell's tau-b. Someone else previously suggested computing a polychoric (ordinal) correlation using jamovi's Seolmatrix module, but I see that Seolmatrix sometimes chooses not to return a result (though R's "polycor" package does). See attached screen shot.
- seolm-Untitled.png (99.84 KiB) Viewed 2591 times
Re: correlation with ordinal variables
Hey @bev,
I see @reason180 has already given you some great suggestions, which I wouldn't hesitate to follow, especially if you're on the jamovi cloud, where you currently can't take advantage of modules in the library, like you would on the jamovi desktop version.
So from the "Analysis > Regression > Correlation Matrix" ribbon you can get Spearman's Correlation and/or Kendall's tau correlation for your ordinal qualitative variables.
*Allow me to suggest a preference towards Kendall, especially if your sample is small because the estimate of the correlation index calculated with this index is more precise than Spearman's.
Cheers,
Maurizio
*Note (if useful):
The difference between the Spearman and Kendall indices lies in the formulas used to calculate them.
Spearman's correlation coefficient uses the same formula as Pearson's correlation, simply applying it to the ranks of the variables.
This formula is based on the deviations of the data from the mean values of the two variables.
For this reason, Spearman's correlation index value is usually more similar to Pearson's than Kendall's.
In fact, Kendall uses a totally different formula that is based on the agreement or disagreement between the pairs of observations.
Usually the value of the Kendall index is smaller than that of the Spearman index calculated on the same data.
However, this does not mean that it is less accurate, but only that the relationship is being evaluated from a different point of view.
I see @reason180 has already given you some great suggestions, which I wouldn't hesitate to follow, especially if you're on the jamovi cloud, where you currently can't take advantage of modules in the library, like you would on the jamovi desktop version.
So from the "Analysis > Regression > Correlation Matrix" ribbon you can get Spearman's Correlation and/or Kendall's tau correlation for your ordinal qualitative variables.
*Allow me to suggest a preference towards Kendall, especially if your sample is small because the estimate of the correlation index calculated with this index is more precise than Spearman's.
Cheers,
Maurizio
*Note (if useful):
The difference between the Spearman and Kendall indices lies in the formulas used to calculate them.
Spearman's correlation coefficient uses the same formula as Pearson's correlation, simply applying it to the ranks of the variables.
This formula is based on the deviations of the data from the mean values of the two variables.
For this reason, Spearman's correlation index value is usually more similar to Pearson's than Kendall's.
In fact, Kendall uses a totally different formula that is based on the agreement or disagreement between the pairs of observations.
Usually the value of the Kendall index is smaller than that of the Spearman index calculated on the same data.
However, this does not mean that it is less accurate, but only that the relationship is being evaluated from a different point of view.