What it says! The catch is that there are so many of them and that they have a relationship with dissimilarity/difference/distance coefficients and indices (doh!)
Details #
Well they are what the name suggests and they are related to correlation coefficients (q.v.), in fact, a fair summary is that correlation coefficients are similarity coefficients for continous measures or measures with a fair few clearly ordered levels. Similarity coefficients are what they say for the remaining sets of variables: binaries (including “yes”/”no”), very short ordinal variables and categorical variables. It might seem that there would only be one or a very small selection but the reality is that there are very many ways of understanding what is “similarity” (and dissimilarity) for such data and there are coefficients that simply describe the observed relationships in the data and others that are estimates from the data of population relationships. The latter usually give statistical significance tests, p-values, for the level of similarity and confidence intervals around the observed value, the former may allow creation of confidence intervals around the observed values by bootstrapping (or other methods) giving some indication of the precision of estimation, i.e. the precision of reasonable generalisation from the data.
These coefficients are important in inter-rater reliability/agreement measurement and underpin multidimensional scaling (MDS).
Try also #
- Bootstrapping & bootstrap methods
- Cohen’s kappa
- Confidence intervals
- Correlation
- Distance measures/metrics
- Estimation
- Inter-rater reliability/agreement
- Multi-dimensional scaling
Chapters #
Not covered in the OMbook.
Online resources #
I can see this is too big to cover well here, definitely needs at least one Rblog post and perhaps some shiny apps. If I ever get the time!
Dates #
First created 4.ii.25.
