ordvar calculates measures of ordinal consensus and dispersion. These include lsq and 1-lsq, which are 0/1 normed ordinal consensus and dispersion statistics described in Blair and Lacy (2000). The ...latter statistic is essentially identical to the IOV measure of Berry and Mielke (1992). The non-normed version of the dispersion statistic, termed d-squared by Lacy (2006), is also calculated, as are delta-method standard errors for these statistics. These measures do not rely on assumptions about intercategory distances or distributional form.
r2o calculates the ordinal explained variation statistic (i.e., R-squared) described by Lacy (2006), which is used to summarize the fit of a regression model for an ordinal response. It rests on an ...ordinal variation measure that entails no assumptions about intercategory distances or distributional form. This measure is valid regardless of the method used to estimate the model, and was shown to outperform various pseudo-R-squared measures in estimating the value of the true R-squared for a regression model for an underlying continuous response, even though its sense does not require such. -r2o- is to be used after a relevant categorical response model has been run, while the e() list is still intact. By default, the program recognizes the following as relevant response models: -ologit-, -oprobit-, -mlogit-, and -gologit2-. However, -r2o- should work after any estimation command for which –predict- p1,..., pk will calculate predicted probabilities, and which follows official Stata's conventions for naming items in the e() list.
PDF