7) The establishment-level EMSPS data in Table 2a suggest that unionized establishments are some 17% more likely to provide training for manual workers in the year preceding the survey; they are about 7% more likely to have provided training for non-manual workers.
In the EMSPS models, unionized establishments provide about 0.
Using the EMSPS data, for both probit models and, in the case of manual workers, for the Tobit model, we could not reject the hypothesis that the coefficients for the union and nonunion segments were equal.
With the EMSPS data, the result for the model of training intensity for non-manual workers was a marginal union impact of 0.
We are fortunate that the EMSPS survey contains details regarding training practices in British workplaces (see Dench [1993a, 1993b, 1993c] or Green, Machin, and Wilkinson  for more details).