Re-open the ‘fail xls’ spreadsheet for modelling the probability of MSc failure and do the following:

(a) Take the country code series and construct separate dummy variables for each country. Re-run the probit and logit regression above with all of the other variables plus the country dummy variables. Set up the regression so that the UK becomes the reference point against which the effect on failure rate in other countries is measured. Is there evidence that any countries have significantly higher or lower probabilities of failure than the UK, holding all other factors in the model constant? In the case of the logit model, use the approach given in box 11.1 to evaluate the differences in failure rates between the UK and each other country.

(b) Suppose that a fellow researcher suggests that there may be a non-linear relationship between the probability of failure and the age of the student. Estimate a probit model with all of the same variables as above plus an additional one to test this. Is there indeed any evidence of such a nonlinear relationship?

AppendixLO1