*The first approach, being a kind of compendium model presently exercised by all UNESCO world reports with the exceptions of the World Education Report and the World Culture Report, is a major activity of the Organization and should be maintained through a reporting mechanism at longer intervals (e.g. every four to six years), whereas the second approach demands an appropriate timing as well as addressee, such as the General Conference of UNESCO, at two-year-intervals. unesdoc.unesco.org*

# Comment faites-vous tirages au sort

There is a little-known phenomenon for binomial GLMs that was pointed out by Hauck & Donner (1977: JASA 72:851-3). The standard errors and t values derive from the Wald approximation to the log-likelihood, obtained by expanding the log-likelihood in a second-order Taylor expansion at the maximum likelihood estimates. If there are some \hat\beta_i which are large, the curvature of the log-likelihood at \hat{\vec{\beta}} can be much less than near \beta_i = 0, and so the Wald approximation underestimates the change in log-likelihood on setting \beta_i = 0. This happens in such a way that as |\hat\beta_i| \to \infty, the t statistic tends to zero. Thus highly significant coefficients according to the likelihood ratio test may have non-significant t ratios.