We have had three very interesting and very different papers this morning. So to be fair, I thought that I would raise for discussion three issues that are problems for both the data collectors and the transmission modellers.

The first problem is how to classify individuals into risk groups on the basis of their reported sexual behaviour. An individual's sexual behaviour is not a stable, easily measurable characteristic like gender or age. Furthermore, it cannot be viewed in isolation, because it involves the formation of partnerships and consequently is affected by the behaviour of others in the population. It may be extremely difficult to decide how to characterize an individual on the basis of reported risk behaviour at any one time; for example, should this be regarded as a constant or as a random variable (so that the reported behaviour at any moment is an observation on an underlying stochastic process)? Deciding upon a suitable classification scheme may require an analysis of the variability of the risk behaviour of each individual over a long period of time. It may be more appropriate to characterize this by averaging over a period of several years, rather than a particular, short time interval. It has been shown that the behaviour of individuals with high levels of sexual activity tends to be highly variable over time. Hence if individuals are characterised by their behaviour in a short time interval, then low activity individuals will generally be correctly classified (because of their low variability), but high-activity individuals may be misclassified.