In order to set quotas for the catch of commercial fish species such as cod, complex population dynamics models have been developed. An essential input to these models is the age distribution of the fish caught every year. This is estimated from fish sampled from boats as they land their catch. The major problem with this data is that although the number of fish sampled is large, there is a very large within-boat correlation, and the number of boats sampled is relatively small. Also the age distribution changes in space and time, and with the type of gear used by the boats.

We analyse the data by developing a hierarchical model. We consider the numbers of fish at each age from an individual boat to have a multinomial distribution, the parameters of which are themselves random variables. These variables are correlated in space and time, and between different fishing gears. This analysis allows us not only to estimate the age distribution in areas where there are few or no samples, but also allows us to give a measure of the uncertainty in the estimation.

The model is applicable to several fish species. Improvements of the model includes simultaneous modelling of North-East Atlantic cod and coastal cod as separate stocks, handling of several sampling designs used in various European countries, and simultaneous modelling of landed and discarded fish. This is joint work with the Institute of Marine Research.