NR has worked with traffic modeling for several years, supported by various public authorities. One project deals with estimation of the effect of introducing a toll cordon in Oslo, other projects consider estimation of the total traffic volume over a year, based on a few counts during the year.

This work has been done for both cars and bikes. As it turns out, estimating the traffic volume of cyclists is more challenging than for cars. This is a result of the larger variation in the counts due to the lower traffic volume of cyclists. In addition, the yearly variation is larger for cyclists than for cars, as a lot more people tend to use their bikes in the summer, than in the winter.

The projects are founded on the same modeling approach, with elements from multivariate regression and time series modeling. The traffic at each count station is decomposed into level, trend, seasonal variation, variation due to day of week, special days and statistical error. The model is estimated from traffic counts at each station.

Some stations may have data covering several years, while others may have data for a few weeks. The model is estimated simultaneously for all stations, taking into account the inter-relationships between the stations. This means that reasonable estimates can be found even for stations with very few data.