This paper concerns the exploration of statistical modelling tools for the analysis of observational freeway flow data, and the development of empirical models to capture and predict short-term changes in traffic flow characteristics on sequences of links in a partially detectorised freeway network. A first set of analyses explores regression models for minute-by-minute traffic flows, taking into account time of day, day of the week, and recent upstream detector-based flows. Relationships between days are captured through the use of day- and link-specific random effects in an hierarchical statistical modelling framework. A second set of analyses refines these models to include parameters that may vary moderately throughout the course of the day to capture day-specific idiosyncracies in traffic patterns. These are found to improve model fit and short-term predictions of flows significantly. A third set of analyses includes further model extensions to include recent downstream flows as additional predictors, and these are found to provide further improvements that, though marginal in most cases, can be quite radically useful in cases of very marked breakdown of freeway flows on some links. Each of these modelling stages is described and developed in analyses of observational flow data from a set of links on Interstate Highway 5 (I-5) near Seattle. Summary conclusions and open questions conclude the paper.

Claudia Tebaldi, PhD, is research fellow at the National Center for Atmospheric Research, Boulder CO. Mike West is Professor and Director of the Institute of Statistics and Decision Sciences, Duke University. Alan Karr is Associate Director of the National Institute of Statistical Sciences. The research reported here was partially supported by NSF grant DMS-9313013 to the National Institute of Statistical Sciences. The corresponding author is Claudia Tebaldi at NCAR, Climate Analysis Section/Geophysical Statistics Project, Boulder, CO, 80307.

The manuscript is available in either postscript or pdf