78 Transportation Research Record: Journal of the Transportation Research Board, No. 2443, Transportation Research Board of the National Academies, Washington, D.C., 2014, pp. 78–87. DOI: 10.3141/2443-09 Transportation engineers and researchers heavily use traffic data, which are generally aggregated by predetermined time intervals (e.g., 5 to 15 min). The aggregation process often discards essential information of traffic state transition (e.g., breakdowns). However, the transition of traffic conditions within an aggregation interval is not well understood. This study explored traffic state transition from uncongested to congested regimes that occurred within a predetermined time interval. From two urban freeway locations in Norfolk, Virginia, traffic data archived at 15-min intervals were obtained. A heuristic method based on a Gaussian mixture model was developed to detect the aggregate traffic data that exhibit the transition of traffic states as well as to partition the data sta- tistically into uncongested and congested traffic states. Results show a substantial difference in travel speed (approximately 20 mph) between the two states. In addition, these results illustrate that aggregating these different traffic conditions can cause substantial traffic data aggregation bias by lowering travel speed and flow rates, especially in high traffic flow situations. Finally, new insights into valid traffic data aggregation and speed–flow–concentration relationship development are discussed. Transportation engineers and researchers heavily utilize traffic data. In practice, traffic data are collected from traffic sensors and then aggregated by predetermined time intervals (e.g., 5 to 15 min). While the aggregation process stabilizes traffic flow rates and pro- vides a manageable size of data points, the process often discards some important information of traffic flow such as traffic state transition, at all aggregation intervals. The transition of speed and concentration may exist within an aggregation interval, while the transition can take place anytime in the middle of the interval (1, 2). However, the transition occurring within a data aggregation interval has not been explicitly or empirically studied in the literature. This paper explores the transition of traffic conditions, e.g., from uncongested to congested regimes, within a traffic data aggregation interval. Notably, the transition usually occurs when traffic flow is higher and near-capacity; therefore, the differences in speed and flow between the two regimes are expected to be substantially large. For this reason, it is interesting to analyze the validity of aggregat- ing the two traffic states. To examine the transition and validity, empirical traffic data aggregated at 15-min intervals were obtained from two freeway sites in Norfolk, Virginia. In addition, this study developed a heuristic procedure with a Gaussian mixture model to detect the aggregate traffic data containing the transition of traffic states. The aggregate traffic data were statistically partitioned into uncongested and congested flows. Finally, this study compared the uncongested and congested conditions and analyzed the differences within the aggregation intervals. The paper can benefit both transportation researchers and engi- neers by providing a better understanding of traffic flow operations and dynamics that can be observed within a data aggregation inter- val. In addition, this paper provides new insights into fitting and interpreting speed–flow–concentration relationships more accu- rately. The remainder of the paper is organized as follows: it begins with a synthesized literature review, which is followed by descrip- tions of the two study sites. After that, the Gaussian mixture model- ing technique is introduced, and the procedure to extract aggregate traffic data that exhibit traffic state transitions is explained. Then, the results of comparisons between uncongested to congested traffic con- ditions are presented and illustrated. Finally, the paper discusses the implications for valid aggregation of traffic data and development of speed–flow–concentration relationships. LITERATURE REVIEW Traffic data have been widely used by transportation engineers and researchers. While examples include traffic speed and density estimations (3, 4), a prominent case is the development of speed– flow–concentration relationships. Generally, these relationships are developed with aggregated traffic data. Since Greenshields’ seminal work on developing relationships among traffic flow vari- ables, the relationships have provided a fundamental understanding of traffic flow operations and dynamics (5, 6). For example, driving behavior (car following) is modeled and validated by the speed–flow– concentration diagrams in microscopic or mesoscopic traffic simu- lations or both (7 ). Moreover, the relationships are the essential building blocks for some of the concepts in the Highway Capacity Manual (HCM), which adopts the relationships to define capacity and level of service of basic freeway and multilane highway seg- ments (8). Furthermore, many transportation applications in transpor- tation planning rely on the relationships, including a volume–delay function in travel demand forecasting models (9, 10). Knowing traffic conditions (e.g., congested or uncongested) under which the data are collected is essential. In addition, the traffic conditions need to be identified when the data are aggre- gated. The development of speed–flow–concentration relationships requires ensuring no influence from downstream traffic operation. In HCM 2010, capacity and speed–flow curves are developed under unsaturated conditions (8). Moreover, volume–delay curves need to Exploring Bias in Traffic Data Aggregation Resulting from Transition of Traffic States Sanghoon Son, Mecit Cetin, and Asad Khattak S. Son, Regional Planning and Environment Division, Jeju Development Institute, 253 Ayeon-Ro, Jeju-Si, Jeju-Do 690-162, South Korea. M. Cetin, Department of Civil and Environmental Engineering, Old Dominion University, 135 Kaufman Hall, Norfolk, VA 23529-0241. A. Khattak, Department of Civil and Environmental Engineering, University of Tennessee, 322 J. D. Tickle Building, Knoxville, TN 37996-2010. Corresponding author: A. Khattak, akhattak@utk.edu.