There is a need for fair and statistically valid assessments of traveler information data services. For measurement of the accuracy of a traveler information service, ground truth data must be collected and compared with the data service being evaluated. Past assessments of data quality have typically used floating-car runs or travel time estimates derived from loop detector speed data to determine ground truth benchmarks. Newer approaches such as measurement of travel time from Bluetooth samples can provide statistically valid sample sizes. However, determining the minimum sample size before data collection is critical to minimize data collection costs and maximize accuracy of ground truth data. An analysis of travel time data from Houston, Texas, is presented. The distribution of travel time variance is explored and a method for network stratification with a classification model is developed. The method can be applied to classify freeway links based on the travel time coefficient of variation. The method will help engineers in determining which links in a network to test and in estimating the minimum sample size per link.