If traffic volume on an interstate is mistakenly recorded too high, then unnecessary adjustments may occur, causing extreme fiscal waste. On the other hand, if traffic volume is mistakenly recorded too low, then needed adjustments will not be initiated and the pavement will deteriorate more quickly than expected; repair costs exceed rehabilitation costs, so again there is fiscal waste. The need for reliable, edited, and validated traffic count data is well acknowledged by transportation officials, but ‘best practices’ are not typically available for: first, deciding which counts are possibly in error and need to be re-observed; second, accounting for spatial correlation and continuity along roadway networks when trying to determine which counts are in error; and third, determining how to edit counts when re-counting is not feasible. In order to improve the entire editing and validation process by increasing the accuracy of reported counts, by reducing the time delay between data collection and reporting, and by providing easily customized reports of traffic counts, spatial analysis can be used.
  The common practice for editing and validating count data is to manually and visually compare current counts to counts from previous years and neighboring stations. If a count is considered unusual, it is often modified to make it more similar to neighboring counts. This process is very slow, is prone to individual subjectivity and bias. Therefore, we need to develop a statistical model that characterizes traffic counts as a function of route characteristics for the count station, route interconnectivity and distances to other count stations, land use surrounding the count station, and census demographics in the area surrounding the count station. This statistical model can provide predictions of traffic counts, which are actually annual average daily traffic counts (AADT), at any location along a primary or secondary road segment in research area. It can even provide predicted AADTs at sites that have never been sampled. For the purpose of quality control, a simple approach is to compare an observed AADT with the prediction obtained from our model for that same count station. ‘Large’ differences may suggest a problem with the count.
  Generally, limited budgets only permit data collection to occur at a limited number of facilities, but one may need predictions at many other (unsampled) facilities. In order to make predictions at unsampled facilities, one may usually do ‘Kriging’ to obtain these prediction. Kriging is a method for predicting the unknown value of a variable at an unsampled site by using a weighted average of the same variable as observed at ‘nearby’ sampled sites. Kriging has several advantages. First, it provides linear unbiased predictors that have the smallest mean squared prediction errors among the class of linear unbiased predictors. Second, kriging is readily available through a number of statistical software packages. In environmental science, kriging is used quite often to obtain a smooth spatial surface from using only data from a relatively small set of irregularly spaced sampling sites.