Considerations in the Use of the NHTSA Cardfile 910894
The CARDfile was developed by NHTSA to aid problem identification and countermeasure development in the field of crash avoidance research. This paper summarizes the results of an evaluation of selected CARDfile data elements. Many of the variables of interest to crash avoidance, such as precrash stability, rural/urban, roadway separation, avoidance attempt, and VIN, are entirely missing for one or more of the CARDfile states. Such non-random missing data are likely to bias the resulting distributions. Other examples are described of missing and incompatible code values. Although five of the six states appear to code precrash stability, in fact there is not a single category on this variable that is recorded in every state, and no single state has all the categories. Thus the CARDfile cannot reflect the true distribution of these attributes for the aggregate of the six states. The authors conclude that perhaps the most significant disadvantage of the CARDfile is the appearance of uniformity that is created when, for many variables, this is not the case. In order to make valid inferences from the CARDfile, it is necessary to be familiar not only with the coding of the original state files, but also the additional translations that produce the CARDfile. These problems illustrate the advantages that more uniform accident reporting at the state level would provide.