A basic fact in statistics is that the more data one has, the more certain one can be of rejecting a false null hypothesis, no matter how small the discrepancy with reality. Thus, in transportation investigations, where large quantities of data in the form of counts are frequently available, the likelihood that the chi squared test will reject good (but not perfect) models becomes a major problem. To avoid this large sample problem, this paper looks at an alternative method of analysis for transportation data in the form of counts. Instead of examining whether the data could have been generated by a particular model or whether a particular model explains a significant amount of the data's variability, the analysis seeks to test whether the model is a sufficiently good approximation to reality. The concept of approximation is introduced to allow the investigator to specify how good the model needs to be. See also TRIS 368051. (Author/TRRL)

Media Info

  • Features: References; Tables;
  • Pagination: p. 71-77
  • Serial:

Subject/Index Terms

Filing Info

  • Accession Number: 00392339
  • Record Type: Publication
  • Source Agency: ARRB
  • Files: ITRD, TRIS, ATRI
  • Created Date: Mar 29 1985 12:00AM