Standard Article

Chemometrics and Statistical Considerations in Forensic Science

Forensic Science

  1. Grzegorz Zadora

Published Online: 15 MAR 2010

DOI: 10.1002/9780470027318.a9122

Encyclopedia of Analytical Chemistry

Encyclopedia of Analytical Chemistry

How to Cite

Zadora, G. 2010. Chemometrics and Statistical Considerations in Forensic Science. Encyclopedia of Analytical Chemistry. .

Author Information

  1. Institute of Forensic Research, Krakow, Poland

Publication History

  1. Published Online: 15 MAR 2010


Questions have been raised by the police, prosecutors, and the courts concerning identification and classification of objects into certain categories and/or an association between two or more items. Most of the samples analyzed by forensic experts are very small. Therefore addressing the above-mentioned problems requires application of various analytical methods which yield various kinds of information, i.e. about the morphology of analyzed objects and the qualitative and quantitative sample composition. When forensic analysis requires interpretation of physicochemical data, then it is common nowadays to apply statistical methods. Measurements on crime scene evidence and evidence from a suspect may be deemed to be “similar” or “very similar” and may be deemed to be “rare” or “common”. The choice of adjective may well be a personal choice and lead to a debate among experts and confusion for the people involved in the investigation. Therefore, two experts could use different adjectives having the same data when they do not use the objective method in their interpretation. This is because they could have different experiences in the field of interpretation of this particular type of evidence material (data, e.g. elemental composition of glass).

Objective assessment which takes account of the underlying variation requires the use of appropriate models. The choice of the model is, of course, a subjective one. However, once the model is chosen, the results will be determined. Different people with the same data and same model will produce the same assessment of the evidence. There is a debate here too, but this debate is about the assumptions underlying the choice of model. Are the observed variances of refractive index (RI) values equal for compared glass samples? If yes, then a Student's t-test could be used to test the null hypothesis that ȳ1 = ȳ2. If not, then a modification of the Student's t-test, called Welch test, should be used. The debate is about science, and not about the choice of adjective.

Applications of the most commonly used statistical and chemometrics methods for solving the classification and comparison problems are presented in this article. A special emphasis is made on methods based on calculation of likelihood ratio models as they seem to be the most suitable form of expression to the role of forensic expert in evaluating evidence E (e.g. physicochemical data), in the context of the prosecution proposition H1 and the defense proposition H2.


  • forensic sciences;
  • chemometrics;
  • evaluation of evidence;
  • likelihood ratio;
  • physicochemical data