Title: Multivariate Statistical Methods in Evaluation of Concrete and Aggregate Properties

Author(s): Peter P. Hudec

Publication: Symposium Paper

Volume: 171

Issue:

Appears on pages(s): 531-550

Keywords: Aggregates; concretes; specifications; statistics; tests.

DOI: 10.14359/6114

Date: 8/1/1997

Abstract:
Personal computers (PCs) now adorn almost every office desktop and lab counter. A variety of programs are available to handle and manage data of various types. Statistical manipulation, once considered to be an arcane art of mathematicians, is now available to anyone with the inclination to learn and use it. Simple univariate statistical analyses such as mean, standard deviation, and bivariate regression, and correlation etc. are in widespread use in concrete technology. However, they only give results concerning at most two variables at a time. In the real world, the properties such as strength, frost resistance, alkali reactivity, etc depend on several mutually dependent variables. For instance, multivariate statistical techniques such as cluster analysis can group all aggregates with similar properties; factor analysis can discern what combination of tests best describe a desired property of concrete; and stepwise regression analysis can be used to predict serviceability (projected life) of concrete or aggregate based on several standard tests. Examples of the above statistical techniques are presented in a non-technical format, based on research into concrete aggregate properties as related to their service record.