Particle Size Differentiation of Some Coastal Sands: A Multinomial Logit Regression Approach


  • Peter Vincent


Multinomial logit regression, discrimination, coastal sediments


This paper describes the application of multinomial logit regression to the discrimination of three groups of sediments at Drigg, Cumbria, UK. Moment measures of particle size were obtained for 150 sand samples taken from the present-day beach, erosion scarps in Holocene dunes, and from glacial-fluvial horizons in eroding Devensian drumlins. Particle size was measured using a Coulter laser diffraction instrument. A multinomial regression model including kurtosis, skewness and standard deviation correctly assigned 86 per cent of the sand samples. Multinomial logit regression is easy to use, provides probabilistic assignments and uses the independent variables in a direct manner, unlike many alternative statistical methods.