Geographia Polonica (1983) vol. 45

Full text

The classification of morphological features — a logistic regression approach

Peter Vincent, Jean Haworth, Joan Victoria Clarke

Geographia Polonica (1983) vol. 45, pp. 5-19


Many features observed by the Geomorphologist raise an interesting question, namely: do all the observations belong to one or more morphological populations? Detailed field measurements of form may help solve this type problem, but in doubt-ful cases one may have to resort to some statistical device which allows a certain probability to be attached to an individual that will help decide to which of two possible populations it belongs.

In this paper we shall describe an example of a problem raised during the map-ping of terracette forms near Salford where there was good documentary evidence that the forms belonged to two populations, natural and man made. Visually, the two populations of terracettes appeared similar and it seemed reasonable to suppose that only careful measurement of the actual morphology might distinguish them.

In dubious cases assignment to one population or the other can be made via logistic regression, which because of its properties is advantageous to other met-hods which might be used to discriminate between the two genetic types of terracette. Hitherto this type of statistic has been little used in Geography, but as we shall indicate it has considerable advantages over other commonly used methods such as ordinary least squares regression and discriminant analysis.


Peter Vincent, University of Lancaster, Lancaster, United Kingdom
Jean Haworth, University of Salford, Salford, United Kingdom
Joan Victoria Clarke, University of Salford, Salford, United Kingdom