Present address: School of Biological Sciences, University of Sussex, Falmer, Brighton, Sussex, BN1 9QG.
SOME STATISTICAL PROBLEMS ASSOCIATED WITH DETERMINATIONS OF POPULATION PARAMETERS FOR HERBACEOUS PLANTS IN THE FIELD
Article first published online: 2 MAY 2006
Volume 74, Issue 2, pages 349–363, March 1975
How to Cite
HUTCHINGS, M. J. (1975), SOME STATISTICAL PROBLEMS ASSOCIATED WITH DETERMINATIONS OF POPULATION PARAMETERS FOR HERBACEOUS PLANTS IN THE FIELD. New Phytologist, 74: 349–363. doi: 10.1111/j.1469-8137.1975.tb02622.x
- Issue published online: 2 MAY 2006
- Article first published online: 2 MAY 2006
- Received 21 August 1974
The use of regressions to predict weights of plants and methods for the assessment of skew-ness of weight and height distributions in populations of Mercurialis perennis (L.) are discussed. Regression methods for estimation of plant weight are non-destructive so that several weight estimates can be made at intervals on growing plants. Highly accurate predictions can be made from relationships between weight and dimensions. Non-linear (quadratic) regressions based on the relationship between weight and hd2 (where h= stem height; d= basal stem diameter) were found to be most accurate in weight prediction. Use of the estimate of relative error to determine error in prediction is considered; when errors are normally distributed and emphasis is placed upon mean weights rather than individual weights, the degree of error is considerably reduced. Because the relationship between weight and hd2 changes during the season, a sequence of regressions should be used for weight prediction. The relationship may be disturbed through time by environmental fluctuations and will vary spatially with habitat and genotype. The problem of inflexions in the use of quadratics is encountered and a solution suggested which assumes that other regressions without inflexions in the sequence are correct.
Histograms of weight and height distributions in plant populations do not allow accurate monitoring of changes in distribution through time. The moment coefficient is suggested as the best numerical estimate of skewness, since it is a sensitive measure and all the available data is utilized in its calculation. Pearson's second coefficient is less sensitive, and only utilizes the population mean and its median–a statistic which may be not sufficiently representative of the whole population to use in the calculation of skewness. Reasons are presented for the differences in the skewness values obtained by using these two coefficients.