CDESCRIBE procedure

Calculates summary statistics and tests of circular data (P.W. Goedhart & R.W. Payne).


Options

PRINT = strings
What to print (summary, fittedvalues); default summ

SEGMENT = scalar
Width of sectors (in degrees) into which to group an ANGLES variate for calculation of the test of randomness and the chi-square goodness of fit statistic for the von Mises distribution; default 20

MSEGMENT = scalar
Defines the centre (in degrees) of the sectors; default 0

DIRECTION = scalar
Direction (in degrees) of the unimodal alternative distribution for the Rayleigh test; default * i.e. not known


Parameters

ANGLES = factors or variates
Directional observations (in degrees)

RESULTS = variates
Saves the summary statistics

VONMISESCOUNTS = pointers
Saves structures relevant for calculation of the the chi-square goodness of fit statistic for the von Mises distribution


Description

CDESCRIBE summarizes data values that consist of directional observations recorded as angles between 0 and 360 degrees. These are supplied using the ANGLES parameter, in either a variate or a factor. The procedure mainly uses the methods presented in the book by Fisher (1993). The various statistics are cross-referenced below with the relevant page numbers.

   CDESCRIBE prints the following summary statistics: number of observations, mean direction (page 31), circular standard deviation (page 32), mean resultant length (page 32), skewness (page 34) and estimate of the parameter Kappa (which provides the concentration parameter of the von Mises distribution for circular data; pages 39 and 88). If the angles are supplied in a factor, a grouping correction is applied to the mean resultant length and to the skewness (page 35).

   Two tests of uniformity are presented. The null hypothesis for both of these is that the observations come from a uniform distribution around the circle. The first is a test of randomness against any alternative model. The test is based on counts of the number of observations in a set of angular sectors of equal size (page 67). If ANGLES is set to a variate, the width of the sectors is defined by the SEGMENT option (in degrees), with centres defined by the MSEGMENT option. The sectors are centred at MSEGMENT, MSEGMENT+SEGMENT, MSEGMENT+2*SEGMENT, and so on. The default values for SEGMENT and MSEGMENT are 20 and 0 respectively. If ANGLES is set to a factor with equidistant levels, it is assumed that the levels define the centres of the segments and that the limits of the sectors are at the midpoints between each pair of factor levels. If ANGLES is set to factor with non-equidistant levels, the SEGMENT and MSEGMENT options are used to define the angular sectors.

   The second is Rayleigh's test of uniformity against a unimodel alternative. The test is based on the mean resultant length and has two forms which differ according to whether or not the mean direction of the alternative distribution is known (pages 69 and 70). The direction, if known, is specified using the DIRECTION option.

   Finally a goodness of fit test is calculated to assess whether the observations follow a von Mises distribution. This is a chi-square test, which compares the observed distribution with the expected distribution from a von Mises distribution with mean direction and concentration parameter (kappa) taking the values estimated from the observations. The observed and expected values are calculated for grouped directional data defined by the (M)SEGMENT options for a variate or by the factor levels if ANGLES is set to a factor.

   The PRINT options controls whether the summary statistics are printed and whether a table of observed and expected counts for the fit of the von Mises distribution is printed. The summary statistics can be saved by means of the RESULTS parameter. The VONMISESCOUNTS parameter saves the grouped directional data used for calculation of the chi-square goodness of fit test and tables of observed and expected counts. Note that when ANGLES is set to factor, the saved grouped directional data is identical to ANGLES.

 

Options: PRINT, SEGMENT, MSEGMENT, DIRECTION.

Parameters: ANGLES, RESULTS, VONMISESCOUNTS.


Method

CDESCRIBE uses methods described by Fisher (1993). A private version (_SPECIALFUNCTION) of the Biometris procedure SPECIALFUNCTION is used to calculate modified Bessel functions and related functions.


Action with RESTRICT

If ANGLES is restricted, only the unrestricted units are analysed.


Reference

Fisher, N.I. (1993). Statistical Analysis of Circular Data. Cambridge University Press, Cambridge.