A maximum likelihood technique is presented for the problem of fitting the ratio of two distributions, as a function of the sampling variable x, which may be multi-dimensional. No binning of the data is involved. A computer program, RATFIT, which implements this algorithm is described.
On the Distribution of the Ratio of the Measures of Divergence Between two Multivariate Populations
β Scribed by Prem Chandra Consul
- Publisher
- John Wiley and Sons
- Year
- 1966
- Tongue
- English
- Weight
- 246 KB
- Volume
- 32
- Category
- Article
- ISSN
- 0025-584X
No coin nor oath required. For personal study only.
β¦ Synopsis
I n this paper we define a new statistic W as a logarithm of the ratio of measures of divergence (as defined by MAHALANOBIS or as defined by BHATTACHARYYA) between two multivariate populations and then we obtain the exact distribution of W by using Characteristic functions, Inversion Theorem and some other results of Operational Calculus. Some of the properties of W-distribution have also been studied with the help of moments.
Coii\ul. Jlulti\ariate Populations
π SIMILAR VOLUMES
## Abstract The degree of agreement between two raters is reβexamined. An alternative statistic which uses the chiβsquare distribution is proposed. We conclude that this statistic is better than the usual __k__βstatistic when the classification variable is at least ordinal.