- Title
- Five strategies for accommodating overdispersion in correspondence analysis
- Creator
- Beh, Eric J.; Lombardo, Rosaria
- Relation
- Advanced Studies in Classification and Data Science p. 117-129
- Relation
- Studies in Classification, Data Analysis, and Knowledge Organization 1431-8814
- Publisher Link
- http://dx.doi.org/10.1007/978-981-15-3311-2_10
- Publisher
- Springer
- Resource Type
- book chapter
- Date
- 2020
- Description
- Traditionally, simple correspondence analysis applied to a two-way contingency table is performed by decomposing a matrix of standardised residuals using singular value decomposition where the sum-of-squares of these residuals gives Pearson's chi-squared statistic. Such residuals, which are treated as being asymptotically normally distributed, arise by assuming that the cell frequencies of the table are Poisson random variables so that their expectation and variance are equivalent. However there is clear evidence in the statistics literature that suggests that the variance of these residuals is less than one. Thus, we observe overdispersion in the table. Various strategies can be undertaken to study, and deal with, overdispersion. In this paper we shall briefly review five possible strategies. Although we conceed that the purpose of this paper is not to provide a comprehensive examination of their utility - future investigations of their properties will confirm any further benefits in their use.
- Description
- 1
- Subject
- cluster analysis; classification; marketing science; visualization; multidimensional representation of data; complex data
- Identifier
- http://hdl.handle.net/1959.13/1442568
- Identifier
- uon:41726
- Identifier
- ISBN:9789811533105
- Language
- eng
- Hits: 325
- Visitors: 321
- Downloads: 0
| Thumbnail | File | Description | Size | Format |
|---|