|WikiProject Statistics||(Rated Start-class, Low-importance)|
|WikiProject Mathematics||(Rated Start-class, Low-importance)|
Now, if we are using the definition of normal distribution having a kurtosis of zero, while many other sources - outside of wikipedia - use 3, shouldn't we, for the case of clarity offer two alternative specifications for this test? Psw2xx 18:53, 12 March 2006 (UTC)
- Why? The definitions differ by exactly three; anyone capable of calculating kurtosis in the first place should be able to cope with the substitution involved. Septentrionalis 06:30, 13 March 2006 (UTC)
Shouldn't this formula use the unbiased versions of sample movements, i.e. (n-1) instead of n? 220.127.116.11 17:23, 31 March 2007 (UTC)
Critical values for small samples
Critical values exposed differ much from chi square values with 2 degrees of freedom. Wuertz and Katzgraber on their paper "Precise �finite-sample quantiles of the Jarque-Bera adjusted Lagrange multiplier test" (2009) (avaiable at http://www.brunel.ac.uk/9379/efwps/04-03.PDF) shows critical values obtained using 10^7 replications of monte carlo simulations, using "R" (GPL software).
changed formula (specifically the K, kurtosis, part, it should be K-3, not K on its own). The formula now follows the common notation for Jarque-Bera.
The R (programming language) documentation about this test (function jarque.bera.test) has this reference:
- J. B. Cromwell, W. C. Labys and M. Terraza (1994): _Univariate Tests for Time Series Models_, Sage, Thousand Oaks, CA, pages 20-22.
but no other reference. Since R and Wikipedia are both GPL projects, I think they should have similar references :-)
Is it appropriate to add a section about programming languages for each statistical test or function or distribution or whatever? Albmont 11:02, 14 November 2006 (UTC)
- Re the reference: That's clearly a secondary source, whereas the present revision of the article here lists the primary source. Re programming languages: I'm not entirely sure how appropriate and informative it is to list programming languages or statistical packages here. For unusual procedures, perhaps, but for something as basic as normality tests, one would think that all the usual suspects (S+/R, SAS, SPSS, Stata, ...) either have functionality for it or allow such functionality to be implemented straightforwardly. This is the sort of information that can easily be looked up in the manual of whatever statistics package you're using. --MarkSweep (call me collect) 06:42, 15 November 2006 (UTC)
The notation be might be improved : at present there is no clear distinctuion between the population and the sample versions of the moments and k-statistics. However there may be a benefit from keeping the present brevity. Melcombe 14:18, 10 July 2007 (UTC)