Changes for version 0.20

  • corrected calculation of circularized psi-sq: was implemented from Good (1953, Eq. 2), the original paper on this test. However, it was seen (after multiple use) to sometimes yield negative values - because it does not square the observed-expected count difference in the sum-of-squares. This was (silently) corrected in Good & Gover (1957), which is now implemented.
  • psisq() called in list context now also returns df and p_value
  • removed z_value method - only appropriate if df = 1, and requires test of validity
  • removed from code undocumented and inappropriate option to set "tails" for p_value (error of code-recycling from other Sequence modules)
  • removed "nstates" arg from nnomes() method; just supply aref of "states" if necessary, as for other methods
  • croak re undefined or zero length argument rewritten, made method-specific
  • removed croak from observed() if a defined length of v is not valid so that frequencies can still be obtained
  • removed doc of excessive aliases for observed() and other methods - support remains but will be removed in next version
  • removed NIST examples from testing as these appear to be basd on circularizing the sequence but then calculating the expectation value in the sum-of-squares in a way that is appropriate for uncircularized sequences.
  • added DIAGNOSTICS section to POD


Serial Test psi-square for equiprobability of v-nomes (or Ngrams) (Good's and Kendall-Babington-Smith's tests)