Web site logical path: [www.psy.gla.ac.uk] [~steve] [this page]

Stats

This page is to collect a few pointers on stats issues, point to MCMD's work, etc.

Underlying points / arguments

• All ANOVAs, t-tests really do is test the ordinal relationships between 2 or more group means. (Denis (2003) suggests this argument.) Thus statistical interactions can be misleading if the lines don't cross, and the units in which the means are measured is non-linear (Loftus 1978; Wagenmakers 2012).

• GLM: anova and regression are the same approach underneath

• Types of range/box plots:
  • Median, quartiles, max&min.
  • Mean and SErr
  • 95% confidence limits (max&min).

• Dunlop fig.1

• xx
• xx
• xx
• xx

  Papers

• E.J.Wagenmakers
• Site for attempted replication experiments
• Reproducibility project
• Dennett on Austin's Putt.
• Wagenmakers (2012) On the interpretation of removable interactions: A survey of the field 33 years after Loftus
• Bargh non-replication debate   2   3: Bargh's spluttering   4   5
• Mark. D. Dunlop and Mark Baillie (2009) "Rejected (p>0.05): An Introduction to the Debate on Appropriateness of Null-Hypothesis Testing" Int. j. mobile HCI Volume 1, Issue 3. download
• Cohen, J. (1994). 'The earth is round (p < .05)" American Psychologist, vol.49, pp.997- 1003
• Sterling, T. D., Rosenbaum, W. L., & Weinkam, J. J. (1995) "Publication decisions revisited: The effect of the outcome of statistical tests on the decision to publish and vice versa" The American Statistician vol.49 no.1 pp.108- 112 [Shows 97% of published psychology articles show a SigDiff effect.]
• Hubbard,R. & Armstrong,J.S. (1997) "Publication Bias Against Null Results" Psychological Reports vol.80 pp.337-338 [Argues against publishing SigDiffs, and instead confidence intervals and effect sizes.]
• Denis,D.J. (2003) "Alternatives to Null Hypothesis Significance Testing" Theory & Science vol.4 no.1 pp.?

  Zipf

• WikiP on zipf distribution

• Power law, but discrete valued only.
• Linear on log-log scales. I.e. log (rank order) vs. log (frequency in a set/corpus).
• Zipf is one case (all power laws) of a more general rule: the rank-size rule/distribution.
• "Zipf's law states that given some corpus of natural language utterances, the frequency of any word is inversely proportional to its rank in the frequency table. Thus the most frequent word will occur approximately twice as often as the second most frequent word, three times as often as the third most frequent word, etc."
• Hence only the top N cases account for half all cases. For English vocab, top 135 account for half the Brown corpus.

  Other links

• Site arguing for Varimin, not Varimax, in factor analysis
• SUMS: Help teaching and learning stats
• Stats jokes
• SPSS help
• xx

Web site logical path: [www.psy.gla.ac.uk] [~steve] [this page]
[Top of this page]