A recent paper by Horton  attracted 1,576 views and 123 tweets within ten days of its publication. A major scientific breakthrough? Hardly, the paper reported failure to replicate an earlier finding. Whose earlier finding? Horton’s! Horton suggests two reasons for his failure to replicate earlier results: lack of generalisability or type 1 error (false positive result arising through the play of chance). Both papers dealt with memory clues through associations – the idea that a clue, such as a place or a person that was present when the memory was stored in the brain, could prompt its recall.
The CLAHRC WM director thinks Type 1 error is the likely explanation. A positive P value is much less impressive evidence against the null hypothesis than many suppose. A Bayesian approach  would have forestalled such a volte-face.
— Richard Lilford, CLAHRC WM Director
- Brown-Schmidt S, Horton WS. The Influence of Partner-Specific Memory Associations on Picture Naming: A Failure to Replicate Horton (2007). PLoS ONE. 2014; 9(10): e109035.
- Reas E. This Week’s Most Discussed PLOS Neuroscience Article: The Influence of Partner-Specific Memory Associations on Picture Naming: A Failure to Replicate. [Online]. 2014.
- Goodman S. A Dirty Dozen: Twelve P-Value Misconceptions. Semin Hematol. 2008; 45(3): 135-40.
- Goodman SN. Toward Evidence-Based Medical Statistics. 2: The Bayes Factor. Ann Intern Med. 1999; 130: 1005-13.