Object recognition testing: Statistical considerations

The object recognition task is a commonly used test for the assessment of memory functions in rodents. In this paper different aspects concerning the analysis of object recognition data are discussed. Using a set of experimental studies and fictive data sets, it was shown that the absolute discrimination measure (d1) behaves different from the ratio measures (d2 and d3). Furthermore, it is suggested that, besides group differences, one should also examine whether each individual group actually discriminates between the novel and the familiar object. For this purpose, the use of a fictive group showing no discrimination is advisable. Furthermore, on basis of 48 object recognition task studies it is shown that discrimination performance does not fall significantly below zero. Therefore one-sided testing is allowed, provided that place- or object biases can be ruled out. Finally, it was shown that differences in exploration levels may affect the statistical evaluation of group differences. Several suggestions for statistical analysis are given.

► Relative discrimination measure and investigation ratio are statistically identical.
► Absolute discrimination measures are more vulnerable to exploratory differences.
► Presence of memory should be always confirmed by comparison with random exploration.
► One-sample t-tests are not a realistic way to compare with random exploration.
► When testing under proper conditions, one-sided testing is allowed/recommended.