Dear All, I am writing to you on behalf of Colin Davis and myself because we believe you have a strong interest and expertise in masked form priming, mega-studies, or both, and hope that you might be persuaded to collaborate with us on the collection of a multi-centre mega-study database of form priming. --- We believe that such a study would substantially advance our empirical knowledge of these priming phenomena, and enhance our ability to assess and improve models. --- We hope that many of you will agree, and be able to contribute participants to the study, but from those who can not we would nevertheless welcome comments on the design to maximise theoretical relevance. More detail on our motivation and plans follow if you might be interested. 1. Motivation There is a growing database of effects in masked form priming spread over a large number of experiments, laboratories and languages. These studies --- in which many of you have been involved --- have, of course, contributed greatly to our understanding of the front-end of visual word recognition, and are used to assess theoretical models of the processes involved in priming. Despite our differing theoretical perspectives on masked priming (Adelman, 2011/submitted: http://www.warwick.ac.uk/~pssgar/ltrs/ ; Davis, 2010: http://www.pc.rhul.ac.uk/staff/c.davis/SpatialCodingModel/ ), Colin and I agree that a mega-study of masked priming is needed to improve our ability to test models. There are two main limitations to using existing small-scale studies. * Precision of estimates Traditional small-scale experiments are designed to efficiently determine whether effects are present or not, rather than produce a precise estimate of the amount of priming. The effects we are interested in are usually under 50 ms in magnitude, but 95% confidence intervals whose widths are 20 ms or more (for either RTs or priming) are not uncommon. This opens up the strong possibility that one model will appear to produce better fits than another because it is capturing noise idiosyncratic to the particular experiments, rather than a systematic effect of the manipulation. This problem can be addressed by increasing the numbers of participants and items. * Between-participant between-laboratory comparisons The vast majority of small-scale experiments are run within-subjects and with careful experimental controls for a very good reason: This ensures that the data in each condition are comparable. When experiments are combined to produce assessments of models, such control is discarded. To account for the effects of individual differences and differences in laboratory conditions, it might be reasonable to allow models to have different parameters for each experiment, but doing so would massively weaken the constraints placed on models by the data. This problem can be addressed by increasing the number of conditions to which each participant contributes data, retaining a within-subject design. Each participant acts as his/her own control, so each laboratory also acts as its own control. 2. Method Once the design is agreed, we will prepare files for DMDX to distribute to those who have agreed to contribute participants, and set up a repository for results files. Our current plans, which are not set in stone, are (briefly, and in reverse order): 2.1. Procedure We anticipate a standard three-field masked form priming lexical decision experiment, with participants completing 1000 trials (500 word, 500 nonword). Our experience suggests this will take the slowest participants an hour, which is comparable to previous mega-studies, as well as being the size of a unit of participation for course credit for us (though we also anticipate some participants will be paid). (Other procedures, such as Lupker & Davis's, 2009, sandwich priming technique are possibilities for follow-up studies.) 2.2. Design/Stimuli Up to 25 prime types will be compared in a completely within-subjects design. Each prime type will be instantiated with each of 500 low-frequency word targets, eight letters in length. Each participant will contribute 20 trials to each prime type condition, seeing each target once. A simple counterbalancing scheme will therefore produce 25 counterbalancing conditions. The particular prime types will be a matter for discussion on the project mailing list; we, of course, already have some ideas, but these are not final. 2.3. Participants Each contributor will supply as many English-speaking participants as they like, subject to the constraint that within any one site, the counterbalancing is complete (i.e., the number will need to be a multiple of 25 [or other number the design requires]). 3. Going forward If you might like to be involved, please respond to this e-mail, and I will add you to the project mailing list; this does not require a commitment to providing participants at this stage (though we would be delighted to receive such a commitment). Once a design is agreed, contributors will receive files to run experiments, and details of a repository to submit the data. A web site will be set up to distribute the data to the research community at large, and we will submit a paper (with everyone as an author) about the database to Behavior Research Methods, which will be the primary source for the data. Contributors will be free to use their advance access to the data in any way that does not jeopardise this initial publication. Thank you for reading on this far; we hope to be working with you in the near future. Regards, -- James S. Adelman, Department of Psychology, University of Warwick, COVENTRY, CV4 7AL, UK.