On Fri, 21 Jun 2002 12:08:33 +0100, Jonathan Robbins <[EMAIL PROTECTED]> wrote:
> At the risk of posing a dumb question from a non-statistician could > someone explain if generalizability analysis makes similar assumptions > to or is the same as multi-level or hierarchical modelling. As someone > who is called upon to be a 'consumer' of statistical information in > order to help in decision making by others, clarification of > similarities or differences would be very helpful. I've read the papers > by Cronbach and others on its application to educational assessments and > looked at the original theory work but an informed view would be useful. One perspective: I think the similarities are mathematical rather than practical. However, I do speak about "reliability" rather than some particular system called 'generalizability.' Yes, Reliability looks a lot like 'repeated measures' because it *is* repeated measures. From my viewpoint, there are different concerns: - what are your assumptions? - what are your questions? - what difficulties arise? Reliability testing is concerned with achieving *high* correlations (at least, part-whole). From the start, you often have the ability to manipulate the 'parts', in addition to testing them against each other. Hierarchical models are concerned with achieving valid comparisons of some awkwardly defined groups. You have to design what you are doing, pretty early-on, and you are stuck with the original quality of the data. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
