Error Components in Grouped Data: Is It Ever Worth Weighting?
Researchers estimating models with grouped data typically weight each observation by the square root of group size assuming that the model error term in the individual data is independently and identically distributed so that the error in the grouped data is heteroskedastic with variance proportional to group size. I argue that individual error terms are likely to be correlated due to group specific error component so weighting by the square root of group size is inappropriate. Tests for the presence of group error components and methods for obtaining efficient estimates of the parameters and consistent estimates of their standard errors are presented. Copyright 1990 by MIT Press.
Volume (Year): 72 (1990)
Issue (Month): 2 (May)
|Contact details of provider:|| Web page: http://mitpress.mit.edu/journals/|
|Order Information:||Web: http://mitpress.mit.edu/journal-home.tcl?issn=00346535|
When requesting a correction, please mention this item's handle: RePEc:tpr:restat:v:72:y:1990:i:2:p:328-33. See general information about how to correct material in RePEc.
For technical questions regarding this item, or to correct its authors, title, abstract, bibliographic or download information, contact: (Anna Pollock-Nelson)
If references are entirely missing, you can add them using this form.