Weighting and cell-specific fitting

• Weighting out the diagonal is exactly equivalent to fitting it exactly.
• To do this, add new values to the symm variable, one for each diagonal cell, and fit without the /cstructure command:

  compute symm2=symm.
  if avote=evote symm2=avote+6.
  factor symm2 d 10.
  
  genlog avote evote with d2 d3 d4 d5 d6 d7 d8 d9 d10
   /print=est/plot=none
   /design=avote evote d2 d3 d4 d5 d6 d7 d8 d9 d10.
  

• The $G^2$ and degrees of freedom will be the same, and the residuals will be identical (diagonal residuals will be either 0.0 or missing).
• The parameter estimates will be different because there is a different set of parameters (but they combine to the same predicted values).