st: Tests for parameters' bias in weighted and unweighted survey analytic models

[Ángel Rodríguez Laso <angelrlaso@gmail.com>]

Dear Statalisters,

The use of weights in analytic models for survey data can produce
great inefficiency. Weights could be skipped for this type of analysis
if biases of parameters calculated with and without weights are not
statistically significant. In the document "The use of survey weights
in regression analysis" by I. Faiella (Temi di discusione (Working
Papers) nº739. January 2010. Banca d'Italia. Eurosistema) four methods
are proposed to ascertain this:

1) Incorporate to an unweighted model that includes the variables
which were used to build the weights, new variables that are the
weights themselves and the interactions weights*independent variables,
and see if the group of new variables is statistically significant.

2) Plotting the residuals of the unweighted regression against survey
weights or design variables.

3) Conduct a Hausman test.

4) Conduct Lagrange Multiplier statistics.

-hausman- help (Stata 9.2) states:  "force specifies that the Hausman
test be performed, even though the assumptions of the Hausman test
seem not to be met, for example, because the estimators are pweighted
or the data are clustered.", which are usual design elements of a
survey. I haven't found a Lagrange Multiplier statistic in Stata.

I've written to Ivan Faiella who has told me that "unfortunately the
tests I propose in the paper are non standard and so you need to write
down the code to compute them". He has sent to me the R codes he has
developed to conduct Hausman and Lagrange Multiplier tests in this
scenario, but I'm not able to translate them into Stata.

Are you aware of something equivalent in Stata?

Thank you very much.

Angel Rodriguez-Laso

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/