[Limdep Nlogit List] Help with citation about random selection model

[GIMENO Javier]

Dear LIMDEPers, 
 
I recall reading -- several years ago -- about a survey technique used
for eliciting truthful responses about private behaviors (I don't recall
if it was drug use, sexual behavior, or something like that).
 
The idea was that the respondent would draw a random binary process
(e.g, coin toss, dice, etc), and depending on the result it would either
respond truthfully to the question, or would enter some randomly
generated (but sensible) answer. The idea was that it would be
impossible to infer back from the individual answer whether it was true
or randomly generated. Yet, because the probabilities of the random
processes were pre-determined, they would add a known parameter shift in
the likelihood function, and therefore the coefficients of a discrete
response model could be estimated through a ML process. That's my
recollection, but memory is imperfect and I probably got some details
wrong.  
 
Does anyone recall seeing a paper using an econometric technique like
this? Essentially it is a selection model where the selection criteria
is exogeneous but unobserved. I've been looking for a reference, but my
memory and my library fail me. I would like to show evidence that this
model has been used before.   
 
My objective is to use this to allow me to merge two datasets, one of
which has confidentiality concerns. By applying a random filter to the
confidential data, it would be impossible to infer back whether the
answer is the right one for a particular firm, but I would still be able
to do meaningful analysis with this variable. 
 
Any guidance would be appreciated.
 
Javier
 
______________________________
Javier Gimeno
Professor of Strategy
INSEAD
Boulevard de Constance
77305 Fontainebleau Cedex
FRANCE
Tel: (+33) 16072 4513
email: javier.gimeno at insead.edu <mailto:javier.gimeno at insead.edu>