El 05/05/11 05:16, Allin Cottrell escribió: According to the gretl users guide, gretl uses "double-precision floating-point numbers", and looking at wikipedia, if I am not understanding badly, it seems that the maximum number we may use is 1.7976931348623157 E+308 (Max Double) I see in the gretl console that with negative sign, the minimum number that gretl accepts is -1.7976931348623156e+308 so the better approximation to what you want in gretl would be: w = (z==1)? y : -1.7976931348623156e+308 #### some gretl results: scalar x=-1.7976931348623157e+308 Se ha reemplazado el escalar x = NA ? scalar x=-1.7976931348623156e+308 Se ha reemplazado el escalar x = -1.79769e+308 -- Ignacio Diaz-Emparanza DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA) UPV/EHU Avda. Lehendakari Aguirre, 83 | 48015 BILBAO T.: +34 946013732 | F.: +34 946013754 www.ea3.ehu.es

On Thu, 5 May 2011, Sven Schreiber wrote: Well, I think so. After the lengthy discussion of this (around April 2010) the conclusion I came to is that it's pretty much impossible to set up a fully coherent system that encompasses infinities, NaNs and true "missing data" (NA). If you're willing to map NA to NaN you can then rely on IEEE arithmetic to do the right thing with NaNs and infinities, but otherwise NA doesn't fit in the IEEE scheme and you get insoluble conundrums in conversion/propagation. Allin