Next to Sawtooth (v7.0) I work with NLogit (v3.0) for different model estimation. In the last project I double checked effects. Using effect-coding everything is fine and effects of all attributes and none parameter are same in both outputs. But there is an issue with linear estimation. Effects over all attributes (one effect for one attribute) show same outputs. Only the none parameter (I wouldnt choose...) seems to be very different (-1.22 vs -3.33). The none parameter in NLogit is coded as seen below as zeros over attributes (in rows) and alternative specific constant with 1 for none and zeros for alternatives.
Alt,Cset,Att1,Att2,Att3,None,Choice
1,3,2,3,3,0,1
2,3,1,1,1,0,0
3,3,0,0,0,1,0
How does Sawtooth handle the none parameter?! And how is it calculated? I am wondering why I got different effects for none parameter. Even when the effect of none parameter I guess is not that important I'd like to know how it works.
Any idea?
How is the none parameter coded and calculated when attributes treated as linear?
Resolved
2 replies