[Limdep Nlogit List] Truncated Independant Variable
Tan Yaw
tanyaw at pilotmm.com
Wed Mar 12 14:05:28 EST 2008
What is the implication on the coefficients if I have an independent
variable which has been truncated ?
How should it be resolved ?
Specifically,
1. We had developed a logit model, where the RHS was a financial ratio
and LHS Default/Non Default
2. The model was piecewise linear, meaning we split the ratios by
using interactive dummies where the RatioX1 and RatioX2 was basically RatioX
but with two different slopes.
3. Only RatioX1 was found to be statistically significant, which means
RatioX2 should be dropped. However, some people are uncomfortable with
keeping one-half and dropping the other half.
4. We have been asked to have a truncated RatioX, where RatioX is now
capped at the kink point where X2 would otherwise be 0.
5. The way I see it the difference between The interactive dummy and
the truncation Ratio X is that the Interactive dummy would have given me 0
at the kink point, while the truncation would have given me a max limit.
6. My concern is that the truncated RatioX would give rise to a bias
coefficient ? (as per the discussion surrounding limited dependant
variables).
Does anyone have any opinions on the above topic. Comments much appreciated.
Thanks
- Tan Yaw
More information about the Limdep
mailing list