[Limdep Nlogit List] Krinsky and Robb confidence intervals EC model
maria korjeh
maria.korjeh at gmail.com
Tue Dec 7 23:05:54 EST 2010
Dear all,
I am trying to estimate Krinsky and Robb confidence intervals for WTP ratios
in Nlogit 4.0.1.
It is working perfectly well for a MNL model (I'm including the code below
but omitting the modelling of U since it is lenghty and not relevant to the
problem):
Nlogit
; Lhs = CHOICE;
; choices = alt1, alt2, alt3;
; model:
U(alt1) = xxx
U(alt2) = xxx
U(alt3) = xxx
$
WALD; Start = B;Var=VARB
; Labels =
a1,a2,a3,a4,a5,a6,a7,a8,a9,a10,a11,a12,a13,a14,a15,a16,a17,a18...
; Fn1 = a14/a17
; Fn2 = a18/a17
; Fn3 = a15/a16
; Fn4 = a19/a16
; k&r$
Which gives the result:
+-----------------------------------------------+
| WALD procedure. Estimates and standard errors |
| for nonlinear functions and joint test of |
| nonlinear restrictions. |
| Wald Statistic = 186.81428 |
| Prob. from Chi-squared[ 4] = .00000 |
| Krinsky-Robb method used with 1000 draws |
+-----------------------------------------------+
+--------+--------------+----------------+--------+--------+
|Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]|
+--------+--------------+----------------+--------+--------+
Fncn(1) | -.00975416 .00169695 -5.748 .0000
Fncn(2) | -.00118653 .00062392 -1.902 .0572
Fncn(3) | -.00612822 .00109797 -5.581 .0000
Fncn(4) | -.00126124 .00042306 -2.981 .0029
However when I try with an EC model with additional variability for the
non-status quo alternatives the K&R procedure is unable to calculate the
confidence intervals for the WTP ratios.
Nlogit
; Lhs = CHOICE;
; choices = alt1, alt2, alt3
; rpl
; Halton
; fcn = no_sq (n,*,0)
; pds = 10
; pts = 500
; maxit = 200
; parameters
; model:
U(alt1) = altsq * ...
U(alt2) = no_sq * ...
U(alt3) = no_sq * ...
$
When, as before, I try to obtain a K&R for the value of time (tid/peng) but
get the following message;
WALD; Start = B;Var=VARB
; Labels =
a1,a2,tid,peng,a5,a6,a7,a8,a9,a10,a11,a12,a13,a14,a15,a16,a17,...
; Fn1 = tid/peng
; k&r$
+-----------------------------------------------+
| WALD procedure. Estimates and standard errors |
| for nonlinear functions and joint test of |
| nonlinear restrictions. |
| VC matrix for the functions is singular. |
| Standard errors are reported, but the |
| Wald statistic cannot be computed. |
| Krinsky-Robb method used with 1000 draws |
+-----------------------------------------------+
+--------+--------------+----------------+--------+--------+
|Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]|
+--------+--------------+----------------+--------+--------+
Fncn(1) | .00775481 ......(Fixed Parameter).......
The strange thing is that if I increase the K&R draws to 2000 the message
changes, but the statistic still is not calculated (the standard error is
very very low...).
+-----------------------------------------------+
| WALD procedure. Estimates and standard errors |
| for nonlinear functions and joint test of |
| nonlinear restrictions. |
| Wald Statistic =************* |
| Prob. from Chi-squared[ 1] = .00000 |
| Krinsky-Robb method used with 2000 draws |
+-----------------------------------------------+
+--------+--------------+----------------+--------+--------+
|Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]|
+--------+--------------+----------------+--------+--------+
Fncn(1) | .00775481 .250766D-08 ******** .0000
If I try the delta method the standard error is calculated just fine!
WALD; Start = B;Var=VARB
; Labels =
a1,a2,tid,peng,a5,a6,a7,a8,a9,a10,a11,a12,a13,a14,a15,a16,a17,...
; Fn1 = tid/peng$
+-----------------------------------------------+
| WALD procedure. Estimates and standard errors |
| for nonlinear functions and joint test of |
| nonlinear restrictions. |
| Wald Statistic = 93.16149 |
| Prob. from Chi-squared[ 1] = .00000 |
+-----------------------------------------------+
+--------+--------------+----------------+--------+--------+
|Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]|
+--------+--------------+----------------+--------+--------+
Fncn(1) | .00775481 .00080344 9.652 .0000
I really need to compute a simulated confidence interval to validate the
reliability of the Delta result shown above but can't figure out why the K&R
is only working with the MNL not the EC model.
I would really appreciate advice on how to carry out simulated confidence
intervals. Would it be possible to use boot-strapping for this aim? Would
anyone be willing to give me some hints on how to write the code for this?
Best regards
-Maria
More information about the Limdep
mailing list