[Nlogit] MNL error message on IIA test

#1
[Nlogitv4] MNL error message on IIA test

Hello,

I am using Nlogit software to do MNL. I created a DCE experiment about smartwatch and collected some data from my friends. I only have 18 responses but each person went through 10 or 11 choice sets. Due to my small sample size, I am not surprised that most of my parameters are not significant. I tried to run IIA test just to see if it runs. And I got error messages. I wonder if it is because I have too few responses or my codes are wrong. I really appreciate your advice, thanks in advance!

Here is some basic info about my DCE experiment before I show you the codes and results.
  • Each respondent went through 10 or 11 choice sets
  • Each choice set has 4 options: samsung, apple and google smartwatches and none option (this is an alternative specific design as the options are branded/labelled)
  • Each product has 4 attributes:
    [*]Phone compatibility (2 levels)
    [*]Whether it measures health metrics (2 levels)
    [*]Whether the watch band is detachable (2 levels)
    [*]Price (4 levels)​

Here are the codes:
nlogit
;lhs= choice, cset, altij
;choices = sams, apple, google, none
;ias=sams
;model:
U(sams) = sams + scomp*compd + shealth*healthd + sband*bandd + sp*price/
U(apple) = apple + acomp*compd + ahealth*healthd + aband*bandd + ap*price/
U(google)= google + gcomp*compd + ghealth*healthd + gband*bandd + gp*price$

Please note: the code ';ias = sams' is used to remove the samsung observations to test IIA

Here are the results and error messages:
| Discrete choice and multinomial logit models|
+---------------------------------------------+

+------------------------------------------------------+
|WARNING: Bad observations were found in the sample. |
|Found 34 bad observations among 179 individuals. |
|You can use ;CheckData to get a list of these points. |
+------------------------------------------------------+

Hessian is not positive definite at start values.
Error 803: Hessian is not positive definite at start values.
B0 is too far from solution for Newton method.
Switching to BFGS as a better solution method.
Normal exit from iterations. Exit status=0.
Error 585: Matrix being moved is too large for target.

+---------------------------------------------+
| Discrete choice (multinomial logit) model |
| Maximum Likelihood Estimates |
| Model estimated: Oct 16, 2013 at 11:20:33AM.|
| Dependent variable Choice |
| Weighting variable None |
| Number of observations 145 |
| Iterations completed 14 |
| Log likelihood function -132.7923 |
| Number of parameters 15 |
| Info. Criterion: AIC = 2.03851 |
| Finite Sample: AIC = 2.06418 |
| Info. Criterion: BIC = 2.34645 |
| Info. Criterion:HQIC = 2.16364 |
| R2=1-LogL/LogL* Log-L fncn R-sqrd RsqAdj |
| Constants only. Must be computed directly. |
| Use NLOGIT ;...; RHS=ONE $ |
| Chi-squared[12] = 23.11360 |
| Prob [ chi squared > value ] = .02678 |
| Response data are given as ind. choice. |
| Number of obs.= 179, skipped 34 bad obs. |
+---------------------------------------------+
| Could not carry Hausman test for IIA. |
| Difference matrix is not positive definite. |

+---------------------------------------------+
| Notes No coefficients=> P(i,j)=1/J(i). |
| Constants only => P(i,j) uses ASCs |
| only. N(j)/N if fixed choice set. |
| N(j) = total sample frequency for j |
| N = total sample frequency. |
| These 2 models are simple MNL models. |
| R-sqrd = 1 - LogL(model)/logL(other) |
| RsqAdj=1-[nJ/(nJ-nparm)]*(1-R-sqrd) |
| nJ = sum over i, choice set sizes |
+---------------------------------------------+
+--------+--------------+----------------+--------+--------+
|Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]|
+--------+--------------+----------------+--------+--------+
SAMS | .000000 1.00000000 .000 1.0000
SCOMP | .000000 ......(Fixed Parameter).......
SHEALTH | .000000 ......(Fixed Parameter).......
SBAND | .000000 ......(Fixed Parameter).......
SP | .000000 .01271025 .000 1.0000
APPLE | 1.57287869 ......(Fixed Parameter).......
ACOMP | -.09410830 .00378327 -24.875 .0000
AHEALTH | .50683557 .01689470 30.000 .0000
ABAND | .68988054 ......(Fixed Parameter).......
AP | -.00417422 ......(Fixed Parameter).......
GOOGLE | .90069676 ......(Fixed Parameter).......
GCOMP | -.12415649 ......(Fixed Parameter).......
GHEALTH | -.67359215 ......(Fixed Parameter).......
GBAND | -1.42521262 ......(Fixed Parameter).......
GP | .00212798 ......(Fixed Parameter).......
 
Last edited: