[Limdep Nlogit List] Exploded and negative wtp estimates
John C. Whitehead
john.c.whitehead at gmail.com
Sun Oct 20 00:09:21 AEDT 2024
That’s a complicated model. What do the results look like when you estimate
a basic mixed logit.
RPLOGIT
; Lhs = choice ; Choices = A, B, C
; Rhs = fortific,weight,certific,twosezon,manysezo,nprice
;Pds = 4
;halton ;pts=100
; Fcn = fortific(n),weight(n),certific(n),twosezon(n),manysezo(n),
nprice(n)$
On Sat, Oct 19, 2024 at 5:11 AM medard kakuru via Limdep <
limdep at mailman.sydney.edu.au> wrote:
> Dear All,
> Here is the syntax and the output of the generalized mixed logit model in
> preference space, using MNL starting values. As you can see, the wtp
> estimates are in hundreds of thousand yet my price attribute levels are in
> thousands. When I add "userp" (using ML starting values) to the syntax, I
> get reasonable wtp estimates (some in hundreds others in thousands) but
> three out five are negative and the standard deviations for all are
> extremely big- in hundreds of thousand.
> Please advise.
>
> CREATE ; nprice = -price $ ? make the price attribute negative
> |-> calc ; ran(10000) $ ?seeding
> |-> GMXLOGIT
> ; Lhs = choice ; Choices = A, B, C
> ; Rhs = fortific,weight,certific,twosezon,manysezo,nprice
> ;Pds = 4
> ;gmx
> ;tau=0.1
> ;gamma=0.1
> ;halton ;pts=100
> ; corr ;Parameters
> ; Fcn = fortific(n),weight(n),certific(n),twosezon(n),manysezo(n),
> nprice(n)
> ; WTP =
>
> fortific/nprice,weight/nprice,certific/nprice,twosezon/nprice,manysezo/nprice
> $ ? estimate individual-specific wtp estimate
> Iterative procedure has converged
> Normal exit: 6 iterations. Status=0, F= .9000622D+03
>
>
> -----------------------------------------------------------------------------
> Start values obtained using MNL model
> Dependent variable Choice
> Log likelihood function -900.06222
> Estimation based on N = 1076, K = 6
> Inf.Cr.AIC = 1812.1 AIC/N = 1.684
> ---------------------------------------
> Log likelihood R-sqrd R2Adj
> Constants only -985.1678 .0864 .0739
> Note: R-sqrd = 1 - logL/Logl(constants)
> Warning: Model does not contain a full
> set of ASCs. R-sqrd is problematic. Use
> model setup with ;RHS=one to get LogL0.
> ---------------------------------------
> Response data are given as ind. choices
> Number of obs.= 1076, skipped 0 obs
>
> --------+--------------------------------------------------------------------
> | Standard Prob. 95% Confidence
> CHOICE| Coefficient Error z |z|>Z* Interval
>
> --------+--------------------------------------------------------------------
> FORTIFIC| .32357*** .08072 4.01 .0001 .16537 .48178
> WEIGHT| -.27733*** .09413 -2.95 .0032 -.46182 -.09285
> CERTIFIC| 1.60380*** .11305 14.19 .0000 1.38223 1.82536
> TWOSEZON| .78121*** .12545 6.23 .0000 .53533 1.02708
> MANYSEZO| 1.44881*** .11970 12.10 .0000 1.21419 1.68342
> NPRICE|-.61331D-04* .3452D-04 -1.78 .0756 -.12899D-03 .63316D-05
>
> --------+--------------------------------------------------------------------
> nnnnn.D-xx or D+xx => multiply by 10 to -xx or +xx.
> ***, **, * ==> Significance at 1%, 5%, 10% level.
> Model was estimated on Oct 19, 2024 at 11:49:37 AM
>
> -----------------------------------------------------------------------------
>
> Line search at iteration 53 does not improve the function
> Exiting optimization
>
>
> -----------------------------------------------------------------------------
> Generalized Mixed (RP) Logit Model
> Dependent variable CHOICE
> Log likelihood function -826.91886
> Restricted log likelihood -1182.10682
> Chi squared [ 29](P= .000) 710.37593
> Significance level .00000
> McFadden Pseudo R-squared .3004703
> Estimation based on N = 1076, K = 29
> Inf.Cr.AIC = 1711.8 AIC/N = 1.591
> ---------------------------------------
> Log likelihood R-sqrd R2Adj
> No coefficients -1182.1068 .3005 .2909
> Constants only -985.1678 .1606 .1492
> At start values -899.3489 .0805 .0680
> Note: R-sqrd = 1 - logL/Logl(constants)
> Warning: Model does not contain a full
> set of ASCs. R-sqrd is problematic. Use
> model setup with ;RHS=one to get LogL0.
> ---------------------------------------
> Response data are given as ind. choices
> Replications for simulated probs. = 100
> Used Halton sequences in simulations.
> RPL model with panel has 269 groups
> Fixed number of obsrvs./group= 4
> BHHH estimator used for asymp. variance
> Number of obs.= 1076, skipped 0 obs
>
> --------+--------------------------------------------------------------------
> | Standard Prob. 95% Confidence
> CHOICE| Coefficient Error z |z|>Z* Interval
>
> --------+--------------------------------------------------------------------
> |Random parameters in utility
> functions..............................
> FORTIFIC| 1.50696*** .54611 2.76 .0058 .43662 2.57731
> WEIGHT| -.20769 .56298 -.37 .7122 -1.31111 .89573
> CERTIFIC| 9.07085*** 1.59825 5.68 .0000 5.93834 12.20335
> TWOSEZON| 5.26384*** 1.05771 4.98 .0000 3.19076 7.33692
> MANYSEZO| 8.40225*** 1.58085 5.32 .0000 5.30384 11.50067
> NPRICE| .00063*** .00024 2.63 .0086 .00016 .00110
> |Diagonal values in Cholesky matrix,
> L...............................
> NsFORTIF| 4.97379*** 1.17328 4.24 .0000 2.67420 7.27337
> NsWEIGHT| .41995 1.33312 .32 .7528 -2.19292 3.03281
> NsCERTIF| 4.06231*** 1.47632 2.75 .0059 1.16877 6.95584
> NsTWOSEZ| 5.15553*** 1.34628 3.83 .0001 2.51686 7.79420
> NsMANYSE| 1.40139 1.40002 1.00 .3168 -1.34259 4.14538
> NsNPRICE| .00101*** .00029 3.54 .0004 .00045 .00157
> |Below diagonal values in L matrix. V =
> L*Lt.........................
> WEIG:FOR| -1.48938* .88668 -1.68 .0930 -3.22725 .24849
> CERT:FOR| -2.51743 1.59438 -1.58 .1143 -5.64236 .60749
> CERT:WEI| 4.19613** 1.80622 2.32 .0202 .65600 7.73626
> TWOS:FOR| 2.95810*** 1.06047 2.79 .0053 .87962 5.03659
> TWOS:WEI| .95649 1.46272 .65 .5132 -1.91039 3.82337
> TWOS:CER| -2.25893* 1.29302 -1.75 .0806 -4.79321 .27535
> MANY:FOR| -.32904 1.13783 -.29 .7724 -2.55915 1.90107
> MANY:WEI| 3.19079* 1.85861 1.72 .0860 -.45202 6.83360
> MANY:CER| -6.35832*** 1.52035 -4.18 .0000 -9.33814 -3.37849
> MANY:TWO| -7.79234*** 1.74477 -4.47 .0000 -11.21203 -4.37265
> NPRI:FOR| -.00020 .00025 -.80 .4223 -.00068 .00029
> NPRI:WEI| .00121** .00047 2.57 .0101 .00029 .00214
> NPRI:CER| -.00072* .00042 -1.69 .0908 -.00155 .00011
> NPRI:TWO| -.00051 .00037 -1.37 .1714 -.00123 .00022
> NPRI:MAN| .00027 .00043 .62 .5331 -.00058 .00112
> |Covariances of Random
> Parameters....................................
> TauScale| 1.13115*** .10543 10.73 .0000 .92452 1.33779
> |Weighting parameter gamma in GMX
> model..............................
> GammaMXL| -.22613** .09949 -2.27 .0230 -.42113 -.03113
> | Sample Mean Sample
> Std.Dev.....................................
> Sigma(i)| .97097 1.32745 .73 .4645 -1.63079 3.57272
> |Standard deviations of parameter
> distributions......................
> sdFORTIF| 4.97379*** 1.17328 4.24 .0000 2.67420 7.27337
> sdWEIGHT| 1.54746 .97110 1.59 .1110 -.35587 3.45078
> sdCERTIF| 6.35982*** 1.75250 3.63 .0003 2.92499 9.79465
> sdTWOSEZ| 6.43020*** 1.24563 5.16 .0000 3.98881 8.87159
> sdMANYSE| 10.6490*** 1.97084 5.40 .0000 6.7863 14.5118
> sdNPRICE| .00184*** .00036 5.15 .0000 .00114 .00253
> |Covariances of Random
> Parameters....................................
> WEIG:FOR| -7.40788 5.39137 -1.37 .1694 -17.97476 3.15901
> CERT:FOR| -12.5212 10.38866 -1.21 .2281 -32.8826 7.8402
> CERT:WEI| 5.51158 7.31301 .75 .4510 -8.82166 19.84481
> TWOS:FOR| 14.7130* 7.61270 1.93 .0533 -.2076 29.6336
> TWOS:WEI| -4.00408 4.20688 -.95 .3412 -12.24941 4.24125
> TWOS:CER| -12.6097 9.77976 -1.29 .1973 -31.7777 6.5582
> MANY:FOR| -1.63656 5.54775 -.29 .7680 -12.50996 9.23684
> MANY:WEI| 1.83003 5.29230 .35 .7295 -8.54268 12.20274
> MANY:CER| -11.6121 11.85646 -.98 .3274 -34.8504 11.6261
> MANY:TWO| -23.7320* 13.89019 -1.71 .0875 -50.9563 3.4923
> NPRI:FOR| -.00099 .00132 -.75 .4552 -.00358 .00161
> NPRI:WEI| .00080 .00187 .43 .6677 -.00287 .00448
> NPRI:CER| .00267 .00449 .60 .5518 -.00612 .01146
> NPRI:TWO| -.00042 .00295 -.14 .8871 -.00620 .00536
> NPRI:MAN| .01282** .00637 2.01 .0443 .00033 .02531
>
> --------+--------------------------------------------------------------------
> ***, **, * ==> Significance at 1%, 5%, 10% level.
> Model was estimated on Oct 19, 2024 at 11:54:35 AM
>
> -----------------------------------------------------------------------------
>
>
>
> --------+-----------------------------------------------------
> Cor.Mat.|FORTIFIC WEIGHT CERTIFIC TWOSEZON MANYSEZO NPRICE
> --------+-----------------------------------------------------
> FORTIFIC| 1.00000 -.96247 -.39583 .46003 -.03090 -.10825
> WEIGHT| -.96247 1.00000 .56003 -.40240 .11105 .28338
> CERTIFIC| -.39583 .56003 1.00000 -.30834 -.17146 .22877
> TWOSEZON| .46003 -.40240 -.30834 1.00000 -.34658 -.03546
> MANYSEZO| -.03090 .11105 -.17146 -.34658 1.00000 .65585
> NPRICE| -.10825 .28338 .22877 -.03546 .65585 1.00000
>
> Saved Individual Estimates of WTP in matrix WTP_I [ 269x5]
> Alternative Attribute Income/Cost
> Chosen FORTIFIC NPRICE
> Chosen WEIGHT NPRICE
> Chosen CERTIFIC NPRICE
> Chosen TWOSEZON NPRICE
> Chosen MANYSEZO NPRICE
> (Saved absolute values. Check signs of coefficients.)
> |-> MATRIX ; List ; 1/269*1'wtp_i $ ?mean wtp
>
> RESULT| 1 2 3 4
> 5
>
> --------+----------------------------------------------------------------------
> 1| -208387. 63355.3 -155431. -109576.
> -33803.6
>
>
> On Fri, Oct 18, 2024 at 6:11 PM Arthur Caplan via Limdep <
> limdep at mailman.sydney.edu.au> wrote:
>
> > Hello Medard,
> >
> > It might help if you shared the code you are using.
> >
> > Arthur Caplan
> >
> >
> > Department of Applied Economics
> > Utah State University
> > 4835 Old Main Hill
> > Logan, Utah 84322-4835
> > tel: 435-797-0775
> > web: https://url.au.m.mimecastprotect.com/s/7A4aCp81lrtxz9QB8iPf1FGeER7?domain=sites.google.com
> > ________________________________
> > From: Limdep <limdep-bounces at mailman.sydney.edu.au> on behalf of medard
> > kakuru via Limdep <limdep at mailman.sydney.edu.au>
> > Sent: Thursday, October 17, 2024 11:58 PM
> > To: Limdep and Nlogit Mailing List <limdep at mailman.sydney.edu.au>
> > Cc: medard kakuru <medakseth at gmail.com>
> > Subject: Re: [Limdep Nlogit List] Exploded and negative wtp estimates
> >
> > Greetings to you all!
> > I estimated wtp using three models: RPL, Generalised mixed logit in
> > preference space and in wtp space. My estimates are extremely big - I
> have
> > tried the normal, triangular and log normal distributions. They become
> > bigger if I use more than 100 random draws, halton draws don't make them
> > any better. Secondly, they are negative yet the MNL estimates are
> positive
> > and my price parameter is negative. What could be the reason/problem for
> > negative and explosive wtp estimates?
> >
> >
> > Best regards,
> >
> > Medard
> >
> > >
> > >
> > >
> > _______________________________________________
> > Limdep site list
> > Limdep at mailman.sydney.edu.au
> > http://limdep.itls.usyd.edu.au
> >
> > _______________________________________________
> > Limdep site list
> > Limdep at mailman.sydney.edu.au
> > http://limdep.itls.usyd.edu.au
> >
> >
> _______________________________________________
> Limdep site list
> Limdep at mailman.sydney.edu.au
> http://limdep.itls.usyd.edu.au
>
>
More information about the Limdep
mailing list