From david.hensher at sydney.edu.au Fri Feb 10 12:42:57 2017 From: david.hensher at sydney.edu.au (David Hensher) Date: Fri, 10 Feb 2017 01:42:57 +0000 Subject: [Limdep Nlogit List] testing the list is now operational again Message-ID: <589D1AA1.6070808@sydney.edu.au> This is a test only David -- DAVID HENSHER FASSA, PhD| Professor and Founding Director Institute of Transport and Logistics Studies | The University of Sydney Business School THE UNIVERSITY OF SYDNEY Rm 201, Building H73| The University of Sydney | NSW | 2006 Street Address: 378 Abercrombie St, Darlington NSW 2008 T +61 2 9114 1871 | F +61 2 9114 1863 | M +61 418 433 057 E David.Hensher at sydney.edu.au | W sydney.edu.au/business/itls Celebrating 25 years of ITLS: 1991-2016 http://youtu.be/s2D0T1crZwY ERA Rank 5 (Transportation and Freight Services) Co-Founder of the International Conference Series on Competition and Ownership of Land Passenger Transport (The 'Thredbo' Series) http://www.thredbo-conference-series.org/ Join the ITLS group on LinkedIn Second edition of Applied Choice Analysis now available at www.cambridge.org/9781107465923 CRICOS 00026A This email plus any attachments to it are confidential. Any unauthorised use is strictly prohibited. If you receive this email in error, please delete it and any attachments. Please think of our environment and only print this e-mail if necessary. From cem.payasli at emu.edu.tr Wed Feb 15 20:20:41 2017 From: cem.payasli at emu.edu.tr (Cem Payasli) Date: Wed, 15 Feb 2017 09:20:41 +0000 Subject: [Limdep Nlogit List] Same predictors in both regression equation and probability equation Message-ID: <336437B377B73841AA857EA0A13A3C3A0BD680C0@DAUMAIL.emu.edu.tr> Dear All, I have fitted the following specification for negative binomial latent class model with 2 classes. My questions: 1) Is it possible to compare several non-nested alternatives on the basis of AIC or log-likelihood ? 2) How can I impose restrictions on both mean(regression) part and class membership probability part ? 3) What would be the interpretation of second class variables with fixed values always set to "0" ? 4) In limdep manual LR test example is provided so the restriction of extended LC (predictors as class probablities) versus constant probablities ( no predictors as class probablities) is given. In th example below Asian variable is signficant in mean regression models but insignficant as class probablity predictor. Supoosing it were signficant there as well, could we say it better belongs to eithe mean part of probability again ? Thanks. for any suggestion. Cem Payasl?o?lu Department of Economics Eastern Mediterranean University, Magusa,TRNC Mersin 10 Turkey Latent Class / Panel NegBnReg Model Dependent variable PATENT Log likelihood function -2772.13293 Restricted log likelihood -386994.81568 Chi squared [ 15 d.f.] 768445.36551 Significance level .00000 McFadden Pseudo R-squared .9928368 Estimation based on N = 566, K = 21 Inf.Cr.AIC = 5586.3 AIC/N = 9.870 Model estimated: Feb 14, 2017, 20:49:28 Sample is 1 pds and 566 individuals Negative binomial regression model --------+-------------------------------------------------------------------- | Standard Prob. 95% Confidence PATENT| Coefficient Error z |z|>Z* Interval --------+-------------------------------------------------------------------- |Model parameters for latent class 1 LRD| .42219*** .07490 5.64 .0000 .27540 .56899 LSALES| .27586* .14920 1.85 .0645 -.01656 .56828 LCAPEX| -.20480* .11319 -1.81 .0704 -.42664 .01705 R1| .88042 .70745 1.24 .2133 -.50616 2.26700 R2| 1.22297* .73600 1.66 .0966 -.21955 2.66550 R3| 1.36867* .72247 1.89 .0582 -.04735 2.78469 R4| 1.32462* .69588 1.90 .0570 -.03929 2.68853 ASIAN| .50393*** .17740 2.84 .0045 .15624 .85163 Alpha| 1.19664*** .19265 6.21 .0000 .81906 1.57422 |Model parameters for latent class 2 LRD| .21053*** .05043 4.17 .0000 .11169 .30938 LSALES| .09020 .05760 1.57 .1174 -.02269 .20310 LCAPEX| -.02938 .05241 -.56 .5751 -.13210 .07334 R1| 1.48180*** .30280 4.89 .0000 .88832 2.07528 R2| 2.17540*** .37662 5.78 .0000 1.43724 2.91356 R3| 1.77508*** .25976 6.83 .0000 1.26596 2.28419 R4| 1.50124*** .27273 5.50 .0000 .96670 2.03578 ASIAN| .19233** .07907 2.43 .0150 .03735 .34732 Alpha| 5.12409*** .98185 5.22 .0000 3.19970 7.04849 |Estimated prior probabilities for class membership ONE_1| -8.52758*** 1.45420 -5.86 .0000 -11.37775 -5.67741 LSALES_1| .95047*** .16183 5.87 .0000 .63329 1.26766 ASIAN_1| .64236* .35511 1.81 .0705 -.05364 1.33836 ONE_2| 0.0 .....(Fixed Parameter)..... LSALES_2| 0.0 .....(Fixed Parameter)..... ASIAN_2| 0.0 .....(Fixed Parameter)..... --------+-------------------------------------------------------------------- Note: ***, **, * ==> Significance at 1%, 5%, 10% level. Fixed parameter ... is constrained to equal the value or had a nonpositive st.error because of an earlier problem From zein.kallas at upc.edu Fri Feb 17 00:27:32 2017 From: zein.kallas at upc.edu (Zein Kallas Calot) Date: Thu, 16 Feb 2017 14:27:32 +0100 Subject: [Limdep Nlogit List] testing the list is now operational again In-Reply-To: <589D1AA1.6070808@sydney.edu.au> References: <589D1AA1.6070808@sydney.edu.au> Message-ID: <010501d28858$6db0f960$4912ec20$@upc.edu> Dear all, I followed the Nlogit 5 reference guide manual on page N-567 to calculate the conditional confidence interval of parameters for each individual after estimating an RPL model. This is the code that I used (example with 2 random attribute for simplicity) CALC ; Ran (12345) $ NLOGIT;Lhs=YC1;Choices=ALT1,ALT2,NO; Rhs=A1,A2,A3,A4 ;RPL;Fcn=A1(n),A2(n) ;PDS=8 ;halton ;parameters $ the conditional confidence interval of parameters MATRIX ; bn = beta_i ; sn = sdbeta_i $ CREATE ; bA1n = 0 ; bA2n = 0 $ CREATE ; sA1n = 0 ; sA2n = 0 $ NAMELIST ; betan = bA1n, bA2n $ NAMELIST ; sbetan = sA1n, sA2n $ CREATE ; betan = bn $ CREATE ; sbetan = sn $ CREATE ; vbA1n = sA1n^2 ; vbA2n = sA2n^2 $ CALC ; List ; sdbA1n = Sqr(xbr(vbA1n) + Var(bA1n)) ; sdbA2n = Sqr(xbr(vbA2n) + Var(bA2n)) $ CREATE ; lowerbA1 = bA1n - 2*sA1n ; upperbA1 = bA1n + 2*sA1n $ CREATE ; lowerbA2 = bA2n - 2*sA2n ; upperbA2 = bA2n + 2*sA2n $ CREATE ; person = Trn(1,1) $ CALC ; meanbA1n = Xbr(bA1n) $ CALC ; highbA1n = meanbA1n + 2*sdbA1n $ CALC ; lowbA1n = meanbA1n - 2*sdbA1n $ CALC ; meanbA2n = Xbr(bA2n) $ CALC ; highbA2n = meanbA2n + 2*sdbA2n $ CALC ; lowbA2n = meanbA2n - 2*sdbA2n $ PLOT ; Lhs = person ; Rhs = lowerbA1,upperbA1 ; Centipede ; Title = Confidence Limits for b_A1 for normal Model ; Bars = meanbA1n,highbA1n,lowbA1n $ PLOT ; Lhs = person ; Rhs = lowerbA2,upperbA2 ; Centipede ; Title = Confidence Limits for b_A2 for normal Model ; Bars = meanbA2n,highbA2n,lowbA2n $ Questions: 1. Is this code correct? Or I missed something? 2. The Figures were plotted for each row in the data and not each person (i.e. for the total number of rows: Number of persons * 8 choice sets * 3 alternatives). How can I correct for that to obtain a similar figure of the Manual (Figure N29.7) 3. Is it possible to extract the values of the confidence interval by person and attribute in a separate data set? Best regards Zein Kallas CREDA Barcelona -----Mensaje original----- De: Limdep [mailto:limdep-bounces at itls.usyd.edu.au] En nombre de David Hensher via Limdep Enviado el: 10 February 2017 02:43 Para: limdep at itls.usyd.edu.au CC: David Hensher Asunto: [Limdep Nlogit List] testing the list is now operational again This is a test only David -- DAVID HENSHER FASSA, PhD| Professor and Founding Director Institute of Transport and Logistics Studies | The University of Sydney Business School THE UNIVERSITY OF SYDNEY Rm 201, Building H73| The University of Sydney | NSW | 2006 Street Address: 378 Abercrombie St, Darlington NSW 2008 T +61 2 9114 1871 | F +61 2 9114 1863 | M +61 418 433 057 E David.Hensher at sydney.edu.au | W sydney.edu.au/business/itls Celebrating 25 years of ITLS: 1991-2016 http://youtu.be/s2D0T1crZwY ERA Rank 5 (Transportation and Freight Services) Co-Founder of the International Conference Series on Competition and Ownership of Land Passenger Transport (The 'Thredbo' Series) http://www.thredbo-conference-series.org/ Join the ITLS group on LinkedIn Second edition of Applied Choice Analysis now available at www.cambridge.org/9781107465923 CRICOS 00026A This email plus any attachments to it are confidential. Any unauthorised use is strictly prohibited. If you receive this email in error, please delete it and any attachments. Please think of our environment and only print this e-mail if necessary. _______________________________________________ Limdep site list Limdep at itls.usyd.edu.au http://limdep.itls.usyd.edu.au From wgreene at stern.nyu.edu Fri Feb 17 07:01:27 2017 From: wgreene at stern.nyu.edu (William Greene) Date: Thu, 16 Feb 2017 15:01:27 -0500 Subject: [Limdep Nlogit List] testing the list is now operational again In-Reply-To: <010501d28858$6db0f960$4912ec20$@upc.edu> References: <589D1AA1.6070808@sydney.edu.au> <010501d28858$6db0f960$4912ec20$@upc.edu> Message-ID: Zein. Before your plot, use SAMPLE to set the sample to the rows you want to plot. Since you extracted the values from matrices, that would be 1 to N where you have 1 to 24*N. /Bill Greene On Thu, Feb 16, 2017 at 8:27 AM, Zein Kallas Calot wrote: > Dear all, > I followed the Nlogit 5 reference guide manual on page N-567 to calculate > the conditional confidence > interval of parameters for each individual after estimating an RPL model. > This is the code that I > used (example with 2 random attribute for simplicity) > > CALC ; Ran (12345) $ > NLOGIT;Lhs=YC1;Choices=ALT1,ALT2,NO; > Rhs=A1,A2,A3,A4 > ;RPL;Fcn=A1(n),A2(n) > ;PDS=8 > ;halton > ;parameters > $ > > the conditional confidence interval of parameters > > MATRIX ; bn = beta_i ; sn = sdbeta_i $ > > CREATE ; bA1n = 0 ; bA2n = 0 $ > CREATE ; sA1n = 0 ; sA2n = 0 $ > NAMELIST ; betan = bA1n, bA2n $ > NAMELIST ; sbetan = sA1n, sA2n $ > CREATE ; betan = bn $ > CREATE ; sbetan = sn $ > > CREATE ; vbA1n = sA1n^2 ; vbA2n = sA2n^2 $ > > CALC ; List ; sdbA1n = Sqr(xbr(vbA1n) + Var(bA1n)) > ; sdbA2n = Sqr(xbr(vbA2n) + Var(bA2n)) $ > > CREATE ; lowerbA1 = bA1n - 2*sA1n ; upperbA1 = bA1n + 2*sA1n $ > CREATE ; lowerbA2 = bA2n - 2*sA2n ; upperbA2 = bA2n + 2*sA2n $ > > CREATE ; person = Trn(1,1) $ > > CALC ; meanbA1n = Xbr(bA1n) $ > CALC ; highbA1n = meanbA1n + 2*sdbA1n $ > CALC ; lowbA1n = meanbA1n - 2*sdbA1n $ > > CALC ; meanbA2n = Xbr(bA2n) $ > CALC ; highbA2n = meanbA2n + 2*sdbA2n $ > CALC ; lowbA2n = meanbA2n - 2*sdbA2n $ > > PLOT ; Lhs = person ; Rhs = lowerbA1,upperbA1 > ; Centipede > ; Title = Confidence Limits for b_A1 for normal Model > ; Bars = meanbA1n,highbA1n,lowbA1n $ > > PLOT ; Lhs = person ; Rhs = lowerbA2,upperbA2 > ; Centipede > ; Title = Confidence Limits for b_A2 for normal Model > ; Bars = meanbA2n,highbA2n,lowbA2n $ > > Questions: > 1. Is this code correct? Or I missed something? > 2. The Figures were plotted for each row in the data and not each > person (i.e. for the total > number of rows: Number of persons * 8 choice sets * 3 alternatives). How > can I correct for that to > obtain a similar figure of the Manual (Figure N29.7) > 3. Is it possible to extract the values of the confidence interval by > person and attribute in a > separate data set? > > Best regards > > Zein Kallas > CREDA > Barcelona > > -----Mensaje original----- > De: Limdep [mailto:limdep-bounces at itls.usyd.edu.au] En nombre de David > Hensher via Limdep > Enviado el: 10 February 2017 02:43 > Para: limdep at itls.usyd.edu.au > CC: David Hensher > Asunto: [Limdep Nlogit List] testing the list is now operational again > > This is a test only > > David > > -- > DAVID HENSHER FASSA, PhD| Professor and Founding Director Institute of > Transport and Logistics > Studies | The University of Sydney Business School > > THE UNIVERSITY OF SYDNEY > Rm 201, Building H73| The University of Sydney | NSW | 2006 Street > Address: 378 Abercrombie St, > Darlington NSW 2008 T +61 2 9114 1871 | F +61 2 9114 1863 | M +61 418 433 > 057 E > David.Hensher at sydney.edu.au | W sydney.edu.au/business/itls > > Celebrating 25 years of ITLS: 1991-2016 http://youtu.be/s2D0T1crZwY ERA > Rank 5 (Transportation and > Freight Services) Co-Founder of the International Conference Series on > Competition and Ownership of > Land Passenger Transport (The 'Thredbo' Series) > http://www.thredbo-conference-series.org/ > Join the ITLS group on LinkedIn > Second edition of Applied Choice Analysis now available at > www.cambridge.org/9781107465923 > > > CRICOS 00026A > This email plus any attachments to it are confidential. Any unauthorised > use is strictly prohibited. > If you receive this email in error, please delete it and any attachments. > Please think of our environment and only print this e-mail if necessary. > > _______________________________________________ > Limdep site list > Limdep at itls.usyd.edu.au > http://limdep.itls.usyd.edu.au > > > _______________________________________________ > Limdep site list > Limdep at limdep.itls.usyd.edu.au > http://limdep.itls.usyd.edu.au > > -- William Greene Department of Economics Stern School of Business, New York University 44 West 4 St., 7-90 New York, NY, 10012 URL: http://people.stern.nyu.edu/wgreene Email: wgreene at stern.nyu.edu Ph. +1.212.998.0876 From zein.kallas at upc.edu Sat Feb 18 21:46:28 2017 From: zein.kallas at upc.edu (Zein Kallas Calot) Date: Sat, 18 Feb 2017 11:46:28 +0100 Subject: [Limdep Nlogit List] testing the list is now operational again In-Reply-To: References: <589D1AA1.6070808@sydney.edu.au> <010501d28858$6db0f960$4912ec20$@upc.edu> Message-ID: <01b601d289d4$42505a40$c6f10ec0$@upc.edu> Dear professor Greene Adding sample from 1 to N works well. Thank you. Regards Zein Kallas Calot, Ph.D. Universitat Polit?nica de Catalunya-UPC/Escola Superior d?Agricultura de Barcelona-ESAB Professor Agregat- Departament d'Enginyeria Agroaliment?ria i Biotecnologia-DEAB Investigador- Centre de Recerca en Economia i Desenvolupament Agroalimentari CREDA-UPC-IRTA Zein.kallas at upc.edu http://www.creda.es/personnel/en -----Mensaje original----- De: Limdep [mailto:limdep-bounces at limdep.itls.usyd.edu.au] En nombre de William Greene Enviado el: 16 February 2017 21:01 Para: Limdep and Nlogit Mailing List Asunto: Re: [Limdep Nlogit List] testing the list is now operational again Zein. Before your plot, use SAMPLE to set the sample to the rows you want to plot. Since you extracted the values from matrices, that would be 1 to N where you have 1 to 24*N. /Bill Greene On Thu, Feb 16, 2017 at 8:27 AM, Zein Kallas Calot wrote: > Dear all, > I followed the Nlogit 5 reference guide manual on page N-567 to > calculate the conditional confidence interval of parameters for each > individual after estimating an RPL model. > This is the code that I > used (example with 2 random attribute for simplicity) > > CALC ; Ran (12345) $ > NLOGIT;Lhs=YC1;Choices=ALT1,ALT2,NO; > Rhs=A1,A2,A3,A4 > ;RPL;Fcn=A1(n),A2(n) > ;PDS=8 > ;halton > ;parameters > $ > > the conditional confidence interval of parameters > > MATRIX ; bn = beta_i ; sn = sdbeta_i $ > > CREATE ; bA1n = 0 ; bA2n = 0 $ > CREATE ; sA1n = 0 ; sA2n = 0 $ > NAMELIST ; betan = bA1n, bA2n $ > NAMELIST ; sbetan = sA1n, sA2n $ > CREATE ; betan = bn $ > CREATE ; sbetan = sn $ > > CREATE ; vbA1n = sA1n^2 ; vbA2n = sA2n^2 $ > > CALC ; List ; sdbA1n = Sqr(xbr(vbA1n) + Var(bA1n)) > ; sdbA2n = Sqr(xbr(vbA2n) + Var(bA2n)) $ > > CREATE ; lowerbA1 = bA1n - 2*sA1n ; upperbA1 = bA1n + 2*sA1n $ CREATE > ; lowerbA2 = bA2n - 2*sA2n ; upperbA2 = bA2n + 2*sA2n $ > > CREATE ; person = Trn(1,1) $ > > CALC ; meanbA1n = Xbr(bA1n) $ > CALC ; highbA1n = meanbA1n + 2*sdbA1n $ CALC ; lowbA1n = meanbA1n - > 2*sdbA1n $ > > CALC ; meanbA2n = Xbr(bA2n) $ > CALC ; highbA2n = meanbA2n + 2*sdbA2n $ CALC ; lowbA2n = meanbA2n - > 2*sdbA2n $ > > PLOT ; Lhs = person ; Rhs = lowerbA1,upperbA1 ; Centipede ; Title = > Confidence Limits for b_A1 for normal Model ; Bars = > meanbA1n,highbA1n,lowbA1n $ > > PLOT ; Lhs = person ; Rhs = lowerbA2,upperbA2 ; Centipede ; Title = > Confidence Limits for b_A2 for normal Model ; Bars = > meanbA2n,highbA2n,lowbA2n $ > > Questions: > 1. Is this code correct? Or I missed something? > 2. The Figures were plotted for each row in the data and not each > person (i.e. for the total > number of rows: Number of persons * 8 choice sets * 3 alternatives). > How can I correct for that to obtain a similar figure of the Manual > (Figure N29.7) > 3. Is it possible to extract the values of the confidence interval by > person and attribute in a > separate data set? > > Best regards > > Zein Kallas > CREDA > Barcelona > > -----Mensaje original----- > De: Limdep [mailto:limdep-bounces at itls.usyd.edu.au] En nombre de David > Hensher via Limdep Enviado el: 10 February 2017 02:43 > Para: limdep at itls.usyd.edu.au > CC: David Hensher > Asunto: [Limdep Nlogit List] testing the list is now operational again > > This is a test only > > David > > -- > DAVID HENSHER FASSA, PhD| Professor and Founding Director Institute of > Transport and Logistics Studies | The University of Sydney Business > School > > THE UNIVERSITY OF SYDNEY > Rm 201, Building H73| The University of Sydney | NSW | 2006 Street > Address: 378 Abercrombie St, > Darlington NSW 2008 T +61 2 9114 1871 | F +61 2 9114 1863 | M +61 418 > 433 > 057 E > David.Hensher at sydney.edu.au | W sydney.edu.au/business/itls > > Celebrating 25 years of ITLS: 1991-2016 http://youtu.be/s2D0T1crZwY > ERA Rank 5 (Transportation and Freight Services) Co-Founder of the > International Conference Series on Competition and Ownership of Land > Passenger Transport (The 'Thredbo' Series) > http://www.thredbo-conference-series.org/ > Join the ITLS group on LinkedIn > Second edition of Applied Choice Analysis now available at > www.cambridge.org/9781107465923 > > > CRICOS 00026A > This email plus any attachments to it are confidential. Any > unauthorised use is strictly prohibited. > If you receive this email in error, please delete it and any attachments. > Please think of our environment and only print this e-mail if necessary. > > _______________________________________________ > Limdep site list > Limdep at itls.usyd.edu.au > http://limdep.itls.usyd.edu.au > > > _______________________________________________ > Limdep site list > Limdep at limdep.itls.usyd.edu.au > http://limdep.itls.usyd.edu.au > > -- William Greene Department of Economics Stern School of Business, New York University 44 West 4 St., 7-90 New York, NY, 10012 URL: http://people.stern.nyu.edu/wgreene Email: wgreene at stern.nyu.edu Ph. +1.212.998.0876 _______________________________________________ Limdep site list Limdep at limdep.itls.usyd.edu.au http://limdep.itls.usyd.edu.au From Lixian.Qian at xjtlu.edu.cn Fri Feb 24 17:36:34 2017 From: Lixian.Qian at xjtlu.edu.cn (Lixian Qian) Date: Fri, 24 Feb 2017 06:36:34 +0000 Subject: [Limdep Nlogit List] Exit Optimization without Status Code Message-ID: Hello, we are running a mixed logit using NLogit. The model runs well with coefficients with normal distributions (status =0), but if we change the distribution of monetary attributes to lognormal, the estimation ends with this statement. "Line search at iteration 46 does not improve fn. Exiting optimization." Can anyone please explain the issue? Thank you. Best, Lixian From david.hensher at sydney.edu.au Fri Feb 24 17:49:24 2017 From: david.hensher at sydney.edu.au (David Hensher) Date: Fri, 24 Feb 2017 06:49:24 +0000 Subject: [Limdep Nlogit List] Exit Optimization without Status Code In-Reply-To: References: Message-ID: <58AFD773.20709@sydney.edu.au> This tells you that it cannot find a solution. Thts is quite common with lognormal. A lognormal parameter is created as beta(i) = exp(b + sigma*w(i)) What you are looking at in your output are b and sigma. Sigma has to be positive. b can be positive or negative. The lognormal distribution is used in RP models to constrain the sign of a coefficient. The specification will appear as in ; Fcn = price(l). However, this forces the coefficient to be positive. To force it to be negative the well known ?trick? is to multiply the price variable by minus one, then force the coefficient to be positive as usual. Alternatively, the sign can be built into the coefficient by using ; Fcn = -price(l). David On 24/02/2017 5:36 PM, Lixian Qian wrote: Hello, we are running a mixed logit using NLogit. The model runs well with coefficients with normal distributions (status =0), but if we change the distribution of monetary attributes to lognormal, the estimation ends with this statement. "Line search at iteration 46 does not improve fn. Exiting optimization." Can anyone please explain the issue? Thank you. Best, Lixian _______________________________________________ Limdep site list Limdep at limdep.itls.usyd.edu.au http://limdep.itls.usyd.edu.au -- DAVID HENSHER FASSA, PhD| Professor and Founding Director Institute of Transport and Logistics Studies | The University of Sydney Business School THE UNIVERSITY OF SYDNEY Rm 201, Building H73| The University of Sydney | NSW | 2006 Street Address: 378 Abercrombie St, Darlington NSW 2008 T +61 2 9114 1871 | F +61 2 9114 1863 | M +61 418 433 057 E David.Hensher at sydney.edu.au | W sydney.edu.au/business/itls Celebrating 25 years of ITLS: 1991-2016 http://youtu.be/s2D0T1crZwY ERA Rank 5 (Transportation and Freight Services) Co-Founder of the International Conference Series on Competition and Ownership of Land Passenger Transport (The 'Thredbo' Series) http://www.thredbo-conference-series.org/ Join the ITLS group on LinkedIn Second edition of Applied Choice Analysis now available at www.cambridge.org/9781107465923 CRICOS 00026A This email plus any attachments to it are confidential. Any unauthorised use is strictly prohibited. If you receive this email in error, please delete it and any attachments. Please think of our environment and only print this e-mail if necessary. From Lixian.Qian at xjtlu.edu.cn Fri Feb 24 17:59:07 2017 From: Lixian.Qian at xjtlu.edu.cn (Lixian Qian) Date: Fri, 24 Feb 2017 06:59:07 +0000 Subject: [Limdep Nlogit List] Exit Optimization without Status Code In-Reply-To: <58AFD773.20709@sydney.edu.au> References: <58AFD773.20709@sydney.edu.au> Message-ID: Dear David, Thank you for your prompt response. Yes, we did transform the price variable by multiplying minus one and then ran the model with lognormal distribution assumption. We are very intrigued by your comment that "this is quite common with lognormal". It will be highly appreciated if you can kindly expand on that comment in the context of using different distributions for price in mixed logit model. Thank you. Best, Lixian From: David Hensher [mailto:david.hensher at sydney.edu.au] Sent: 24 February 2017 14:49 To: Limdep and Nlogit Mailing List Cc: Lixian Qian Subject: Re: [Limdep Nlogit List] Exit Optimization without Status Code This tells you that it cannot find a solution. Thts is quite common with lognormal. A lognormal parameter is created as beta(i) = exp(b + sigma*w(i)) What you are looking at in your output are b and sigma. Sigma has to be positive. b can be positive or negative. The lognormal distribution is used in RP models to constrain the sign of a coefficient. The specification will appear as in ; Fcn = price(l). However, this forces the coefficient to be positive. To force it to be negative the well known 'trick' is to multiply the price variable by minus one, then force the coefficient to be positive as usual. Alternatively, the sign can be built into the coefficient by using ; Fcn = -price(l). David On 24/02/2017 5:36 PM, Lixian Qian wrote: Hello, we are running a mixed logit using NLogit. The model runs well with coefficients with normal distributions (status =0), but if we change the distribution of monetary attributes to lognormal, the estimation ends with this statement. "Line search at iteration 46 does not improve fn. Exiting optimization." Can anyone please explain the issue? Thank you. Best, Lixian _______________________________________________ Limdep site list Limdep at limdep.itls.usyd.edu.au http://limdep.itls.usyd.edu.au -- DAVID HENSHER FASSA, PhD| Professor and Founding Director Institute of Transport and Logistics Studies | The University of Sydney Business School THE UNIVERSITY OF SYDNEY Rm 201, Building H73| The University of Sydney | NSW | 2006 Street Address: 378 Abercrombie St, Darlington NSW 2008 T +61 2 9114 1871 | F +61 2 9114 1863 | M +61 418 433 057 E David.Hensher at sydney.edu.au | W sydney.edu.au/business/itls Celebrating 25 years of ITLS: 1991-2016 http://youtu.be/s2D0T1crZwY ERA Rank 5 (Transportation and Freight Services) Co-Founder of the International Conference Series on Competition and Ownership of Land Passenger Transport (The 'Thredbo' Series) http://www.thredbo-conference-series.org/ Join the ITLS group on LinkedIn Second edition of Applied Choice Analysis now available at www.cambridge.org/9781107465923 CRICOS 00026A This email plus any attachments to it are confidential. Any unauthorised use is strictly prohibited. If you receive this email in error, please delete it and any attachments. Please think of our environment and only print this e-mail if necessary. From david.hensher at sydney.edu.au Fri Feb 24 18:13:37 2017 From: david.hensher at sydney.edu.au (David Hensher) Date: Fri, 24 Feb 2017 07:13:37 +0000 Subject: [Limdep Nlogit List] Exit Optimization without Status Code In-Reply-To: References: <58AFD773.20709@sydney.edu.au>, Message-ID: Lognormals have very long tails and this is a concern if you want to get WTP that are sensible, and often issue is data. If you want a specific sign then try a constrained normal or triangular. This does not solve the issue but experience suggests we resolve issue of very high or low estimates on the distribution. There are arbitrary analytical distributions and will always have issues and you need to be aware of this in trying different distns to try and explore an your data. This in art as much as science and one reason people use LCM Sent from my iPhone 0418 433 057 David A Hensher Note: do not use hgroup at optusnet.com.au from now on, and instead use hgroup at hensher.com.au or David.hensher at bigpond.com.au or David.hensher at sydney.edu.au These emails are linked so use one only On 24 Feb 2017, at 5:59 pm, Lixian Qian > wrote: Dear David, Thank you for your prompt response. Yes, we did transform the price variable by multiplying minus one and then ran the model with lognormal distribution assumption. We are very intrigued by your comment that ?this is quite common with lognormal?. It will be highly appreciated if you can kindly expand on that comment in the context of using different distributions for price in mixed logit model. Thank you. Best, Lixian From: David Hensher [mailto:david.hensher at sydney.edu.au] Sent: 24 February 2017 14:49 To: Limdep and Nlogit Mailing List Cc: Lixian Qian Subject: Re: [Limdep Nlogit List] Exit Optimization without Status Code This tells you that it cannot find a solution. Thts is quite common with lognormal. A lognormal parameter is created as beta(i) = exp(b + sigma*w(i)) What you are looking at in your output are b and sigma. Sigma has to be positive. b can be positive or negative. The lognormal distribution is used in RP models to constrain the sign of a coefficient. The specification will appear as in ; Fcn = price(l). However, this forces the coefficient to be positive. To force it to be negative the well known ?trick? is to multiply the price variable by minus one, then force the coefficient to be positive as usual. Alternatively, the sign can be built into the coefficient by using ; Fcn = -price(l). David On 24/02/2017 5:36 PM, Lixian Qian wrote: Hello, we are running a mixed logit using NLogit. The model runs well with coefficients with normal distributions (status =0), but if we change the distribution of monetary attributes to lognormal, the estimation ends with this statement. "Line search at iteration 46 does not improve fn. Exiting optimization." Can anyone please explain the issue? Thank you. Best, Lixian _______________________________________________ Limdep site list Limdep at limdep.itls.usyd.edu.au http://limdep.itls.usyd.edu.au -- DAVID HENSHER FASSA, PhD| Professor and Founding Director Institute of Transport and Logistics Studies | The University of Sydney Business School THE UNIVERSITY OF SYDNEY Rm 201, Building H73| The University of Sydney | NSW | 2006 Street Address: 378 Abercrombie St, Darlington NSW 2008 T +61 2 9114 1871 | F +61 2 9114 1863 | M +61 418 433 057 E David.Hensher at sydney.edu.au | W sydney.edu.au/business/itls Celebrating 25 years of ITLS: 1991-2016 http://youtu.be/s2D0T1crZwY ERA Rank 5 (Transportation and Freight Services) Co-Founder of the International Conference Series on Competition and Ownership of Land Passenger Transport (The 'Thredbo' Series) http://www.thredbo-conference-series.org/ Join the ITLS group on LinkedIn Second edition of Applied Choice Analysis now available at www.cambridge.org/9781107465923 CRICOS 00026A This email plus any attachments to it are confidential. Any unauthorised use is strictly prohibited. If you receive this email in error, please delete it and any attachments. Please think of our environment and only print this e-mail if necessary. From Lixian.Qian at xjtlu.edu.cn Fri Feb 24 18:18:12 2017 From: Lixian.Qian at xjtlu.edu.cn (Lixian Qian) Date: Fri, 24 Feb 2017 07:18:12 +0000 Subject: [Limdep Nlogit List] Exit Optimization without Status Code In-Reply-To: References: <58AFD773.20709@sydney.edu.au>, Message-ID: Dear David, Thank you for your response again. Best, Lixian From: David Hensher [mailto:david.hensher at sydney.edu.au] Sent: 24 February 2017 15:14 To: Lixian Qian Cc: Limdep and Nlogit Mailing List Subject: Re: [Limdep Nlogit List] Exit Optimization without Status Code Lognormals have very long tails and this is a concern if you want to get WTP that are sensible, and often issue is data. If you want a specific sign then try a constrained normal or triangular. This does not solve the issue but experience suggests we resolve issue of very high or low estimates on the distribution. There are arbitrary analytical distributions and will always have issues and you need to be aware of this in trying different distns to try and explore an your data. This in art as much as science and one reason people use LCM Sent from my iPhone 0418 433 057 David A Hensher Note: do not use hgroup at optusnet.com.au from now on, and instead use hgroup at hensher.com.au or David.hensher at bigpond.com.au or David.hensher at sydney.edu.au These emails are linked so use one only On 24 Feb 2017, at 5:59 pm, Lixian Qian > wrote: Dear David, Thank you for your prompt response. Yes, we did transform the price variable by multiplying minus one and then ran the model with lognormal distribution assumption. We are very intrigued by your comment that "this is quite common with lognormal". It will be highly appreciated if you can kindly expand on that comment in the context of using different distributions for price in mixed logit model. Thank you. Best, Lixian From: David Hensher [mailto:david.hensher at sydney.edu.au] Sent: 24 February 2017 14:49 To: Limdep and Nlogit Mailing List Cc: Lixian Qian Subject: Re: [Limdep Nlogit List] Exit Optimization without Status Code This tells you that it cannot find a solution. Thts is quite common with lognormal. A lognormal parameter is created as beta(i) = exp(b + sigma*w(i)) What you are looking at in your output are b and sigma. Sigma has to be positive. b can be positive or negative. The lognormal distribution is used in RP models to constrain the sign of a coefficient. The specification will appear as in ; Fcn = price(l). However, this forces the coefficient to be positive. To force it to be negative the well known 'trick' is to multiply the price variable by minus one, then force the coefficient to be positive as usual. Alternatively, the sign can be built into the coefficient by using ; Fcn = -price(l). David On 24/02/2017 5:36 PM, Lixian Qian wrote: Hello, we are running a mixed logit using NLogit. The model runs well with coefficients with normal distributions (status =0), but if we change the distribution of monetary attributes to lognormal, the estimation ends with this statement. "Line search at iteration 46 does not improve fn. Exiting optimization." Can anyone please explain the issue? Thank you. Best, Lixian _______________________________________________ Limdep site list Limdep at limdep.itls.usyd.edu.au http://limdep.itls.usyd.edu.au -- DAVID HENSHER FASSA, PhD| Professor and Founding Director Institute of Transport and Logistics Studies | The University of Sydney Business School THE UNIVERSITY OF SYDNEY Rm 201, Building H73| The University of Sydney | NSW | 2006 Street Address: 378 Abercrombie St, Darlington NSW 2008 T +61 2 9114 1871 | F +61 2 9114 1863 | M +61 418 433 057 E David.Hensher at sydney.edu.au | W sydney.edu.au/business/itls Celebrating 25 years of ITLS: 1991-2016 http://youtu.be/s2D0T1crZwY ERA Rank 5 (Transportation and Freight Services) Co-Founder of the International Conference Series on Competition and Ownership of Land Passenger Transport (The 'Thredbo' Series) http://www.thredbo-conference-series.org/ Join the ITLS group on LinkedIn Second edition of Applied Choice Analysis now available at www.cambridge.org/9781107465923 CRICOS 00026A This email plus any attachments to it are confidential. Any unauthorised use is strictly prohibited. If you receive this email in error, please delete it and any attachments. Please think of our environment and only print this e-mail if necessary.