[Limdep Nlogit List] Fwd: Question about Convolution test

[Thao Thai]

Hi Nlogit users,

Following Michael's instructions, I am able to generate Poe test statistics
for individual attributes as below

|-> CALC ;List; Poe(PRLH2,FRLH2,1) $
[CALC]         =      264.1756004
|-> CALC ;List; Poe(PFL,FFL,1) $
[CALC]         =      223.8058368
|-> CALC ;List; Poe(PCR1,FCR1,1) $
[CALC]         =     1079.2828222
|-> CALC ;List; Poe(PCR2,FCR2,1) $
[CALC]         =      135.7597390
|-> CALC ;List; Poe(PLO1,FLO1,1) $
[CALC]         =     3489.5048159
|-> CALC ;List; Poe(PRLC1,FRLC1,1) $
[CALC]         =      235.9289009
|-> CALC ;List; Poe(PRLC2,FRLC2,1) $
[CALC]         =      522.8063550

Please let me know if these huge figures look sensible in nlogit and how to
obtain p-values or level of significance for these figures? Please help or
point me to the available instructions. Please help!
Thank you in advance.
Best regards,
Thao

On Thu, 22 Apr 2021 at 15:54, Danne, Michael <
michael.danne at agr.uni-goettingen.de> wrote:

> Hi Thao,
>
> Run the model for the first subset with including the ;parameter command.
> Than you can copy/paste the individual WTPs in Excel. Do it in the same way
> with the second subset. In your Excel, each attribute is in a single
> column. Attribute 1 of subset one and attribute one of subset 2 should have
> different names in excel, for example car1 and car2. If you than use the
> poe test it is than CALC ; List ; Poe(car1,car2,1) $
> You have to be careful which vector you list first in the command. As I
> remember the higher vector (higher mean) has to be the first.
>
> I don’t know if there is an easier way, but it should work.
>
> Best regards
> Michael
>
> -----Ursprüngliche Nachricht-----
> Von: Limdep <limdep-bounces at mailman.sydney.edu.au> Im Auftrag von Thao
> Thai via Limdep
> Gesendet: Donnerstag, 22. April 2021 07:41
> An: Limdep and Nlogit Mailing List <limdep at mailman.sydney.edu.au>
> Cc: Thao Thai <Thao.T.Thai at monash.edu>
> Betreff: [Limdep Nlogit List] Fwd: Question about Convolution test
>
> Hi again,
>
> I found one relevant previous message with instructions on the Poe test
> below (
> https://protect-au.mimecast.com/s/8VIQC91WPRTk21ABziopbLO?domain=mailman.sydney.edu.au).
> However, I am still not sure how to test WTP estimates from WTP-space
> models for individual attributes from two datasets.
>
> I tried to replicate the suggested command as below:
> 1. I ran the first WTP-space model on one data subset and defined the
> vector of WTP estimates of 12 attributes:
> MATRIX ; vector1 = b (6:20)$
>
> 2. I ran the second WTP-space using another data subset and defined the
> vector of WTP estimates (two data subsets have information on the same
> attributes)
> MATRIX ; vector2 = b (6:20)$
>
> 3. I then ran the Poe test command as instructed.
>  CALC ; List ; Poe(vector1,vector2,1) $
> This command produced one single test statistic. However, I would like to
> have Poe test statistics for 12 individual attributes.
>
> Your suggestions would be appreciated. Please help!
> Best wishes,
> Thao
>
>
> "CALC ; List ; Poe(vector1,vector2) $
> This produces the statistic based on the full convolution. Vectors must be
> defined with a MATRIX command.  If x1 and x2 are two variables, use MATRIX
> ; vector1 = x1 ; vector2 = x2 $ then CALC Poe.  The test can be based on
> the difference of the vectors if they are the same length.  Use CALC ; List
> ; Poe(vector1,vector2,1) $ Regards Bill Greene"
>
>
> ---------- Forwarded message ---------
> From: Thao Thai <Thao.T.Thai at monash.edu>
> Date: Wed, 21 Apr 2021 at 16:47
> Subject: Question about Convolution test
> To: Limdep and Nlogit Mailing List <limdep at mailman.sydney.edu.au>
>
>
> Hi Nlogit user community,
>
> I would like to compare the WTP estimates from two WTP-space models (all
> random coefficients) using two different datasets. I am wondering what is
> the detailed procedure to conduct a Convolution test (Poe test) on the
> distributions of WTPs estimates in Nlogit?
>
> I could not find the procedure in the Manual, so any suggestions would be
> appreciated.
>
> Poe test: Poe GL, Giraud KL, Loomis JB (2005). Computational methods for
> measuring the difference of empirical distributions. American Journal of
> Agricultural Economics, 87, 353–365.
>
> Thank you in advance.
> Best wishes,
> Thao
> _______________________________________________
> Limdep site list
> Limdep at mailman.sydney.edu.au
> https://protect-au.mimecast.com/s/O2wyC0YKPviGm8Q3kfD00Yy?domain=limdep.itls.usyd.edu.au
>
>