Cluster-robust standard error

27 views
Skip to first unread message

Atefeh Fakourrad

unread,
Sep 28, 2021, 2:33:09 AMSep 28
to Biogeme
Dear Prof. Bierlaire,

I am estimating a panel model and would like to know if Biogeme is able to estimate and report a cluster-robust standard error? 
I was also wondering if it is possible to conduct a part-worth analysis using Bioegeme?

Best regards,
Ati. 

Bierlaire Michel

unread,
Sep 28, 2021, 2:40:55 AMSep 28
to a.fak...@gmail.com, Bierlaire Michel, Biogeme

On 28 Sep 2021, at 08:04, Atefeh Fakourrad <a.fak...@gmail.com> wrote:

Dear Prof. Bierlaire,

I am estimating a panel model and would like to know if Biogeme is able to estimate and report a cluster-robust standard error? 

Yes. If the data is panel, bootstrapping samples individuals with replacement, and not observations. 

I was also wondering if it is possible to conduct a part-worth analysis using Bioegeme?

You mean that you want to calculate the utility of each alternative, for each observation? Yes. You need to use the simulation feature. 


Best regards,
Ati. 

--
You received this message because you are subscribed to the Google Groups "Biogeme" group.
To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/34628b99-d357-4fe9-883c-a5f69b78065bn%40googlegroups.com.

Atefeh Fakourrad

unread,
Sep 28, 2021, 12:48:47 PMSep 28
to Biogeme
Dear Prof. Bierlaire,

Thanks for your reply. Regarding the cluster-robust standard error, I am not sure if I completely understand your point. I am not sure what should be replaced with observations? Is there any example of this? 

Thanks,

Best,
Ati. 

Bierlaire Michel

unread,
Sep 28, 2021, 12:51:24 PMSep 28
to a.fak...@gmail.com, Bierlaire Michel, Biogeme
Bootstrapping is about sampling with replacement. 
Panel data consist in a list of individuals, each associated with a list of observations. 
My point is that Biogeme samples individuals, so that the standard errors calculated with bootstrapping are “cluster-robust”. If the sampling was done on observations instead, it would not be cluster-robust. 

On 28 Sep 2021, at 11:47, Atefeh Fakourrad <a.fak...@gmail.com> wrote:

Dear Prof. Bierlaire,

Thanks for your reply. Regarding the cluster-robust standard error, I am not sure if I completely understand your point. I am not sure what should be replaced with observations? Is there any example of this? 

Thanks,

Best,
Ati. 

On Tuesday, September 28, 2021 at 8:40:55 AM UTC+2 michel.b...@epfl.ch wrote:

On 28 Sep 2021, at 08:04, Atefeh Fakourrad <a.fak...@gmail.com> wrote:

Dear Prof. Bierlaire,

I am estimating a panel model and would like to know if Biogeme is able to estimate and report a cluster-robust standard error? 

Yes. If the data is panel, bootstrapping samples individuals with replacement, and not observations. 

I was also wondering if it is possible to conduct a part-worth analysis using Bioegeme?

You mean that you want to calculate the utility of each alternative, for each observation? Yes. You need to use the simulation feature. 


Best regards,
Ati. 

--
You received this message because you are subscribed to the Google Groups "Biogeme" group.
To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/34628b99-d357-4fe9-883c-a5f69b78065bn%40googlegroups.com.


--
You received this message because you are subscribed to the Google Groups "Biogeme" group.
To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.

Atefeh Fakourrad

unread,
Oct 1, 2021, 6:44:19 AMOct 1
to Biogeme
Dear Prof.Bierlaire,

Thanks for your reply. In conclusion, when a panel model is estimated, Biogeme automatically calculates and reports the clustered-robust standard error, right? 

Regarding the part-worth (conjoint) analysis, I am referring to a method that allows comparing the importance of attributes and find the most/least important attributes.  Is there any way to find this out? 

Best regards,
Ati. 

On Tuesday, September 28, 2021 at 8:40:55 AM UTC+2 michel.b...@epfl.ch wrote:

Bierlaire Michel

unread,
Oct 1, 2021, 11:32:50 AMOct 1
to a.fak...@gmail.com, Bierlaire Michel, Biogeme

On 1 Oct 2021, at 11:20, Atefeh Fakourrad <a.fak...@gmail.com> wrote:

Dear Prof.Bierlaire,

Thanks for your reply. In conclusion, when a panel model is estimated, Biogeme automatically calculates and reports the clustered-robust standard error, right? 

Yes, the robust ones. 

To convince yourself, calculate the std error using bootstrapping and compare the results. 


Regarding the part-worth (conjoint) analysis, I am referring to a method that allows comparing the importance of attributes and find the most/least important attributes.  Is there any way to find this out? 

If you divide each coefficient by the cost coefficient, you obtain something in currency units that you ca compare. It works only for linear specifications.  


Best regards,
Ati. 

On Tuesday, September 28, 2021 at 8:40:55 AM UTC+2 michel.b...@epfl.ch wrote:

On 28 Sep 2021, at 08:04, Atefeh Fakourrad <a.fak...@gmail.com> wrote:

Dear Prof. Bierlaire,

I am estimating a panel model and would like to know if Biogeme is able to estimate and report a cluster-robust standard error? 

Yes. If the data is panel, bootstrapping samples individuals with replacement, and not observations. 

I was also wondering if it is possible to conduct a part-worth analysis using Bioegeme?

You mean that you want to calculate the utility of each alternative, for each observation? Yes. You need to use the simulation feature. 


Best regards,
Ati. 

--
You received this message because you are subscribed to the Google Groups "Biogeme" group.
To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/biogeme/34628b99-d357-4fe9-883c-a5f69b78065bn%40googlegroups.com.


--
You received this message because you are subscribed to the Google Groups "Biogeme" group.
To unsubscribe from this group and stop receiving emails from it, send an email to biogeme+u...@googlegroups.com.

Atefeh Fakourrad

unread,
Oct 7, 2021, 9:17:30 AMOct 7
to Biogeme
Dear Prof. Michel Beirlaire,

I am conducting a choice-based conjoint (CBC) analysis where I need to define a separate parameter for every single attribute level. To do this, I am going to use effect coding for creating and incorporating M-1 (M is the number of attribute levels) parameters in the model. After estimating the model, the coefficient of the reference attribute level, which is left out of the estimation, can be calculated by minus the sum of all other levels of that attribute ( the average across all utilities is zero in effect coding). I wonder how I can calculate the standard error and t-value of this parameter?

Kind regards,
Ati. 

Bierlaire Michel

unread,
Oct 7, 2021, 9:28:01 AMOct 7
to a.fak...@gmail.com, Bierlaire Michel, Biogeme
It is not clear what you want to do. It is not a parameter of the model. There is no such thing as a coefficient for the reference level. This is why it is called a reference level. 

Reply all
Reply to author
Forward
0 new messages