[Limdep Nlogit List] Question on missing values in N-logit

Christoph Buschmann christoph.buschmann at thuenen.de
Thu May 20 18:03:33 AEST 2021


Dear David and Bill,

thank you very much for your answers. 

Just to be on the safe side: I understand that I should either remove the eight respondents or fill their missing data on covariates (characteristics) with the mean, median or so, because:

- Coding missing covariate data with -888 is problematic, as you described Bill.

- Coding missing covariate data with -999 is also problematic: that is the situation I described in my first e-mail: Per choice task one row is removed, but “NLOGIT does not create a new choice task by removing rows from the old one” as you said Bill. Also, to be on the safe side, I send you attached a screenshot of NLogit’s report and the dataset. For example, the report says that row 1561 and row 1564 are skipped. In the dataset we see that row 1561 and 1564 describe Alternative 1 out of three Alternatives (Alternative three is Opt-out). So, it is problematic that the first alternative is skipped, but the information on the other two alternatives are still in the data set. But Nlogit does not create a new choice task out of the remaining two alternatives.

Thank you again for your help.

Kind regards, 

Christoph Buschmann



----- Ursprüngliche Mail -----
Von: "Limdep and Nlogit Mailing List" <limdep at mailman.sydney.edu.au>
An: "Limdep and Nlogit Mailing List" <limdep at mailman.sydney.edu.au>
CC: "William Greene" <wgreene at stern.nyu.edu>
Gesendet: Mittwoch, 19. Mai 2021 14:07:58
Betreff: Re: [Limdep Nlogit List] Question on missing values in N-logit

David and all.  Be careful in using the -888 form for missing data.  To
reiterate, if one of the 3 alts has
missing values for an attribute (-999), the entire choice task has to be
removed from the sample.
Christoph, your interpretation is not correct.  NLOGIT does not create a
new choice task by removing
rows from the old one.  -888 does not operate on the data.  It operates on
the coefficient vector.  For
observations with x(j,k) = -888, where j is the alt and k is the
characteristic, when computing beta'x(j,k)
for that alt j, beta(k) is set equal to zero - not x(j,k).  This matters.
You don't want to set the x(j,k) to
zero, especially if it is a price.  Nonattendance (-888) means the marginal
utility is zero, not that the
attribute is zero.
Cheers
Bill Greene

On Wed, May 19, 2021 at 4:04 AM David Hensher via Limdep <
limdep at mailman.sydney.edu.au> wrote:

> Missing data can exist for  a number of reasons. While not being sure what
> is happening here there are two codes of value. -999 will remove an entire
> choice set and -888 will ignore an attribute level associated with an
> alternative or individual you have in the data. So if a covariant is say
> income, code as -888 to retain the choice set but note this amounts to it
> being missing which can be problematic if you too many of these are missing
> and it is retained in the model. Some people either replace it with the
> mean or mode or run an auxiliary regression to try and predict it based on
> other covariates.
> David
>
> Sent from my iPhone
> 0418 433 057
> David A Hensher
> Choice modellers prefer Nlogit: See https://protect-au.mimecast.com/s/uMhOC4QOPEiB6118RhOsFga?domain=limdep.com
> ITLS celebrating 30 years
>
>
> Note:
> hgroup at hensher.com.au David.hensher at bigpond.com
> David.hensher at sydney.edu.au
> are linked so use one only
>
>
> On 19 May 2021, at 5:51 pm, Christoph Buschmann <
> christoph.buschmann at thuenen.de> wrote:
>
> Dear group,
>
> I have a question on how N-logit deals with missing data in the
> covariates. My choice experi-ments were carried out completely by all
> respondents, but 8 respondents did not answer all questions about the
> covariates.
>
> I have ten choice cards per respondent and 3 alternatives in each choice
> situation. So, my dataset comprises three rows for each choice situation.
> N-Logit reports that, for the 8 respondents in question, per choice
> situation one missing value has been found and refers to the first row per
> choice situation, i.e. the row that gives information about the choice of
> alternative 1.
>
> I assume that, for the 8 respondents in question, N-logit skips one row
> per choice situation (the one for alternative 1), including the information
> about covariates, and keeps the other two rows (for alternatives 2 and 3).
>
> This would mean that for the 8 respondents in question, some of the
> information is retained and so it makes sense to keep them in the data set.
> In fact, if I remove the 8 respondents from the dataset by hand, the model
> deteriorates (AIC).
>
> What is your experience with missing data? Is my interpretation correct?
>
> Thank you very much for your help.
>
>
> Kind regards,
>
> Christoph Buschmann
>
> --
> Christoph Buschmann, M.Sc.
>
> Stabsstelle Klima/ Coordination Unit Climate
> Thünen-Institut
> Bundesallee 49
> D-38116 Braunschweig
> Germany
>
> Im Home-Office erreichbar per e-mail
> E-mail: christoph.buschmann at thuenen.de
> Homepage: [ https://protect-au.mimecast.com/s/7CwcC5QPXJiZRGGlwCO9o2L?domain=thuenen.de | https://protect-au.mimecast.com/s/7CwcC5QPXJiZRGGlwCO9o2L?domain=thuenen.de ]
> Twitter: @ThuenenClimSoil
>
> _______________________________________________
> Limdep site list
> Limdep at mailman.sydney.edu.au
> https://protect-au.mimecast.com/s/VUG_C6XQ4LfrMKKWZCmq4jK?domain=limdep.itls.usyd.edu.au
>
> _______________________________________________
> Limdep site list
> Limdep at mailman.sydney.edu.au
> https://protect-au.mimecast.com/s/VUG_C6XQ4LfrMKKWZCmq4jK?domain=limdep.itls.usyd.edu.au
>
>

-- 
William Greene
Department of Economics, emeritus
Stern School of Business, New York University
44 West 4 St.
New York, NY, 10012
URL: https://protect-au.mimecast.com/s/nAXXC71R2NTAykkRnIN6c5y?domain=people.stern.nyu.edu
Email: wgreene at stern.nyu.edu
Editor in Chief: Journal of Productivity Analysis
<https://protect-au.mimecast.com/s/vDR1C81V0PT6QRRkBIomzNG?domain=springer.com>
<https://protect-au.mimecast.com/s/vDR1C81V0PT6QRRkBIomzNG?domain=springer.com>
Editor in Chief: Foundations and Trends in Econometrics
Associate Editor: Economics Letters
Associate Editor: Journal of Business and Economic Statistics
_______________________________________________
Limdep site list
Limdep at mailman.sydney.edu.au
https://protect-au.mimecast.com/s/VUG_C6XQ4LfrMKKWZCmq4jK?domain=limdep.itls.usyd.edu.au


More information about the Limdep mailing list