R help - Jul 2015 - “Error in if (abs(x - oldx) < ftol)” when using “lognormal” distribution in mixed logit

Hi, I have a question about how to use the mlogit package in R to do analysis
of discrete choice survey data. Our survey is about asking people to choose
from different insurance policies(with two attributes of deductible and
premium).

The code I used to fit mixed logit is:


[1] ml <- mlogit.data (mydata, choice="choice", shape =
"wide", id "individual",
               opposite =c ('deductible', 'premium'),varying =
5:10)

[2] ml.w5 <- mlogit (choice~deductible+premium|0, ml, panel = TRUE, 
             rpar = c(deductible='ln', premium='ln'), 
             R = 100, halton = NA, print.level=0)

I try to use lognormal because we hope the coefficients for both deductible
and premium are negative. And I use "opposite" in [1] to reverse the
sign
because lognormal is always positive.

But I always get the error warning:

"Error in if (abs(x - oldx) < ftol) { : missing value where TRUE/FALSE
needed
In addition: Warning message: In log(start[ln]) : NaNs produced"
I double check the data and am sure there isn't any missing data. And if I
change the lognormal "ln" to "n" or "cn", it will
work without any warning.

Does anyone know how to deal with this? Thank you for your help.



--
View this message in context:
http://r.789695.n4.nabble.com/Error-in-if-abs-x-oldx-ftol-when-using-lognormal-distribution-in-mixed-logit-tp4710284.html
Sent from the R help mailing list archive at Nabble.com.

Den 2015-07-24 kl. 00:33, skrev Dera:
> Hi, I have a question about how to use the mlogit package in R to do
analysis
> of discrete choice survey data. Our survey is about asking people to choose
> from different insurance policies(with two attributes of deductible and
> premium).
>
> The code I used to fit mixed logit is:
>
>
> [1] ml <- mlogit.data (mydata, choice="choice", shape =
"wide", id > "individual",
>                 opposite =c ('deductible',
'premium'),varying = 5:10)
>
> [2] ml.w5 <- mlogit (choice~deductible+premium|0, ml, panel = TRUE,
>               rpar = c(deductible='ln', premium='ln'),
>               R = 100, halton = NA, print.level=0)
>
> I try to use lognormal because we hope the coefficients for both deductible
> and premium are negative. And I use "opposite" in [1] to reverse
the sign
> because lognormal is always positive.
>
> But I always get the error warning:
>
> "Error in if (abs(x - oldx) < ftol) { : missing value where
TRUE/FALSE
> needed
> In addition: Warning message: In log(start[ln]) : NaNs produced"
> I double check the data and am sure there isn't any missing data. And
if I
> change the lognormal "ln" to "n" or "cn", it
will work without any warning.
>
> Does anyone know how to deal with this? Thank you for your help.
I've never used the 'mlogit' package but that error message has
likely
nothing to do with missingness in your data.

The error message is pretty clear: for some reason 'if (abs(x - oldx) < 
ftol)', which looks like a check for convergence, doesn't evaluate to 
TRUE or FALSE as expected.  Perhaps due to the warning message about 
problems with NaN in the starting values?  Have you checked your data? 
That warning message is typical for when you try to take logs of a 
negative number:

 > log(-1)
[1] NaN
Warning message:
In log(-1) : NaNs produced

Unless this helps you identify the problem, we need a (preferably small) 
reproducible example demonstrating the issue to be able to provide 
further help.

BTW, have you asked the package maintainer (as per the posting guide)?


Henric Winell


>
>
>
> --
> View this message in context:
http://r.789695.n4.nabble.com/Error-in-if-abs-x-oldx-ftol-when-using-lognormal-distribution-in-mixed-logit-tp4710284.html
> Sent from the R help mailing list archive at Nabble.com.
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>

Hi Henric,

Thank you for your reply! It really helps. I should have noticed this
earlier. I don't have a negative input, the error is because our survey
allowed people to choose neither of the choices we gave, so we added a
choice "Neither" with both deductible and premium being 0, which
caused
this problem. I tried to delete the "0" entries just now and it
worked! I
think I need to find a way do deal with the "Neither" choice in
another way.

Thank you so much and have a nice weekend!

Best, Dong

On Fri, Jul 24, 2015 at 7:16 AM, Henric Winell <nilsson.henric at
gmail.com>
wrote:
> Den 2015-07-24 kl. 00:33, skrev Dera:
>
>  Hi, I have a question about how to use the mlogit package in R to do
>> analysis
>> of discrete choice survey data. Our survey is about asking people to
>> choose
>> from different insurance policies(with two attributes of deductible and
>> premium).
>>
>> The code I used to fit mixed logit is:
>>
>>
>> [1] ml <- mlogit.data (mydata, choice="choice", shape =
"wide", id >> "individual",
>>                 opposite =c ('deductible',
'premium'),varying = 5:10)
>>
>> [2] ml.w5 <- mlogit (choice~deductible+premium|0, ml, panel = TRUE,
>>               rpar = c(deductible='ln', premium='ln'),
>>               R = 100, halton = NA, print.level=0)
>>
>> I try to use lognormal because we hope the coefficients for both
>> deductible
>> and premium are negative. And I use "opposite" in [1] to
reverse the sign
>> because lognormal is always positive.
>>
>> But I always get the error warning:
>>
>> "Error in if (abs(x - oldx) < ftol) { : missing value where
TRUE/FALSE
>> needed
>> In addition: Warning message: In log(start[ln]) : NaNs produced"
>> I double check the data and am sure there isn't any missing data.
And if I
>> change the lognormal "ln" to "n" or "cn",
it will work without any
>> warning.
>>
>> Does anyone know how to deal with this? Thank you for your help.
>>
>
> I've never used the 'mlogit' package but that error message has
likely
> nothing to do with missingness in your data.
>
> The error message is pretty clear: for some reason 'if (abs(x - oldx)
<
> ftol)', which looks like a check for convergence, doesn't evaluate
to TRUE
> or FALSE as expected.  Perhaps due to the warning message about problems
> with NaN in the starting values?  Have you checked your data? That warning
> message is typical for when you try to take logs of a negative number:
>
> > log(-1)
> [1] NaN
> Warning message:
> In log(-1) : NaNs produced
>
> Unless this helps you identify the problem, we need a (preferably small)
> reproducible example demonstrating the issue to be able to provide further
> help.
>
> BTW, have you asked the package maintainer (as per the posting guide)?
>
>
> Henric Winell
>
>
>
>
>>
>>
>> --
>> View this message in context:
>>
http://r.789695.n4.nabble.com/Error-in-if-abs-x-oldx-ftol-when-using-lognormal-distribution-in-mixed-logit-tp4710284.html
>> Sent from the R help mailing list archive at Nabble.com.
>>
>> ______________________________________________
>> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
>> https://stat.ethz.ch/mailman/listinfo/r-help
>> PLEASE do read the posting guide
>> http://www.R-project.org/posting-guide.html
>> and provide commented, minimal, self-contained, reproducible code.
>>
>>
>
	[[alternative HTML version deleted]]