R help - Aug 2012 - Questionnaire Analysis virtually without continuous Variables

Hello!
I am doing an analysis on a questionnaire of hunters taken in 4 
different districts of some mysterious foreign country. The aim of the 
study was to gather info on the factors that determine the hunting 
success of a peculiarly beautiful bird in that area. All variables are 
factors, i.e. they are variables such as "Use of Guns - yes / no",
"Use
of Dogs - yes / no" and the likes. The response is upposed to be
"number
of Birds caught", which was designed to be the only continuous variable. 
However, in reality the number of caught birds is between 0 and 1, 
sometimes hunters answered with 2. Unfortunately, it is not the 
questioner who is burdened with the analysis, but me. I am struggling to 
find an appropriate approach to the analysis. I don't really consider 
this as count data, since it would be very vulnerable to overinflation 
(and a steep decline for counts above 0). I can't really suggest 
binomial models either, since the lack of explanatory, continuous data 
renders such an approach quite vague. I also struggle with the random 
design of the survey (households nested within villages nested within 
districts). Adding to that, hunters don't even target the bird as their 
prime objective. The bird is essentially a by-catch, most often used for 
instant consumption on the hunting trip. I therefore doubt that any 
analysis makes more than a little sense, but I will not yet succumb to 
failure. Any ideas?

Thanks in advance!

PS: I just realized that this is not a question related to R but to 
statistics in general. Apologies for that!

On Sat, Aug 4, 2012 at 9:12 AM, Sacha Viquerat
<dawa.ya.moto at googlemail.com> wrote:
> Hello!
> I am doing an analysis on a questionnaire of hunters taken in 4 different
> districts of some mysterious foreign country. The aim of the study was to
> gather info on the factors that determine the hunting success of a
> peculiarly beautiful bird in that area. All variables are factors, i.e.
they
> are variables such as "Use of Guns - yes / no", "Use of Dogs
- yes / no" and
> the likes. The response is upposed to be "number of Birds
caught", which was
> designed to be the only continuous variable. However, in reality the number
> of caught birds is between 0 and 1, sometimes hunters answered with 2.
> Unfortunately, it is not the questioner who is burdened with the analysis,
> but me. I am struggling to find an appropriate approach to the analysis. I
> don't really consider this as count data, since it would be very
vulnerable
> to overinflation (and a steep decline for counts above 0). I can't
really
> suggest binomial models either, since the lack of explanatory, continuous
> data renders such an approach quite vague. I also struggle with the random
> design of the survey (households nested within villages nested within
> districts). Adding to that, hunters don't even target the bird as their
> prime objective. The bird is essentially a by-catch, most often used for
> instant consumption on the hunting trip. I therefore doubt that any
analysis
> makes more than a little sense, but I will not yet succumb to failure. Any
> ideas?
>
> Thanks in advance!

Hi Sacha,

This sounds a good deal like homework to me ("some mysterious foreign
country") and this list has a "no homework" policy so
unfortunately, I
don't think you'll be able to get much help here.

Best of luck with your analysis however!

Michael
>
> PS: I just realized that this is not a question related to R but to
> statistics in general. Apologies for that!
>
> ______________________________________________
> R-help at r-project.org mailing list
> 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 Sacha,

You're right that this is not an R related question really (would be better
somewhere like crossvalidated.com).

If basically everyone catches 0/1 birds, then I would consider dichotomizing:

Y <- as.integer(caught >= 1)

then check cross tabs to make sure there are no zero cells between predictors
and outcome:

xtabs(~Y + dogs + guns, data=yourdata)

then use the glmer() function to model the nested random effects.

m <- glmer(Y ~ dog + gun + (1 | household) + (1 | village) + (1 | district),
data = yourdata, family=binomial)

summary(m)

Cheers,

Josh

On Aug 4, 2012, at 7:12, Sacha Viquerat <dawa.ya.moto at googlemail.com>
wrote:
> Hello!
> I am doing an analysis on a questionnaire of hunters taken in 4 different
districts of some mysterious foreign country. The aim of the study was to gather
info on the factors that determine the hunting success of a peculiarly beautiful
bird in that area. All variables are factors, i.e. they are variables such as
"Use of Guns - yes / no", "Use of Dogs - yes / no" and the
likes. The response is upposed to be "number of Birds caught", which
was designed to be the only continuous variable. However, in reality the number
of caught birds is between 0 and 1, sometimes hunters answered with 2.
Unfortunately, it is not the questioner who is burdened with the analysis, but
me. I am struggling to find an appropriate approach to the analysis. I don't
really consider this as count data, since it would be very vulnerable to
overinflation (and a steep decline for counts above 0). I can't really
suggest binomial models either, since the lack of explanatory, continuous data
renders such an approach quite vague. I also struggle with the random design of
the survey (households nested within villages nested within districts). Adding
to that, hunters don't even target the bird as their prime objective. The
bird is essentially a by-catch, most often used for instant consumption on the
hunting trip. I therefore doubt that any analysis makes more than a little
sense, but I will not yet succumb to failure. Any ideas?
> 
> Thanks in advance!
> 
> PS: I just realized that this is not a question related to R but to
statistics in general. Apologies for that!
> 
> ______________________________________________
> R-help at r-project.org mailing list
> 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.

You may be able to get around zero cells using a an MCMC approach such as with
MCMCglmm.

On Aug 4, 2012, at 15:30, Sacha Viquerat <dawa.ya.moto at googlemail.com>
wrote:
> On 08/04/2012 07:57 PM, Joshua Wiley wrote:
>>  Hi Sacha,
>> 
>> You're right that this is not an R related question really (would
be better somewhere like crossvalidated.com).
>> 
>> If basically everyone catches 0/1 birds, then I would consider
dichotomizing:
>> 
>> Y <- as.integer(caught >= 1)
>> 
>> then check cross tabs to make sure there are no zero cells between
predictors and outcome:
>> 
>> xtabs(~Y + dogs + guns, data=yourdata)
>> 
>> then use the glmer() function to model the nested random effects.
>> 
>> m <- glmer(Y ~ dog + gun + (1 | household) + (1 | village) + (1 |
district), data = yourdata, family=binomial)
>> 
>> summary(m)
>> 
>> Cheers,
>> 
>> Josh
>> 
>> On Aug 4, 2012, at 7:12, Sacha Viquerat <dawa.ya.moto at
googlemail.com> wrote:
>> 
>>> Hello!
>>> I am doing an analysis on a questionnaire of hunters taken in 4
different districts of some mysterious foreign country. The aim of the study was
to gather info on the factors that determine the hunting success of a peculiarly
beautiful bird in that area. All variables are factors, i.e. they are variables
such as "Use of Guns - yes / no", "Use of Dogs - yes / no"
and the likes. The response is upposed to be "number of Birds caught",
which was designed to be the only continuous variable. However, in reality the
number of caught birds is between 0 and 1, sometimes hunters answered with 2.
Unfortunately, it is not the questioner who is burdened with the analysis, but
me. I am struggling to find an appropriate approach to the analysis. I don't
really consider this as count data, since it would be very vulnerable to
overinflation (and a steep decline for counts above 0). I can't really
suggest binomial models either, since the lack of explanatory, continuous data
renders such an approach quite vague. I also struggle with the random design of
the survey (households nested within villages nested within districts). Adding
to that, hunters don't even target the bird as their prime objective. The
bird is essentially a by-catch, most often used for instant consumption on the
hunting trip. I therefore doubt that any analysis makes more than a little
sense, but I will not yet succumb to failure. Any ideas?
>>> 
>>> Thanks in advance!
>>> 
>>> PS: I just realized that this is not a question related to R but to
statistics in general. Apologies for that!
>>> 
>>> ______________________________________________
>>> R-help at r-project.org mailing list
>>> 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.
> I did exactly what you proposed already (since the binomial model seemed
obvious to me), however, of course there are zero cells. I was thinking someone
more accustomed to doing questionnaire analysis could unveil some mysterious
approach common to sociologists but occluded from the naturalists eyes (hardened
after years of dealing with exact science ;)
> I think I will expand the binomial approach and just try to find fancy
graphics that make up for the low value of the actual results (maybe with
colours). :D
> Thank you for the reply (do they really give such tasks for homework these
days? These kids must be awesome statisticians!)
>