[R-sig-ME] Help with mixed model design

[Thierry Onkelinx]

Dear Chad,

it seems like you want predictions for specific combinations of covariates
or testing for specific contrasts of the parameters.

Best regards,

ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature and
Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
Kliniekstraat 25
1070 Anderlecht
Belgium

To call in the statistician after the experiment is done may be no more
than asking him to perform a post-mortem examination: he may be able to say
what the experiment died of. ~ Sir Ronald Aylmer Fisher
The plural of anecdote is not data. ~ Roger Brinner
The combination of some data and an aching desire for an answer does not
ensure that a reasonable answer can be extracted from a given body of data.
~ John Tukey

2017-09-05 21:49 GMT+02:00 Chad Newbolt <newboch op auburn.edu>:

> All,
>
>
>
> I'm having some difficulty envisioning how to address a question with
> mixed model and thought might find some assistance here.  I have a dataset
> generated from a survey where individual were asked to view a set of images
> and classify animals in these images according to sex/age groups.  We
> wanted to investigate the effects of observer experience, profession, etc.,
> and factors related to the image itself (day/night, animal group, etc.) on
> how accurately observers could classify animals.  Responses in our survey
> could be either correct/incorrect (we knew truth in our test images) which
> could be input as a simple binary response; however, viewers also had the
> option of responding  "Unknown" for images they did not feel comfortable
> answering due to image quality, etc.  This was done to mimic what is done
> in practice in our field of work but was a real pain in our current
> efforts.  Since "Unknown" responses in our particular application are
> technically neither correct nor incorrect (but d
>  o influence census results in practice), I decided to split my analysis
> into two models: 1) A correct/incorrect response model (I removed the
> Unknown responses in this data set), and 2) An Unknown response model (I
> created a new binary variable for Unknown/Any other response).  These
> models answer two very different questions which we believe are both of
> value to us.
>
>
>
> My difficulty now concerns the Incorrect responses.  I would like to find
> a way to determine probabilities of the kinds of error that we see in our
> data.  For example, images that contained male animals that were
> incorrectly answered, how likely were those responses female vs. young?
> This information would be valuable in predicting how the errors that we
> observed might be influencing population surveys.
>
>
>
> My thought is to first subset the data, maintaining only incorrect
> responses.  I assume I could then create a new binary factor where one
> level of observer responses (Male, Female, Young) is 1 and all others are
> 0.  I could then model this binary response with a fixed effect for the
> answer I know to be true (also Male, Female, Young) along with my random
> effects.  This would be repeated three times for each kind of incorrect
> response.  Something like:
>
>
>
> Response"Male 1 or 0" ~  True Answer + (1| Observer ID) + (1|Question)
>
> Response"Female 1 or 0" ~  True Answer + (1| Observer ID) + (1|Question)
>
> Response"Young 1 or 0" ~  True Answer + (1| Observer ID) + (1|Question)
>
>
>
> Does that make any sense?  Any better ideas?  It's probably very simple
> but I've really struggled with this one.
>
>
>
> Thanks
>
>         [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models op r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>

	[[alternative HTML version deleted]]