Structuring Missing Data
Dan
I have two survey, double-observer data and am having difficulty running models because of missing data.
I was under the impression that even if count data was present and an observation-level covariate was missing for a particular site, that that occasion would be omitted. E.g.:
Site y1.1 y1.1 y.2.1 y2.2 obscov1.1 obscov1.2 obscov2.1 obscov2.2
1 X X X X X X X X
2 X X X X NA NA X X
- where obsvar1.1 is the variable value for the first survey and first observer
However, this produces the following warning when I run p(obsvar1):
Warning message:
Some observations have been discarded because corresponding covariates were missing.
Thinking that this just notified me that some rows were dropped, I went on to put models in a FitList and received this error:
Error in validityMethod(object) :
Data are not the same among models due to missing covariate values. Consider removing NAs before analysis.
In addition: Warning message:
Some observations have been discarded because corresponding covariates were missing.
If I code the corresponding ‘y’ data to NA’s, p(obsvar) works fine, e.g.:
Site y1.1 y1.1 y.2.1 y2.2 obscov1.1 obscov1.2 obscov2.1 obscov2.2
1 X X X X X X X X
2 NA NA X X NA NA X X
Although this may work for “obsvar”, what if I had a second, third, fourth… obs-level covariate that I actually had data for. By coding the ‘y’ data to NA’s, it would omit these data despite the fact that I may have data for the other covariates. I don’t believe this method is correct. For example, if I had 5 obs-level covariates and at each site 1 of the 5 was always missed, I would not be able to model any of the obs-level covariates. This is a bit of an extreme example, but you can see how quickly you would lose a lot of information. Or am I wrong?
So the question is how should this be structured so that I can model multiple observation-level covariates?
Any insight is greatly appreciated.
Cheers,
Dan
Richard Chandler
Paul J Taillie
Dan
I have been in contact with Richard off-list and he provided the following comments:
"Basically, anytime you have a NA in an obsCov, the corresponding "y" value needs to be removed. This is because there is no likelihood contribution for such an observation. The warning message you see is just telling you this. It isn't a problem except when you have covariates with different patterns of missing values. In such a case, you need to ensure that the missing values are consistent among covariates if you want to compare them using AIC. This isn't an unmarked issue, it is an issue related to the use of AIC in general."
In regards to imputing the missing obs-level covariates, Richard advised against this:
"I don't think that imputing (making up) values for the missing observations is a good idea. But if you do it, you should assess how sensitive the results are to different made-up values."
Cheers,
Dan
Pablo García Díaz
I want to construct a function for calculating and store the log-likelihoods of all the models I'm running, but using the function log-Lik in unmarked just allow me to calculate the values one-by-one (i.e., I cannot do logLik(c(garduna, garduna2, garduna4))
Any suggestion to get the job done?
Many thanks in advance, best regards
Pablo
Richard Chandler
MP
I am working in pcount with avian point-count data that include some missing obs. covariate values. I would like to know if there's a way to directly remove rows containing one or more NA values from the Formal class UnmarkedFramePCount.
Ideally, I'd like to scan each row in the umf for NA values, then omit the row when they're located. In a regular dataframe, I'd use na.omit(), but I'm not sure how best to go about this in a umf.
I am separately analyzing multiple subsets of data, so I need something I can automate.
I guess my other option would be to locate and remove the NAs in the initial covariate dataset, make note of their rows, and then remove rows from the initial bird dataset using that list of rows (all before creating a umf)?
(I hope this table stays properly formatted in this post.)
> umf1 Data frame representation of unmarkedFrame object. y.1 y.2 y.3 DBT.1 DBT.2 DBT.3 JD.1 JD.2 JD.3 ST.1 ST.2 ST.3 PCTSHRB.1 PCTSHRB.2 PCTSHRB.3 YYYY.1 YYYY.2 YYYY.3 V01881 4 0 1 18.3 27.0 30.6 135 154 179 910 812 816 49.40393519 49.40393519 49.40393519 2011 2011 2011 V01884 1 0 1 28.9 26.0 25.0 157 174 180 800 836 700 0.00000000 0.00000000 0.00000000 2011 2011 2011 V01904 2 0 1 26.1 24.0 29.4 157 174 180 656 703 857 NA NA NA 2011 2011 2011 V01906 1 0 1 16.7 30.0 30.0 134 156 177 736 714 818 59.21296296 59.21296296 59.21296296 2011 2011 2011 V02542 1 1 0 15.0 25.0 26.0 134 156 177 614 613 627 20.41087963 20.41087963 20.41087963 2011 2011 2011-Michael
Richard Chandler
--
You received this message because you are subscribed to the Google Groups "unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.