[this is not my question; it's posted on behalf of someone who wants to remain anonymous ...] I am testing the effect of a treatment to reduce bycatch in fishing nets. Note the the design uses paired nets (control vs experiment) soaked simultaneously but of different length (limited budget did no allow to have an experimental net as long as control net). The dependent variable are counts (no. individuals entangled), and I have fishing effort and treatment (control vs experiment) as independent variables. Since bycatch events were rare , the dataset is zero inflated and positive catches are usually of 1 individual, therefore we switched to a binomial model to test the probability of catching an individual where if the catch is zero then probability =0, but if the catch is >0 then probability is a 1. We used this model to predict bycatch probability in control and experimental nets by setting fishing effort = 1. There is an issue being raised, that Fishing effort being significantly higher for control than experimental nets, the binomial model can yield biased estimates of treatment and overestimate treatment efficiency. I thought that including Effort as a fixed effect in the model would mean that the model takes into account the difference in effort when predicting the bycatch probability. Is that true? However, I am not entirely sure HOW the glmer function does it and I would like to know your opinion about the issue being raised."
incorporating effort as an effect in binomial GLMM
2 messages · Ben Bolker, Thierry Onkelinx
Dear Anonymous, Here a few ideas How did you check for zero-inflation? A lot of zero's does not imply zero-inflation. E.g. table(rpois(1e6, lambda = 0.01)) has lots of zero's but no zero-inflation. I'd recommend using a Poisson distribution. Then check for zero-inflation by comparing the distribution of the number of zero's from several datasets simulated based on the model with the observed number of zero's. The logit-link complicates the interpretation of the fishing effort in the binomial model. I suggest using a Poisson model with log(length) of the nets as a fixed effect to the model to correct from fishing effort. Then you can get predictions in terms of number per unit length the net. Best regards, ir. Thierry Onkelinx Statisticus / Statistician Vlaamse Overheid / Government of Flanders INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND FOREST Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance thierry.onkelinx at inbo.be Havenlaan 88 bus 73, 1000 Brussel www.inbo.be /////////////////////////////////////////////////////////////////////////////////////////// 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 /////////////////////////////////////////////////////////////////////////////////////////// <https://www.inbo.be> Op vr 17 jan. 2020 om 22:01 schreef Ben Bolker <bbolker at gmail.com>:
[this is not my question; it's posted on behalf of someone who wants to remain anonymous ...] I am testing the effect of a treatment to reduce bycatch in fishing nets. Note the the design uses paired nets (control vs experiment) soaked simultaneously but of different length (limited budget did no allow to have an experimental net as long as control net). The dependent variable are counts (no. individuals entangled), and I have fishing effort and treatment (control vs experiment) as independent variables. Since bycatch events were rare , the dataset is zero inflated and positive catches are usually of 1 individual, therefore we switched to a binomial model to test the probability of catching an individual where if the catch is zero then probability =0, but if the catch is >0 then probability is a 1. We used this model to predict bycatch probability in control and experimental nets by setting fishing effort = 1. There is an issue being raised, that Fishing effort being significantly higher for control than experimental nets, the binomial model can yield biased estimates of treatment and overestimate treatment efficiency. I thought that including Effort as a fixed effect in the model would mean that the model takes into account the difference in effort when predicting the bycatch probability. Is that true? However, I am not entirely sure HOW the glmer function does it and I would like to know your opinion about the issue being raised."
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models