Timeline for Small Sample Sizes and Zero Inflated Count Data in R
Current License: CC BY-SA 4.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 22 at 11:14 | vote | accept | newspice | ||
| Mar 22 at 8:46 | answer | added | Gordon Smyth | timeline score: 8 | |
| Mar 22 at 1:26 | comment | added | Stefan |
Yes give it a try! I am not sure though if there is a package that can deal with complete separation and also fits Hurdle models... in this case you might have to go Bayesian (see the first link I posted) and explore the brms package. Good luck!
|
|
| Mar 22 at 0:40 | comment | added | newspice |
Thanks so much Stefan. I wasn't aware there was a term for it! My random effect was originally the species for the overall model fitted by adding (1 | species) however for an individual species set like the example, I thought this was unnecessary to include given the species remains the same for the entire set. I will look to implementing a two step model like you mentioned and will post my results.
|
|
| Mar 21 at 23:57 | comment | added | Stefan |
What do you think about analyzing this as a two step model, i.e. a hurdle model? First you ask what is the probability of germination and then using a zero-truncated poisson, what's rate of germination for those seeds that germinated? All of this can be done in one model using the glmmTMB package.
|
|
| Mar 21 at 23:51 | comment | added | Stefan | Thanks! When you fit the glmm, what was your random effect? Why did you move away from it? Also why wouldn't you want to add species into the model as a fixed effect? The problem you are describing is called complete separation, i.e. when you have all zeros or ones for a given treatment combination. See here and here. There are many more examples here on CV regarding this. | |
| Mar 21 at 23:30 | history | edited | newspice | CC BY-SA 4.0 |
added 7 characters in body
|
| Mar 21 at 23:30 | comment | added | newspice | Thanks Stefan. Ideally, I would like to model the probability that seeds will germinate but really I'm just looking to model the effect of treatments on germination so the response variable is flexible. I have edited my question to include the sample data I used in the example model outputs. My overall dataset has 11 different seed species so to see the effect of treatments on an individual seed I filter the dataset to just 1 seed. | |
| Mar 21 at 23:25 | history | edited | newspice | CC BY-SA 4.0 |
Added sample data
|
| Mar 21 at 23:07 | comment | added | Stefan | Hi newspice! What are you modeling exactly? Number of seeds germinated? Probability of seeds germinated? How many seeds per treament? Can share a subset of your data structure? What does this exactly mean "For example the below is the summary output of the model when only filtered to one seed with around 24 observations." | |
| S Mar 21 at 22:46 | review | First questions | |||
| Mar 21 at 23:20 | |||||
| S Mar 21 at 22:46 | history | asked | newspice | CC BY-SA 4.0 |