Parameterizing seed dispersal based on seed trap data

Hi all,

I am trying to get started on modeling some seed dispersal data, but am having a hard time getting started and looking for some help.

I have a dataset that contains information about seeds caught in a seedtrap. Basically a container was placed under X number of seeds on an individual shrub and, over N time steps, the number of (1) predated seeds (2) whole seeds, and (3) seeds still in the fruit that were found in the container were recorded. The data look like this:


The total seeds were not observed until all were gone, and the timesteps were not uniform. I am basically interested in (a) how many seeds did not fall into the trap (%), and of those that fell into the trap (b) how many were predated (%), whole seeds (without fruit) (%) and whole fruit (%).

In my thinking, this can be thought of as either a binomial test (successes and failures), but also has some elements of survival analysis (though I am not really interested in time to failure (or dispersal).

Ultimately, I have thought about collapsing the data, but this seems to be losing some information. Also, the observer made observational errors, so this could likely only be included if we keep each timestep.

Any help on getting started would be appreciated. Sorry for the lack of code, but if I get pointed in the right direction, I can provide code in the next step.

Have a great weekend.

Can you share some plots of the data? That might help to narrow down the model choices.

Ideally can you think of the simplest model that might explain the data, like it is ok to think of this as a linear model just to start with?

Are you looking to code this up in R or python? If in R rstanarm, rstan, brms, or something else?

Hey, thanks for the reply. I am actually working on a toy data set and reworking the example so it is more general. I will be post before the end of the day.

1 Like

Also what’s your comfort level/knowledge/practice with building models? I always forget to ask this.

I’ll keep an eye for the toy data set, thanks @Wade

I am okay with building models. My approach is to cut and paste, hack at it for awhile, and hopefully stumble upon the answer! They should really teach this in grad school, since this is what most people do.

Seriously, I do generally simulate data first, and have a pretty basic understanding of Stan.

1 Like