search

weecology / bbs-forecasting

Research on forecasting using Breeding Bird Survey data
MIT License
7 stars 9 forks source link

misc updates/responses #213

Closed davharris closed 6 years ago

davharris commented 6 years ago

Mostly minor, except where I tried to clarify what the mixed model was doing.

sdtaylor commented 6 years ago

Wouldn't all 3 environmental models be robust to uninformative covariates? If BRT are considered robust than random forests should be as well, and since it's based on a neural net I would guess MistNet is as well.

ethanwhite commented 6 years ago

It does read a little oddly to call out a single method as being robust and not include the others. If describing all three this way is justified then I like @sdtaylor's suggestion of mentioning them all. If not I think it's OK to say that this is complicated by the model fitting procedure and (if we all agree that what follows is correct) that variable selection is typically less important for prediction than for inference.

The reality is that we've done a massive amount of comparative modeling here and reviewers will always be interested in seeing you go one step further than you have. We've already made a lot of big steps forward in this paper and it's fine to stop there.

davharris commented 6 years ago

I had thought he was just asking about the "macroecological model" based on the line numbers, but maybe I misinterpreted. Will take another stab at this soon.

davharris commented 6 years ago

You two were right and I misread: he did ask about all environmental models, not just boosting. Fixed now.

sdtaylor commented 6 years ago

:+1: LGTM