jpwrobinson
commented
6 years ago On the modelled/non-modelled stuff - modelling is only used to scale up trade volumes from monthly to yearly estimates. The actual species presence/absence can't be modelled. So this means we can use the non-modelled data, because partial years still contain information about which species countries exports.
It does also mean that partial years ('modelled') have lower sample effort than full years (2008,09,10), so we might miss the true number of species exported by each country, from 2000-2010.
To me, the options are:
1) aggregate across years and assume that countries export the common species every year, so we have reasonably detected the true number of countries that exports each species
2) Control for partial sampling with an offset/random covariate that identifies when species are sampled in full years (2008-10) or partial years (2000, 2004, 2005).
It would be good to check how species often appear in both partial + full datasets, per country. Would give an idea of the consistency of export records. Can't do this until Rhyne gets back with the weirdness of non-modelled records
JamieMcDevittIrwin
Huge outliers in the volume of certain species --> decided to just use species occurrences instead of volume. So our response variable will be: If a species was exported or not (0,1) (binomial).
In addition, we decided to only look at the top 100 species that are exported by volume (just the full years 2008, 2009, 2010, no modelled data).
However, should we keep the dataframe like it is with species being exported by multiple countries in multiple years? Currently the top 100 csv is saved where only country is included, so not looking at species occurrences in years.