Demystifying fixed/random UPG EBV estimation w/wo model fixed effects

[janaobsteter]

This is the results for one replicate of the three population simulation, where target population receives imports from import populations 1&2. There is a non-1 genetic correlation between populations. We run 20 years of burn-in with no import and 20-years of import. The EBVs are estimated with blupf90+, we include UPGs (one UPG per population) as either fixed (through UPG_TYPE option) or as random (by adding them to the pedigree), but with no records. We run four estimation models regarding fixed effects - no fixed effect / mean / year / mean+year. Each is ran with UPGs are fixed / random.

1. GV trends in each population for the target-pop trait.
   All looks ok, the target population has the highest GV since it's directly selecting on the trait - the other two populations are indirectly selection on the trait as well (import_pop1 has a cor = 0.9 and import_pop2 has a cor = 0.8)

2. Gene proportions and GV partitions
   With UPGs as fixed (meaning they are not in the pedigree as individuals), the GVs are centered onto founders. With UPGs as random (meaning they are in the pedigree with year = year(founders) - 1), the GVs are centered onto the GV of UPGs.
   The GV of UPGs are calculated as the mean of the associated founders.

   UPG1     UPG2     UPG3 
363.7854 359.7571 349.7763 

3. EBVs - accuracy and distribution
   a) Accuracy

b) Distribution of EBVs

Here, Ind means BVs for individuals that are not UPGs. UPG1 = UPG target pop, UPG2 = UPG import_pop1, UPG3 = UPG import_pop2.

4. Partitions of pedigree BVs

5. Estimated of the fixed effects
   a) Plotted with the partitions

b) Text file
Fixed_effects.csv