When you look into a chip, you have certain behavior. When you look into the device (that is the chip + package), the behavior may change.
If you have a chip input with 0.5 pF input capacitance and the package adds 0.2 pF, the device has 0.7 pF input capacitance. So when specifying a device, one has to include device parasitics.
When dealing with fast changing signals, the package (with bonding wires) can make the difference between a solution or no solution. So it is important to know the effect of the package on device performance.
Instead of supplying a model for each device, one can use a two-step approach.
A certain chip can be housed in various packages. A manufacturer may publish simulation models for the chip (without package), and (RLC) models for the package. To get a device in simulation, you just have to merge the chip model with the package model.
Once you have a good model for the package, you can reuse this for other chips.
Even for relatively slow signals the package can be of importance as bonding wires may generate heat and therefore may limit maximum current (though the chip itself may handle more current).
The other way around: If you have a device, and you want to characterize the chip, you need to know the behavior of the package.