Semi Parametric/Fully Parametric and Non-Parametric - When do you prefer which?

Hello there! I'm trying to write a portion of an essay covering the question "Under what circumstances might a semi-parametric specification be preferred to a fully parametric specification and vice versa? When might a non-parametric model be preferred?"

As far as I'm aware, semi-parametric models are preferred to fully parametric models - due to less assumptions that need to be validated in semi-parametric models and hence, more flexibility. However, if the distributional assumptions can be met, Fully parametric models are more efficient. Semi parametric models such as Cox PH tends are more applicable to more distributions.

The last piece to evaluate is how Non-parametric models fit in. Non parametric models are used when you aren't concerned over covariates and the advantage of NP models is flexibility of the baseline hazard not being constrained.

However health sciences tend to prefer fully parametric and SP models since you can make inferences on covariates - which you can'd to with NP models.

I just wanted to check if I'm missing any crucial points? Alternatively if anyone could suggest better ways to get my ideas across it would be greatly appreciated!

Cheers in advance


Omega Contributor
How are you defining these three variables? A measure is a parameter, so in particular what would be a semi -measure?