Global Model to asses c-hat? Trend and/or limited age models

[jclaisse]

I am trying to asses the effect on age on daily survival of juvenile fish. I do not have enough data (especially for the oldest indivduals) to use a full age, meaning a seperate Phi for every age(days after recruitment) as my global model (many of the parameters can't be estimated). Based on the liturature from similar species, mortality can be very high for the just first few days, or the first week or or the first 2 weeks. Based on this I've settled on a set of candidate models that includes age in the form of a Trend model and then modeled as sets of ages or days after recruitment that i've tried to nest:

1) Trend (intercept + 1 parameter)
2) first 3 days + remaining days (2 parameters)
3) first 3 days + next 4 days + remaining days (3 parameters)
4) first 3 days + next 4 days + next 7 days + remaining days (4 parameters)

Does it make sense to use a global model to estimate c-hat which includes age in both forms:
3 days + 4 days + 7 days + remaining days + Trend

since the trend together with the other 4 parameters is somewhat redundant?

or is it better to use just the form from model #4 with 4 paramters or just the Trend model?