Why is there more than one estimation model?

At various stages of maturity, a software project may exhibit 3 different defect rates.   Increasing, constant or decreasing.  Estimation models are all probabilistic models that each have assumptions about the defect rate. Within each probabilistic model, there are assumptions about whether the inherent defects in the software are fixed, not fixed, presumed infinite or presumed finite.  Also, there are 2 measurement domains.   The time domain is when time between defects is observed.  The defect domain is when the defects over time are observed. Below is a short summary of when some of the models are best used:

• Time and fault based exponential - useful for later testing when software is operational or in a customer acceptance state

• Weibull model - useful for all phases of testing
• Logarithmic models - useful for early testing when some bugs are more likely to be found then others

During the 1980's and 1990's there was an effort to pick just ONE estimation model.   However, this effort was invalid as there is no probabilistic model that is flexible enough to be used on all defect profiles.  There is also no good reason to have just ONE model.  There isn't just one model for hardware so why should there be one model for software?