Why does step input lead to better fit in system identification as oppose to a chirp signal input?

I am using the 'ident' tool to fit a first order transfer function onto experiment data that I generated using a second order transfer function.

I apply a chirp input to a 2nd order lag which I am trying to identify as a 1st order system. I use an IDDATA sink block to capture the data input to and output from this model which is then saved in a data object in the workspace called chirp_data.

In the 'ident' tool, I then select Estimate -> Transfer function models -> 1 pole and 0 zeros and click estimate which returns a 1st order transfer function with 60% accuracy as shown in the screen shot. This Ident session is saved and called chirp_data_ident.sid.

This is not a very accurate fit. However, if I instead apply a step input to the 2nd order system instead of a chirp and use that data for identification, I get 90% accuracy. Similarly, applying a sine input and identifying yields a 58% accuracy.

Question 1: Why does the identification yield a better % fit when a step input is used rather than a sine or chirp input? (Surprisingly, I would have expected the chirp to yield the best result since a wider range of input frequencies are tested compared with the step input.)

Question 2: When I compare the bode plots of the true system with that of the identified systems I find that the bode plots do not match very well (as illustrated when Bode_Comparison.m is run). Why do the bode plots of the identified systems differ so much from the true system and what must I do to get better matching bode plots with that of the real system? (besides increasing the order of the estimated transfer functions)

Question 3: What does the "% fit to estimation data" represent? i.e. how does MATLAB calculate that value?