The accuracy of order of magnitude estimations comes from the balance of under estimates and over estimates. Work through a few Fermi questions and purposely skew some of your responses. Does this affect your final estimation?
The University of Maryland maintains a site dedicated to Fermi Problems. You can see them here: http://www.physics.umd.edu/perg/fermi/fermi.htm
Another example of a Fermi-like-problem is the Drake Equation. It tries to estimate the number of intelligent life in the universe. See a TED-Ed lesson about it here: http://ed.ted.com/lessons/calculating-the-odds-of-intelligent-alien-life