Neil and Chad have some good points to consider for anyone doing analysis. What they mention are essentially best practices in organizing data for analysis. By doing these best practices, one will avoid many headaches and help insure the results you get are valid.
To summarize:
1. Start out by making sure all the inputs have projections and coordinate system before anything else. If they already have a projection, but it is not he one desired, use a process to convert them to the one desired. As Chad alludes to, avoid the pitfall of assigning a desired projection to something that already has a projection - instead, use one of the several methods to convert from the current projection to the desired one.
2. Expanding on what Neil mentioned, it's usually best to take a feature class with joined data and combine it into one dataset before running analysis. Take the joined dataset and create a new feature class from it. This new feature class will permanently have the joined data in it, eliminating potential issues due to the Join. Then check the result. Note that it is common for issues in field names to crop up here, so checking it (and then editing your input Excel table to resolve them, if needed) is important. For example, look out for field names in Excel that contain restricted characters, have spaces, are very long, etc. Also, check that the data types are as expected. For example, did a string field become a number type when it should of stayed string? The bigger reason for doing this is that it identifies potential common issues before the model runs and blows up, or returns unexpected results.
Chris Donohue, GISP
