I addition to ^this^:
I think that the thing that a lot of folks forget is that a Mass Air Sensor takes only a SAMPLE of the air that is flowing through the housing. Let's take this one for instance:
Let's assume that these two hosings (for the purpose of discussion) are exactly the same size and say that they are each 70mm housings.
Let's also assume for this discussion, that they flow the same CFM under identical circumstances.
Each of these housings sample air passing though it, at a different location. The one on the left has an obvious sample port and is designed and tuned to calculate (the best that it can), the total air volume passing through it as a formula of the whole.
Now the one on the right has the sample port in a different location. Air movement through this housing is going to differ from air movement for the housing on the left. As a result, a different calculation is used to determine the total volume of air passing though it.
What happens if we just swap the sensors on these two housings?
The answer is that the each of them is going to be less accurate because it is not sitting within the housing that it was designed and tuned for.
As JR and Mike have illustrated, the sensors themselves are designed to provide a specific output for a specific volume of air that passes over the sensor. This volume of air is 'calculated' to be an accurate representation of the total mass of air (versus being an actual calculation of the total mass of air).
Changing even the pattern of air as it passes over the sample port, can provide you with some erratic results. If you've ever had to re-clock a MAF because of some sillyness, then you know what I'm taking about. Swapping sensors into NEW/DIFFERENT housings is problematic when trying to uses these kinds of "assemblies" with an OEM EEC.
Aftermarket and/or standalone units are slightly different. They can look at input from many sensors and re-calculate the transfer function of an unknown MAF. Some do it better than others. OEM computers are not designed to compensate for this specifically but can adjust within a tolerance of roughly 20% (that's 10% above and below the base line). That's not very much. It's enough to allow for slight variations in assembly tolerances. Most stock vehicles are generally into and make use of about 5% of the available adjustment just because of differences in pump gas alone. For simplicity, I'll leave engine condition out of this scenario.
Now consider for a moment, that you've discovered a an assembly and sensor that is dead on at idle but varies ever so slightly as air-flow increases. At the top of your transfer function (before you get into the WOT fuel table), your sensor output is going to be either above or below expected output as volume increases. Exactly how much depends on the sensor/housing combo. The point in this event, is whether or not the output signal is still within that +/- 10% of the adaptive strategy.
You really want to screw things up? Use replaceable sample tubes. Their primary function is to cause aggravation and thoughts of suicide. Why? Because small changes in airflow can have large effects when you are 'sampling' and not measuring the total volume of air.