The following write-up is summarized from this DPReview thread, with thanks to all who commented and helped me think about this issue. I will update as I continue to learn more about this and other metering issues.
While Experimenting with the Nikon D80's Matrix Metering system, I confirmed a Digital Outback Photo post that claimed matrix metering "averages" exposures. This, along with the fact that Matrix Metering biases toward the AF focus point is an important concept to understand and keep in mind while shooting.
Setup:
Equipment: Nikon D80, with 24-85mm AF f/2.8-4 lens, mounted on tripod (sorry Ken Rockwell) to produce consistently framed shots.
AF mode: AF-Single (others have noted the behavior described in this write-up does not occur when using AF-A or AF-C).
For this test, I shot using ISO 400 and f/5.6 in Aperture priority mode.
The first shot shows the composition as I want it: vase on lower right third, and the center of the frame on the back-lit window. Here, I focused on the window, dead center in the frame (a bright spot) and got an exposure of 1/50 sec.
Next, I paned left, focused on the vase, left the shutter half-pressed, then paned back to the desired composition. Matrix metering gave a 1/30 sec exposure.
Now, I repeated the pan, except this time I also locked the exposure after focusing on the vase. Matrix metering returned a 1/15 sec exposure.
Finally, I added a 4th case, locking the composition as in the first case above (where I focused on the window), but moving the focus point to the vase (without panning, e.g., by selecting from the 11 focus points a point that fell on the vase at the same spot where I focused for shots 2 & 3). Would I get a 1/15 sec exposure or a 1/50 sec exposure? The exposure came back at 1/50 sec exposure.
The key here is in the results for the 2nd shot. The resulting exposure lies nearly halfway between the 1st and the 2nd, suggesting, as the Outback post indicated, that Matrix Metering has performed some sort of averaging or compromise between the areas of high and low contrast. Or to think about it another way, Matrix Metering has used 2 scenes to evaluate lighting rather than the final one: one scene had the vase in the center (prior to panning back to the window), and the other had the window in the center. Interesting and useful information, I think.
But what will happen if the scene doesn't have high contrast between dark and light areas?
For this test I went outdoors during an overcast day and picked a scene that had some shadow areas to be sure, but nothing too dark. I also selected my focus points so that I wouldn't be picking any spot that was too dark or light. Here's how it went.
With Aperture priority at f/8.0, the first shot focuses on the lighter green bush in the center of the frame, with an exposure of 1/125 sec.
The second shot focuses on the brick above the back of the chair, then with shutter half-pressed, pans to the left to compose the scene: exposure = 1/200 sec.
The third shot mimics shot #2, except that it locks exposure before panning: exposure = 1/200 sec.
The fourth shot shifts the AF focus point to the left and at the top of the back of the chair: exposure = 1/200 sec.
It would seem that with lower contrast, even for a scene with mixed light values, Matrix Metering produces more consistent results. This would seem to match up with comments made in this thread about MM being better suited for scenes with lower dynamic range (1-2 stops?).
It is useful to note that the behavior demonstrated in Test #1 occurs only with AF-Single mode. Neither Auto nor Continuous mode exhibit this behavior. Also, rather than calling this a bug, if we think about it, this is a useful feature, as IMO, the exposure where "averaging" took place actually gives the more balanced results in terms of lighting and detail. It would be good if Nikon documented this feature to eliminate confusion among users.