Unlike jpg files, raw files have a greater capacity for capturing highlight detail. My preference for exposing raw files is to push the exposure of the significant highlight areas as close to the right end of the histogram as possible without clipping them. Speculars will clip at anything but radical underexposure so identifying them and ignoring them in the process is important. It is also important to realize that the camera histogram and highlight warnings are based on a camera processed jpg, not the raw file. Eliminating all of the blinking warnings will usually result in an underexposed raw file.
The reason for this is inherent in the digital capture itself which is not the logarithmic capture we see with our eyes. Linear digital capture records twice the data each time the light doubles, so underexposing by one stop is essentially throwing away as much as half the image information compared to a brighter exposure. After processing the information is measured against the inherent noise floor of the camera. This “signal to noise” ratio hides the noise under the important information, so the more information the less noise.
The best exposure for a full range image is an accurate exposure that uses as much of the capture capabilities of the camera. With an image with less contrast, processing a slightly brighter than centered image to a lower value pushes down the noise floor, and the added initial exposure results in more information in the shadow areas of the image. Newer cameras have lower noise floors than older digital models do, so noise is less than in the past. The latest raw processors are also better at reducing both luminosity and color noise. That does not mean it is not possible to add noise by expanding an underexposed image.
Typically I determine exposure with an incident light meter set to 1/3 lower ISO than the camera setting (meter at 160 – camera at 200). This has proven to give me a more accurate exposure, and I suggest testing your meter against your camera to see if that is true for you as well. This is essentially correcting the meter for an accurate exposure. This “meter calibration” is sufficient for most work, but you need to carefully analyze what the tones in each image require.
If I compare my incident exposure settings against the in camera meter I get a considerable difference in many cases, which is not surprising. I need to “overexpose” by 2/3 to 1 full stop when using my in-camera meter compared to the incident meter to get the same results. I blame the camera manufacturer for this discrepancy as I suspect in-camera meters and associated Program mode and other “automatic” exposure methods are designed to protect highlight detail when exposing jpg images rather than to properly expose raw files. I find TTL readings to be consistently underexposed.
My images come into the raw processor slightly bright using this technique, but rarely does anything but a specular reach the clipping point. Most exposure compensation is less than a half stop, and slight modifications to the highlight controls restore a normal looking image and shadow detail is rich and open.
The issue with talking about this technique is the necessity to use the term “overexpose” to describe the process. I am really compensating for the hardware involved rather than making a deliberate mistake. Learning to read the histogram and use it to modify exposure is a valuable technique. Remember that good exposure is different than good lighting. A well exposed image with bad light will still be a bad image. With good light and a proper exposure you will be able to accomplish more in your post processing, including rich tonal detail in both the highlights and the shadows.