Underexposure or overexposure?
It all depends the purpose of the pictures. If you just want to have some photos from your holiday or from a party with some friends, one tends to shoot in JPEG i.s.o. RAW. In many cases you would select a (semi-)auto setting. And typically you want to give it some extra EV push for snowy conditions and some less EV for taking a picture of your black cat in a dark brown sofa. Just for the reason not to end up with a grey cat in a grey sofa as the camera tends to level everything back to grey :-)
But what with pictures that you want to sell or print in large formats? Well, in most cases you will shoot them in RAW to allow some good post processing. Do you still need to follow the standard rules from above? Nope. You can forget about your grey cat in a grey sofa :-) Another way of working applies here. But before going into that, first some technical stuff to explain the background.
First keep in mind that a camera works in a linear way in contrary to humans. An histogram of an 8 bit picture has typically 256 values, with 255 being white an 0 being black. Say that you need a certain light condition to get to 255. What happens when you decrease the amount of light with 50%? You can chop the x-axis of the histogram also in 2 (value 128). If you again reduces the amount of light with 50% the left hand side of the x-axis is divided again in 2 (linear, right). If you continue like this, you get closer and closer to the black point on the histogram and this means in terms of images and image quality that at the lightest conditions, more detail is available in the picture than at the darker areas. If this is all too abstract or technical, we can better move on to the next paragraph :-)
How to show this in practice?
Let's take an underexposed RAW image and post process this to an acceptable format. Do the same with an overexposed RAW image and make sure that there is no clipping. After post processing the 2 images, zoom in on one of the dark areas to 100% or 200%. You will definitely see more noise in the image that was initially underexposed.
Since it's kinda impossible to upload noisy images here, you can always find the examples of what I explain below, at the link in my profile and go to the "RAW and EV setting" article.
The image at the left was taken with -2EV, while the image at the right side was taken at an +2EV. The images are then leveled back so that their histograms are similar and both are a crop of a 100% view, which allows you to easily see how much noise is generated by post processing the underexposed images up to an acceptable level. So even Raw is not sacred here...
So, I hope that you found this article of any use. Thanks for reading and keep 'em uploading!