Raw files contain much more information, stored in much finer increments, than 8-bit JPEGs do. Not only are raw files 12-bit or 14-bit, but they are the actual luminance values collected by the imaging sensor.
From this answer to Filter for RGB separation and its effect on the image:
Raw files don't really store any colors per pixel. They only store a single brightness value per pixel.
It is true that with a Bayer mask over each pixel the light is filtered with either a Red, Green, or Blue filters over each pixel well. But there's no hard cutoff where only green light gets through to a green filtered pixel or only red light gets through to a red filtered pixel. There's a lot of overlap. A lot of red light and some blue light gets through the green filter. A lot of green light and even a bit of blue light makes it through the red filter, and some red and green light is recorded by the pixels that are filtered with blue.
Since a raw file is a set of single luminance values for each pixel on the sensor there is no actual color information to a raw file. Color is derived by comparing adjoining pixels that are filtered for one of three colors with a Bayer mask. But just like putting a red filter in front of the lens when shooting black and white film didn't result in a monochromatic red photo, the Bayer mask in front of monochromatic pixels doesn't create color either. What it does is change the tonal value (how bright or how dark the luminance value of a particular color is recorded) of various colors by differing amounts. When the tonal values (gray intensities) of adjoining pixels filtered for the three different colors used in the Bayer mask are compared then colors may be interpolated from that information. This is the process we refer to as demosaicing.
The idea that you can view raw image files in any way "without applying any editing" is a myth.
Anytime you open a raw image file using an application to view it as an image on a monitor, there are development settings applied to the raw data. If you don't specify particular development settings, LR will use it own default settings. There's no such thing as a "straight out of camera" raw file that looks anything like we would expect it to look.
Here is what a demosaiced raw file (of a properly exposed scene) with the linear values recorded by the sensor uncorrected and converted to a jpeg looks like:
Here's the thumbnail preview image generated by the camera's raw conversion algorithm embedded in the same raw file:
Anytime you open a raw file and look at it on your screen, you are not viewing "THE raw file."¹ You are viewing one among a near-countless number of possible interpretations of the data in the raw file. The raw data itself contains a single (monochrome) brightness value measure by each pixel well. With Bayer masked camera sensors (the vast majority of color digital cameras use Bayer filters) each pixel well has a color filter in front of it that is either 'red', 'green', or 'blue' (the actual 'colors' of the filters in most Bayer Masks are anywhere from a slightly yellowish-green to an orange-yellow for 'red", a slightly yellow green for 'green' and a slightly violet blue for 'blue' - these colors more or less correspond to the center of sensitivity for the three types of cones in our retinas). For a more complete discussion of how we get color information out of the single brightness values measured at each pixel well, please see RAW files store 3 colors per pixel, or only one?
When you change the white balance, contrast, or many other parameters of a raw file you're not making changes to the 8-bit interpretation of the raw file you see on your screen, you are making changes to the way the linear 14-bit monochromatic raw data is interpreted and then displayed on your screen with the updated white balance, contrast, or other parameter. That is, you're using the full advantage of those 16,384 discrete monochromatic linear steps that the raw file contains for each pixel, not the 256 discrete gamma corrected steps in three color channels for each pixel that you see on your 8-bit screen as a representation of that raw file. You're also taking advantage of all the other information contained in the raw image data, including such things as masked pixels and other information that is discarded when the file is converted to an 8-bit format to be displayed on your screen.
How the image you see on your monitor when you open a raw file will look is determined by how the application you used to open the file interprets the raw data in the file to produce a viewable image. But that is not the "only" way to display "THE original raw file." It's just the way your application - or the camera that produced the jpeg preview attached to the raw file - has processed the information in the raw file to display it on your screen.
Each application has its own set of default parameters that determine how the raw data is processed. One of the most significant parameters is how the white balance that is used to convert the raw data is selected. Most applications have many different sets of parameters that can be selected by the user, who is then free to alter individual settings within the set of instructions used to initially interpret the data in the raw file. Many applications will use the white balance/color channel multipliers estimated by the camera (when using AWB in-camera) or entered by the user (when using CT + WB correction in-camera) at the time the photo was taken. But that is not the only legitimate white balance that can be used to interpret the raw data.
With a 14-bit raw file, there are 16,384 discrete values between 0 (pure black) and 1 (pure white). That allows very small steps between each value. But these are monochrome luminance values. When the data is demosaiced, gamma curves are applied, and conversion to a specific color space is done, the WB conversion multipliers are usually applied to these 14-bit values. The final step in the process is to remap the resulting values down to 8-bits before doing lossy file compression. 8-bits only allows 256 discrete values between 0 (pure black) and 1 (pure white). Thus each step between values is 64X larger than with 14-bits.
Once we have used the raw data to create one possible interpretation of that data and save it as an 8-bit JPEG, most of the information in the original raw file is not preserved in the JPEG. Only the information needed to produce our single interpretation of the raw data is preserved in the JPEG! If all we have is a JPEG, there's absolutely no way to recover all of the other information that was contained in the raw file. Almost all of the decisions we (or our camera's automatic routines) used to process the wealth of information contained in the raw data are "baked in" and are irreversible.
If, for example, we then try to change the WB with these much courser gradations, the areas we try to expand push each of the steps in the data we're using further than a single step in the resulting file. So the gradations in those areas become even coarser. The areas we shrink push each of those steps into a smaller space than a single step in the resulting file. But then those steps all get realigned to fit the 256 step gradation between '0' and '1'. This often results in banding or posterization instead of smooth transitions.
¹ Please see: Why are my RAW images already in colour if debayering is not done yet?
Examples
As the saying goes, "The proof of the pudding is in the taste."
Below are some examples of images shot under difficult conditions that benefited from the increased flexibility that raw processing allows.
Concert/Theatrical
Theatrical/concert photography is one of the most challenging kind there is, both in terms of pushing the equipment you use to the absolute edge of their capabilities and in terms of requiring every bit of skill and experience you might have as the photographer.
Photography is the art of capturing light. Most concerts don't offer much light to capture and what light there is to capture is changing rapidly and the subjects are usually very animated. So the traditional solution to not much light (longer shutter speed using a tripod to hold the camera still) doesn't work because nobody on stage stands still for 10-15 seconds while you take a picture. The traditional solution to capturing motion (faster shutter speeds) doesn't usually work because there isn't enough light to capture a good image on a small sensor with a narrow aperture. In the end you have to balance the two as best you can AND use gear that allows you to capture as much of the scarce light that is present in the scene in as fast a time as possible. That means fast lenses (wide apertures), larger sensors, and cameras that are highly responsive (fast handling).
Due to the nature of the less-than-full-spectrum lighting used at many concerts, post processing is a necessary step to get optimal results. Although you can use custom white balance an/or white balance correction in-camera, the range of adjustment they give you in-camera is very often not enough to fully compensate for the deficiencies in the lighting.
Straight-out-of-camera JPEG under difficult LED stage lighting that is now quite common in small bars and nightclubs:
Color correction using the "eyedropper" color picker tool applied to the jpeg:
Color correction using the "eyedropper" color picker tool (on the same spot in the scene as above with the JPEG), as well as simple contrast, highlight, shadow, and saturation slider adjustments, applied to the raw image data:
Post processing raw files with many applications will give you more room to adjust the white balance and also give you the power of an HSL (Hue-Saturation-Luminance) tool that lets you adjust each of about eight different color bands independently of the others. Please note that White Balance encompasses more than just Color Temperature. Color temperature is but a single axis in the two dimensional color wheel. (Brightness/saturation of any particular hue is yet a third dimension). White balance includes adjustments along the Green ←→ Magenta axis as well as color temperature adjustments along the Blue ←→ Amber axis.
The accepted answer to How to cancel purple stage lighting on subjects? covers white balance for photographing stage acts in smaller clubs that almost exclusively use LED lighting these days. This answer to Blown out blue/red light making photos look out of focus specifically talks about how to deal with LED lighting when only the blue and red lights are up and the green lights are dark.
Image with standard 'Auto white balance' and other in-camera settings applied:
The same image with significant WB correction, contrast, and HSL adjustments:
Canon EOS 5D Mark III + EF 50mm f/1.4
Sports under dim, flickering lighting
In-camera produced JPEG:
Edited JPEG using the above JPEG as the source:
Edited CR2 file of the same image:
For more about how this image was produced, including raw processing steps applied to the data from the raw file, please see: Lots of noise in my hockey pictures. What am I doing wrong?
High Contrast Scenes
RAW file with Canon's "neutral" in-camera processing applied using Canon's Digital Photo Professional (ver. 3). This is pretty much identical to what an out of camera JPEG would have looked like:
The same .CR2 file after extensive processing and tone mapping using the raw image data:
For more about how this image was produced, please see this answer to: How to make camera LCD show true RAW data in JPG preview and histogram?
Low CRI Lighting
Some forms of lighting are designed with energy efficiency as the primary consideration. They have poor CRI (Color Rendering Index) performance. This means that, unlike most natural light sources, they don't produce a broad spectrum of visible wavelengths, but only emit a narrow spectrum of wavelengths of light.
Three versions of the same image. The one on the left is an unedited conversion of the raw image opened using default settings. The one in the middle is a color corrected conversion made using the raw image data. The one on the right is an attempt at color correcting a JPEG version of the image on the left.
For more about how this image was produced, please see: Why can software correct white balance more accurately for RAW files than it can with JPEGs?
One thing that all of these examples have in common is that someone had to apply their expertise to process the raw image data to produce a better image than what the camera's automatic processing routines, the photographer's manual settings done in camera, or the default interpretation of the raw image data by a raw convertor such as Lightroom, could produce. It doesn't happen automatically when a raw image file is first opened in LR very often.
From the comments:
... raw files are 16 bit formats, just as dng's and tiffs are. They may contain 12 or 14 bits of information depending on the ADC process used; but that is only true when the sensor is generating 12 or 14 bits of data. At high ISOs the sensor may be generating less than 8 bits of data, in which case the file format is irrelevant because even the 8 bit jpeg format has enough accuracy/capacity to record it. But what is not irrelevant is the lossy processing the camera will apply to jpegs regardless of how much data was originally available.
"... but that is only true when the sensor is generating 12 or 14 bits of data." At the most fundamental level (each sensel), sensors do not generate bits of anything. They generate analog voltages. There can be a very large number of potential voltages generated between "0" and full well capacity.²
Even high ISO files can use the full range of 16,384 discrete values between 0-16,383 to numerically assign luminance values to the analog signal when it is converted to a 14-bit digital file. The highlights, represented by any voltage over the "gate voltage" are still "16,383". Due to the analog amplification applied, it takes much less than full well capacity to produce enough analog energy to "peg the meter" and everything past that is, of course, clipped. But all of those voltages above the threshold are assigned the same value (16,383). None of the other 16,383 values between 0-16,382 are wasted on those clipped values. Just because there may be 8 Ev or less of DR in the information between darkest (lowest voltage) and brightest (highest voltage) after an analog amplification of 64X (six stops, or ISO 6400 for a sensor with a base sensitivity of ISO 100) does not mean that only 256 discrete values are used to numerically represent that information. All 16,383 discrete values are still available when encoding those analog voltages to 14-bit numbers.
"... in which case the file format is irrelevant because even the 8 bit jpeg format has enough accuracy/capacity to record it."
This understanding also fails to grasp that raw values are linear monochrome luminance values. If you stretch only 8-bits (256 discrete values) of linear raw luminance information when applying light curves to approximate the logarithmic (with "shoulders" at the extremes) response of human vision, you'll get far rougher transitions from dark to light than an 8-bit jpeg, which is typically produced by converting to 8-bit and then compressing only after demosaicing and gamma curves have been applied to 12-bit or 14-bit numbers, will demonstrate.
It doesn't matter how many digits it's written with, nor how much amplification is applied to get there. If the sensor is only generating 8ev/bits you cannot increase the accuracy beyond that.
Again, there is not a necessary direct relationship between "1 Ev" and "1-bit" any more than there is a direct relationship between "one stop" and "one zone" in Adams' zone system. That was the entire point of what Adams did: to squeeze more than 11 stops into the 11 zones (0-10 inclusive) for high contrast scenes and to stretch fewer than 11 stops to the full 11 zones for low contrast scenes. When we amplify analog voltages before converting them to digital, we are doing the equivalent of the latter.
One can represent more than 8 stops of DR using an 8-bit JPEG. We do it all of the time. To do it, we do have to give up smooter gradations using only the 256 discreet numbers available (0-255) to fit more dynamic range into 8-bits.
It the same thing as converting an 8bit jpeg to a 16bit file/space, or
increasing the exposure in post... the numbers are different but
nothing else has changed, and there is no increase in the accuracy of
the original data.
No it isn't. Smooth slopes are not converted to "stairsteps" until after digitization occurs. When you amplify an analog sine wave, you do not get a "stair step" pattern, you get a sine wave with a steeper slope, but it is still a continuous slope. If we collect an image that contains a smooth range of analog voltages with at least 16,383 different evenly spaced levels between "0" and one-fourth of the sensor's full well capacity, when we amplify that analog voltage by a factor of 4X (two stops) we still have 16,383 different evenly spaced voltages that are 4X stronger than they were before. The slope between the lowest and highest levels is not sliced into discrete stairsteps until we do analog-to-digital conversion.
All values below one-quarter full well capacity (FWC) can be smoothly digitized to all 16,384 values of a 14-bit file when the analog signal is amplified by two stops (4X) the native sensitivity of the sensor. All values over 0.25 FWC will be clipped at "16,383". Values beneath 0.25 FWC can be smoothly distributed along the entire 16,384 values possible using 14 bits.
"At the most fundamental level (each sensel), sensors do not generate bits of anything. They generate analog voltages." And that's the point... high ISO's are used to compensate for collecting lower levels of light and generating lower voltages. And when that is done there is less difference between the min and max voltages generated by the sensor... I.e. less accuracy/information.
There's certainly less difference in total brightness range recorded at higher analog amplifications, which means finer gradations for each value between "0" and "16,383". Only digital amplification of already digitized information will skip some of the values between "0" and "16,383". For example, if you raise a raw image 2 Ev digitally, you're multiplying all of the values by "4X" and the resulting values used will be 0, 4, 8, 12, 16, etc. while values not used will be 1, 2, 3, 5, 6, 7, 9, 10, 11, 13, etc. The analog information prior to ADC does not suffer from this limitation. Within the range of voltages that are not over the system limit, smaller differences in the pre-amplified voltage can be assigned to each of the 16,383 "steps" of the digitized values. All values between "0" and "16,383" are still possible even after the analog signal has been amplified 4X base ISO. It's just that anything over one-fourth of full well capacity before amplification will then be over the maximum voltage threshold and equally clipped.
The biggest problem with doing this is that all "shot noise" (Poisson distribution) and the mostly thermally related "read noise" in the analog voltages read off the sensor are amplified as well, so when we start processing it digitally, we tend to raise the noise floor by cutting off everything below a higher value as "black." (We cut off a good bit, even at base ISO, but we tend to cut off more as the analog amplification is raised.) But that does not mean we have to. The information is still there in the digitized numbers of our 14-bit file.
It is irrelevant what number is used to describe a value... and all values are relative. I.e. I can use a ruler with 1/32 increments or 1/4 increments, but if what I am measuring is 1/1 it makes no difference.
And there is a direct relationship between 1 bit and 1 EV... the digitized value must be 2x the previous in order for it to represent/record a 1stop/EV difference.
There is no direct relationship between bits and Ev until you have digitized things into bits!
The increments on your ruler do not create discrete steps until digitization! You can measure any number of objects that are any (theoretically infinite) number of lengths between 1/2 inch and 17/32 inch. The objects will all fall between the two marks on your ruler and be recorded (digitized) with the same number. If you magnify the objects by 4X before you measure them, the lengths of the objects between 1/2" and 17/32" when magnified by 4X will range from 2" to 2 1/8" (2 4/32"). A step of 1/32" after magnification (analog amplification) allows for four discrete recorded values between 2" and 2 1/8" instead of one.
There will now be four discrete values of the magnified object between 2" and 2 1/8" using increments of 1/32" where before there was only one discrete value between 1/2" and 7/32".
² The Canon EOS 1D X Mark II, for instance, has a full well capacity of 103,999 e-. Since each increase of one electron registered by the sensor would give a corresponding increase in voltage, that means such a sensor has the potential for 103,999 different voltage levels for each sensel at base ISO. This is more than 6X the 16,384 discrete values possible with a 14-bit digital numeric value. So only at an amplification of more than 6.34X would it not be potentially possible to have an amplified voltage for every discrete value between 0-16,383 when the analog voltages are digitized. That's 2.67 stops, or ISO 640 for a sensor with base ISO 100. 103,999 potential different voltage levels is 406 times the number of discrete values possible with an 8-bit digital numerical value. A factor of 406X is 8.67 stops (2^8.6666=406). That equates to ISO 40,960 before a difference of one electron is theoretically equal to a single discrete step with an 8-bit digitized numerical value!
