Once measurements are good enough (below threshold of hearing), other things (which we cannot measure, or more precisely, not easily measure, or not easily present or visualize) come into play. Also, there are many measurement secrets used by people who design DACs and amps that do correlate to sound quality. If you have not made stuff, it would be totally understandable to think of this as voodoo.

Basically, measurements are only good enough to tell you if something is shitty. The basically measurements that you see on this site are very coarse. Many people fail to realize that measurements are wave to visualize extremely phenomena. The very process of having to present something complex in an understandable visual sense results in the throwing away of a lot of data.

That's being said, there is absolutely nothing wrong to look at measurements and buy audio gear based on good measurements. The problem is what measurements and availability of measurements and myopic take on measurements.

There is also nothing wrong in buying products based on how something subjectively sounds to you.

There is also nothing wrong in buying products because they cost a lot of money, are really cool, and you can easily afford to do so.

Again, measurements can tell you if something is really off or shitty. They cannot necessarily tell you if something is awesome.

Again, measurements can tell you if something is really off or shitty. They cannot necessarily tell you if something is awesome.

And this is the problem with the current suite of audio measurements: they will help find fault, but they won't necessarily tell you what something sounds like. Also, if the fault is illuminated by measurements, you may end up having to use something that looks a little like voodoo to un-do it.

Let's look at the current suite of measurements people tend to fixate on: THD, IMD, noise, power output into various loads, output impedance (AKA damping factor), crosstalk, jitter (on the digital side.)

Are these valuable measurements? Sure. But many of them are limited.

THD is measured at a single frequency, or stated over a range, based on a sweep. It does not show you the harmonic components of the distortion. The harmonic distribution is important, since higher-order harmonics are usually sonically more objectionable. IMD is closely correlated with THD, and has the same limitations.

Both THD and IMD have to be defined in-circuit, on production boards, not quoting some manufacturers' number. THD and can go up several orders of magnitude due to one mis-routed trace or on-the-edge compensation. It can also vary in production. Even if one product measures 0.0002%, do you want to quote that for everything? Do you want to set your automated test limits that low? Do you want to provide a FFT with every product, so people can argue over "hey, I got one with -107dB 2nd harmonic, but yours is -108dB, what gives?"

Noise, similarly, doesn't get into the spectral component. Is it in the audio band? Out of the audio band? Both can be bad, one from a usability perspective and one from a "smoking the tweeters" perspective. What's the weighting? What's the reference? SNR should always be specified as something like, "-102dB, referenced to 1VRMS, unweighted," to be meaningful. Throw in an FFT? Same problems. "Hey, you're -104dB, I'm -102.5!"

Power output into various loads is definitely something that should be spec'd, and it is meaningful. However, it should be quoted as RMS per channel, or it is meaningless. An amp that does 100W at 8 ohms sounds impressive, until you dissect the claim and it is actually summing the total (so 50W per channel) and it is not RMS (so 25W per channel.)

Output impedance is also meaningful (or, in the case of a speaker amp, damping factor.) These are relatively clear measurements, no arguments here. Of note, though, is the relative ease of getting low output impedance with a high-feedback solid state design, and the difficulty of achieving the same with tube OTL output. There are limits here.

Crosstalk? Sure, why not, though it varies with frequency (and can vary with input), so spec this conservatively as well...claims of 100+dB better have discrete connectors spaced several inches apart.

Jitter? Argh. Claims abound these days, typically of fantastical 1ps numbers that aren't really measurable without a dedicated $32,000 piece of gear (which only measures jitter.) And don't move the cable, because that will change the measurement by a factor of 10.

So how does this correlate? Well, it allows you to find the gross errors. It gives you some data with which to correct the errors. But it does not tell you how to lay out the PC board for best measurements--and it does not tell you what it will sound like.

So what do we do? We use the measurements, and we listen. And, we actually use one non-standard multitone test that goes beyond simple IMD and l ook at both distortion products at the output and at the feedback summing junction (if there is feedback applied.) This test appears to provide some correlation with perceived SQ, and I believe it is the reason we are getting more towards consistently producing neutral amplifiers with good resolving ability than before. However, the correlation is not 100% (and certainly too debatable for, say, publication), so we continue to measure, listen, and tweak as necessary to achieve the results we're looking for.

My 2 cents, the normal suite of electrical measurements as quoted by manufacturers ain't enough to discern any useful information. Proprietary methods utilising multitones or overlaying info or stuff related to even the electrical parameters gets you a little bit of info like how different filters sound/ zero feedback amps might sound like.