My two cents...people always say "don't mix with your eyes," but the more senses you can get involved in a process, the better. Devices like spectrum analyzers can be an important part of ear training, like when your ears tell you that something sounds "muddy"...but the spectrum analyzer shows you the mud is somewhere around 300-400Hz. You're now in a better position to correlate EQ settings to what you hear, and dial in the right setting without having to refer to the spectrum analyzer.
Another example of ear training is delay. Back in the days of the Echoplex, if you'd asked me how much delay I was hearing, my answer would have been "I dunno." But since the advent of digital delays, I've learned what delay times sound like and can come pretty close to identifying delay times.
A final example is modeling with EQ. Give me nine bands of parametric EQ, and I can make guitar A sound like guitar B to a degree of accuracy where a guitarist can't tell the difference in a blind test. But give me a spectrum analyzer, and I can do it in 15 minutes. If I have to do it by ear, it takes a
whole lot longer - but I've already gotten much better about getting frequencies "in the ballpark" because of what I've learned from using spectrum analysis.
So, while a spectrum analyzer won't necessarily help you create a better mix, it can definitely help train your ears by correlating what you hear to specific physical frequencies and characteristics...and that
will help you create a better mix eventually.