I suppose we should start with the differences between real world and digital recording.
The real world is measured in dBSPL or Sound Pressure Level. This is a measurement of the actual movement of air. There are two variations A and C. "C" is flat. If the air moves with sound, the air moves. "A" is biased to pay attention to the air movement the same way your ear does which is decidedly not flat. Fingernails on blackboard may make you hold your ears in pain whereas a light, but much more powerful earthquake (low rumble) will not.
So your ear works in the A curve and the microphones work in C.
http://en.wikipedia.org/wiki/Weighting_filter
There are average measurements published for "Train Going By" and "Airplane Going Over." The system is open ended at least going up. I believe physical pain is 130 dBA-SPL. That's where hearing loss can occur--ask any rock concert goer.
So that's that one. The microphone's job is to translate moving air into volts. That's not open ended. You can damage a microphone by subjecting it to loud noises. You also can't get sound too low or the microphone won't generate any volts. The job of the recordist is to put the mic where it will do the most good without breaking anything.
Once the sound becomes volts--hopefully accurately, an Analog to Digital converter creates a digital bitstream according to another dB range. Zero dB in sound is the maximum level that the digital system can handle without distortion. This system is open ended going down. -60 dB is the limit of human hearing. 16 bit sound channels run out of sound management at -96 dB. If you record in 24 bit, the theoretical bottom is -144dB.
Let's bring all that home. Even though sound signals half and double every 6dB, human hearing halves and doubles about every 18. This is where the Audacity percentage waveform scale falls apart really fast. I always convert to dB.
http://kozco.com/tech/audacity/AudacityPanelFull.jpg
So if your comfortable middle for recording is -18 (Europe) or -20 (US), then you have the ability to double the volume of the show for a special effect (add 18) without damaging the sound. In 16 bit audio (96 total range) that means you can go down to -66 from there without distortion, well below the range of human hearing.
Let me take that again. The show is shuffling along with normal dialog around -18 or so when somebody fires a gun to -1. As the gun shot echoes to nothing -50, a woman screams -3 and the room erupts in excited loud chatter -8.
So there is no direct comparison between real world dBSPL and dB inside Audacity. It's whatever you want it to be. If a room full of people talking peaks about -32 on the Audacity peak audio meters and someone with dialog peaks about-12, they will appear about twice as loud as the background and clearly audible.
This is where the BBC tells us strongly to use the speakers instead of just staring constantly at the meters.
Oh, stop using the spectrum view to measure loudness. That's hopeless.
Koz