Ok I'm old to the music biz, but new to mixing,  I have been messing around with side chain compression and have a couple of questions.
 
1) I set up a mix where when the singer comes in the background tracks dip about 3db.  Is this a good idea?
2) When you set up a side chain thing for "ducking" like when a narrator is over a music bed,  the louder the narrator speaks, the more the music bed gets compressed downward.  Would you not want the opposite?  when the narrator  is soft you want more compression,  when the narrator is loud you need less compression.  how would you handle this??
 
THANKS!!!! 
 

1.  With sidechain compression, small gaps in the vocal can cause the background track to "pump".  Cool effect if you're going for that techno/house sound.  If you want something more natural, volume automation for the background track is probably the way to go.
2.  Ducking works best when the sound used to trigger the ducking has a fairly consistent volume.  Once again, volume automation for the music bed might be a better choice.

I have never used side chain compression in my recordings.  There is a danger that if it's not set right, and even if it is, levels can tend to pump as mentioned above.  I don't want that in my music.
 
I normally set up volume envelopes for control of levels but to answer the first question...... NO... I do not pull the backing tracks down at all when I have the vocals in the mix and the singer is singing.
 
I control the mix by leaving the bass and drums at the same general level throughout the mix..... I will use the volume envelopes to bring the other instruments up and down....in and out of the mix. If I want the singer to shine in a verse,,,,, I simply pull the mix back to bass, drums, and something like an acoustic guitar... as the chorus comes, I might add another instrument or two. Listen to a few of my tunes to see what I mean.
 
As far as #2... I don't do this type of mixing... narration over a music bed.... but in this instance I can see the usefulness of side chaining.  I see this as a see-saw effect.... the voice is given precedence. And any time it is present the music, which is secondary, is turned down so the voice can be heard clearly. Many radio talk shows use this in hardware format so that when the host talks, the callers are reduced in volume. 

I use gentle sidechain compression for things it's good at e.g. pulling back a bass transient that masks kick drum hits. Using it to drop the backing level behind vocals however can sound very artificial indeed, pump like mad and take the 'drive' out of the backing as well. Structuring the backing in some way to leave 'space' for the vocal,works much better.

As for sidechaining background music behind narration, I think you generally need a more even and slower response than a compressor, which even at its slowest release is still pretty fast acting and can obtrusively pump up the music in short gaps between words or phrases. Common practice seems to be to use a volume envelope on the background music track instead, so that the music is dropped throughout spoken passages and shifts in music volume can be slow and smooth. Keeping the background track sparse and not too ear-catching helps as well. While the music can go a long way to setting an emotional atmosphere, you want people to listen to the words not focus on the music.

Try using a side-chainable gate instead of a compressor. This alleviates the problem you describe, wherein the amount of compression is proportionate to the sidechain level. With a gate, it only cares if the sidechain signal exceeds the threshold and then dips by a fixed amount that's not proportional to the sidechain.
 
Ducking with a compressor works OK if the sidechain signal is pre-compressed. You can, for example, clone the lead vocal for the sole purpose of providing a sidechain to the background vocals. Because the cloned track isn't used for anything else, you can squash it flat, thus providing a level key. But in the end it's pretty much the same result as using a gate.
 
You will, however, need a gate that lets you adjust attack, and release times like a compressor. The good ol' Sonitus gate does this.

Does the latest version of legacy Splat include a side chain gate plugin?

Sonitus Gate

using this more and more:
 
https://www.waves.com/plugins/h-comp-hybrid-compressor
 
has a great wet/dry mix function for easy parallel compression

You can also use a dynamic EQ to duck only some frequencies, which avoids the pumping effect. Waves' F6 has a preset that's a good starting point. I'm also fond of automation.