Getting more overload peaks in mp3 AFTER I reduce snare level - why?

I notice this quite often - I reduce levels on snare or other item and re-render as mp3 but after the reduction I get MORE or larger overload peaks. Why is this? (I mix to mp3 so I can do previews in my car/etc so I'm not sure if this would happen when rendering to WAV with a post convert to mp3 - but I only do that at the very end of my mixing). Does mp3 consider what sound preceded any given moment and compress these differently .. I guess it must .. so in effect it's more mpeg than jpeg. Is that correct?

Take a moment of your audio without snare. Say, peak value at this moment is +1(voltage scaled to [-1...+1]). You also have a snare at this moment, which have value -1. When rendering that values summed. So (-1) + 1 = 0 (silence) If you take snare away, the result will be +1 (loudest peak) In case of mp3, it use spectral processing, so it doesn't care about out level, but levels of spectral bins. After decoding phase of spectral parts can be different from what it was before. Some frequencies reduced, as a result, you will have different sum of phases of spectral bins. Very stupidly explained (I'm not got at English), but hope you get this.

Thanks both.. So MPL I think you mean because audio is effectively a/c that is how a snare could have a minus value. And of course it will have a minus value for half of its existence - it's not 'absolute' addition is it - ahah.. weird stuff.