I think top 13 bits are the same, below that data is hijacked for MQA. As with most things MQA related, then it isn’t quite a simple as saying the end result is only 13 bit, supposedly the encoding process remains sonically equivalent to 14-15 bit (I think).
Personally I dispute that - I did run some test on various MQA files I could get my hands on to see how they sounded in the absence of any MQA decoding and rendering and I was not impressed. I thought it was generally quite obviously harsh. Some thread (or maybe another forum) on here has photos of what my cat thought of it - he was repeatably looking a bit stressed with the MQA version as a reasonable listening level, but wasn’t bothered by a non MQA version at the same level or MQA decoded version.
[Edit - turns out it was here High-end DAC really necessary?. Sadly he passed away early this year].