[Discuss-gnuradio] why use 8 bits sample is not as accurate as use 16 bits sample shift right 8 bits then shift left 8bits?

[James Jordan]

why use 8 bits sample is not as accurate as use 16 bits sample shift right 8 bits then shift left 8bits?
I use 8 bits sample and the process do not work fine, then I change to 16 bits sample and shift sampled data 
right 8 bits ( which must the same logic in usrp hardware to sample 8 bits data ) and then shift the data left 8 bits.
Then this works fine. 
So what usrp produce 8 bits sample data? Does not it sample 16 bits data then shift it right 8 bits then send it to host
through usb?