Simply, if you have, say, two antennas seperated several times the wavelength, then the sub-channels of these two antennas will be independent. So, if you transmit the same data stream on the two antennas, then you decrease the probability of error by a power of 2, and you got MIMO diversity .
However, if you send different data streams on the two antennas, then you got MIMO multiplexing by which you increase the data rate.
Of courcse, which to use depends on the enviroment itself, and you may switch between the two schemes of MIMO adaptively.