If you are using a CPU to address one 2MB memory module, then since 1MB = 220 and 2·220 = 221 = 2MB, you need 21 address lines.
If you use a 2MB 16-bit memory module instead, you can remove address line 0. Why is that?
With 16-bit wide memory a read of address xxx0 gives you two bytes: the byte at address xxx0 and the byte at address xxx1. So the addresses ending in "1" are superfluous; you only read address xxx0, never xxx1. Now, since the last address bit will always be zero, we can drop it, ending up with an address 1 bit shorter.
One way to think about this is that the memory module consists of a set of blocks, and that the address identifies a specific block in memory.
So, for memory with an 8 bit block, an address would define 1 byte, and in the 2MB memory case you would need the 21 address lines for 2^21 addressable blocks. An address of '4' points to the 4th byte in memory.
If instead you have a 16 bit block, an address would define 2 bytes, and there would be only 2^20 addressable blocks. An address of 4 points to bytes 7 and 8 in memory.
That said, I would suspect that any decrease in address lines is more then made up for by the doubling of data lines.