This concept should be described in any steel text. Additionally, the design criteria are in the tension members chapter of the AISC steel spec.
Essentially, it's a phenomenon affecting tension members at the connection. If one part of a member is connected, and another is not, the portion of the gross area at and around the connected area becomes overloaded and the part that is not connected is not fully stressed. To accomodate this, we reduce the effective area at the connection.
Check out my spiffy braced frame isometric. If you were designing the HSS braces, you could take the full gross area for the tension member away from the connection. However, since it's only connected by four welds at the gusset plate (2 on each side), the portion of the gross area near the welds is fully effective, and the portion away from the welds (facing HSS walls in the isometric) are not as stressed.
Shear lag is inversely proportional to the length of the connection -- i.e. if the welds in the picture are longer, there is less shear lag.
You should rarely design tension members to their full capacity, because it will require doubler plates or additional reinforcement at the connection, which is generally much higher fabrication cost than upsizing the member to accomodate shear lag.
This is shear lag. Hope it helps.