A note on Southwood's instability criterion