Why might changes in the percentage of regional populations in extreme poverty not accurately represent changes in the number of actual people in poverty?