I'll try and do this using ice creams, so as not to cause confusion about definitions
you have X people and not enough ice creams to go round, then the chance of getting an ice cream <100% (call it z%).
If you take a bunch of the "premium" people and give them all an ice cream, for the remainder the chance of getting an ice cream is now worse (call it y%). y<z.
i.e. when you remove a portion of the people from a ratio calculated using them (and say they were not affected by the rationing) the ratio can only change in one direction. In the table the change goes in both directions.
The assumption is that the same thing is being measured across the whole table.