Then for each observation we can calculate what the normalised deviation would be corresponding to that rank if the data were indeed normally distributed: this is d(calculated), and comes from the cumulative normal distribution.
A point in the middle of the list (rank number n/2) should have a value of d of 0.0, while those at the beginning will have d around -4 to -3, and those at the end of the list have d around +4 to +3
A normal probability plot then plots d(observed) against d(calculated). Each point on the plot is a single observation. For data following normal distribution, the plot should be diagonal with a slope of 1.0, with the points clustering densely around the centre, and sparse at the edges, since in this case d(observed) = d(calculated)