We have seen already the bearing for the full adjustment (in
section 5.5) and we can calculate the bearing for the
adjustment solely on 
, giving a vector essentially in the
direction of 
:
The difference between the two is the bearing for the partial adjustment:
It is solely in this direction that expectations can change according to
the new information contained in 
. Clearly these two directions
(which aggregate to form the overall bearing) are different, so that the
partial adjustment is telling us something extra about a different
direction. We can summarise this by evaluating the path
correlation: the prior correlation between the previous and partial
adjustment bearings. Here the path correlation is
showing that from the point of view of revising expectations, the data are partly contradictory. Had this correlation been positive, we would have argued that the data complemented each other, with the magnitude of correlation indicating the degree of consistency. Our interpretation follows from our being able to write
Thus, it is the negative covariance (which we see summarised as
a path correlation of 
) between
the bearings for the previous and partial adjustments which serves to
diminish the size of the joint adjustment: the changes in
expectation for the initial and partial adjustments are in different
directions, and thus tend to cancel out each other.