Metapp,
Good discussion here.
The cancelled, withdrawn and expireds are searched for relevant indicators and if one (or more) of those is a good match it will end up in the report as a photo and caption indicator, not on grid as there is no sale price, only asking price which may be way out of line; also, the seller motivation and circumstances of a listing come into question such as a listing cancelled which then becomes a short sale or REO later on.
This is what I mean by inconclusive, there is no settled price to adjust against. But as you do could certainly be reported though migh bear little or no weight. Therefore these are left out of the number count for the 1004MC. Even though these might become shadow inventory of which there are many (but not as many as was the case 2 years ago) in most markets now.
You mention a bifucated market area with older OR new/newer homes; I would leave out of the 1004MC and also at top of page 2 1004 the properties not a reasonably good match for your subject as to age even though a buyer might consider either/or in the specific school area. Can be explained easily enough in market area comments why you did this.
ResGuy mentions trending, I get my trends from a longer look at listing prices, supply,
demand, SP as % of LP, and also days on market, not from 1004MC stats, such as they are. Because as mentioned these end up being too small a sample as these are "comparables only". That's my take.