First-linkage of U.S. Census Data Households

A VoxEU note recently presented an apparently novel use of U.S. Census data: to measure racial segregation in American neighborhoods. Simply put, if almost all households of a particular declared race are adjacent to another household of the same race, there is clumping or clustering, hence segregation.

Whence the Claim : "Federal census enumerators would go door-to-door when enumerating the residents of a community. Consequently, next-door neighbours appear adjacent to one another on the census manuscript pages. With the availability of complete census manuscripts for 1880 and 1940, this allows us to identify the races of all household heads and of their next-door neighbours. "

There are some possible problems with this, but I don't know that they invalidate the conclusions one would (and the authors do) draw from using the measure as proposed. I have browsed census data from 1940, and a few others, in the course of genealogical curiosity.  Much of the data I've seen was already transcribed, the 1940 data was not.

1. The 1940 data in the neighborhood I studied did not have all the households reported consecutively with their physically adjacent neighbors;  when nobody was home, there were gaps filled later on other sheets.
• it may be that either the transcribers have re-sorted the data to put households back in physical adjacency order, or that it does not really matter (it might not, but I haven't thought it through so cannot say).  
• Would the phenomenon have changed in its frequency or handling (by the census takers and processors) between the two censuses in a way that distorts the apparent trend? It might have; urbanisation had increased, perhaps household size had decreased, the latter of which might increase the number of first-pass gaps (nobody home when the enumerator knocked on the door). If list adjacency does not reliably indicate residential adjacency, the statistic is not what it is meant to be.
• Does it matter whether the enumerators' routes crossed the street or turned the corner? It might, depending on the street to cross. A household on a corner would, I believe, feel the house around the corner to be closer than the house across the street, particularly if the street is one that serves as a sort of demarcation of neighborhood boundaries: the street on which side one resides matters.
2. Household and dwelling place are not the same thing.  Many households are in multiple-household dwelling places, such as apartment buildings or  boarding houses (not only for singles), or residential hotels.  How then to meaningfully apply the concept of next-door neighbor?
• I do not know how they handled this, I've only read the VoxEU piece, not the paper it summarizes and pitches. Perhaps I shall.
• Given the intended use, I would expect that dwelling in a place where any of the other heads of household "under the same roof" is of a different declared race, whether or not they live on the same floor (who know? not the census) counts as having a different-race neighbor.
• It is not so easy to decree that someone living in a multiple household dwelling place that is mono-racial (according to declarations of heads of household) next door to one that is mono-racial of a different race has a next-door neighbor of a different race.

I think the proposed statistic is interesting, and possibly useful.  But it needs some vetting.