At Virginia Tech I hired some really great research assistants to do data entry. They entered population data for more than a thousand of these sheets for Austin — more than 53,000 residents for the city in 1930.
All of these sheets were collected and documented by Enumeration District, the areas that individual census enumerators were assigned to canvass in the city. Doing some research, I found records from UT-Austin indicating the geography of 1930 EDs. Plugging that map into ArcGIS, I was able to match up the 1930 EDs with the 1940 Census Tract shapefile and georeference it. You can see that these mostly match up with the Census Tract boundaries. They are not one-to-one, but rather 3 or 4 EDs put together generally make up a CT. Here the treatment was easy. If the boundaries are the same, just sum up the individual level data into the enumeration districts, then sum up the ED-level data into the Census Tracts. In several instances this is not the case, where one ED boundary does not line up. In these cases I had a research assistant track down the blocks and shift the data over to the other, appropriate census tract. Then I summed up the data from the Census Tract.
Having all of this data hand-entered brought some problems — non-readability of the hand-written folios and simple entry errors. I controlled for this by having other assistants spot-checking (or checking it myself) or re-reading the Census folios. In the case of countable information (sex, race, home value, etc.) the entries were much more standardized and easy to read than individual names, for example.
Then, with this spreadsheet full of data tallied, I assigned the 1930 equivalents to the 1940 Census Tracts a new field facilitating the ArcGIS join function, as NHGIS does, “GISJOIN.” I took it over to ArcGIS, performed the Join function, and now have meaningful tract-level Census data for Austin in 1930 with almost direct comparability to 1940 data. Here are two maps illustrating racial distribution in the city. The first is the African American population; the second is all non-whites (including Latinos).
Here is the comparable map I created from 1940 data, along with student residence data. Here you can quite clearly see the effects of the city effort to drain the minority population from around the city and concentrate it in East Austin. This was the intent of the 1928 Koch and Fowler city plan (Austin’s first) and the 1928 municipal infrastructure program to implement the plan. The thinking was to maintain segregation but not create dual facilities throughout the city. By concentrating African Americans and Latinos in East Austin, the city would only have one set of black facilities and the black population would choose to move there in order to use them. We can see that the proportion of minorities in other parts of the city did indeed decrease throughout the 1930s.
This is a fairly labor-intensive process, I admit. It took months to get all of the data entered and to check it, and this is only version 1 of the data (we will no doubt do some more revisions with more checking). In addition, the city boundaries changed a bit as the city annexed land, so at the edges of development (see the northwest area of the city), there was some development outside the city that was not counted in 1930, but that would skew a comparison to 1940 data. However, this is fairly limited and it is a risk I am willing to accept in order to be able to make some comparison at all.
In addition, it opens up possibilities for nearly any city in the 20th century. If the data can be entered (and for 1940, the Minnesota Population Center is undertaking that right now), then anyone can compare a city’s 1940 data to later censuses. In other cities without such growth at the borders, the Austin boundary problem is much less of an issue. We can effectively push back the boundaries of the U.S. Census as far back as addresses were taken in any U.S. city.*
*This is not possible for 1930 or 1940 in Blacksburg, VA, for example, where I teach at Virginia Tech. The town was so small in 1930 and 1940 that buildings did not have street numbers, according to Sanborn maps and town directories of the period.