NHGIS, one of the digital efforts of the Minnesota Population Center, is totally wonderful. Since I learned about it as a graduate student, it has been an essential source when I need demographic data from the U.S. Census and to help me think geospatially.
As a 20th century historian, and an urban historian in particular, I have run up against its limitations several times, which are really the limitations of the U.S. Census. The Census Bureau created the Census Tract framework and began implementing it in New York City with the 1900 census (augmenting and really replacing the Enumeration District). In 1910 it expanded to several other, slightly smaller cities and kept expanding to new cities every ten years. During the course of my masters thesis research on Ann Arbor, I was frustrated by the lack of Census Tract data before 1960 — the same goes for Berkeley in my book project.
In the course of my research on Austin, Texas, in the 1930s and 1940s, I realized that new digital methods and Census privacy law would allow me to break out of those limitations. Austin was tracted by the Census in 1940, but not in 1930. I realized that, because the 1930 manuscript Census — the individual-level records — are publicly available, I could plug them into 1940 Census Tracts and push the boundaries of the Census back in time. Here is how: