Lab 14 - Spatial Data Aggregation

Lab 14 was a doozy. Maybe it was being in the middle of trying to move or being near the end of the semester, but this lab was certainly challenging. The topic was spatial data aggregation. We started with readings and lecture by learning more about the Modifiable Areal Unit Problem (MAUP) which is the basic idea that different units of aggregation can affect statistical and spatial analysis results. There are two main issues related to the MAUP: scale effect, which is where results can differ depending on the number of spatial units used (e.g. population data within a block group versus a tract) and zonation effect, which is where results can differ depending on the variation of equal area units in a study area (e.g. a study area that is first divided into three north/south rectangles versus the study area being divided into three east/west rectangles). After lecture, we carried out two different parts to the lab to practice the concepts of result variation based on modified areal units.

In the first part of the lab, I carried out three OLS regressions to look at the relationship between nonwhites and poverty in Florida. Each regression used a different areal unit: first block groups, then zip code polygons, house voting district polygons, and finally counties. With each regression I also produced a scatterplot and table to identify how the statistics changed with the different unit sizes. In general, as the unit size increased and sample size decreased, the correlation increased due to a decrease in variable variation.

In the second part of the lab, the longest and most challenging part, I used congressional district and county data for the contiguous US to explore the practice of gerrymandering, or the intentional drawing of district boundaries in a way that favors a certain party or group. The first part to this section was determining multipart districts and which ones would be more appropriately one unit. The second and more intense part was trying to determine which districts in the study area were the "worst" offenders in district delineation. Two ways to measure the effects of gerrymandering are to determine how compact a district is (more compact would be more appropriate), and if the district minimizes dividing counties up between multiple districts (i.e. community).

My process for testing compactness was to calculate the area divided by the square of perimeter. If a circle has an area to perimeter ratio of 1, then counties with values near 1 would be more compact, further from one would be less compact. My process for testing community involved doing an Intersect overlay to find the counties that were broken up by districts, then do summaries and look at how many counties out of total counties in a district were broken up. The more that were broken, the worse the district was in terms of "gerrymandering."

Example of an district that is not compact, with high perimeter to area ratio. 

Example of a district that has poor "community" or high numbers of counties broken by it's boundaries (as well as not being very compact).