4.5 Assignment 4

The BART station example from this chapter is just one example of using available public Census datasets to create a quasi-experimental design in the Bay Area. One of the clear insights from class this week is that good quasi-experimental designs (in absence of the ability to conduct RCTs in many urban systems contexts) require intentional tact and creativity to construct. Consider the datasets that are publicly available to us for observing outcomes in the Bay Area over time (such as 5-yr and 1-yr ACS and PUMS data, PG&E energy data, air quality data, etc.), with all of their various limitations, and explore one question of interest using a difference-in-difference analysis. This brainstorming should be guided by some open-ended searching for research papers online, as well as searching for notable events in the Bay Area (within the time period for which we have data) that had a focused geographic effect that may be observable through our longitudinal data.

A possibly useful constraint is to focus on a year that is right between two 5-yr ACS datasets, which would let you examine outcomes at the CBG level. For example, using 2014-2018 ACS data, one could examine the outcomes of a Bay Area event around late-2013 early-2014 that affected a limited set of block groups, compared to matched CBGs with similar outcomes in 2009-2013 and 2005-2009 data. (When 2015-2019 ACS data is released on December 10, 2020, then for the first time, it will be possible to line up 3 separate 5-yr summaries: 2005-2009, 2010-2014, and 2015-2019.)

Keep in mind all the personal judgment involved in constructing a difference-in-differences analysis, and explain your reasoning and methodology clearly. It is perfectly acceptable for your analysis to yield no rejection of the null hypothesis.