1. Structured parking costs less than underground parking, but in areas with height limits devoting aboveground space to parking can be costly in terms of buildable area.

2. One can have earnings without a job, of course.

3. All data, as well as survey questions, are available from the author.

4. There is no single definition of downtown Los Angeles, but the general consensus is that downtown is the area bounded by the 10, 5, 101, and 110 freeways. This area is encompassed in Census tracts 2062, 2063, 2073, 2074, 2075, 2077, 2079, and 2260. My estimate is conservative both because I probably missed some buildings and also because two ARO projects, the 1010 and 1100 Wilshire buildings, lie just outside these traditional boundaries (they are slightly west of the 110 Freeway), so in comparing the ARO with residential construction in previous decades, I err on the conservative side and remove their 455 housing units from the total.

5. Total building activity exceeded 9,000 units because considerable demolition took place as well.

6. This increase was large enough to warrant the U.S. Census Bureau splitting the tract in two in 2010.

7. These t tests were done for the price, the price per square foot, and the natural log of both.

8. Using the asking price rather than sale price allows me a greater sample size. Across the 877 units for which I have sales data, the average sale price is 99% of the average asking price.

9. There is some disagreement about the proper functional form for housing-price regressions, but the semi-log transformation remains the most common practice (Cassel & Mendleson, 1985 Cassel, E. and Mendleson, R. 1985. The choice of functional forms for hedonic price equations: Comment. Journal of Urban Economics, 18(2): 135–142. [Crossref], [Web of Science ®] , [Google Scholar]; Diewert, 2003 Diewert, E. 2003. Hedonic regressions: A review of some unresolved issues, (National Bureau of Economic Research Monograph). Retrieved from http://www.nber.org/CRIW/papers/diewert.pdf [Google Scholar]).

10. Logit regressions are a standard approach for binary variables, and Poisson regressions are standard for count data. When the dependent variable is overdispersed, a negative binomial regression is sometimes preferred to the Poisson. But the overdispersion associated with bundled parking is modest, and using a negative binomial regression does not change the coefficient. See Long and Freese (1997 Long, S. and Freese, J. 1997. Regression models for categorical and limited dependent variables, Thousand Oaks, CA: Sage.  [Google Scholar]).

11. The insignificance of the bathroom variable in the apartment equations appears to be driven by high collinearity with square footage (r = 0.75).

12. I consulted for the developer during one of the variance proceedings.

13. The Factory Place Lofts were progressively redeveloped. I only count those buildings that used the ARO.

14. Neither of these buildings added new parking during ARO conversion.