Robustness Checks
| Variable | Dependent Variable: Log Sales Price | |||||
|---|---|---|---|---|---|---|
| Repeat Sales | Include Time Invariant BDI total | Tract × Year FE | Presample Sales Price as Predictor | Use Distance Squared in BDI Calculation | Use Root Distance in BDI Calculation | |
| (1) | (2) | (3) | (4) | (5) | (6) | |
| BDI | –0.0134*** (0.0031) | –0.0103*** (0.0023) | –0.0085** (0.0035) | –0.0187*** (0.0031) | –0.0022*** (0.0004) | –0.0190*** (0.0023) |
| BDI total | –0.0004 (0.0021) | |||||
| Housing and neighborhood characteristics | Yes | Yes | Yes | Yes | Yes | Yes |
| Year-month FE | Yes | Yes | Yes | Yes | Yes | Yes |
| Census block group FE | No | Yes | Yes | Yes | Yes | Yes |
| Town × Year FE | Yes | Yes | No | Yes | Yes | Yes |
| Presample mean price × Year-quarter FE | Yes | Yes | No | No | Yes | Yes |
| Property FE | Yes | No | No | No | No | No |
| Tract × Year FE | No | No | Yes | No | No | No |
| Presample sales price × Year-quarter FE | No | No | No | Yes | No | No |
| Observations | 17,506 | 43,787 | 35,711 | 9,836 | 43,742 | 43,787 |
| Adjusted R-squared | 0.859 | 0.723 | 0.717 | 0.789 | 0.723 | 0.724 |
Note: The Column 1 sample includes only properties that transact more than once, and the model includes property-specific fixed effects. Column 2 includes as an independent variable BDI total, which is the sum of the inverse distances to all brownfields, remediated and unremediated, within 2 km of a housing transaction. Column 3 adds census tract by year fixed effects to the Table 2 Column 5 specification and removes year-town fixed effects and presample neighborhood mean price interacted with year-quarter fixed effects. Column 3 includes only tracts with more than 200 observations in the sample. The Column 4 sample includes only properties that transacted January 2000–September 2003 as well as the sample time frame October 2003–March 2013; then the presample sales price (inflation adjusted and logged) is interacted with year-quarter fixed effects. For Columns 5 and 6, the definition of BDI is altered. Modifying equation [1] to have an exponent on inverse distance,
, Column 5 sets p = 2 and Column 6 sets p = 0.5, as compared to the main results that set p = 1. Squaring inverse distance produces a distribution of BDI with a long right tail; Column 5 drops transactions in the top 0.1% of the BDI distribution to reduce influence of outliers.↵**, *** Significance at 5% and 1%, respectively.
