Table S1: Validity Checks of Validity of Wind Instruments

Table S1: Validity Checks of Validity of Wind Instruments

Supplementary Information

Table S1: Validity checks of validity of wind instruments

(1) / (2) / (3) / (4) / (5) / (6) / (7) / (8)
30km / 50km / 80km / 100km / 30km / 50km / 80km / 100km
VARIABLES / log(PM2.5) / log(PM2.5) / log(PM2.5) / log(PM2.5) / log(PM2.5) / log(PM2.5) / log(PM2.5) / log(PM2.5)
Fire hotspots from SE direction / 0.024*** / 0.024*** / 0.025*** / 0.027***
(0.000) / (0.000) / (0.000) / (0.000)
Fire hotspots from S direction / 0.005 / 0.001 / -0.004 / -0.004
(0.588) / (0.937) / (0.713) / (0.714)
Fire hotspots from N direction / 0.015* / 0.013* / 0.012* / 0.010*
(0.061) / (0.059) / (0.076) / (0.097)
Constant / 9.077*** / 9.078*** / 9.091*** / 9.097*** / 9.094*** / 9.097*** / 9.103*** / 9.108***
(0.000) / (0.000) / (0.000) / (0.000) / (0.000) / (0.000) / (0.000) / (0.000)
Observations / 440 / 440 / 440 / 440 / 440 / 440 / 440 / 440
F-test statistics / 25.54 / 24.93 / 19.94 / 13.46 / 3.767 / 3.852 / 3.360 / 2.934
R-squared / 0.087 / 0.083 / 0.072 / 0.061 / 0.027 / 0.022 / 0.017 / 0.013
p-value in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Standard errors clustered at district level

Table S2: Second stage of sorting model with different exclusion distance

(1) / (2) / (3)
30 km / 100 km / 50km (only districts with upwind neighbors)
VARIABLES / Adjusted utility / Adjusted utility
log(PM2.5) / -0.087* / -0.101 / -0.108***
(0.067) / (0.407) / (0.000)
Crime index / -0.000 / -0.000 / 0.006***
(0.255) / (0.699) / (0.000)
Health facilities / 0.008 / 0.007 / 0.007
(0.219) / (0.244) / (0.342)
Schools / 0.000 / 0.000 / 0.004
(0.977) / (0.977) / (0.791)
log(population density) / 0.041*** / 0.042*** / 0.043***
(0.000) / (0.000) / (0.000)
Observations / 174 / 174 / 161
1st stage F-statistics / 40.8 / 22.6 / 67.3
R-squared / 0.780 / 0.777 / 0.79
p-value in parentheses
*** p<0.01, ** p<0.05, * p<0.1

Table S3: Property hedonic OLS estimation using same data

VARIABLES / Log(rent)
Log PM2.5 / 0.083***
(0.000)
Size (m2) / 0.001***
(0.000)
Number of rooms / 0.034***
(0.000)
House type 1 (Single unit) a / -0.227
(0.154)
House type 2 (Duplex)a / -0.239
(0.133)
House type 3 (Multiple units) a / -0.244
(0.126)
House type 4 (On stilts)a / -0.278*
(0.082)
House type 5 (Apartments)a / -0.223
(0.218)
Floor type 1 (Ceramic/Marble) b / 0.163***
(0.000)
Floor type 2 (Tiles) b / 0.070***
(0.000)
Floor type 3 (Cement/Brick) b / 0.029**
(0.033)
Floor type 4 (Lumber) b / 0.012
(0.534)
Floor type 5 (Bamboo)b / -0.017
(0.696)

Table S4: Property hedonic estimation using same data (Log PM2.5 instrumented using source-, wind-, and distance- based instrument)

VARIABLES / Log(rent)
Log PM2.5 / -0.169***
(0.002)
Size (m2) / 0.000***
(0.000)
Number of rooms / 0.017***
(0.000)
House type 1 (Single unit) a / -0.127
(0.116)
House type 2 (Duplex)a / -0.137*
(0.091)
House type 3 (Multiple units) a / -0.135*
(0.095)
House type 4 (On stilts)a / -0.145*
(0.073)
House type 5 (Apartments)a / -0.133
(0.151)
Floor type 1 (Ceramic/Marble) b / 0.083***
(0.000)
Floor type 2 (Tiles) b / 0.033***
(0.000)
Floor type 3 (Cement/Brick) b / 0.012*
(0.071)
Floor type 4 (Lumber) b / 0.010
(0.314)
Floor type 5 (Bamboo)b / -0.004
(0.861)
Wall type 1 (Masonry) c / 0.067***
(0.000)
Wall type 2 (Lumber)c / 0.017**
(0.030)
Roof type 1 (Concrete) d / 0.081***
(0.000)
Roof type 2 (Wood) d / 0.067***
(0.000)
Roof type 3 (Metal) d / 0.039***
(0.002)
Roof type 4 (Shingles) d / 0.054***
(0.000)
Roof type 5 (Asbestos)d / 0.057***
(0.000)
Owner / 0.013***
(0.010)
Constant / 0.924***
(0.000)
Observations / 11,883
R-squared / 0.251
p-value in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Province FEs included
Standard errors clustered at district level
a Excluded category is multi-story units which are the most valuable housing type; b Excluded category is dirt; c Excluded category is bamboo/mat
d Excluded category is foliage/grass

Table S5: Results from sorting model without moving costs (Left table – First stage; Right table – Second stage)

1

Figure S1: Map showing which districts in the sample do not have upwind districts in the South or Southeast direction

Figure S2: Map of averaged PM2.5 (ug/m3) across choice-set districts from 2001 to 2006. Note that northeast Sumatra and southeast Kalimantan have the worst air quality. This is because they are a lot of forest fires in these places.

F Duke Research Dissertation Migration sorting GIS stuffs Figure 1 tif

Figure S3: Map showing distribution of households at district-level

C Users Jie Sheng AppData Local Microsoft Windows INetCache Content Word Households map png

1