1# RECORD 2ce98014780d8a025c9acf3e943a33fd71e8d5b62# TITLE: Forest pests and home values: The importance of accuracy in damage assessment and geocoding of properties3# YEAR: 2017.0 VENUE: nan DOI: 10.1016/j.jfe.2017.02.0024# AUTHORS: K. Moeltner; C. Blinn; T. Holmes5# PAGES: 10 CANDIDATE TABLE PAGES (1-based): [6, 8]6# ABSTRACT: We examine the impact of measurement errors in geocoding of property locations and in the assessment of Mountain Pine Beetle-induced tree damage within the proximity of a given residence on estimated losses in home values. For our sample of homes in the wildland-urban interface of the Colorado front range and using a novel matching estimator with Bayesian regression adjustment we find that both types of errors can lead to substantial biases in estimated losses. Our results confirm that the Forest Service's Aerial Detection Survey is generally too coarse to be informative for property valuation that depends on highly localized spatial data.789===== PAGE 1 (score 11, numeric-paren 0, table-words 0, var-words 11) =====1011Journal12 of13 Forest14 Economics15 2616 (2017)17 46–5518Contents19 lists20 available21 at22 ScienceDirect23Journal24 of25 Forest26 Economics27jo28 ur29 nal30 ho31 me p32 age:33 www.elsevier.com/locate/jfe34Forest35 pests36 and37 home38 values:39 The40 importance41 of42 accuracy43 in44 damage45assessment46 and47 geocoding48 of49 properties50Klaus51 Moeltner a,∗, Christine52 E. Blinn b, Thomas53 P.54 Holmes c55a Department56 of57 Agricultural58 and59 Applied60 Economics,61 Virginia62 Tech,63 Blacksburg,64 VA,65 USA66b Department67 of68 Forest69 Resources70 and71 Environmental72 Conservation,73 Virginia74 Tech,75 Blacksburg,76 VA,77 USA78c USDA79 Forest80 Service,81 Southern82 Research83 Station,84 Research85 Triangle86 Park,87 NC,88 USA89a90 r t91 i c l e 92i n f o93Article94 history:95Received96 697 December98 201699Accepted100 2101 February102 2017103Keywords:104Mountain105 Pine106 Beetle107Matching108 estimator109Measurement110 error111Aerial112 Detection113 Survey114a115 b116 s117 t r a118 c t119We examine120 the impact of measurement121 errors in geocoding122 of property123 locations124 and in the assessment125of126 Mountain127 Pine Beetle-induced128 tree damage129 within the proximity130 of a given131 residence132 on estimated133losses134 in home values. For our sample135 of homes136 in the wildland-urban137 interface138 of the Colorado139 front140range141 and using a novel142 matching143 estimator144 with Bayesian145 regression146 adjustment147 we find that both148 types149of errors can lead to substantial150 biases151 in estimated152 losses.153 Our results confirm154 that the Forest Service’s155Aerial156 Detection157 Survey is generally158 too coarse to be informative159 for property160 valuation161 that depends162 on163highly164 localized165 spatial166 data.167© 2017 Department168 of Forest Economics,169 Swedish170 University171 of Agricultural172 Sciences, Ume ˚a.173Published174 by Elsevier175 GmbH. All rights176 reserved.177Introduction178The179 accuracy180 of181 address-based182 geocoding183 (ABG),184 as185 used186 in187 many188geographic189 information190 system191 (GIS)192 software193 packages,194 has195 been196subject197 to198 scrutiny199 in200 recent201 years202 by203 spatially204 focused205 disciplines206such207 as208 geography209 and210 epidemiology.211 The212 emerging213 consensus214 in215these216 areas217 is218 that219 ABG220 can221 lead222 to223 measurable224 positional225 errors,226when227 compared228 to229 ground-truthing230 or231 interpretation232 via233 aerial234imagery.235 Furthermore,236 these237 errors238 tend239 to240 be241 more242 pronounced243in244 rural245 areas246 with247 lower248 population249 density.250For251 example,252 Cayo253 and254 Talbot255 (2003)256 examine257 3000258 addresses259of260 residential261 properties262 in263 upstate264 New265 York266 and267 find268 a269 median270error,271 in272 terms273 of274 the275 Euclidean276 distance277 between278 the279 true280 and281 the282presumed283 location284 of285 the286 actual287 housing288 structure,289 of290 38291 meters292(m),293 78294 m,295 and296 201297 m298 in299 urban,300 suburban,301 and302 rural303 settings,304 respec-305tively.306 Schootman307 et308 al.309 (2007), in310 turn,311 report312 a313 mean314 error315 of316 42317 m318and319 a320 median321 deviation322 of323 31324 m325 for326 299327 addresses328 in329 urban330 St.331 Louis,332Missouri.333 Zandbergen334 (2007), considering335 over336 100,000337 addresses338in339 urban/suburban340 Orange341 County,342 Florida,343 discovers344 a345 mean346 error347of348 66349 m,350 with351 a352 standard353 deviation354 of355 over356 430357 m.358As359 mentioned360 in361 Burra362 et363 al.364 (2002), and365 discussed366 in367 detail368 in369Zandbergen370 (2007)371 these372 positional373 errors374 can375 introduce376 serious377bias378 into379 an380 underlying381 statistical382 analysis383 if384 they385 are386 systematic387 in388nature,389 which390 is391 often392 the393 case394 in395 a396 given397 empirical398 application. For399∗Corresponding400 author.401example,402 in403 Zandbergen404 (2007)’s405 context406 ABG407 generally408 “moves”409residential410 locations411 closer412 to413 major414 traffic415 arteries416 than417 they418 really419are.420 This421 leads422 to423 an424 over-estimation425 of426 the427 number428 of429 children430exposed431 to432 traffic-related433 air434 pollution.435 Similar436 biases437 are438 reported439in440 Harada441 and442 Shimada443 (2006)444 and445 in446 Hay447 et448 al.449 (2009)450 in451 the452 context453of454 spatial455 densities456 and457 hot458 spots459 for460 crime461 locations.462Somewhat463 surprisingly464 given465 the466 recent467 proliferation468 in469 spa-470tially471 explicit472 property473 valuation474 studies,475 the476 potential477 inaccuracy478of479 ABG480 has481 largely482 been483 ignored484 in485 the486 applied487 economics488 literature.489As490 we illustrate491 in492 this493 study,494 pinpointing495 the496 accurate497 location498 of499a500 private501 residence502 can503 be504 very505 important506 if507 the508 effect509 of510 environ-511mental512 changes513 in514 the515 immediate516 vicinity517 of518 a519 home520 constitutes521 the522central523 focus524 of525 a526 given527 analysis.528In529 our530 application531 we532 consider533 tree534 mortality535 induced536 by537 the538Mountain539 Pine540 Beetle541 (MPB)542 within543 a544 close545 perimeter546 of547 a548 given549residence550 in551 the552 wildland-urban553 interface554 (WUI)555 of556 the557 Colorado558front559 range.560 The561 MPB562 (dentroctonus563 ponderosae) is564 a565 native566 forest567pest568 that569 has570 shown571 explosive572 population573 growth574 in575 recent576 years,577with578 annual579 forest580 destruction581 rates582 in583 North584 America585 and586 Canada587comparable588 to589 that590 of591 all592 wildfires593 combined594 (Carroll595 et596 al.,597 2004;598Bentz599 et600 al.,601 2010;602 Sims603 et604 al.,605 2010;606 Man,607 2012).608This609 poses610 a611 second612 spatially613 explicit614 risk615 of616 bias617 in618 estimated619 real620estate621 capitalization622 rates623 if624 tree625 mortality626 itself627 is628 assessed629 with630systematic631 error.632 Specifically,633 we examine634 the635 home-level636 accu-637racy638 of639 MPB640 damage641 as642 determined643 by644 the645 U.S.646 Forest647 Service’s648Aerial649 Detection650 Survey651 (ADS)652 relative653 to654 case-by-case655 photo-656interpretation657 by658 a659 remote660 sensing661 expert.662http://dx.doi.org/10.1016/j.jfe.2017.02.0026631104-6899/©664 2017665 Department666 of667 Forest668 Economics,669 Swedish670 University671 of672 Agricultural673 Sciences,674 Ume ˚a.675 Published676 by677 Elsevier678 GmbH.679 All680 rights681 reserved.682683684===== PAGE 6 (score 44, numeric-paren 0, table-words 12, var-words 9) =====685686K.687 Moeltner688 et689 al.690 / Journal691 of692 Forest693 Economics694 26695 (2017)696 46–55697 69851699Table 2700Sample701 statistics.702Treated703 observations704 705Eligible706 control707 observations708Variable709 710Mean711 / pct.712 713std714 715min716 717max718 719mean720 721std722 723min724 725max726Full727 sample728 (treated729 =730 304,731 eligible732 controls733 = 604)734Price735 (000s)736 737$355738 739$191740 741$77742 743$1,300744 745$460746 747$364748 749$55750 751$4,313752Age753 75430.67755 75616.61757 7580759 76092761 76230.87763 76423.53765 7660767 768151769sqft770 (00s)77118.80772 7738.46774 7753.68776 77766.24778 77921.29780 7819.57782 7834.71784 78571.47786Lot787 sqft.788 (000s)789 790114.96791 792163.19793 7940.87795 7961590.30797 798101.68799 800259.06801 8020.00803 8042132.70805Bedrooms806 8073.02808 8090.94810 8110812 8137814 8153.24816 8170.96818 8191820 8219822Bathrooms823 8242.37825 8260.97827 8280829 8306831 8322.75833 8341.18835 8360837 8389839Near840 fire841 84211.84%843 84423.84%845Sold846 1st847 quarter848 84921.71%850 85125.00%852Sold853 2nd854 quarter85531.25%856 85725.17%858Sold859 3rd860 quarter86129.93%862 86334.11%864Sold865 4th866 quarter867 8688.22%869 8708.61%871Golden872 (treated873 =874 161,875 eligible876 controls877 = 349)878Price879 (000s)880 881$413882 883$208884 885$78886 887$1,300888 889$469890 891$255892 893$81894 895$2,200896Age897 89830.92899 90016.61901 9023903 90489905 90628.93907 90821.56909 9100911 912136913sqft914 (00s)915 91620.77917 9189.11919 9205.76921 92266.24923 92423.10925 9269.56927 9284.71929 93071.47931Lot932 sqft.933 (000s)934 935100.53936 937154.02938 9390.87940 9411590.30942 94366.24944 945188.45946 9470.00948 9492132.70950Bedrooms951 9523.27953 9540.93955 9561957 9587959 9603.42961 9620.96963 9641965 9669967Bathrooms968 9692.68970 9710.92972 9731974 9756976 9773.04978 9791.16980 9811982 9839984Near985 fire9866.83%987 98822.64%989Sold990 1st991 quarter992 99319.88%994 99523.50%996Sold997 2nd998 quarter999 100032.92%1001 100227.22%1003Sold1004 3rd1005 quarter1006 100727.95%1008 100934.38%1010Sold1011 4th1012 quarter1013 10148.07%1015 10168.60%1017Boulder1018 (treated1019 =1020 20,1021 eligible1022 controls1023 =1024 86)1025Price1026 (000s)1027$3151028 1029$1081030 1031$1591032 1033$5891034 1035$7371036 1037$7051038 1039$1081040 1041$4,3131042Age1043 104440.301045 104615.191047 104891049 1050701051 105238.131053 105425.851055 105631057 10581311059sqft1060 (00s)1061 106215.551063 10645.221065 10667.381067 106824.571069 107021.111071 107211.611073 10745.861075 107661.271077Lot1078 sqft.1079 (000s)1080 1081154.821082 1083232.831084 108517.091086 1087916.501088 1089158.971090 1091393.201092 10930.711094 10951742.401096Bedrooms1097 10983.051099 11001.051101 110211103 110451105 11063.211107 11080.931109 111011111 111251113Bathrooms1114 11152.001116 11170.861118 111911120 112131122 11232.771124 11251.221126 112711128 112971130Near1131 fire1132 113345.00%1134 113566.28%1136Sold1137 1st1138 quarter113930.00%1140 114123.26%1142Sold1143 2nd1144 quarter1145 114615.00%1147 114827.91%1149Sold1150 3rd1151 quarter1152 115340.00%1154 115534.88%1156Sold1157 4th1158 quarter1159 116015.00%1161 11629.30%1163std:1164 standard1165 deviation;1166 min1167 (max):1168 minimum1169 (maximum);1170 pct.:1171 percent.1172than1173 treated1174 properties,1175 and1176 control1177 homes1178 in1179 Boulder1180 have1181 higher1182average1183 square1184 footage1185 than1186 treated1187 residences.1188 We use1189 the1190 basic1191housing1192 attributes1193 captured1194 in1195 the1196 table1197 in1198 our1199 balanced1200 matching1201procedure,1202 and1203 add1204 to1205 these1206 the1207 fire1208 and1209 sales1210 quarter1211 indicators1212 in1213the1214 auxiliary1215 regression1216 model.1217Table1218 31219 shows1220 sample1221 counts1222 before1223 and1224 after1225 matching,1226 as1227 well1228as1229 post-matching1230 counts1231 for1232 the1233 three1234 types1235 of1236 errors1237 that1238 can1239 occur1240using1241 one1242 of1243 the1244 flawed1245 interpretation1246 methods.1247 The1248 first1249 block1250 of1251rows1252 refers1253 to1254 the1255 full1256 sample.1257 The1258 first1259 two1260 rows,1261 which1262 give1263 the1264number1265 of1266 identified1267 treated1268 and1269 eligible1270 control1271 properties,1272 are1273repeated1274 from1275 Table1276 1. The1277 third1278 row1279 shows1280 the1281 remaining1282 control1283observations1284 after1285 matching,1286 that1287 is1288 the1289 set1290 of1291 controls1292 that1293 actually1294feed1295 into1296 the1297 econometric1298 analysis1299 and1300 the1301 estimated1302 treatment1303effect.1304 In1305 all1306 cases1307 the1308 number1309 of1310 these1311 matched1312 controls1313 is1314 smaller1315than1316 the1317 number1318 of1319 treated.1320 This1321 is1322 as1323 expected,1324 since1325 the1326 balancing1327and1328 matching1329 algorithms1330 allow1331 for1332 replacement1333 in1334 the1335 selection1336 of1337controls,1338 following1339 best1340 practices1341 in1342 the1343 matching1344 literature1345 (e.g.1346 Ho1347et1348 al.,1349 2007;1350 Abadie1351 and1352 Imbens,1353 2011).1354Rows1355 four1356 through1357 six1358 provide1359 a1360 detailed1361 look1362 at1363 omission1364and1365 swapping1366 errors.1367 As1368 is1369 evident1370 from1371 the1372 table,1373 all1374 three1375 non-1376benchmark1377 approaches1378 introduce1379 all1380 three1381 types1382 of1383 errors,1384 although1385error1386 counts1387 are1388 substantially1389 smaller1390 for1391 the1392 “PI,1393 ABG”1394 version1395compared1396 to1397 the1398 ADS-based1399 approaches.1400 Specifically,1401 the1402 latter1403drop1404 over1405 50%1406 of1407 actually1408 treated1409 from1410 the1411 analysis.1412 As1413 discussed1414above1415 this1416 not1417 only1418 hurts1419 efficiency1420 via1421 reduced1422 sample1423 sizes,1424 but1425can1426 introduce1427 selection1428 bias1429 if1430 this1431 attrition1432 is1433 non-random.1434 Post-1435matching1436 swapping1437 errors1438 are1439 also1440 considerable1441 for1442 the1443 ADS-based1444strategies,1445 with1446 approximately1447 17%1448 of1449 actually1450 treated1451 moved1452 to1453the1454 bin1455 of1456 matched1457 controls,1458 and1459 approximately1460 10%1461 of1462 controls1463interpreted1464 as1465 treated.1466Analogous1467 results1468 for1469 Golden1470 are1471 captured1472 in1473 the1474 second1475 block1476of1477 rows.1478 As1479 can1480 be1481 seen1482 from1483 the1484 table,1485 the1486 benchmark1487 remote1488sensing1489 approach1490 identifies1491 1611492 treated1493 observations,1494 which1495 leaves14963491497 eligible1498 control1499 homes.1500 After1501 balanced1502 matching1503 1131504 of1505 these1506are1507 actually1508 selected1509 as1510 matched1511 controls,1512 with1513 some1514 serving1515 as1516matches1517 for1518 multiple1519 treated.1520 In1521 comparison,1522 the1523 “PI,1524 ABG”1525 inter-1526pretation1527 completely1528 omits1529 five1530 treated1531 homes1532 from1533 the1534 analysis,1535erroneously1536 sorts1537 three1538 treated1539 homes1540 into1541 the1542 bin1543 of1544 matched1545 con-1546trols,1547 and1548 misinterprets1549 two1550 control1551 homes1552 as1553 treated.1554 As1555 for1556 the1557sample1558 at1559 large,1560 errors1561 introduced1562 by1563 the1564 ADS-based1565 methods1566 are1567more1568 staggering,1569 leading1570 to1571 the1572 omission1573 of1574 close1575 to1576 80%1577 of1578 actually1579treated1580 in1581 both1582 cases.1583 In1584 addition,1585 several1586 treated1587 homes1588 (121589 and159015,1591 respectively)1592 are1593 misplaced1594 into1595 the1596 matched1597 control1598 bin,1599 and1600 a1601comparable1602 number1603 of1604 control1605 cases1606 (111607 and1608 10,1609 respectively)1610 are1611falsely1612 used1613 as1614 treated1615 properties.1616Results1617 for1618 Boulder1619 are1620 given1621 in1622 the1623 bottom1624 block1625 of1626 rows1627 of1628Table1629 3.1630 The1631 benchmark1632 interpretation1633 produces1634 201635 treated1636 homes,1637861638 eligible1639 controls,1640 and1641 151642 matched1643 controls.1644 Working1645 with1646 flawed1647geocodes1648 in1649 the1650 “PI,1651 ABG”1652 interpretation1653 drops1654 four1655 treated1656 alto-1657gether,1658 and1659 misplaces1660 another1661 treated1662 and1663 three1664 controls.1665 While1666these1667 figures1668 may1669 seem1670 diminutive,1671 they1672 still1673 constitute1674 a1675 sizable1676share1677 of1678 all1679 available1680 observations1681 for1682 this1683 small1684 sample.1685 As1686 is1687 the1688case1689 for1690 the1691 full1692 sample1693 and1694 Golden,1695 the1696 ADS-based1697 approaches1698 drop1699even1700 more1701 treated1702 (eight1703 in1704 both1705 cases)1706 and1707 induce1708 some1709 erroneous1710swapping1711 of1712 actually1713 treated1714 (one1715 in1716 each1717 case)1718 and1719 actual1720 controls1721(101722 in1723 each1724 case).172517261727===== PAGE 7 (score 21, numeric-paren 0, table-words 5, var-words 11) =====17281729521730 1731K.1732 Moeltner1733 et1734 al.1735 / Journal1736 of1737 Forest1738 Economics1739 261740 (2017)1741 46–551742Table1743 31744Sorting1745 effects1746 based1747 on1748 remote1749 sensing1750 approaches.1751PI,1752 vis.1753 1754PI,1755 ABG1756 1757ADS,1758 vis.1759 1760ADS,1761 ABG1762Full1763 sample1764Treated1765 17663041767 17682931769 17701461771 17721391773Eligible1774 controls1775 17766041777 17786151779 17807621781 17827691783Matched1784 controls1785 17862061787 17882001789 17901201791 17921141793Unused1794 actually1795 treated1796 1797–1798 1799101800 18011731802 18031831804Actually1805 treated1806 used1807 as1808 matched1809 controls1810 1811–1812 1813111814 1815501816 1817451818Actual1819 controls1820 used1821 as1822 treated1823 1824–1825 1826101827 1828651829 1830631831Golden1832Treated1833 18341611835 18361551837 1838321839 1840281841Eligible1842 controls1843 18443491845 18463551847 18484781849 18504821851Matched1852 controls1853 18541131855 18561141857 1858301859 1860281861Unused1862 actually1863 treated1864 1865–1866 186751868 18691281870 18711281872Actually1873 treated1874 used1875 as1876 matched1877 controls1878 1879–1880 188131882 1883121884 1885151886Actual1887 controls1888 used1889 as1890 treated1891 1892–1893 189421895 1896111897 1898101899Boulder1900Treated1901 1902201903 1904181905 1906211907 1908211909Eligible1910 controls1911 1912861913 1914881915 1916851917 1918851919Matched1920 controls1921151922 1923151924 1925181926 1927181928Unused1929 actually1930 treated1931 1932–1933 193441935 193681937 193881939Actually1940 treated1941 used1942 as1943 matched1944 controls1945 1946–1947 194811949 195011951 195211953Actual1954 controls1955 used1956 as1957 treated1958 1959–1960 196131962 1963101964 1965101966PI:1967 damage1968 assessment1969 via1970 expert1971 photo-interpretation;1972 ADS:1973 damage1974 assessment1975 via1976 Aerial1977 Detection1978 Survey;1979 ABG:1980 address-based1981 geocoding1982 using1983 generic1984 GIS1985 software;1986vis.:1987 expert1988 geocoding1989 using1990 photo-imagery.1991As1992 discussed1993 previously,1994 swapping1995 effects,1996 by1997 themselves,1998should1999 unambiguously2000 bias2001 the2002 ATT2003 towards2004 zero.2005 In2006 theory,2007 the2008omission2009 effect,2010 by2011 itself,2012 can2013 be2014 bias-neutral2015 if2016 omissions2017 are2018 com-2019pletely2020 random.2021 However,2022 the2023 bias2024 can2025 go2026 towards2027 either2028 direction2029if2030 omitted2031 treated2032 homes2033 differ2034 from2035 the2036 sample2037 at2038 large2039 in2040 relevant2041dimensions.2042 In2043 our2044 case,2045 such2046 a2047 dimension2048 could2049 be2050 the2051 intensity2052 of2053infestation,2054 that2055 is2056 the2057 actual2058 count2059 or2060 proportion2061 of2062 affected2063 trees2064surrounding2065 a2066 property.2067 For2068 example,2069 ADS2070 fly-overs2071 may2072 focus2073 pri-2074marily2075 on2076 the2077 most2078 heavily2079 affected2080 areas2081 and2082 miss2083 low-intensity2084impacts.2085 This2086 would2087 bias2088 the2089 ATT2090 away2091 from2092 zero2093 and2094 lead2095 to2096 an2097over-estimation2098 of2099 losses.2100 Similarly,2101 if2102 homes2103 are2104 “moved”2105 from2106their2107 woody2108 surroundings2109 to2110 the2111 center2112 of2113 a2114 highway2115 by2116 ABG,2117 it2118is2119 feasible2120 that2121 the2122 erroneous2123 new2124 location2125 will2126 “pick2127 up”2128 some2129 tree2130damage2131 only2132 in2133 the2134 most2135 heavily2136 affected2137 areas.2138 Our2139 expert2140 assess-2141ment2142 did2143 not2144 capture2145 actual2146 counts2147 of2148 diseased2149 trees2150 due2151 to2152 budget2153and2154 time2155 limitations,2156 which2157 preempts2158 a2159 rigorous2160 quantitative2161 ver-2162ification2163 of2164 intensity-based2165 selection.2166Econometric2167 analysis2168As2169 mentioned2170 above,2171 we2172 match2173 homes2174 on2175 age2176 (in2177 years2178 since2179construction),2180 square2181 footage2182 (in2183 units2184 of2185 1002186 square2187 feet),2188 lot2189 size2190(in2191 1000s2192 of2193 square2194 feet),2195 number2196 of2197 bedrooms,2198 and2199 number2200 of2201 bath-2202rooms.2203 To2204 check2205 on2206 balance2207 within2208 the2209 GenMatch2210 algorithm,2211 we2212also2213 include2214 all2215 possible2216 squared2217 and2218 interaction2219 terms2220 for2221 these2222variables,2223 as2224 suggested2225 in2226 Diamond2227 and2228 Sekhon2229 (2013). Example2230balancing2231 results2232 for2233 the2234 benchmark2235 case2236 are2237 given2238 in2239 Appendix2240 B.2241The2242 original2243 set2244 of2245 matching2246 variables2247 are2248 also2249 included2250 in2251 the2252 aux-2253iliary2254 regression2255 that2256 underlies2257 the2258 regression-adjusted2259 ATT,2260 with2261the2262 addition2263 of2264 binary2265 indicators2266 for2267 the2268 occurrence2269 of2270 a2271 wildfire2272within2273 five2274 kilometers2275 and2276 the2277 past2278 five2279 years,2280 as2281 well2282 as2283 the2284 annual2285quarter2286 in2287 which2288 a2289 give2290 sale2291 occurred.2292 The2293 full-sample2294 regression2295also2296 contains2297 sub-regional2298 fixed2299 effects.2300As2301 described2302 above,2303 the2304 auxiliary2305 regression2306 is2307 estimated2308 in2309 a2310Bayesian2311 framework,2312 with2313 diffuse2314 normal2315 priors2316 for2317 all2318 coefficients2319(mean2320 =2321 0,2322 variance2323 =2324 109),2325 and2326 a2327 flat2328 inverse-gamma2329 prior2330 for2331 the2332variance2333 (shape2334 =2335 scale2336 =2337 1/2).2338 We let2339 our2340 Gibbs2341 Sampler2342 burn2343 in2344 for234550002346 draws2347 to2348 reduce2349 the2350 effect2351 of2352 starting2353 values,2354 and2355 retain2356 10,0002357additional2358 draws2359 of2360 parameters2361 for2362 inference.2363 Specifically,2364 we use2365each2366 draw2367 of2368 regression2369 coefficients2370 to2371 generate2372 an2373 estimate2374 of2375 pre-2376dicted2377 prices2378 for2379 a2380 given2381 treated2382 home2383 and2384 its2385 matched2386 control,2387that2388 is2389 the2390 terms2391 ˆ(xi)2392 and2393 ˆ(xj) in2394 Eq.2395 (3). These2396 predicted2397 prices2398then2399 feed2400 into2401 an2402 estimate2403 of2404 the2405 overall2406 ATT2407 via2408 Eq.2409 (2). Ultimately,2410therefore,2411 our2412 Bayesian2413 model2414 de2415 facto2416 produces2417 10,0002418 draws2419 to2420characterize2421 the2422 distribution2423 of2424 the2425 sought2426 ATT.2427Estimation2428 results2429 are2430 captured2431 in2432 Table2433 4. As2434 before,2435 the2436 table2437exhibits2438 three2439 blocks2440 of2441 rows2442 for2443 the2444 full2445 sample,2446 Golden,2447 and2448 Boul-2449der,2450 respectively.2451 The2452 columns2453 show2454 the2455 posterior2456 mean,2457 standard2458deviation,2459 proportion2460 of2461 posterior2462 distribution2463 to2464 the2465 right2466 of2467 zero2468(p2469 >2470 0),2471 and2472 the2473 lower2474 and2475 upper2476 bound2477 of2478 the2479 95%2480 highest2481 posterior2482density2483 interval2484 (HPDI).2485 The2486 p2487 >2488 02489 statistic2490 provides2491 an2492 at-a-glance2493assessment2494 if2495 the2496 ATT2497 is2498 unambiguously2499 negative2500 (p2501 >2502 02503 is2504 close2505 to2506zero),2507 positive2508 (p2509 >2510 02511 is2512 close2513 to2514 one),2515 or2516 indeterminate2517 in2518 sign2519 (p2520 >2521 02522is2523 close2524 to2525 0.5).2526 Thus,2527 it2528 can2529 be2530 used2531 as2532 an2533 informal2534 indication2535 of2536significance,2537 in2538 slight2539 abuse2540 of2541 the2542 classical2543 definition.2544 The2545 HPDI,2546 in2547turn,2548 gives2549 the2550 narrowest2551 possible2552 set2553 of2554 bounds2555 that2556 contain2557 95%2558 of2559the2560 posterior2561 distribution.2562 While2563 conceptually2564 and2565 computationally2566different,2567 it2568 can2569 be2570 interpreted2571 akin2572 to2573 a2574 classical2575 confidence2576 interval2577in2578 the2579 sense2580 that2581 there2582 is2583 a2584 95%2585 chance2586 that2587 the2588 (presumable2589 fixed2590but2591 unknown)2592 ATT2593 lies2594 between2595 these2596 bounds.2597 The2598 last2599 column2600 in2601the2602 table2603 gives2604 percentage2605 deviation2606 of2607 estimates2608 flowing2609 from2610 the2611flawed2612 remote2613 sensing2614 approaches2615 relative2616 to2617 those2618 generated2619 by2620the2621 benchmark.2622As2623 can2624 be2625 seen2626 from2627 the2628 table,2629 all2630 three2631 benchmark2632 ATTs2633 are2634clearly2635 negative,2636 though2637 the2638 HPDI2639 does2640 include2641 zero2642 in2643 the2644 case2645of2646 Boulder,2647 which2648 has2649 the2650 smallest2651 sample2652 size.2653 Generally,2654 though,2655this2656 supports2657 existing2658 research2659 that2660 finds2661 that2662 MPB-induced2663 tree2664damage2665 has2666 a2667 detrimental2668 effect2669 on2670 home2671 values2672 (Price2673 et2674 al.,26752010;2676 Cohen2677 et2678 al.,2679 2016).2680 For2681 the2682 sample2683 at2684 large,2685 the2686 average2687loss2688 is2689 $10,4712690 per2691 home,2692 which2693 amounts2694 to2695 approximately2696 2.5%2697of2698 the2699 sample2700 mean,2701 counting2702 all2703 treated2704 and2705 all2706 eligible2707 controls.2708In2709 contrast,2710 average2711 losses2712 in2713 Golden2714 amount2715 to2716 over2717 $23,000,2718 or27195.1%2720 of2721 the2722 sample2723 mean.2724 Boulder2725 posts2726 the2727 highest2728 losses2729 with2730a2731 posterior2732 mean2733 of2734 over2735 $124,000,2736 or2737 close2738 to2739 19%2740 of2741 the2742 sample2743mean.2744Turning2745 to2746 the2747 “PI,2748 ABG”2749 remote2750 sensing2751 version,2752 the2753 swapping2754effect2755 dominates2756 any2757 potential2758 selection2759 effects2760 through2761 omission2762for2763 Golden2764 and2765 Boulder,2766 and2767 thus2768 introduces2769 a2770 pronounced2771 positive2772deviation2773 from2774 the2775 benchmark2776 for2777 the2778 posterior2779 mean2780 for2781 both2782 areas2783(108%2784 and2785 35%,2786 respectively).2787 The2788 opposite2789 holds2790 for2791 the2792 full2793 sample,2794where2795 the2796 “PI,2797 ABG”2798 approach2799 produces2800 a2801 posterior2802 mean2803 that2804 shows2805a2806 negative2807 deviation2808 of2809 −79%2810 from2811 the2812 benchmark.2813 Thus,2814 in2815 this2816 case281728182819===== PAGE 8 (score 30, numeric-paren 12, table-words 1, var-words 10) =====28202821K.2822 Moeltner2823 et2824 al.2825 / Journal2826 of2827 Forest2828 Economics2829 262830 (2017)2831 46–552832 2833532834Table 42835Estimated2836 losses2837 in2838 home2839 values2840 ($000’s).2841Method2842 2843Mean2844 2845std2846 2847p2848 > 02849 2850Low2851 2852Up2853 2854Diff.2855 bench.2856 (%)2857Full2858 sample2859PI,2860 visual2861−10.4712862(2.926)2863 28640.0002865 2866−16.5172867 2868−5.1102869 2870–2871PI,2872 ABG2873 2874−18.7462875 2876(3.192)2877 28780.0002879 2880−24.8282881 2882−12.3872883 2884−79.03%2885ADS,2886 visual2887 2888−23.8352889 2890(2.299)2891 28920.0002893 2894−28.3122895 2896−19.3762897 2898−127.63%2899ADS,2900 ABG2901 290215.5062903 2904(55.595)2905 29060.6072907 2908−91.2652909 2910123.9712911 2912248.09%2913Golden2914PI,2915 visual2916 2917−23.0972918 2919(5.871)2920 29210.0002922 2923−34.7912924 2925−11.5622926 2927–2928PI,2929 ABG29301.9122931 2932(6.252)2933 29340.6212935 2936−10.563293713.8982938 2939108.28%2940ADS,2941 visual2942−14.4302943 2944(5.130)2945 29460.0042947 2948−24.8692949 2950−4.5732951 295237.52%2953ADS,2954 ABG2955 2956−91.3142957 2958(112.915)2959 29600.2042961 2962−311.2402963 2964136.4972965 2966−295.35%2967Boulder2968PI,2969 visual2970 2971−124.3382972 2973(81.445)2974 29750.0582976 2977−285.0752978 297940.8012980 2981–2982PI,2983 ABG2984 2985−81.5032986 2987(40.797)2988 29890.0242990 2991−164.9312992 2993−0.7922994 299534.45%2996ADS,2997 visual2998 2999−173.5823000 3001(26.180)3002 30030.0003004 3005−227.7003006 3007−124.0263008 3009−39.60%3010ADS,3011 ABG3012 3013−159.5673014 3015(22.131)3016 30170.0003018 3019−201.5983020 3021−114.2803022 3023−28.33%3024PI:3025 damage3026 assessment3027 via3028 expert3029 photo-interpretation;3030 ADS:3031 damage3032 assessment3033 via3034 Aerial3035 Detection3036 Survey;3037 ABG:3038 address-based3039 geocoding3040 using3041 generic3042 GIS3043 software;3044vis.:3045 expert3046 geocoding3047 photo-imagery;3048 p3049 >3050 0:3051 proportion3052 of3053 posterior3054 distribution3055 greater3056 than3057 zero;3058 low3059 (up):3060 lower3061 (upper)3062 bound3063 of3064 95%3065 highest3066 posterior3067 density3068 interval;3069diff.3070 bench:3071 %3072 difference3073 relative3074 to3075 benchmark3076 (PI,3077 visual).3078intensity3079 effects3080 via3081 involuntary3082 selection3083 of3084 retained3085 treated3086 appear3087to3088 outweigh3089 swapping3090 effects.3091The3092 “ADS,visual” approach3093 leads3094 to3095 negative3096 deviations3097 from3098 the3099benchmark3100 for3101 the3102 full3103 sample3104 (−128%)3105 and3106 Boulder3107 (−40%),3108 and3109 a3110positive3111 deviation3112 for3113 Golden3114 (38%).3115 The3116 “ADS,ABG”3117 version,3118 in3119 turn,3120yields3121 a3122 negative3123 deviation3124 for3125 Boulder3126 (−28%)3127 and3128 Golden3129 (−295%,3130although3131 with3132 20%3133 of3134 posterior3135 draws3136 in3137 the3138 positive3139 domain),3140 and3141a3142 positive3143 deviation3144 for3145 the3146 full3147 sample3148 (248%,3149 though3150 40%3151 of3152 the3153posterior3154 is3155 located3156 in3157 the3158 negative3159 domain).3160 Thus,3161 while3162 some3163 of3164these3165 estimates3166 exhibit3167 less-than-desirable3168 posterior3169 precision,3170 we3171obtain3172 a3173 mixed3174 picture3175 for3176 these3177 ADS-based3178 effects,3179 with3180 selection3181effects3182 potentially3183 dominating3184 over3185 swapping3186 effects.3187Our3188 benchmark3189 results3190 for3191 MPB-induced3192 losses3193 in3194 home3195 values3196are3197 within3198 the3199 ballpark3200 range3201 of3202 those3203 found3204 by3205 Cohen3206 et3207 al.3208 (2016)3209for3210 Boulder3211 county3212 and3213 its3214 northern3215 neighbor,3216 Larimer3217 county.3218Specifically,3219 Cohen3220 et3221 al.3222 (2016)3223 estimate3224 MPB-induced3225 losses3226 of322712.5%3228 for3229 the3230 typical3231 Boulder3232 county3233 home3234 located3235 near3236 host3237 trees3238(damaged3239 or3240 not)3241 compared3242 to3243 a3244 home3245 outside3246 of3247 the3248 host3249 tree3250 zone3251for3252 the3253 20113254 calendar3255 year.3256 In3257 contrast,3258 our3259 analysis3260 de3261 facto3262 com-3263pares3264 homes3265 with3266 damaged3267 host3268 trees3269 to3270 homes3271 with3272 healthy3273 host3274trees.3275 To3276 the3277 extent3278 that3279 homes3280 in3281 a3282 healthy3283 forest3284 surrounding3285 are3286typically3287 more3288 highly3289 priced3290 than3291 homes3292 outside3293 the3294 host3295 tree3296 zone3297in3298 that3299 area,3300 our3301 larger3302 loss3303 estimate3304 of3305 19%3306 for3307 Boulder3308 appears3309 rea-3310sonable.3311 It3312 also3313 comes3314 close3315 to3316 Cohen3317 et3318 al.3319 (2016)’s3320 estimate3321 of3322 a332322%3324 erosion3325 of3326 home3327 values3328 in3329 20113330 for3331 the3332 typical3333 host-tree-zone3334property3335 in3336 adjacent3337 Larimer3338 county.3339Conclusion3340We3341 take3342 a3343 closer3344 look3345 at3346 the3347 possible3348 estimation3349 biases3350 than3351 can3352occur3353 in3354 a3355 property3356 valuation3357 study3358 with3359 focus3360 on3361 localized3362 for-3363est3364 health3365 due3366 to3367 imprecise,3368 address-based3369 geocoding3370 and3371 overly3372coarse3373 tree3374 damage3375 assessment3376 via3377 the3378 Forest3379 Service’s3380 Aerial3381Detection3382 Survey.3383For3384 our3385 sample3386 of3387 over3388 9003389 single3390 family3391 residences3392 in3393 the3394Colorado3395 front3396 range,3397 we3398 find3399 that3400 both3401 geocoding3402 and3403 damage3404assessment3405 errors3406 can3407 introduce3408 substantial3409 biases3410 for3411 residential3412capitalization3413 rates.3414 In3415 theory,3416 either3417 type3418 and3419 combination3420 of3421 spa-3422tial3423 mis-information3424 can3425 introduce3426 selection3427 bias3428 by3429 retaining3430 only3431a3432 subset3433 of3434 treated3435 that3436 is3437 no3438 longer3439 representative3440 of3441 the3442 broader3443population,3444 as3445 well3446 as3447 swapping3448 bias3449 by3450 sorting3451 homes3452 into3453 the3454wrong3455 bin3456 in3457 an3458 econometric3459 matching3460 framework.3461 In3462 our3463 appli-3464cation,3465 we3466 find3467 that3468 ADS-based3469 approaches3470 to3471 damage3472 assessment3473are3474 susceptible3475 to3476 both3477 types3478 of3479 errors,3480 leading3481 to3482 substantial3483 over-3484or3485 under-estimation3486 of3487 lost3488 values.3489While3490 our3491 empirical3492 application3493 is3494 germane3495 to3496 forest3497 health,3498the3499 importance3500 of3501 accurate3502 geocoding3503 likely3504 carries3505 over3506 to3507 other3508property3509 valuation3510 contexts3511 that3512 rely3513 on3514 localized3515 distances3516 to3517environmental3518 (dis-)amenities3519 or3520 hot3521 spots,3522 such3523 as3524 polluted3525 river3526segments3527 (Cho3528 et3529 al.,3530 2011),3531 hazardous3532 waste3533 sites3534 (Guignet,3535 2013;3536Mastromonaco,3537 2014),3538 noise3539 sources3540 (Pope,3541 2008),3542 urban3543 recreation3544sites3545 (Phaneuf3546 et3547 al.,3548 2008),3549 or3550 air3551 and3552 water3553 quality3554 monitor-3555ing3556 stations3557 (Leggett3558 and3559 Bockstael,3560 2000;3561 Poor3562 et3563 al.,3564 2007;3565 Neill3566et3567 al.,3568 2007;3569 Anselin3570 and3571 Lozano-Gracia,3572 2008).3573 It3574 may3575 be3576 pru-3577dent3578 in3579 such3580 cases3581 to3582 obtain3583 a3584 small3585 sample3586 of3587 expertly3588 geocoded3589properties3590 and3591 decide3592 if3593 ABG-related3594 position3595 errors3596 are3597 small3598and/or3599 random3600 enough3601 to3602 be3603 ignorable3604 for3605 a3606 primary3607 valuation3608analysis.3609A3610 natural3611 extension3612 of3613 our3614 work3615 would3616 be3617 a3618 more3619 careful,3620imagery-based3621 measurement3622 of3623 the3624 intensity3625 of3626 infestation3627 sur-3628rounding3629 a3630 given3631 property,3632 in3633 terms3634 of3635 the3636 number3637 or3638 proportion3639of3640 diseased3641 trees3642 within3643 a3644 close3645 buffer.3646 This3647 would3648 allow3649 for3650 a3651more3652 rigorous3653 examination3654 and3655 quantification3656 of3657 our3658 presumed3659selection3660 effect3661 towards3662 more3663 heavily3664 impacted3665 areas3666 associated3667with3668 the3669 ADS3670 efforts.3671 It3672 would3673 also3674 pave3675 ways3676 towards3677 investi-3678gating3679 if3680 the3681 relationship3682 between3683 infested3684 trees3685 and3686 home3687 values3688may3689 be3690 nonlinear3691 or3692 exhibit3693 threshold3694 effects.3695 Another3696 interest-3697ing3698 extension3699 might3700 be3701 the3702 outsourcing3703 of3704 image3705 interpretation3706for3707 dead3708 trees3709 to3710 online3711 labor3712 markets,3713 as3714 recently3715 implemented3716by3717 Yu3718 et3719 al.3720 (2015)3721 in3722 the3723 context3724 of3725 cloud3726 detection3727 in3728 Landsat3729scenes.3730Acknowledgements3731This3732 work3733 was3734 supported3735 by3736 the3737 United3738 States3739 Forest3740 Service,3741Southern3742 Research3743 Station3744 [cooperative3745 agreement3746 number3747 12-JV-374811330143-106].3749Appendix3750 A.3751 Econometric3752 details3753The3754 GenMatch3755 balancing3756 algorithm3757 returns3758 the3759 original3760 sample3761of3762 treated,3763 i =3764 13765 . .3766 . nT, and3767 a3768 matched3769 sample3770 of3771 unique3772 controls,3773j3774 =3775 13776 .3777 . . nM, with3778 possible3779 replications,3780 such3781 that3782 nM ≤3783 nT. Each3784matched3785 control3786 receives3787 a3788 weight3789 Kj that3790 indicates3791 for3792 how3793 many3794treated3795 it3796 serves3797 as3798 a3799 match.3800 For3801 most3802 matched3803 observation,3804 Kj will3805be3806 unity.380738083809===== PAGE 9 (score 17, numeric-paren 0, table-words 3, var-words 8) =====38103811543812 3813K.3814 Moeltner3815 et3816 al.3817 / Journal3818 of3819 Forest3820 Economics3821 263822 (2017)3823 46–553824Next,3825 we3826 build3827 the3828 auxiliary3829 regression3830 model,3831 using3832 only3833 the3834matched3835 controls.3836 The3837 models3838 follows3839 a3840 basic3841 weighted3842 OLS3843 speci-3844fication,3845 that3846 is:3847˜yj = ˜x′3848jˇ3849 +3850 j, 3851with3852˜yj = 3853Kjyj,3854˜xj = 3855Kjxj3856and3857j∼n38580,3859 23860(A.1)3861where3862 yj is3863 the3864 observed3865 sales3866 price3867 of3868 home3869 j,3870 xj is3871 a3872 vector3873 of3874 home3875characteristics,3876 wildfire3877 indicator,3878 and3879 quarter-of-sales3880 indicators3881as3882 described3883 in3884 the3885 main3886 text.3887 The3888 error3889 term3890 j follows3891 the3892 usual3893normal3894 distribution3895 with3896 zero3897 mean3898 and3899 common3900 variance3901 2.3902For3903 the3904 entire3905 sample3906 of3907 matched3908 controls,3909 the3910 likelihood3911 func-3912tion3913 is3914 thus3915 given3916 as:3917p3918˜y|ˇ,3919 2, ˜X3920= (2)−nM /239212−nM /23922exp39233924−139252239263927˜y −˜Xˇ′ 3928˜y −˜Xˇ3929,3930 3931(A.2)3932where ˜y3933 denotes3934 the3935 entire3936 vector3937 of3938 weighted3939 home3940 prices3941 for3942 the3943matched3944 controls,3945 and ˜X is3946 the3947 weighted3948 sample3949 matrix3950 of3951 explana-3952tory3953 variables.3954Next,3955 we3956 employ3957 the3958 standard3959 normal3960 prior3961 for3962 ˇ3963 and3964 inverse-3965gamma3966 prior3967 for3968 2, i.e.:3969ˇ∼n (0,3970 V0) ,3971 3972or3973p3974ˇ3975= (2)−k/2|V0|−1/2 exp39763977−13978239793980ˇ3981 −3982 03983′(V0)−1 3984ˇ3985 −3986 0398739882∼ig (0, 0) ,3989 3990or3991p399223993=39940399503996 (0)399739982−(0+1) exp39994000−04001240024003, 4004with4005E 400624007=4008040090 −4010 1 ,4011 4012V 401324014=40152401604017(0 −4018 1)2 (0 −4019 2)4020,4021(A.3)4022where4023 k4024 is4025 the4026 dimension4027 of4028 ˇ. As4029 is4030 evident4031 from4032 (A.3)4033 we para-4034meterize4035 the4036 inverse-gamma4037 with4038 shape4039 0 and4040 scale4041 0, following4042Gelman4043 et4044 al.4045 (2004).4046Combining4047 the4048 likelihood4049 with4050 these4051 priors4052 and4053 dropping4054 all4055nuisance4056 terms4057 yields4058 the4059 posterior4060 kernel4061p4062ˇ,4063 2|˜y, ˜X4064∝40652 −nM−20−2406624067exp40684069−1407024071 14072240734074˜y4075 −˜Xˇ′ 4076˜y4077 −˜Xˇ4078+4079ˇ4080 −4081 04082′(V0)−1 4083ˇ4084 −4085 040864087(A.4)4088Following4089 standard4090 procedures,4091 we4092 break4093 this4094 posterior4095 kernel4096 into4097conditional4098 densities4099 for4100 ˇ4101 and4102 2,4103 given4104 as:4105p4106ˇ, |2,4107 y, ˜X4108∼n (1,4109 V1) , 4110with4111V1 =41124113V−1411404115+ 141162 ˜X4117′ ˜X4118−14119,4120 4121and41221 =4123 V141244125V−141260 0 + 141272 ˜X4128′˜y41294130(A.5)4131and41322|ˇ, ˜y, ˜X∼ig (1,4133 1) ,4134 4135with41361 = 20 +4137 n413824139,4140 4141and41421 =4143 0 + 14144241454146˜y4147 −˜Xˇ′ 4148˜y4149 −˜Xˇ4150(A.6)4151Our4152 Gibbs4153 Sampler4154 then4155 draws4156 sequentially4157 and4158 repeatedly4159 from4160these4161 two4162 conditionals.4163 as4164 discussed4165 inter4166 alia4167 in4168 Koop4169 (2003)4170 and4171Table4172 B.14173Balancing4174 results4175 for4176 benchmark4177 case.4178Mean4179 4180max4181Variable4182 4183Before4184 4185After4186 4187Before4188 4189After4190Full4191 sample4192Age4193 41946.074195 41960.944197 419859.004199 42005.004201sqft004202 42032.564204 42050.914206 42076.154208 42098.964210lot0004211 421258.864213 421420.634215 4216542.404217 4218152.104219Golden4220Age4221 42225.794223 42241.684225 422647.004227 42285.004229sqft004230 42312.684232 42331.014234 42356.154236 42375.234238lot0004239 424067.404241 424219.104243 4244542.404245 4246114.754247Boulder4248Age4249 425011.434251 42526.154253 425461.004255 425617.004257sqft004258 42596.604260 42610.904262 426336.704264 42652.384266lot0004267 4268120.324269 427054.124271 4272825.904273 4274387.684275mean4276 (max)4277 = mean4278 (maximum)4279 quantile4280 deviation4281 (5%4282 steps);4283 sqft004284 =4285 square4286footage4287 in4288 units4289 of4290 100;4291 lot0004292 = lot4293 size4294 in4295 units4296 of4297 10004298 square4299 feet.4300Koop4301 et4302 al.4303 (2007), after4304 a4305 sufficient4306 number4307 of4308 iterations4309 the4310 sam-4311pler4312 will4313 converge4314 to4315 the4316 desired4317 joint4318 posterior4319 given4320 in4321 (A.4).4322As4323 described4324 in4325 the4326 main4327 text,4328 for4329 each4330 retained4331 draw4332 of4333 ˇr from4334the4335 rth4336 round4337 of4338 the4339 sampler,4340 we4341 compute4342 the4343 predictive4344 constructs4345ˆ4346r(˜xi)4347 = ˜x′4348iˇr and4349 ˆ4350r(˜xj)4351 = ˜x′4352jˇr. These4353 then4354 feed4355 into4356 the4357 rth draw4358of4359 the4360 ATT4361 via4362 Eqs.4363 (2)4364 and4365 (3)4366 of4367 the4368 main4369 text.4370Appendix4371 B.4372 Balancing4373 results4374To4375 check4376 on4377 improvements4378 in4379 distributional4380 similarity4381 between4382treated4383 and4384 controls4385 before4386 and4387 after4388 balancing4389 the4390 sample,4391 we4392compare4393 the4394 difference4395 in4396 5%-incremented4397 quantiles4398 between4399treated4400 and4401 original4402 and4403 matched4404 controls,4405 respectively.4406 The4407 table4408below4409 shows4410 these4411 diagnostics4412 for4413 the4414 benchmark4415 remote4416 sensing4417approaches4418 for4419 each4420 sample4421 used4422 in4423 our4424 main4425 analysis,4426 and4427 for4428 the4429three4430 continuous4431 variables4432 used4433 in4434 matching.4435 Clearly,4436 both4437 mean4438and4439 maximum4440 quantile4441 deviations4442 decrease4443 markedly4444 after4445 balanc-4446ing4447 for4448 all4449 variables4450 and4451 applications.4452The4453 only4454 exception4455 is4456 a4457 slight4458 increase4459 in4460 the4461 maximum4462 deviation4463of4464 square4465 footage4466 (from4467 6.154468 to4469 8.25)4470 for4471 the4472 full4473 sample.4474 How-4475ever,4476 this4477 comes4478 at4479 the4480 gain4481 of4482 dramatic4483 reductions4484 in4485 maximum4486deviations4487 for4488 age4489 and4490 lot4491 size.4492Table4493 B.14494References4495Abadie,4496 A.,4497 Imbens,4498 G.,4499 2006.4500 Large-sample4501 properties4502 of4503 matching4504 estimators4505 for4506average4507 treatment4508 effects.4509 Econometrica4510 74,4511 235–267.4512Abadie,4513 A.,4514 Imbens,4515 G.,4516 2011.4517 Bias-corrected4518 matching4519 estimators4520 for4521 average4522treatment4523 effects.4524 J.4525 Bus.4526 Econ.4527 Stat.4528 29,4529 1–11.4530Abadie,4531 A.,4532 Drucker,4533 D.,4534 Herr,4535 J.,4536 Imbens,4537 G.,4538 20024539 October.4540 Simple4541 and4542bias-corrected4543 matching4544 estimators4545 for4546 average4547 treatment4548 effects,4549 nBER4550Technical4551 Working4552 paper4553 283.4554Abbott,4555 J.,4556 Klaiber,4557 H.,4558 2013.4559 The4560 value4561 of4562 water4563 as4564 an4565 urban4566 club4567 good:4568 a4569 matching4570approach4571 to4572 community-provided4573 lakes.4574 J.4575 Environ.4576 Econ.4577 Manage.4578 65,4579 208–224.4580Anselin,4581 L.,4582 Lozano-Gracia,4583 N.,4584 2008.4585 Errors4586 in4587 variables4588 and4589 spatial4590 effects4591 in4592hedonic4593 house4594 price4595 models4596 of4597 ambient4598 air4599 quality.4600 Empir.4601 Econ.4602 34,4603 5–34.4604Backsen,4605 J.,4606 Howell,4607 B.,4608 2013.4609 Comparing4610 aerial4611 detection4612 and4613 photo4614 interpretation4615for conducting4616 forest4617 health4618 surveys.4619 West.4620 J.4621 Appl.4622 For.4623 28,4624 3–8.4625Bentz,4626 B.,4627 Jacques,4628 R.,4629 Fettig,4630 C.,4631 Hansen,4632 E.,4633 Hayes,4634 J.,4635 Hicke,4636 J.,4637 Kelsey,4638 R.,4639 Negrn,4640 J.,4641Seybold,4642 S.,4643 2010.4644 Climate4645 change4646 and4647 bark4648 beetles4649 of4650 the4651 Western4652 United4653States4654 and4655 Canada:4656 direct4657 and4658 indirect4659 effects.4660 BioScience4661 60,4662 602–613.4663Burra,4664 T.,4665 Jerrett,4666 M.,4667 B.R.T.,4668 Anderson,4669 M.,4670 2002.4671 Conceptual4672 and4673 practical4674 issues4675 in4676the4677 detection4678 of4679 local4680 disease4681 clusters:4682 a4683 study4684 of4685 mortality4686 in4687 Hamilton,4688Ontario.4689 Can.4690 Geogr.4691 46,4692 160–171.4693Carroll,4694 A.,4695 Taylor,4696 S.,4697 Régnière,4698 J.,4699 Safranyik,4700 L.,4701 2004.4702 Effects4703 of4704 climate4705 change4706 on4707range4708 expansion4709 by4710 the4711 mountain4712 pine4713 beetle4714 in4715 British4716 Columbia.4717 In:4718 Shore,4719 T.,4720Brooks,4721 J.,4722 Stone,4723 J.4724 (Eds.),4725 Mountain4726 Pine4727 Beetle4728 Symposium:4729 Challenges4730 and4731Solutions.4732 October4733 30–31,4734 2003,4735 Kelowna,4736 British4737 Columbia,4738 Natural4739 Resources4740Canada,4741 Canadian4742 Forest4743 Service,4744 Pacific4745 Forestry4746 Center,4747 Information4748 Report4749BC-X-399,4750 pp.4751 223–232.4752Cayo,4753 M.,4754 Talbot,4755 T.,4756 2003.4757 Positional4758 error4759 in4760 automated4761 geocoding4762 of4763 residential4764addresses.4765 Int.4766 J. Health4767 Geogr.4768 2,4769 1–12.4770