* stata_codebook.do - attach long-form notes to the .dta files (run once in Stata).
* Generated by build_data_dictionary.py - do not edit by hand.

* ---- bridge_nightlights.dta ----
use "bridge_nightlights.dta", clear
label data "Nighttime lights panel"
note _dta: Luminosity mn is bottom-coded at 1.0, so most rural upazilas sit near the floor; the tutorial uses ln(mn + 1). pop91 is missing for 4 upazilas and rainsd is zero for 28, which is why the estimation sample is 247 rather than 252 upazilas.
note geocode: Bangladesh upazila (subdistrict) identifier, harmonised across census rounds.. Construction: Master administrative code. Upazilas that split or merged between rounds were matched by superimposing digital maps and applying area weights, so the panel is comparable over time.. Units: code. Source: Blankespoor et al. (2021) replication package
note year: Integer period index, NOT a calendar year. Nightlights: 1=1992-94, 2=1995-97, 3=1998-2000, 4=2001-04, 5=2005-07, 6=2008-10, 7=2011-13. Census: 1=1991, 2=2001, 3=2011. Yield: 1=1988-91, 2=1992-94, 3=1995-97, 4=1998-2000, 5=2001-04, 6=2005-07, 7=2008-10, 8=2011-13. DHS: 1=1994, 2=1996, 3=1997, 4=2003, 5=2007, 6=2011, 7=2013.. Construction: Annual satellite and yield series are averaged into three-year blocks to smooth transitory shocks. The bridge opened in June 1998, inside nightlights period 3 and yield period 4.. Units: index. Source: Blankespoor et al. (2021) replication package
note mn: Mean DMSP-OLS digital-number luminosity over the upazila, averaged across the three-year block.. Construction: 1 km satellite pixels averaged to the upazila boundary, then across years. BOTTOM-CODED AT 1.0 - the minimum in the data is exactly 1.0, so most rural upazilas sit at the floor. The tutorial uses ln(mn + 1) for this reason.. Units: digital number (0-63 scale). Source: DMSP-OLS, NOAA Earth Observation Group
note div: Administrative division: 1 Barisal, 2 Chittagong, 3 Dhaka, 4 Khulna, 5 Rajshahi.. Construction: Master administrative code.. Units: code. Source: Blankespoor et al. (2021) replication package
note dist: District identifier. In the yield file this is the unit of observation (former district); elsewhere it is an upper-level grouping.. Construction: Master administrative code.. Units: code. Source: Blankespoor et al. (2021) replication package
note pop91: Total population recorded in the 1991 census - a baseline characteristic fixed before the bridge.. Construction: Used in the propensity model and, interacted with the period index, as a control that lets initially-large units follow a different trajectory.. Units: persons. Source: IPUMS International
note rainm: Average monthly precipitation over the period.. Construction: Extracted at the upazila level from the NOAA PREC/L gridded reconstruction. Enters the regressions as ln(rainm + 1) for nightlights and ln(rainm) elsewhere.. Units: cm per month. Source: NOAA PREC/L
note rainsd: Standard deviation of monthly precipitation over the period - a control for weather risk, which matters directly for agricultural yield.. Construction: Extracted at the upazila level from NOAA PREC/L. Zero values become missing under the log transform, which is what removes 28 nightlights upazilas from the estimation sample.. Units: cm per month. Source: NOAA PREC/L
note jamuna_m: Crow-fly distance from the unit centroid to the Jamuna bridge foot.. Construction: Computed in ArcGIS from upazila-level shapefiles for Bangladesh.. Units: metres. Source: Blankespoor et al. (2021), from Bangladesh shapefiles
note padma_m: Crow-fly distance from the unit centroid to the planned Padma bridge site.. Construction: Computed in ArcGIS from upazila-level shapefiles for Bangladesh.. Units: metres. Source: Blankespoor et al. (2021), from Bangladesh shapefiles
note treat: 1 if the unit lies in the Jamuna hinterland, the region the bridge connected; 0 otherwise.. Construction: Assigned by geography: which side of which river the unit sits on.. Units: 0/1. Source: Blankespoor et al. (2021) replication package
note smp1: Non-missing for the Jamuna and Padma hinterlands; MISSING for the Dhaka-Chittagong core. Filtering on non-missing is how the core is excluded from every regression.. Construction: Missing wherever treatd == 1. In the two DHS files there is no core, so this variable is numerically identical to treat.. Units: 0/1 or missing. Source: Blankespoor et al. (2021) replication package
note treatd: 1 for the Dhaka-Chittagong core and adjacent areas. Used only for descriptive comparison in the paper's Appendix Table AT.1, never as a DiD comparison group.. Construction: The authors dropped core estimates from the published design after a referee pointed out that the Padma hinterland is not a valid comparison for the core.. Units: 0/1. Source: Blankespoor et al. (2021) replication package
save "bridge_nightlights.dta", replace

* ---- bridge_employment.dta ----
use "bridge_employment.dta", clear
label data "Population census panel"
note _dta: Only 1991 is pre-bridge, so no pre-trend test is possible here. 24 rows have zero recorded rainfall and drop when the log controls are built, taking the estimation sample from 248 to 246 upazilas.
note geocode: Bangladesh upazila (subdistrict) identifier, harmonised across census rounds.. Construction: Master administrative code. Upazilas that split or merged between rounds were matched by superimposing digital maps and applying area weights, so the panel is comparable over time.. Units: code. Source: Blankespoor et al. (2021) replication package
note year: Integer period index, NOT a calendar year. Nightlights: 1=1992-94, 2=1995-97, 3=1998-2000, 4=2001-04, 5=2005-07, 6=2008-10, 7=2011-13. Census: 1=1991, 2=2001, 3=2011. Yield: 1=1988-91, 2=1992-94, 3=1995-97, 4=1998-2000, 5=2001-04, 6=2005-07, 7=2008-10, 8=2011-13. DHS: 1=1994, 2=1996, 3=1997, 4=2003, 5=2007, 6=2011, 7=2013.. Construction: Annual satellite and yield series are averaged into three-year blocks to smooth transitory shocks. The bridge opened in June 1998, inside nightlights period 3 and yield period 4.. Units: index. Source: Blankespoor et al. (2021) replication package
note div: Administrative division: 1 Barisal, 2 Chittagong, 3 Dhaka, 4 Khulna, 5 Rajshahi.. Construction: Master administrative code.. Units: code. Source: Blankespoor et al. (2021) replication package
note dist: District identifier. In the yield file this is the unit of observation (former district); elsewhere it is an upper-level grouping.. Construction: Master administrative code.. Units: code. Source: Blankespoor et al. (2021) replication package
note pop: Total population recorded in the census round.. Construction: Individual census records collapsed to the upazila.. Units: persons. Source: IPUMS International
note emp: Total employed population - the denominator of the three sectoral shares.. Construction: Individual census records collapsed to the upazila.. Units: persons. Source: IPUMS International
note pop_agr: Number employed in agriculture. Divided by emp it gives the agriculture share, which falls 1.3 percentage points on average and 5.7 points in the farthest distance band.. Construction: Individual census records collapsed to the upazila.. Units: persons. Source: IPUMS International
note pop_ind: Number employed in industry and manufacturing. The share falls 1.2 percentage points in the long run - roughly a third of a 2.8 percent baseline sector.. Construction: Individual census records collapsed to the upazila. Logged as ln(pop_ind + 1) in the original code because some upazilas have very few industrial workers.. Units: persons. Source: IPUMS International
note pop_serv: Number employed in services - trading, transport and processing, the activities a region takes on when it starts shipping output. The share rises 2.4 percentage points in the long run.. Construction: Individual census records collapsed to the upazila.. Units: persons. Source: IPUMS International
note density: Population per square kilometre. This is the outcome that discriminates between the backwash and comparative-advantage theories: backwash requires it to fall, and it rises 5.9 percent in the long run.. Construction: Census population divided by upazila land area. Enters as ln(density).. Units: persons per sq km. Source: IPUMS International
note pop91: Total population recorded in the 1991 census - a baseline characteristic fixed before the bridge.. Construction: Used in the propensity model and, interacted with the period index, as a control that lets initially-large units follow a different trajectory.. Units: persons. Source: IPUMS International
note rainm: Average monthly precipitation over the period.. Construction: Extracted at the upazila level from the NOAA PREC/L gridded reconstruction. Enters the regressions as ln(rainm + 1) for nightlights and ln(rainm) elsewhere.. Units: cm per month. Source: NOAA PREC/L
note rainsd: Standard deviation of monthly precipitation over the period - a control for weather risk, which matters directly for agricultural yield.. Construction: Extracted at the upazila level from NOAA PREC/L. Zero values become missing under the log transform, which is what removes 28 nightlights upazilas from the estimation sample.. Units: cm per month. Source: NOAA PREC/L
note surban: Share of the upazila population classified as urban.. Construction: Census classification. Used descriptively rather than as a regression outcome.. Units: share (0-1). Source: IPUMS International
note prop_elec: Proportion of households with electricity.. Construction: Census household records collapsed to the upazila. Used in the paper's Appendix Table AT.1 to establish that the Jamuna hinterland lagged the core before the bridge.. Units: share (0-1). Source: IPUMS International
note jamuna_m: Crow-fly distance from the unit centroid to the Jamuna bridge foot.. Construction: Computed in ArcGIS from upazila-level shapefiles for Bangladesh.. Units: metres. Source: Blankespoor et al. (2021), from Bangladesh shapefiles
note padma_m: Crow-fly distance from the unit centroid to the planned Padma bridge site.. Construction: Computed in ArcGIS from upazila-level shapefiles for Bangladesh.. Units: metres. Source: Blankespoor et al. (2021), from Bangladesh shapefiles
note treat: 1 if the unit lies in the Jamuna hinterland, the region the bridge connected; 0 otherwise.. Construction: Assigned by geography: which side of which river the unit sits on.. Units: 0/1. Source: Blankespoor et al. (2021) replication package
note smp1: Non-missing for the Jamuna and Padma hinterlands; MISSING for the Dhaka-Chittagong core. Filtering on non-missing is how the core is excluded from every regression.. Construction: Missing wherever treatd == 1. In the two DHS files there is no core, so this variable is numerically identical to treat.. Units: 0/1 or missing. Source: Blankespoor et al. (2021) replication package
note treatd: 1 for the Dhaka-Chittagong core and adjacent areas. Used only for descriptive comparison in the paper's Appendix Table AT.1, never as a DiD comparison group.. Construction: The authors dropped core estimates from the published design after a referee pointed out that the Padma hinterland is not a valid comparison for the core.. Units: 0/1. Source: Blankespoor et al. (2021) replication package
save "bridge_employment.dta", replace

* ---- bridge_yield.dta ----
use "bridge_yield.dta", clear
label data "Boro rice yield panel"
note _dta: Agricultural statistics in Bangladesh are published at the old district level, so this entire result rests on 9 to 11 clusters. Two districts (Faridpur, Patuakhali) receive negative Oaxaca-Blinder weights and drop from the weighted specifications.
note former_dis: Old-district identifier used by the agricultural statistics, which were never re-issued on the modern district boundaries.. Construction: Value-labelled in the original .dta: 2 Barisal, 3 Bogra, 5 Comilla, 6 Dhaka, 7 Dinajpur, 8 Faridpur, 9 Jamalpur, 10 Jessore, 12 Khulna, 13 Kishoreganj, 14 Kushtia, 15 Mymensingh, 17 Pabna, 18 Patuakhali, 19 Rajshahi, 21 Rangpur, 23 Tangail.. Units: code. Source: Bangladesh Bureau of Statistics
note dist: District identifier. In the yield file this is the unit of observation (former district); elsewhere it is an upper-level grouping.. Construction: Master administrative code.. Units: code. Source: Blankespoor et al. (2021) replication package
note year: Integer period index, NOT a calendar year. Nightlights: 1=1992-94, 2=1995-97, 3=1998-2000, 4=2001-04, 5=2005-07, 6=2008-10, 7=2011-13. Census: 1=1991, 2=2001, 3=2011. Yield: 1=1988-91, 2=1992-94, 3=1995-97, 4=1998-2000, 5=2001-04, 6=2005-07, 7=2008-10, 8=2011-13. DHS: 1=1994, 2=1996, 3=1997, 4=2003, 5=2007, 6=2011, 7=2013.. Construction: Annual satellite and yield series are averaged into three-year blocks to smooth transitory shocks. The bridge opened in June 1998, inside nightlights period 3 and yield period 4.. Units: index. Source: Blankespoor et al. (2021) replication package
note div: Administrative division: 1 Barisal, 2 Chittagong, 3 Dhaka, 4 Khulna, 5 Rajshahi.. Construction: Master administrative code.. Units: code. Source: Blankespoor et al. (2021) replication package
note yld: Yield of Boro rice, the dry-season irrigated crop that dominates Bangladeshi cereal production.. Construction: Annual district figures averaged into three-year blocks. Enters the regressions as ln(yld).. Units: metric tons per hectare. Source: Yearbook of Agricultural Statistics 2016, Bangladesh Bureau of Statistics
note pop91: Total population recorded in the 1991 census - a baseline characteristic fixed before the bridge.. Construction: Used in the propensity model and, interacted with the period index, as a control that lets initially-large units follow a different trajectory.. Units: persons. Source: IPUMS International
note rainm: Average monthly precipitation over the period.. Construction: Extracted at the upazila level from the NOAA PREC/L gridded reconstruction. Enters the regressions as ln(rainm + 1) for nightlights and ln(rainm) elsewhere.. Units: cm per month. Source: NOAA PREC/L
note rainsd: Standard deviation of monthly precipitation over the period - a control for weather risk, which matters directly for agricultural yield.. Construction: Extracted at the upazila level from NOAA PREC/L. Zero values become missing under the log transform, which is what removes 28 nightlights upazilas from the estimation sample.. Units: cm per month. Source: NOAA PREC/L
note jamuna_m: Crow-fly distance from the unit centroid to the Jamuna bridge foot.. Construction: Computed in ArcGIS from upazila-level shapefiles for Bangladesh.. Units: metres. Source: Blankespoor et al. (2021), from Bangladesh shapefiles
note padma_m: Crow-fly distance from the unit centroid to the planned Padma bridge site.. Construction: Computed in ArcGIS from upazila-level shapefiles for Bangladesh.. Units: metres. Source: Blankespoor et al. (2021), from Bangladesh shapefiles
note treat: 1 if the unit lies in the Jamuna hinterland, the region the bridge connected; 0 otherwise.. Construction: Assigned by geography: which side of which river the unit sits on.. Units: 0/1. Source: Blankespoor et al. (2021) replication package
note smp1: Non-missing for the Jamuna and Padma hinterlands; MISSING for the Dhaka-Chittagong core. Filtering on non-missing is how the core is excluded from every regression.. Construction: Missing wherever treatd == 1. In the two DHS files there is no core, so this variable is numerically identical to treat.. Units: 0/1 or missing. Source: Blankespoor et al. (2021) replication package
note treatd: 1 for the Dhaka-Chittagong core and adjacent areas. Used only for descriptive comparison in the paper's Appendix Table AT.1, never as a DiD comparison group.. Construction: The authors dropped core estimates from the published design after a referee pointed out that the Padma hinterland is not a valid comparison for the core.. Units: 0/1. Source: Blankespoor et al. (2021) replication package
save "bridge_yield.dta", replace

* ---- bridge_dhs_household.dta ----
use "bridge_dhs_household.dta", clear
label data "DHS/HIES household questionnaire"
note _dta: No core region and no smp1 filter - smp1 is numerically identical to treat. Clustering is at the District level because upazila identifiers are not released for DHS rounds before 2003. Floor is missing for 183 rows.
note year: Integer period index, NOT a calendar year. Nightlights: 1=1992-94, 2=1995-97, 3=1998-2000, 4=2001-04, 5=2005-07, 6=2008-10, 7=2011-13. Census: 1=1991, 2=2001, 3=2011. Yield: 1=1988-91, 2=1992-94, 3=1995-97, 4=1998-2000, 5=2001-04, 6=2005-07, 7=2008-10, 8=2011-13. DHS: 1=1994, 2=1996, 3=1997, 4=2003, 5=2007, 6=2011, 7=2013.. Construction: Annual satellite and yield series are averaged into three-year blocks to smooth transitory shocks. The bridge opened in June 1998, inside nightlights period 3 and yield period 4.. Units: index. Source: Blankespoor et al. (2021) replication package
note Cluster: DHS primary sampling unit - effectively a village. The unit of observation in both DHS files.. Construction: DHS survey variable.. Units: code. Source: The DHS Program
note Division: Administrative division as coded in the DHS files: 1 Barisal, 2 Chittagong, 3 Dhaka, 4 Khulna, 5 Rajshahi, 6 Sylhet.. Construction: DHS survey variable.. Units: code. Source: The DHS Program
note District: District identifier in the DHS files. This is the clustering and fixed-effect level for the public-goods placebo, because upazila identifiers are not released for survey rounds before 2003.. Construction: DHS survey variable.. Units: code. Source: The DHS Program
note rural: 1 if the survey cluster is rural. Entered as a control in the household regression because it dominates every service-access outcome.. Construction: DHS survey variable.. Units: 0/1. Source: The DHS Program
note female: Share of the village population that is female.. Construction: DHS household records averaged to the survey cluster.. Units: share (0-1). Source: The DHS Program
note age5: Share of the village population below five years of age.. Construction: DHS household records averaged to the survey cluster.. Units: share (0-1). Source: The DHS Program
note age60: Share of the village population above sixty. Significantly lower in the Jamuna hinterland before the bridge - one of the demographic imbalances the paper acknowledges.. Construction: DHS household records averaged to the survey cluster.. Units: share (0-1). Source: The DHS Program
note Age: Average age of household members in the village.. Construction: DHS household records averaged to the survey cluster.. Units: years. Source: The DHS Program
note hsize: Mean number of members per household in the village.. Construction: DHS household records averaged to the survey cluster.. Units: persons. Source: The DHS Program
note DrinkWater: Share of households with piped drinking water. One of the few pre-bridge characteristics on which the two hinterlands differ significantly.. Construction: DHS household records averaged to the survey cluster.. Units: share (0-1). Source: The DHS Program
note Toilet: Share of households with a flush toilet.. Construction: DHS household records averaged to the survey cluster.. Units: share (0-1). Source: The DHS Program
note Electricity: Share of households in the village with electricity - the one household-questionnaire outcome carried into the paper's published Table 3.. Construction: DHS household records averaged to the survey cluster.. Units: share (0-1). Source: The DHS Program
note Floor: Share of households with a cement floor - a standard asset proxy. Missing for 183 rows.. Construction: DHS household records averaged to the survey cluster.. Units: share (0-1). Source: The DHS Program
note Roof: Share of households with a cement or corrugated-iron roof.. Construction: DHS household records averaged to the survey cluster.. Units: share (0-1). Source: The DHS Program
note padma_m: Crow-fly distance from the unit centroid to the planned Padma bridge site.. Construction: Computed in ArcGIS from upazila-level shapefiles for Bangladesh.. Units: metres. Source: Blankespoor et al. (2021), from Bangladesh shapefiles
note jamuna_m: Crow-fly distance from the unit centroid to the Jamuna bridge foot.. Construction: Computed in ArcGIS from upazila-level shapefiles for Bangladesh.. Units: metres. Source: Blankespoor et al. (2021), from Bangladesh shapefiles
note treat: 1 if the unit lies in the Jamuna hinterland, the region the bridge connected; 0 otherwise.. Construction: Assigned by geography: which side of which river the unit sits on.. Units: 0/1. Source: Blankespoor et al. (2021) replication package
note smp1: Non-missing for the Jamuna and Padma hinterlands; MISSING for the Dhaka-Chittagong core. Filtering on non-missing is how the core is excluded from every regression.. Construction: Missing wherever treatd == 1. In the two DHS files there is no core, so this variable is numerically identical to treat.. Units: 0/1 or missing. Source: Blankespoor et al. (2021) replication package
note yr: 1 for survey rounds after the bridge opened (2003 onward); 0 for 1994, 1996 and 1997.. Construction: Pre-coded in the DHS source files.. Units: 0/1. Source: Blankespoor et al. (2021) replication package
save "bridge_dhs_household.dta", replace

* ---- bridge_dhs_village.dta ----
use "bridge_dhs_village.dta", clear
label data "DHS village questionnaire"
note _dta: Missingness is heavy and uneven, which is why the twelve placebo regressions run on samples ranging from 398 to 1,449 observations. Zero of the 21 resulting estimates are significant at 5 percent.
note year: Integer period index, NOT a calendar year. Nightlights: 1=1992-94, 2=1995-97, 3=1998-2000, 4=2001-04, 5=2005-07, 6=2008-10, 7=2011-13. Census: 1=1991, 2=2001, 3=2011. Yield: 1=1988-91, 2=1992-94, 3=1995-97, 4=1998-2000, 5=2001-04, 6=2005-07, 7=2008-10, 8=2011-13. DHS: 1=1994, 2=1996, 3=1997, 4=2003, 5=2007, 6=2011, 7=2013.. Construction: Annual satellite and yield series are averaged into three-year blocks to smooth transitory shocks. The bridge opened in June 1998, inside nightlights period 3 and yield period 4.. Units: index. Source: Blankespoor et al. (2021) replication package
note Cluster: DHS primary sampling unit - effectively a village. The unit of observation in both DHS files.. Construction: DHS survey variable.. Units: code. Source: The DHS Program
note Division: Administrative division as coded in the DHS files: 1 Barisal, 2 Chittagong, 3 Dhaka, 4 Khulna, 5 Rajshahi, 6 Sylhet.. Construction: DHS survey variable.. Units: code. Source: The DHS Program
note District: District identifier in the DHS files. This is the clustering and fixed-effect level for the public-goods placebo, because upazila identifiers are not released for survey rounds before 2003.. Construction: DHS survey variable.. Units: code. Source: The DHS Program
note Urban: Village classification: 1 rural area, 2 SMA, 3 municipality area, 4 other urban.. Construction: DHS village questionnaire.. Units: code. Source: The DHS Program
note grameen_bank: Whether the village has a Grameen Bank branch - a genuine presence indicator.. Construction: DHS village questionnaire. Note that the raw variable carries an occasional code 9, so the mean exceeds a clean 0/1 share.. Units: 0/1. Source: The DHS Program
note co_operative_soc: Whether the village has a cooperative society. Its true short-run estimate is 0.090 (0.108); the published Table 3 prints 0.420, which is in fact the post_office coefficient - see section 18.2 of the companion post.. Construction: DHS village questionnaire.. Units: 0/1. Source: The DHS Program
note madrassa_school: Distance from the village to the nearest madrassa school.. Construction: DHS village questionnaire.. Units: miles. Source: The DHS Program
note primary_school: Distance from the village to the nearest primary school.. Construction: DHS village questionnaire.. Units: miles. Source: The DHS Program
note high_school: Distance from the village to the nearest high school. Its long-run estimate of +0.535 (0.290) is the only marginally significant coefficient among the 21 public-goods tests - and it has the wrong sign for the political-economy story, since it says schools got farther away.. Construction: DHS village questionnaire.. Units: miles. Source: The DHS Program
note post_office: Distance to the nearest post office. NOTE: filed under 'Presence of' in the paper's published Table 3, but labelled as a distance in the source data.. Construction: DHS village questionnaire.. Units: miles. Source: The DHS Program
note cinema: Distance to the nearest cinema. Like post_office, filed under 'Presence of' in the published table but a distance in the data.. Construction: DHS village questionnaire.. Units: miles. Source: The DHS Program
note dist_Thana: Distance from the village to the thana (police-subdistrict) headquarters. Non-missing for only 559 rows.. Construction: DHS village questionnaire, reported in miles.. Units: miles. Source: The DHS Program
note dist_district: Distance from the village to the district headquarters. Non-missing for 618 rows.. Construction: DHS village questionnaire, reported in miles.. Units: miles. Source: The DHS Program
note NGO: Whether an income-generating NGO operates in the village. Non-missing for only 413 rows, and the treatment terms are dropped for collinearity in the placebo regression.. Construction: DHS village questionnaire.. Units: 0/1. Source: The DHS Program
note dist_satellite_clinic: Distance to the nearest satellite health clinic. The sparsest variable in the file, non-missing for only 398 rows.. Construction: DHS village questionnaire.. Units: miles. Source: The DHS Program
note padma_m: Crow-fly distance from the unit centroid to the planned Padma bridge site.. Construction: Computed in ArcGIS from upazila-level shapefiles for Bangladesh.. Units: metres. Source: Blankespoor et al. (2021), from Bangladesh shapefiles
note jamuna_m: Crow-fly distance from the unit centroid to the Jamuna bridge foot.. Construction: Computed in ArcGIS from upazila-level shapefiles for Bangladesh.. Units: metres. Source: Blankespoor et al. (2021), from Bangladesh shapefiles
note treat: 1 if the unit lies in the Jamuna hinterland, the region the bridge connected; 0 otherwise.. Construction: Assigned by geography: which side of which river the unit sits on.. Units: 0/1. Source: Blankespoor et al. (2021) replication package
note smp1: Non-missing for the Jamuna and Padma hinterlands; MISSING for the Dhaka-Chittagong core. Filtering on non-missing is how the core is excluded from every regression.. Construction: Missing wherever treatd == 1. In the two DHS files there is no core, so this variable is numerically identical to treat.. Units: 0/1 or missing. Source: Blankespoor et al. (2021) replication package
note yr: 1 for survey rounds after the bridge opened (2003 onward); 0 for 1994, 1996 and 1997.. Construction: Pre-coded in the DHS source files.. Units: 0/1. Source: Blankespoor et al. (2021) replication package
note dist_hos: Distance from the village to the nearest hospital.. Construction: DHS village questionnaire.. Units: miles. Source: The DHS Program
save "bridge_dhs_village.dta", replace