Research ArticlesENVIRONMENTAL STUDIES

Hydraulic fracturing and infant health: New evidence from Pennsylvania

See allHide authors and affiliations

Science Advances  13 Dec 2017:
Vol. 3, no. 12, e1603021
DOI: 10.1126/sciadv.1603021
  • Fig. 1 Locations of births and fractured wells in Pennsylvania.

    Each square displayed above is 0.25° latitude by 0.25° longitude. We use all birth certificates in Pennsylvania for 2004–2013. They include maternal address which is used to calculate average yearly births per square. Black triangles represent the exact locations of fractured wells, which we observe from the Pennsylvania Department of Environmental Protection (DEP) Internal Operator Well Inventory. These data include all oil and gas wells with a Pennsylvania DEP drilling permit and which are not currently filled in (plugged). We queried this database in November 2014. Fractured wells are those marked “unconventional” in the database. We have dropped any wells with missing American Petroleum Institute numbers, spud or permit date, or location information.

  • Fig. 2 Number of fractured wells and value of all drilling in Pennsylvania (2004–2013).

    The left y axis shows total fractured wells in Pennsylvania by spud year and quarter (that is, the commencement of drilling), and the right y axis reports annual values of gas from fractured wells in Pennsylvania. X axis shows spud year and month (dates of commencement of drilling) that are recorded in the Pennsylvania DEP Internal Operator Well Inventory, which is described in the notes to Fig. 1. Annual gas production per well is recorded by the Pennsylvania DEP in its Oil and Gas Historical Production Report. We merge these data to our Internal Operator Well Inventory data by well identification number and then sum gas production to the year level. To convert production to dollars, we use gas prices from the U.S. Energy Information Administration (EIA), which reports the Henry Hub Natural Gas Spot Price (www.eia.gov/dnav/ng/hist/rngwhhdA.htm). To convert to British Thermal Units annual heating values for Pennsylvania are taken from the EIA.

  • Fig. 3 Effect of fracturing on low–birth weight, county fixed effects.

    The left y axis of the graph indicates coefficients and confidence intervals (CIs) from a version of Eq. 1 in which “Near” is replaced with 15 distance indicators representing the proximity of maternal residence to well sites; the coefficients represent the in utero effect on infant health of hydraulic fracturing (that is, when conception occurs after well spud date) at 1-km intervals from the well site. The data sources for the regression are all birth certificates issued in Pennsylvania from 2004 to 2013 and the Pennsylvania DEP Internal Operator Well Inventory. We exclude births with missing values for gestation length or latitude/longitude of maternal residence. We calculate the distance between maternal residence and well sites using Vincenty’s formula. The specification includes year fixed effects (FE), month of birth FE, and county of maternal residence FE. The following demographic controls are also included: mother is married, marital status missing, maternal race and ethnicity (black, Hispanic, missing), maternal education [no high school (HS), HS diploma, some college, college, advanced degree, missing], maternal age (<20, 20 to 24, 25 to 29, 30 to 34, 35+, missing), child is male, child sex missing, and child parity (first, second, third, fourth born and higher, parity missing). Standard errors are clustered on maternal ID. The right y axis plots average yearly births at each distance from a well site.

  • Fig. 4 Effect of fracturing on infant health index, county fixed effects.

    The left y axis of the graph indicates coefficients and CIs from a version of Eq. 1 in which “Near” is replaced with 15 distance indicators representing the proximity of maternal residence to well sites; the coefficients represent the in utero effect on infant health of hydraulic fracturing (that is, conception occurs after well spud date) at 1-km intervals from the well site. The data sources for the regression are the universe of birth certificates issued in Pennsylvania from 2004 to 2013 and the Pennsylvania DEP Internal Operator Well Inventory. We exclude births with missing values for gestation length or latitude/longitude of maternal residence. We calculate the distance between maternal residence and well sites using Vincenty’s formula. The infant health index ranges from 0 to 1; an increase indicates better health. The regression specification includes year FE, month of birth FE, and county of maternal residence FE. The following demographic controls are also included: mother is married, marital status missing, maternal race and ethnicity (black, Hispanic, missing), maternal education (no HS, HS diploma, some college, college, advanced degree, missing), maternal age (<20, 20 to 24, 25 to 29, 30 to 34, 35+, missing), child is male, child sex missing, and child parity (first, second, third, fourth born and higher, parity missing). Standard errors are clustered on maternal ID. The right y axis plots average yearly births at each distance from a well site.

  • Table 1. Difference in means.

    The data source is the universe of birth certificates in Pennsylvania (2004–2013) matched to the Pennsylvania DEP Internal Operator Well Inventory. Maternal and infant demographic indicators and health outcomes are recorded at the time of birth. “Near, 0–1 km” indicates that the mother lives within 0 to 1 km of at least one well site. “Far, 3–15 km” indicates that the mother lives 3 to 15 km from the nearest well site. Columns (5) to (7) report P values from t tests of equality of means across the different samples indicated. Column (7) tests whether (2) − (1) = (4) − (3). This quantity is referred to as the difference-in-differences, or D-in-D.

    Near, 0–1 kmFar, 3–15 kmP values
    BeforeAfterBeforeAfter(5)(6)(7)
    (1)(2)(3)(4)(1) − (2)(3) − (4)D-in-D
    Mother characteristics
    Married0.680.610.620.630.000.000.00
    Black0.010.020.120.060.010.000.00
    Hispanic0.010.010.020.020.660.000.57
    <High school0.110.120.110.100.270.000.01
    High school0.280.310.260.250.020.000.00
    Some college0.320.320.280.290.890.000.43
    College0.220.170.220.230.000.000.00
    Advanced0.080.080.130.130.740.690.74
    <20 years old0.060.060.060.050.890.000.11
    20–24
    years old
    0.210.260.220.210.000.000.00
    25–29
    years old
    0.300.310.290.290.420.000.86
    30–34
    years old
    0.260.230.270.280.010.000.00
    >35 years old0.170.140.170.160.010.000.07
    Infant characteristics
    Male0.500.520.510.520.240.270.32
    First born0.420.410.430.430.660.150.84
    Second born0.330.340.330.330.460.050.68
    Third born0.160.160.150.150.790.640.73
    Fourth and up0.090.090.090.090.900.510.98
    Health outcomes
    Low birth
    weight
    0.050.070.060.060.040.000.01
    Birth weight3354.353312.813316.943331.080.010.000.00
    Health index0.050.010.010.020.010.000.00
    n48711798133,10778,366
  • Table 2. Effect of fracturing on infant health.

    Each coefficient and SE (shown in parentheses) is from a different regression and represents the effect on the given infant health outcome of in utero exposure to fracturing (when conception occurs after well spud date) within the indicated distance. The data sources for the regression are all birth certificates issued in Pennsylvania from 2004 to 2013 and the Pennsylvania DEP Internal Operator Well Inventory. We calculate the distance between maternal residence and well sites using Vincenty’s formula. The infant health index ranges from 0 to 1; an increase indicates better health. Each regression specification includes region of maternal residence*year FE, year*month of birth FE, and county of maternal residence FE. The following demographic controls are also included: mother is married, marital status missing, maternal race and ethnicity (black, Hispanic, missing), maternal education (no HS, HS diploma, some college, college, advanced degree, missing), maternal age (<20, 20 to 24, 25 to 29, 30 to 34, 35+, missing), child is male, child sex missing, and child parity (first, second, third, fourth born and higher, parity missing). Where indicated, we include a vector of maternal ID fixed effects (“mother FE”). “Under 15 km” indicates the subset of mothers living less than 15 km from the nearest well site. SEs are clustered on maternal ID. +P < 0.10; **P < 0.05; ***P < 0.01.

    Dependent variable(Near, 0–1 km) × after(Near, 1–2 km) × after(Near, 2–3 km) × after
    (1)(2)(3)(4)(5)(6)(7)(8)(9)
    Low birth weight
    (mean, 0.065)
    0.016**
    (0.007)
    0.015**
    (0.007)
    0.012
    (0.014)
    0.006+
    (0.004)
    0.005
    (0.004)
    0.004
    (0.007)
    0.009***
    (0.003)
    0.008***
    (0.003)
    0.007
    (0.005)
    Birth weight
    (mean, 3319.6)
    −38.654**
    (15.558)
    −36.707**
    (15.595)
    −13.034
    (31.137)
    −3.534
    (8.487)
    −2.023
    (8.530)
    −10.439
    (14.349)
    −7.092
    (6.515)
    −5.294
    (6.515)
    0.803
    (10.608)
    Health index
    (mean, 0.000)
    −0.054***
    (0.019)
    −0.052***
    (0.019)
    −0.004
    (0.040)
    −0.020**
    (0.010)
    −0.018+
    (0.011)
    −0.018
    (0.020)
    −0.028***
    (0.008)
    −0.025***
    (0.008)
    −0.015
    (0.015)
    n1,086,917231,578231,5781,102,424247,085247,0851,117,919262,580262,580
    Mother FENoNoYesNoNoYesNoNoYes
    Under 15 kmNoYesYesNoYesYesNoYesYes

Supplementary Materials

  • Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/3/12/e1603021/DC1

    table S1. Comparison of mothers by number of births observed in sample.

    table S2. Effect of fracturing on infant health (including both inactive and active wells).

    table S3. Effect of fracturing on infant health (treatment based on birth date).

    table S4. Effect of fracturing on infant health (controlling for latitude/longitude grid*year controls).

    table S5. Mothers with <2 well sites spudded within 1 km versus mothers with 2+ well sites spudded within 1 km.

    table S6. Effect of fracturing on infant health (logit for low birth weight).

  • Supplementary Materials

    This PDF file includes:

    • table S1. Comparison of mothers by number of births observed in sample.
    • table S2. Effect of fracturing on infant health (including both inactive and active wells).
    • table S3. Effect of fracturing on infant health (treatment based on birth date).
    • table S4. Effect of fracturing on infant health (controlling for latitude/longitude grid*year controls).
    • table S5. Mothers with <2 well sites spudded within 1 km versus mothers with 2+ well sites spudded within 1 km.
    • table S6. Effect of fracturing on infant health (logit for low birth weight).

    Download PDF

    Files in this Data Supplement: