Does Self-Regulation Reduce Pollution? Responsible Care in the US chemicals industry Shanti Gamper-Rabindran Assistant Professor Graduate School of Public.

Does Self-Regulation Reduce Pollution?Responsible Care

in the US chemicals industry

Shanti Gamper-RabindranAssistant Professor

Graduate School of Public & International AffairsUniversity of Pittsburgh

Stephen FingerAssistant Professor

Moore Business SchoolUniversity of South Carolina

Funding: NSF BCS 0351058 U Pitt UCSUR, CRDF, EUCE

Outline

• Why study Self-regulation?

• Method

• Data

• Results

• Conclusion

Self-regulation• Industry associations mandate their members to

attain environmental goals, beyond that specified by existing regulations.

• Widely used.1) Nuclear power-plants in the US– INPO2) Petroleum industry in the US – STEP

Does self-regulation reduce pollution or environmental risks?

Self-regulation• Industry associations mandate their members to

attain environmental goals, beyond that specified by existing regulations.

• Widely used.1) Nuclear power-plants in the US -INPO2) Petroleum industry in the US – STEP

Does self-regulation reduce pollution or environmental risks?

Responsible Care• Launched by American Chemical Council in 1989.• Union Carbide accident killed 20,000+ people.• Stock prices for all chemical firms fell.

Responsible Care• Launched by American Chemical Council in 1989.• Union Carbide accident killed 20,000+ people.• Stock prices for all chemical firms fell.

Responsible Care• Adopted worldwide

Responsible Care• Stated goal – self-regulation to improve environmental

performance of the chemical industry.• Codes of Conduct – waste minimization

& pollution prevention.• Firms submit self-assessment to ACC

Responsible Care• But no third party verification (until 2002).• No expulsion of errant members (as of 2002).

Research question• Did Responsible Care reduce pollution?• Our results: No

Outline

• Why study Responsible Care?

• Literature Review

• Method

• Data

• Results

• Policy conclusion

Literature Review

• Can self-regulation achieve stated goals?

• Maybe yes

• Maybe no

• Empirical evaluation is scarce.

Supporting View: RC create sufficient incentives for plants’ pollution

reduction.• Incentive 1 : Industry self-regulation can pre-empt

stricter government regulation.– Coordination problem for firms in industry.– Comply & reduce pollution? Or Shirk?– Critical number of members will reduce pollution,

even if others free-ride, to maintain overall credibility of the RC program.

– Costs to these firms of reducing pollution under RC < Costs of government regulation if self-regulation fails.

Dawson and Segerson (2008).

Supporting View: RC can create incentives for plants’ pollution

reduction.• Incentive 2: Benefits from Green Reputation. – The RC program, by limiting its membership to firms

that commit to RC’s goals, including pollution prevention, allows member firms to benefit from the positive reputational effect of being socially responsible.

– These firms can benefit from consumers who choose to purchase from, and investors who choose to invest in, firms that establish the reputation of being responsible (Hay, Stavins, and Vietor, 2005).

Supporting View: RC create sufficient incentives for plants’ pollution

reduction.Participation in a program that signals green

- reduces inspections or enforcement actions by the regulatory agencies

(Maxwell & Decker, 2006; Innes & Sam, 2008)

– discourages boycotts by environmental groups or pre-empt their lobbying for stricter regulations (Maxwell et al., 2000; Baron, 2001).

Opposing View: RC is green-wash

• Firms join RC for positive publicity but in reality they do not incur the costs to reduce their pollution.

• Firms have no incentive to reduce pollution.– Firms not subject to sanctions if fail to achieve code of

conduct– Firms not subject to third party verification.

Empirical study

• Lenox and King (2000)

• Pioneering empirical study on self-regulation

Lenox and King (2000)

• Method problem: Ignore self-selection.

• Overstate RC impact on reducing pollution

– If the firms that self-select into RC are those that will reduce their pollution regardless of RC participation.

• Understate RC impact on reducing pollution

– If firms that self-select are those are those that face more difficulties in reducing pollution, and join to benefit from best practices.




– If the firms that self-select into RC are those that will reduce their pollution regardless of RC participation


– If firms that self-select are those are those that face more difficulties in reducing pollution, and join in order to from shared best practices




– If the firms that self-select into RC are those that will reduce their pollution regardless of RC participation.


– If firms that self-select are those are those that face more difficulties in reducing pollution, and join to benefit from best practices.


• Data problem: – TRI “Production Ratio” variables to control for

output – Problematic variable

• We use # employee, imperfect proxy for output

Outline


• Theory

• Method

• Data

• Results

• Conclusion

Did RC reduce pollution?

• Do plants that belong to RC participating firms reduce their pollution relative to statistically equivalent plants that belong to non-RC participating firms?

Method

• “Treatment” groups – plants belonging to RC firms.

• “Control” groups – statistically equivalent plants belonging to non-RC firms.

• Use Instrumental Variables (IV) – to address self-selection into program

• Limitation– Non-RC firms reduce their pollution in response to

RC.

Pollution Equation

• Obs: Plant j (belonging to firm i) at time t.

• yijt = x1 ijt β1+ pijt1 +x3it β3+µijt

• yijt = Log (toxicity weighted air pollution/ # employee)

• pijt = 1 if plant j is owned by firm which participates in RC at time t; 0 otherwise.

• x1 ijt = plant factors that directly affect plant’s pollution.

• x3 it = firm factors that directly affect plant’s pollution.

1 negative => impact of RC.

• Plant j (belonging to firm i) time t

• Pollution Equation

yijt = x1 ijt β1+ pijt1 +x3it β3+µijt

• Participation Equation

Firm: Benefitit* = ji x1 ijt θ1 + x3 itθ3 + z1it θ4

pit = 1 if Benefitit * > 0

Plant-Level Estimating Equation:

Benefitijt* =x1ijtθ1 + -jx1 ijt θ1 + x3it θ3 + z1itθ4 + ijt

Estimation – IV/GMM

yijt = x1 ijt β1+ pijt1 +x3it β3+µijt

• For all plants, use z1it as instrument

• For plants belonging to multi-plant firms, additionally use -jx1 ijt as instrument

What are the instruments?

SampleS1 S2

All plants Plants belonging to Instruments multiplant firms onlyI1: Firm-level variables

I2: Characteristics of other plants belonging to the same firm

Instruments for the ‘all plants’ sampleInstrument 1

• Average RC participation within the same sub-industry

• If the average is high: – there may be features of the sub-industry that

make RC appealing.

Instrument 2: RC participation by firm in previous period

• Persistence in RC participation

• The cost of continuing participation less than the cost of a new member joining – members may have already implemented new

systems and procedures to adhere RC’s standards.

– costly to switch out of the program as it may send a negative signal to their consumers or to regulators about their conduct.

Instrument 3: Firm’s membership in ACC pre-1989

• Firms that were ACC members prior to RC– were more likely to receive a positive net benefit from the trade

association.– After RC, they continue in ACC if RC benefits > RC costs

• Firms that choose not to be members of the ACC prior to RC

– costs of membership exceeded the trade association benefits of the program.

• For these firms to join RC after its inception, they need to:– offset their negative trade association costs and.– generate positive net benefits from RC.

Instruments for multi-plant firms

• If Dow Chemical needs to reduce pollution at a plant in New Jersey due to neighborhood pressure, that factor: – reduces the additional cost for Dow to join RC

and thus affect the likelihood of all Dow plants being in the program.

– does not directly cause Dow to reduce pollution at a plant in Louisiana.

– Caveat – technological spillovers across plants in the same firm.

– Must check if instruments invalid using over-identification test.

Instruments for multi-plant firms• Firm f owns plant j, k, l, m.

• As instrument for plant j, use characteristics of other plants owned by same firm.

• Nevo (2000) uses the average prices of the same product in other cities in the region as instruments for a product’s price in a given city.

• Berry, Levinsohn and Pakes (1995) use characteristics of other products by same producers as instruments for unobserved characteristics of a given product.

Instrument 4: Firm’s HAP to TRI ratio

• Hazardous air pollutants (HAPs) are subject to stricter pollution abatement regulations (the “MACT-hammer”).

• Parent firm w/ plants with high HAP/TRI must reduce pollution, regardless of RC, face lower additional costs in joining RC.

• HAP/TRI of other plants belonging to the same firm affects plant j’s participation in RC, through their effect on the parent firm.

Instrument 5: Firm’s share of production in dirtier sub-

industries• Poll inten for SIC28xx= Pollution/empl in SIC 28xx

Pollution/empl SIC-28

• Less likely to join RC– More costly to reduce pollution when rely on

pollution-intensive production technologies.

• More likely to join RC– Less costly for dirtier firms to reduce pollution if

diminishing return to pollution abatement.


industries• Pollution/empl in SIC 28xx







industries• Pollution/empl in SIC 28xx






Instrument 6: Firm’s plants’ neighborhood characteristics

• Firms face neighborhood pressure to join RC. – % low education– % poor– % white

Method

• Evidence of heteroskedasticity

• Use GMM estimator

• More efficient than standard IV

• We allow errors to be correlated among plants within the same firm.

Control variables

• Larger firms may have greater financial resources to invest in pollution abatement.– Plant’s size [lagged plants’ employees]– Firm’s size [lagged firms’ employees] – The number of plants owned by the firm. – Dummy for single-plant firms

Control variables

• Industry-level variables at SIC-4– Producer price index, shipment quantity index, the

Herfindahl-Hirschman index and SIC-4 dummies.

• Year dummies – changes in federal regulations and available

technologies.

• Neighborhood pressure on plants– the median income, share white, share < high

school education.

• Lagged emissions (instrumented by t-2)

Outline


• Method

• Data

• Results

• Policy conclusion

CHEMICAL SECTOR

TRI-RSEI

Plants in the US chemicals industry

Plant-level toxicity-weighted air emissions

RSEI EPA IDEACENSUS

Toxicity weights for emissions

Clean Air Act EPA Inspection at plants

DATABASE CONSTRUCTION

CHEMICAL SECTOR

TRI-RSEI

Dun & Bradstreet


Plant-level Employment

RSEI


Firm-plant linkages

Mergents & Corporate Affiliations

ACC

RCC membership


CHEMICAL SECTOR

TRI-RSEI

Dun & Bradstreet




RSEI EPA IDEACENSUS


Demographics% poor % minority % low educ % urban at the census tract-level



Firm-plant linkages


ACC

RCC membership

CHEMICAL SECTOR

TRI-RSEI

Dun & Bradstreet




RSEI EPA IDEACENSUS


Demographics% poor % minority % low educ % urban at the census tract-level



Firm-plant linkages


Firms 1,500+ Plants 2,700+ Time 1988-2001

ACC

RCC membership

Outline


• Method

• Data

• Results

• Policy implication

1988 2001Plants # RC participants 804 1199

% participants 24 28Firms # RC participants 126 142

% participants 6.1 6.4

Participation equation

Are instruments correlated with participation?

Probability of RC participation with values of covariates set at the sample mean is

0.13 for all plants.

0.54 for plants owned by multi-plant firms.

Impact on probability of participation from 1 std dev change in variables

Variable Instruments†ACC membership in 1988 & 1989 0.4 ** 0.5 **†RC participation dummy (t-1) 0.9 ** 0.9 **% RC participation in SIC-4 0.8 ** 0.3 **Firm's HAP/TRI -0.01Firm's SIC pollution index -0.01Firms' plants' average neighborhood pressure % white 0.004 % low education -0.06 ** % poverty 0.01 % urban -0.04 ** Non-attainment county dummy -0.03Other co-variatesPlant's pollution (t-1) 0.01 0.01Plant's HAP/TRI (t-1) 0.01 0.02 *Plant's SIC pollution index 0.01 ** 0.02 *

All plants Multiplant firms

[1] [2] [3] [4] All plantsMain spec Vary instruments

RC-status 0.05 0.05 0.05 0.05 (0.03) (0.03) (0.04) (0.03)

[5] [6] [7] [8] Only Plants Owned by Multi-Plant FirmsMain spec Vary instruments

RC-status 0.03 0.03 0.03 0.02 (0.03) (0.05) (0.03) (0.04)

RC did not reduce pollution

Impact of RC

95% C.I. (-0.02, 0.12) β=0.05

+ 12%

- 2%

Non-RC RC Non-RC RC

Comparison: Average plant-level annual pollution decline 6%

Most favorable to finding RC caused pollution reduction

Least favorable to finding RC caused pollution reduction

All plants Multiplant Firms

[1] [2]RC-status 0.05 0.03(i.e. impact of RC on pollution) (0.03) (0.03)Test - statisticsUnder-ID: Kleibergen-Paap 192 135 LM rk statisticsWeak-ID: Kleibergen-Paap 2025 726 Wald rk statistics Stock-Yogo Critical Values 5% Relative Bias 11 19 20% Relative Bias 6 6Hansen J statistic 0.1 7p-value for Hansen J stat 1 0.6Emissions equation: R-squared 0.8 0.8Observations 18,850 12,705

Ho: instruments are not correlated to error in second stage – Fail to Reject

Instruments do not fail validity tests.



Ho: instruments are correlated to RC participation: Fail to Reject

Are instr. “weak enough to imperil inference,” i.e., the bias in

coefficients from the Biv exceeds a

specific percent bias in Bols?



Ho: Instruments are not only weakly correlated to RC participation – Fail to Reject

Are instr. “weak enough to imperil inference,” i.e., the bias in

coefficients from the Biv exceeds a

specific percent bias in Bols?

Yearly effect of participation

[1] [2] [3]Main spec. 6 year blocks 3 year blocksRC 0.05 RC x 0.07 * RC x 0.09 *

(0.03) I(yr<='95) (0.04) I(yr='90-92) (0.05)RC x 0.05 RC x 0.05 I(yr>'95) (0.04) I(yr='93-95) (0.05)

RC x 0.02 I(yr='96-98) (0.04)RC x 0.09 * I(yr='99-01) (0.05)

Robustness check – Denominator

• Dependent var: Pollution/ # employee • Possible bias against finding RC reduced pollution• If plants respond to RC by choosing a production

process that is less labor intensive, but that does not raise pollution per unit of production.

• Should larger plants increase their output at a faster rate than labor, our denominator for large plants may be too small, resulting in too large a measure of pollution intensity

• Given that RC participants typically have larger plants, this mis-measurement of pollution intensity could bias our estimates of the impact of RC.

• Alternative dependent var: pollution


• Dependent var: Pollution/ # employee • Possible bias against finding RC reduced pollution• If plants respond to RC by choosing a production

process that is less labor intensive, but that does not raise pollution per unit of production.

• Should larger plants increase their output at a faster rate than labor, our denominator for large plants may be too small, resulting in too large a measure of pollution intensity.


• Alternative dependent var: pollution


• Dependent var: Pollution/ # employee • Possible bias against finding RC reduced pollution• If plants respond to RC by choosing a production process

that is less labor intensive, but that does not raise pollution per unit of production.

• Should larger plants increase their output at a faster rate than labor, our denominator for large plants may be too small, resulting in too large a measure of pollution intensity


• Alternative dependent var: • pollution & pollution/(employee)2

Robustness check: Alternative denominators

[1] [2] [3]Media air air air

Denominator empl empl2

RC 0.05 0.05 0.05(0.03) (0.03) (0.04)

Observations 18,850 18,850 18,850

Robustness check – Alternative media

[4] [5] [6] [7] [8] [9]Media all all water land onsite offsiteDenominator empl empl empl empl emplRC 0.02 0.02 0.02 -0.01 0.05 0.01

(0.03) (0.02) (0.02) (0.02) (0.04) (0.05)

Robustness check – add regulatory variables

[1] [2] [3] [4] [5] All plants Multiplant firms

Main Inspection Main InspectionSpec Vars spec Vars

added addedFirst stage participation equationFirm's average lagged plant-level 0.3 Clean Air Act inspection (0.2)Dummy for firms with no plants 0.5 * subject to Clean Air Act inspection (0.3)Dummy for plant inspection under the 0.06 -0.05 -0.05 Clean Air Act (t-1) (0.1) (0.1) (0.1)GMM estimationRC participation dummy 0.05 0.06 0.03 0.03 0.03

(0.03) (0.03) (0.03) (0.04) (0.04)No obs. 18850 13564 12705 9702 9702

Robustness check: subsamples

• We cannot differentiate between plants that fail to report pollution and plants that close.

• It could be possible for participating firms to close their dirty plants and open clean ones.

• We would not recognize this as improved performance in our estimation method.

• Subsample - Continuous reporters on pollution.• Subsample – Continuous reporters on

employees.


[1] [2] [3] [4]Sample Main Continuous Continuous RC status spec. reporters reporters for no. changes for pollution of employees RC 0.05 0.03 0.07 * 0.04

(0.03) (0.04) (0.04) (0.06)

Obs 18,850 9,399 13,598 7,747


• Our identification of RC’s impact relies on 2 sources of variation:

• (1) variation in RC status between plants with similar observed characteristics; and

• (2) the intertemporal variation for plants whose RC status changes within our panel.

• Limitation – cross section – bias if our instruments and covariates fail to fully control for the systematic differences between plants that were always RC participants and those that were always non-RC participants.

• Solution - Relating changes in plants' RC status to plants' pollution intensity.


[1] [2] [3] [4]Sample Main Continuous Continuous RC status spec. reporters reporters for no. changes for pollution of employees RC 0.05 0.03 0.07 * 0.04

(0.03) (0.04) (0.04) (0.06)

Obs 18,850 9,399 13,598 7,747

• RC impact for subsets of plants or firms characteristics– Dawson and Segerson (2008) – free-riding problems within voluntary arrangements may be

overcome if sub-groups of firms have incentives to participate.– Plants belonging to firms with larger no of plants and firms with

larger number of employees.

– “Environmental Justice concerns” – plants in poor, low-educated, minority neighborhoods – less likely to reduce pollution.

Robustness check: Heterogeneous program effects

[1] [2] [3] [4]RC 0.04 0.04 0.05 0.05

(0.05) (0.04) (0.03) (0.03)RC x -0.0007 no plants (0.003)RC x dummy for 0.2 0.2 single-plant Firm (0.1) (0.1)RC x firm's # -0.001 employees (t-1) (0.02)RC x plant's # 0.02 employees (t-1) (0.02)RC x % poor in plant's 0.1 neighborhood (0.1)RC x % urban in plant's -0.03 neighborhood (0.04)

Heterogeneous program effects

βOLS vs. βIV

βOLS=0.2** βIV=0.03 or 0.05

Possible explanation – Firms with plants that face more difficulties in

reducing pollution are more likely to self-select and join RC to signal green, but fail to reduce pollution.

– After controlling for self-selection, the pollution increase related to RC participation is less pronounced.

Outline


• Method

• Data

• Results

• Conclusion

What have we learned?

• RC did not lead to overall waste reductions in the chemicals

industry from 1990-2002.• In some specifications, RC participation caused 7-9%

increase in pollution at the average plant.• Self-regulation without third-party verification or enforceable

penalties may not be an effective substitute for formal

regulation.






regulation.






regulation.

Caveats

• We will understate impact of the program.- If non-members respond to presence of RC by reducing pollution- Pollution reduction by RC firms cause technological innovations that reduce pollution abatement costs for all firms.

• We only analyze two of RC's six codes of conduct.

• RC has mandated third party verification post-2002.

Future work: Why did RC not reduce emissions?

• Future work 1:• Did RC participation result in fewer EPA inspections?• Soft-pedaling on regulation?• Sam and Innes (2008)

• Future work 2:• Pre 2002 – no third party verification.• Post 2002 – there was third party verification • Test pre-2002 vs. post-2002?

Thank you!

paper posted at

www.pitt.edu/~shanti1/

http://www.pitt.edu/~shanti1/

Pollution equation• yijt = x1 jt β1 + pjt1 + pjt (x2jt-xx2) β2+ µjt • Pollution is affected by the plant characteristics (xjt), the observed

participation decision (pjt), a subset of plant characteristics which affect the

impact of RC (x2jt), and an unobserved component (µjt).

• The first term (x1 jt β1) accounts for the effect of the covariates on pollution regardless of RC status.

• The second term (pjt1 ) captures the effect of RC on the average plant,

• The third term (pjt (x2jt-xx2) β2) captures the impact of RC that varies by plant characteristics.

• We demean the x2 variables in the third term in order to consistently

estimate the effect of RC on an average plant with the 1 coefficient.

• Firm i plant j time t

• Emissions

yijt = yij(t-1)βy +x1 ijt β1+ pijt1

+pijt (x2ijt-xx2) β2+x3it β3+µijt

• Participation

Firm: Benefitit* = ji x1 ijt θ1 + x3 itθ3 + z1it θ4

pit = 1 if Benefitit * > 0

Plant-Level Estimating Equation:

Benefitijt =x1ijtθ1 + -jx1 ijt θ1 + x3it θ3 + z1itθ4 + ijt

Does Self-Regulation Reduce Pollution? Responsible Care in the US chemicals industry Shanti Gamper-Rabindran Assistant Professor Graduate School of Public.

Documents

self regulation

industry selfregulation

study selfregulation

goal selfregulation

euce slide

acc slide

chemical firms

study responsible care