Changing Corporations: An Assessment of the Indicators of Environmental Behavior in the Chemical Manufacturing Industry By Cristina Turney and Sarah

Winkeller

A practicum submitted in partial fulfillment of the requirements for the degree of Master of Science (Natural Resources and Environment) at the University of Michigan December 2006

Faculty advisors: Professor Andrew J. Hoffman, Ph.D., Holcim (U.S.) Professor of Sustainable Enterprise Professor Thomas P. Lyon, Ph.D., Dow Chemical Professor of Sustainable Science, Technology and Commerce

Table of Contents Abstract.................................................................................................................v Problem Definition..............................................................................................1 Literature Review................................................................................................2 How Our Paper Differs From Prior Scholarship.............................................6 Theoretical Framework......................................................................................7 Methodology.......................................................................................................13 Discussion of Variables.....................................................................................14 Social Drivers.....................................................................................................14 Environmental NGOs......................................................................................14 Press ................................................................................................................15 The Courts .......................................................................................................19 The Community ..............................................................................................20 Academia ........................................................................................................22 Coercive Drivers................................................................................................24 Domestic Regulation ......................................................................................24 Socially Responsible Investing.......................................................................26 Buyers..............................................................................................................27 Market Drivers...................................................................................................28 Trade Associations .........................................................................................28 Other Components of the Model......................................................................30 Toxic Release Inventory Data.........................................................................30 Industry Lobbying Dollars..............................................................................32 Median Income for Family of Four ................................................................33 Dummy variables.............................................................................................34 Creating the Model............................................................................................35 Interpretation of the Model..............................................................................39 Implications of the Model for Actors..............................................................41 ............................................................................................................................41 Implications for Activists................................................................................41 Implications for Corporate Executives............................................................41 Limitations of the Model...................................................................................44 Data limitations................................................................................................44 Limitations of using a statistical model..........................................................45 Suggested Improvements to Our Model for Future Researchers................46 Model Methodology Improvements................................................................49 Conclusion..........................................................................................................50 Bibliography.......................................................................................................51 Appendices..........................................................................................................52 Appendix 1: Model with all Variables.............................................................52 Appendix 2: Eight Variable Model..................................................................53 Fixed-effects (within) regression Number of obs = 400........53 Group variable (i): stateid Number of groups = 50..........53 R-sq: within = 0.0509 Obs per group: min = 8............53 between = 0.0319 avg = 8.0...........................53

ii

max = 8.............................53 F(8,342) = 2.45................................53 corr(u_i, Xb) = -0.2604 Prob > F = 0.0138.................53 ------------------------------------------------------------------------------....................53 | Robust.............................................................................53 normalized~a | Coef. Std. Err. t P>|t| [95% Conf. Interval] ..............................................................................................................................53 -------------+----------------------------------------------------------------...................53 pressartic~s | 188.7636 111.5377 1.69 0.091 -30.62256 408.1498. 53 sri | 6777.162 5710.666 1.19 0.236 -4455.287 18009.61........53 lobbying | -15344.68 13290.77 -1.15 0.249 -41486.61 10797.26. .53 publicopin~t | -2683.439 3034.526 -0.88 0.377 -8652.124 3285.24653 academic | 927.0522 1076.896 0.86 0.390 -1191.12 3045.225...53 ngomembers | 1.942028 1.309168 1.48 0.139 -.6330074 4.517063 ..............................................................................................................................53 westlaw | -5926.027 6662.665 -0.89 0.374 -19030.99 7178.932...53 ofisocerti~s | -2170.648 1558.801 -1.39 0.165 -5236.693 895.3966..53 _cons | 586475.9 328492.2 1.79 0.075 -59643.4 1232595......53 -------------+----------------------------------------------------------------...................53 sigma_u | 345003.06....................................................................................53 sigma_e | 124199.75....................................................................................53 rho | .88527134 (fraction of variance due to u_i)..............................53 ------------------------------------------------------------------------------....................53 Appendix 3: Model with t stats 1 < t < -1........................................................54 Fixed-effects (within) regression Number of obs = 400........54 Group variable (i): stateid Number of groups = 50..........54 R-sq: within = 0.0490 Obs per group: min = 8............54 between = 0.0305 avg = 8.0...........................54 overall = 0.0105 max = 8.............................54 F(5,345) = 3.21................................54 corr(u_i, Xb) = -0.2538 Prob > F = 0.0076.................54 ------------------------------------------------------------------------------....................54 | Robust................................................................................54 normalized~a | Coef. Std. Err. t P>|t| [95% Conf. Interval] ..............................................................................................................................54 -------------+----------------------------------------------------------------...................54 pressartic~s | 175.2665 119.7608 1.46 0.144 -60.28673 410.8197. 54 sri | 5339.994 5233.784 1.02 0.308 -4954.147 15634.14........54 lobbying | -16081.18 13742.26 -1.17 0.243 -43110.33 10947.97. .54 ngomembers | 1.863761 1.317967 1.41 0.158 -.7285011 4.456023 ..............................................................................................................................54 ofisocerti~s | -2118.885 1280.629 -1.65 0.099 -4637.707 399.9375..54 _cons | 510769.7 283446.5 1.80 0.072 -46731.07 1068270.....54 -------------+----------------------------------------------------------------...................54 sigma_u | 344463.45....................................................................................54 sigma_e | 123778.54....................................................................................54

overall = 0.0112

iii

rho | .88564292 (fraction of variance due to u_i)..............................54 ------------------------------------------------------------------------------....................54 Appendix 4: Final Model..................................................................................55 Fixed-effects (within) regression Number of obs = 500........55 Group variable (i): stateid Number of groups = 50..........55 R-sq: within = 0.0373 Obs per group: min = 10...........55 between = 0.0263 avg = 10.0..........................55 overall = 0.0114 max = 10............................55 F(2,448) = 8.47................................55 corr(u_i, Xb) = -0.3479 Prob > F = 0.0002.................55 ------------------------------------------------------------------------------....................55 Robust.................................................................................55 normalized~a | Coef. Std. Err. t P>|t| [95% Conf. Interval] ..............................................................................................................................55 -------------+----------------------------------------------------------------...................55 ngomembers | 4.053103 1.748284 2.32 0.021 .6172474 7.488959 ..............................................................................................................................55 ofisocerti~s | -1125.731 306.8528 -3.67 0.000 -1728.781 -522.6819. 55 _cons | 213051.5 12319.82 17.29 0.000 188839.6 237263.3.....55 -------------+----------------------------------------------------------------...................55 sigma_u | 420766.55....................................................................................55 sigma_e | 191398.43....................................................................................55 rho | .82855799 (fraction of variance due to u_i)..............................55 ------------------------------------------------------------------------------....................55 Appendix 5: Sources of Data for Variables....................................................56 Appendix 6: Additional Data Graphs.............................................................58

iv

ABSTRACT This paper presents a statistical model that quantifies the influence of various stakeholders on chemical manufacturing companies environmental performance. We based the model on a framework outlined by Andrew

Hoffman in Competitive Environmental Strategy: A Guide to the Changing Business Landscape.1 The model uses actors from the following four

categories: Social Drivers, Resource Drivers, Market Drivers and Coercive (Regulatory) Drivers as independent variables to explain the variation in Toxic Release Inventory Emissions (TRI) across the 50 states in the years 1995 through 2004. Our final model indicates that the number of ISO 14001 certified chemical companies contributes to lower TRI emissions and that higher NGO membership contributes to higher TRI emissions: NormalizedTRIdata = 213,051.5 1,125.7*ISOcertifiedcompanies + 4.05* NGOMembers. This model highlights the action channels that most influence companies level of TRI emissions. Our hope is that it may help corporate executives and activists understand how to use their time and resources to effectively influence the environmental performance of chemical manufacturing companies.

Hoffman, Andrew J. Competitive Environmental Strategy: A Guide to the Changing Business Landscape. 2000. Island Press, Washington, D.C. p. 29

PROBLEM DEFINITION Corporate activity can significantly impact the environment. As corporate stakeholders become increasingly concerned with this impact and consequently take action against corporations, corporate executives may struggle to understand when it is necessary to take action to reduce their companies environmental impacts. Similarly, stakeholders attempting to influence

corporate environmental behavior may not understand which action channels will most motivate a company or industry. In this paper, we discuss a statistical model we created to quantify the influence a variety of stakeholders have on environmental performance in the chemical manufacturing industry. We hope that our model helps corporate executives and stakeholders understand the action channels that effectively influence the chemical manufacturing industry to lessen its impact on the environment. Though our model focuses on one industry, we hope that future researchers will use our findings as a basis to examine other industries to reach conclusions regarding the actions that are most relevant in different industry contexts.

LITERATURE REVIEW We are not the first researchers to attempt to identify influencers of corporate environmental behavior. Our work is informed by and augments the work done previously in this area, both when examining company behavior broadly and chemical industry behavior specifically. The section that follows is our review of the articles we consider most relevant and an explanation of how they influenced our work. Our model attempts to capture the factors that lead a firm to reduce pollution. The paper, Stakeholders and Environmental Management Practices: An Institutional Framework, by Magali Delmas and Michael W. Toffel, discusses why firms may adopt environmental programs that exceed regulatory compliance. Delmas and Toffel use an institutional theory framework that is similar to the one we adopted (i.e., factors that include government, regulators, customers, competitors, community and environmental interest groups, and industry associations). They propose that a firms organizational characteristics will cause it to interpret pressures from each of these actors differently, which results in firms adopting different policies towards the environment.2 When examining environmental impact reduction in the chemical industry specifically, we looked to the paper Voluntary Pollution Reductions and the Enforcement of Environmental Law: An Empirical Study of the 33/50 Program by Abdoul Sam and Robert Innes, which identifies correlations

Delmas, Magali and Michael W. Toffel. Stakeholders and Environmental Management Practices: an Institutional Framework. 2004. Bus. Strat. Env. 13, 209-222.

3 between pollution reduction, participation in the EPAs 33/50 voluntary pollution reduction program, regulatory enforcement and boycotts. Though the firms examined in this paper came from a variety of fields, not just the chemical industry, we surmised that its conclusions might stand when applied specifically to the chemical industry. Sam and Innes conclude that important factors in a firms reduction of pollution are participation in the voluntary program, government inspections, preempting regulation, and deterring boycotts. Another factor may be a desire to raise competitors costs. They found no statistically significant link between pollution reduction and a firms proximity to the consumer (green marketing) or between pollution reduction and the presence of strict environmental liability.3 Andy King and Michael J. Lenoxs paper, Industry Self-Regulation Without Sanctions: The Chemical Industrys Responsible Care Program, informed our decision to use Responsible Care as an indicator. Using the Environmental Protection Agencys (EPA) Toxics Release Inventory (TRI) data as a measure of environmental performance and weighting chemicals by level of toxicity, King and Lenox tested eight hypotheses about the effect membership in Responsible Care has on both member and non-member chemical companies. Their most striking conclusion was that members of Responsible Care are improving their relative environmental performance more slowly than are non-members. Though the chemical industrys environmental performance improved more rapidly after the inception of Responsible Care, it
3

Sam, A., and R. Innes, "Voluntary Pollution Reductions and the Enforcement of Environmental Law: An Empirical Study of the 33/50 Program," Working Paper, University of Arizona, 2004, p. 13

4 only did so among companies that were not members of Responsible Care. We were interested to see whether our model would also show a negative impact of Responsible Care membership on environmental performance.4 When developing our model, we also reviewed Wilma Rose Anton, George Deltas, and Madhu Khannas paper entitled Incentives for Environmental SelfRegulation and Implications for Environmental Performance. In this paper, Anton et al develop a quantitative model to examine the factors that caused firms to adopt Environmental Management Systems (EMS) and whether the comprehensiveness of an EMS had any impact on the toxic releases of the firms. Like King and Lenox, Anton et al used the TRI as an indicator,

reinforcing our decision to use this data as our dependent variable. Based on their model, they found that the greatest factors influencing EMS adoption were consumer and investor pressure, as well as future liability and the scale of past emissions.5 We were able to include the first two indicators in our own model. In Self-Regulation and Social Welfare: The Political Economy of Corporate Environmentalism, John W. Maxwell, Thomas P. Lyon, and Steven C. Hackett model corporate environmentalism in the context of voluntary adoption of cleaner products or processes that result in a reduction of toxic chemical emissions. Maxwell et al found that company self-regulation is

induced by the threat of government regulation, a factor we consequently

4

King, Andrew A. and Lenox, Michael J., Industry Self-Regulation Without Sanctions: The Chemical Industrys Responsible Care Program. Academy of Management Journal, 2000, Vol. 43, No. 4, 698-716. 5 Anton, Wilma Rose, George Deltas, and Madhu Khanna. Incentives for Environmental SelfRegulation and Implications for Environmental Performance. Journal of Environmental Economics and Management 48, 1, 632-654. 2004.

5 examined in our own model.6

Maxwell, John W., Thomas P. Lyon, Steven C. Hackett. Self-Regulation and Social Welfare: The Political Economy of Corporate Environmentalism. October 2000. Journal of Law & Economics, vol. XLIII.

HOW OUR PAPER DIFFERS FROM PRIOR SCHOLARSHIP Our unique contribution to the research on this topic is the set of variables we used to explain fluctuations in TRI. Seven of the eleven variables we used in our model - press coverage, ISO memberships, SRI participation, public opinion, academic press, NGO membership and legal cases - have seldom been used to explain changes in TRI data over time. We believe that our work demonstrates that these variables deserve further research to fully understand their impact on TRI emissions. Furthermore, we not only explored the

relationship between these variables and the TRI, but also aggregated the data that comprise these seven variables. Because of our research, these variables are now available for the use of future researchers.

THEORETICAL FRAMEWORK We based our analysis of the potential influencers of corporate environmental behavior on the framework proposed by Andrew J. Hoffman in his book Competitive Environmental Strategy: A Guide to the Changing Business Landscape.7 Hoffmans framework posits that the influencers of

corporate environmental behavior can be grouped into four categories: Social Drivers, Resource Drivers, Market Drivers and Coercive (Regulatory) Drivers. Within these four groups reside numerous actors, enumerated below. Social Drivers include Environmental NGOs, the Press, the Courts, Religious Institutions, the Community and Academia Coercive Drivers include Domestic Regulation and International Regimes Resource Drivers include Suppliers, Banks, Shareholders/Investors, Buyers and Insurance Companies Market Drivers include Consumers, Trade Associations, Competitors and Consultants We decided to narrow our focus to one specific industry when building our model because we believe that the level of influence of these actors varies across industries. We decided that the chemical manufacturing industry would be most appropriate for a first attempt at a model because of the availability and comprehensiveness of a dependent variable in the EPAs TRI data, which annually tracks chemical emissions by members of the chemical manufacturing
7

Hoffman, Andrew J. Competitive Environmental Strategy: A Guide to the Changing Business Landscape. 2000. Island Press, Washington, D.C. p. 29

8 industry. We also decided that we would focus our model on companies in the U.S., both because the TRI is only applicable to companies in the U.S. and because quantifying the differences in political systems and actors across countries would expand beyond the scope of an initial model and dataset. Our next step was to pick the actors within each of Hoffmans four categories that we believed would be the most relevant drivers of environmental action in the chemical manufacturing industry. We based our selection on the relationships each of these actors had with the chemical manufacturing industry over time. We explain in detail below the specific reasons we chose to include each actor. The Social Drivers we measured were Environmental NGOs, the Press, the Courts, the Community, and Academia. Recent campaigns by environmental NGOS against chemical manufacturing companies, such as Greenpeaces campaign against Dow Chemical, demonstrate that environmental NGOs are trying to influence chemical manufacturing companies behavior on the environment. We included NGOs as an actor to try to measure whether these actions truly have an impact on company behavior. Press coverage of chemical manufacturing companies

environmental impacts raises awareness of this behavior among the general public. We wanted to determine whether greater publicity of their actions on the environment caused companies to change their behavior.

9 Courts are an actor whose decisions impact the particular company being tried in court. We wondered if an increase in the number of cases against chemical manufacturing companies would cause companies in the industry to take proactive (or reactive) steps to protect themselves from environmental lawsuits. Community members often provide vocal censure of chemical manufacturing companies environmental behavior. We wondered if the opinion and actions of this actor had a measurable impact on the behavior of the industry. Finally, we heard that King and Lenoxs paper on Responsible Care received notice within the chemical industry. We hoped to measure whether academic opinion in general translated into action by chemical manufacturing companies. We did not study religious institutions because after discussions with companies and individuals at the Interfaith Center on Corporate Responsibility, we determined that religious institutions primary influence on company behavior is through shareholder action.

The Coercive Driver we measured was Domestic Regulation. Maxwell, Lyon and Hackett found that domestic regulation was the biggest driver of TRI emission reduction; we wanted to verify this

10 finding in our model. We did not study international regimes as a coercive driver because we focused exclusively on the chemical manufacturing industry in the U.S; while international regimes may influence the action of U.S. chemical companies, we considered these regimes outside of our project scope. In the category of Resource Drivers, we looked at investors and buyers. Conversations with individuals in the industry indicated to us

that chemical manufacturing industry executives view investors as one of the main influencers of corporate behavior.8 We wanted to see if our model verified that assumption. We viewed buyers as an important group because they are the

immediate customers of the chemical manufacturing industry. As the purchasers of the companies products, we reasoned that buyers environmental beliefs and actions should have significant affects on company environmental behavior. We decided not to examine suppliers to the chemical industry in

our model. Because these suppliers are so dispersed around the world and supply such basic elements, we thought it unlikely that they would have a unified impact on the industry that we could measure in our model. This decision was specific to the chemical manufacturing industry in other industries suppliers may have
8

Phone conversation with Scott Noesen, Director of Sustainable Development, Dow Chemical, October 31st, 2006

11 greater influence. Our research indicated that banks do not currently take

environmental impacts into consideration when lending to chemical manufacturing companies, except with regard to how they impact risk. We thus excluded them from our model. We came to a similar conclusion about insurance companies.

Because both banks and insurance companies measures of risk take so many factors into consideration, we did not think we could measure their specific relationship to the environmental behavior of chemical manufacturing companies. The Market Driver we examined was the chemical manufacturing industry trade association Responsible Care. We chose Responsible Care because of King and Lenoxs

findings about its influence on the chemical industry, which we hoped to replicate or disprove. We chose not to include consumers as a variable because the

chemical manufacturing industry is largely a business-to-business (B2B) supplier, not a business-to-consumer (B2C) retailer to individual consumers; therefore, we did not think consumer pressure would be significant. In addition, we thought that the chemical manufacturing industrys B2B model meant that the buyer category would more directly capture consumer pressure applied to this industry.

12 Finally, we decided not to include competitor pressure as a

separate variable because we studied the industry as a whole (as opposed to one individual company). We also determined that

competitor pressure is likely captured within the trade association data.

METHODOLOGY Once we selected the group of actors, we attempted to find and/or create numerical datasets indicative of the influence of each of these actors on the chemical manufacturing industry. We defined the chemical manufacturing

industry as all companies designated as code 28 by the U.S. Department of Labors Standard Industrial Classification System (SIC). We sought to measure each actor over the ten-year period from 1995 to 2004 on a state-by-state basis. We chose 2004 as the upper limit because when we began working on this project in 2005, 2004 was the most recent year for which full data was reliably available. We tried to get as much data on a state level as possible. However, with certain variables it was only possible to obtain data on a national level. We indicated in the text the variables collected on a state level and those collected on a national level. Each variable is discussed in detail in the text below. Appendix 5: Sources of Data for Variables provides detailed

information on how to access the data comprising each variable. Information availability necessarily constrained our choice of data; most of the data we sought did not exist in extant sets and many of the extant datasets we found were incomplete. In some cases, the constraint of building our own datasets meant that the dataset we chose to represent a certain indicator was the dataset we could obtain, rather than the ideal one. For information on

alternative variables that those who build on this research might obtain, see the section Suggested Improvements to Our Model for Future Researchers.

13

DISCUSSION OF VARIABLES Following is an explanation of how and why we selected the data that comprise each of our variables. Social Drivers Environmental NGOs Money raised, actions taken (boycotts, press, etc), and NGO membership are all potential measures of pressure put on chemical manufacturing companies by environmental NGOs. Of these, we selected NGO membership as our variable because we thought this statistic best reflected the magnitude of the constituency represented by the NGO. We requested membership data by state for the years 1995 2004 from the following seventeen largest environmental NGOs in the U.S.: The Nature Conservancy, National Wildlife Federation, World Wildlife Fund, Greenpeace USA, National Audubon Society, Sierra Club, Environmental Defense Fund, Wilderness Society, Natural Resources Defense Council, Water Environment Federation, National Parks Conservation Foundation, Air & Waste Management Association, Defenders of Wildlife, Friends of the Earth, Izaak Walton League of America, Citizens for a Better Environment, and Environmental Action. Of these seventeen, the Sierra Club was the only NGO that was able to provide us with a complete dataset. We therefore used the Sierra Club membership data as our indicator of NGO pressure.

14

15 The graph below illustrates NGO membership by year across all states. In addition, the first graph in Appendix 6 entitled NGO Membership by State illustrates the NGO membership for each state (in alphabetical order). As both of these graphs illustrate, NGO membership is relatively consistent year over year for most states and many states have low total membership. We therefore suspected this variable would not have a significant impact on the model.
NGO Membership by State: 1995-2004

198590
CA CA CA CA CA CA CA CA CA

CA

NGO Members

Number of NGO Members

NY NY WA IL TX FL PA NJ OH MA MN CO OR NC MD MI VT WV GA CT MO AZ HI IN VA TN NM UT SC KY KS IA OK NV NH ME LA AL RI ID MT AR AK WI NE WY SD ND MS DE NY IL WA TX PA FL NJ OH MA CO NC MI OR MN MD VA AZ GA MO WI CT NM IN UT TN HI VT SC NV ME KY KS IA OK NH LA AL RI ID WV MT AR AK NE WY SD ND MS DE NY IL FL WA TX PA MA NJ MI OH MN NC CO OR VA MD AZ MO WI GA CT NM IN UT TN HI SC NV KY KS IA VT NH ME LA RI ID WV OK MT AR AL NE WY SD ND MS AK DE NY NY FL IL TX MA WA PA NJ MI OH NC MN CO OR MD VA GA AZ CT WI MO NM IN UT TN IA VT SC NV KY KS HI OK NH ME LA AL RI WV MT AR AK NE DE ID WY SD ND MS MA FL PA WA TX IL NJ MI OH MN CO NC OR MD AZ VA GA CT WI MO TN NM IN IA UT SC NV NH ME KY KS AL NE HI VT OK LA RI WV MT MS AR AK DE ID WY SD ND NY MA PA FL TX WA IL NJ MI MN CO OH NC WV OR VA MD GA WI AZ CT MO NM IN TN UT SC NH IA NV KY KS AL HI VT OK ME LA RI MT MS AR AK NE DE ID WY SD ND FL MA WA PA IL TX NJ OR MN MI CO OH NC VA MD GA WI AZ MO CT IN TN NM UT SC NH IA HI NV ME LA KY VT KS AL RI OK MT AR ID WV MS AK NE DE WY SD ND

NY FL WA PA MA IL OR TX NJ MN CO MI OH NC VA MD GA WI AZ CT MO IN TN NM SC NH ME IA UT NV KY KS HI VT LA AL RI OK MT AR NE ID WV MS AK DE WY SD ND

NY NY WA FL TX PA MA IL OR MN CO NJ MI OH VA NC MD GA CT WI AZ MO IN TN IA SC NV NM NH ME UT LA KY KS AL HI VT RI OK MT AR NE ID WY WV MS AK DE SD ND FL WA PA MA IL TX OR NJ MN CO MI OH NC VA MD WI GA AZ CT MO IN TN NM IA SC NH UT NV ME KY KS HI LA AL VT OK RI ID WV MT AR AK NE DE WY MS SD ND

172

1995

YEAR

2004

Year

Press To measure Press influence, we collected the number of articles published each year about the chemical industrys impact on the environment. We

hypothesized that the number of articles might positively correlate with industry action; more articles about the chemical industrys environmental impact could cause more public scrutiny, which in turn might spur companies to act to

16 preempt reputation damage. We searched for articles within the Lexis-Nexis Environmental Database. This database contains searchable information from the following

environmental sources: journals, conference papers and proceedings, federal and state government reports, major daily newspapers, consumer and trade magazines, newsletters, law reviews, administrative codes, case law, regulatory agency decisions and waste site and hazardous material data.9 We conducted our search within all sources in the Environmental News category.10 We chose to search in the Lexis-Nexis Environmental Database rather than the popular press because we considered these environmental news sources as feeders for major newspapers stories on the environment. We modeled our search terms on the methodology used by Christine R. Ader in her article, A Longitudinal Study of Agenda Setting for the Issue of Environmental Pollution.11 This article outlined a study conducted to measure the influence of the New York Times on public opinion. Ader measured

environmental articles by searching the Times index for the years 1970 to 1990 using the following search terms: air pollution, environment, water pollution and waste materials and disposal. She chose those terms in response to an article she read in 1988 that cited disposal of wastes, air quality and water quality as the three most salient environmental issues.
9

University of San Francisco. Gleeson Library, Geschke Center. Lexis Nexis Environmental description. http://138.202.192.14/library/databases/envuniverse.html (viewed October 24, 2006) 10 Note: Since our project began, the school discontinued its subscription to the Lexis-Nexis Environmental database. 11 Ader,Christine R. A Longitudinal Study of Agenda Setting for the Issue of Environmental Pollution, Journalism & Mass Communication Quarterly 72 (Summer 1995):303.

17 We used Aders search terms, but paired each of them with the phrase chemical industry to ensure we retrieved only articles relevant to our model. We used the phrase chemical industry instead of chemical manufacturing industry because we thought that the press would be unlikely to make a distinction between the two in its coverage. We also divided Aders last term into two terms. Our final search term pairs were: air pollution and chemical industry; environment and chemical industry; water pollution and chemical industry; waste materials and chemical industry; and waste disposal and chemical industry. We searched for each of these five pairs of terms from 1/1 to 12/31 for each of the ten years of our model. After obtaining the raw number of articles that each of these pairs of search terms returned, we sorted all of the articles and removed any duplicates. If the same article appeared in a source in different editions or on different days, each of these occasions was counted as a separate article. Because we focused on the chemical manufacturing industry in the U.S., we removed from our count any articles that primarily concerned a foreign country or came from a source based in a foreign country. The number of articles remaining after removing the duplicate and foreign articles was the number we used to represent press influence from that year. Because these sources were national instead of local, we used the same number for each state for that year. As you can see from the graph below, the number of articles shows a clear upward trend, increasing dramatically starting in the year 2001. Unfortunately, we could not identify a specific reason for the increase in press articles at the

18 time, although reasons could range from the policies towards the environment of the administration at the time to a continuing increase in the total number of news publications.
Environmental Press for the Chemical Industry: 1995-2004

1159

Press Articles

Number of Press Articles

482 1995 YEAR 2004

Year

19 The Courts We thought that the number of lawsuits involving chemical manufacturers was the best representation of court pressure on chemical firms to improve their environmental performance. It made intuitive sense to us that an increase in the number of lawsuits against chemical manufacturers would motivate firms to lower their TRI emissions to avoid costly legal battles. To find data on the number of environmental lawsuits against the chemical manufacturing industry, we used the Westlaw Environmental Database, which contains only lawsuits about environmental issues. We searched the database on a state by state basis, using the search term chemical and the date parameters January 1, 1995 to January 1, 2005. We manually sorted the results of this search and counted only those cases that involved a chemical manufacturing company, as defined by SIC Code 28. We used each cases coding by year to create individual data points for each state and each year. As you can see from the graph on the following page, the total number of cases that met our criteria was small. Further, the graph in Appendix 6 entitled Westlaw Chemical Industry Cases by State illustrates that there were no cases that met our criteria for most of the states in the years we measured. Finally, the largest number of cases in any individual year for any state was six.

20
Chemical Industry Cases by State: 1995-2004
6
LA LA

NY

Westlaw
OH IL OH IL PA FL LA CT LA AL TX OH

Number of Cases by State

CA

WA TX SC PA MA CA NJ MI IA

WY MO ME KS NJ MI

PA NY NC MA CT CA AL FL

NY LA CA WI NJ MI

TX MD LA NJ MI IL

TX SC OH ME CA NE MI IN

TX PA MT CT CO CA AR DE

NY CT CA NE

TX GA CO CA MI

CO DE IL

WY WV VT VA UT TN SD OR OK NV NM NH ND NC MT MS MO MN ME MD LA KY KS GA CT CO AR AL AK WI NE FL DE AZ RI IN ID HI

WV WA VT VA UT TX TN SD SC PA OR OK NY NV NM NH ND NC MT MS MN MD MA KY GA CT CO AR AL AK WI NE FL DE AZ RI IN ID IA HI

WY WV WA VT VA UT TX TN SD SC OR OK OH NV NM NH ND MT MS MO MN ME MD LA KY KS GA CO AR AK WI NJ NE MI DE AZ RI IN IL ID IA HI

WY WV WA VT VA UT TX TN SD SC OR OK OH NV NM NH ND NC MT MS MO MN ME MD MA KY KS GA CT CO AR AL AK NE DE AZ RI IN IL ID IA HI

WY WV WA VT VA UT TN SD SC PA OR OK OH NY NV NM NH ND NC MT MS MO MN ME MA KY KS GA CT CO CA AR AL AK WI NE FL DE AZ RI IN ID IA HI

WY WV WA VT VA UT TN SD PA OR OK NY NV NM NH ND NC MT MS MO MN MD MA KY KS GA CO AR AL AK WI NJ FL DE AZ RI IL ID IA HI

WY WV WA VT VA UT TN SD SC OR OK OH NY NV NM NH ND NC MS MO MN ME MD MA KY KS GA AK WI NJ NE MI FL AZ RI IN IL ID IA HI

WY WV WA VT VA UT TN SD SC PA OR OK OH NV NM NH ND NC MT MS MO MN ME MD MA LA KY KS GA CO AR AL AK WI NJ MI FL DE AZ RI IN IL ID IA HI

WY WV WA VT VA UT TN SD SC PA OR OK NY NV NM NH ND NC MT MS MO MN ME MD MA KY KS CT AR AL AK WI NJ NE FL DE AZ RI IN IL ID IA HI

WY WV WA VT VA UT TX TN SD SC PA OR OK OH NY NV NM NH ND NC MT MS MO MN ME MD MA LA KY KS GA CT CA AR AL AK WI NJ NE MI FL AZ RI IN ID IA HI

1995

YEAR

2004

Year

The Community We chose public opinion about the environment as a proxy to quantify community pressure. Of all of our actors, public opinion (community) may be the most challenging to quantify. This is due both to the lack of comprehensive databases spanning our timeframe and to the fact that past studies show that public opinion may be a flawed measure in and of itself because it often has low correlation with actions. In our model, we chose to measure public opinion via the Gallup Poll. The Gallup Poll was the only data source we could find that had a comprehensive database on public opinion about the environment that spanned our models timeframe. Since 1984, the Gallup Poll has asked the following question: With which of these statements would you most agree: Protection of the

environment should be given priority, even at the risk of curbing economic

21 growth (or) economic growth should be given priority, even at the risk of harming the environment?12 We used the percentage of respondents who

answered that protection of the environment should receive priority even at the risk of curbing economic growth as our indicator. Because no percentage was available for 1996, we calculated this data point by taking the average of 1997 and 1995. As illustrated in the graph below, the affirmative answers decreased dramatically over the past ten years. This seems surprising, given the

increasing attention brought to environmental protection in recent years. We suspect this is correlated with the state of the economy, which experienced a dramatic downturn that coincides with the downturn on this graph.
Percentage of Gallup Poll Respondents Stating Environment Should Receive Priority: 1995-2004

69

Public Opinion on the Environmen

Number of Respondents
47 1995 YEAR 2004

Year

12

http://www.galluppoll.com/content/default.aspx?ci=1615 viewed on November 10, 2006

22 Academia We found two potential indicators of academic pressure on the chemical manufacturing industry. The first is the number of academic degrees granted in environmental fields. The second is the number of academic articles dealing with the chemical industry and the environment. We chose to use the latter variable. We did not think that the number of academic degrees in the

environmental field would indicate academic pressure because so many individuals do not pursue careers in the field in which they receive their degrees. We therefore thought academic articles represented a better indicator of academic thought and leadership on this issue. To obtain the number of academic articles, we searched the Jstor database by year, using the same search terms we used to collect the press articles: air pollution and chemical industry; environment and chemical industry; water pollution and chemical industry; waste materials and chemical industry; and waste disposal and chemical industry. We then sorted the articles for duplicates to come up with a single number of articles for each year. The data were measured on a national level and are detailed in the graph on the following page.

23
Academic Articles on the Chemical Industry and the Environment: 1995-2004

74

Academic

Number of Articles

7 1995 YEAR 2004

Year

An interesting result of this search is that the number of academic articles about the chemical industry declined significantly from 1995 to 2004. In 1995, there were 60; in 2004, there were 7 (see graph below). We hypothesized that this decrease could be due to a wide range of factors, such as a specific action taken by the chemical industry (such as increasing self-regulation policies), a specific environmental event that occurred in the industry, a general lack of support for research in this area by universities, or perhaps a decrease in the number of academic positions dedicated to this type of research. Unfortunately, examining the titles of the articles and general news feeds regarding the chemical industry did not yield any insight into the cause for the articles decrease. We were unable to pursue the hypothesis that universities could be

24 decreasing support for academic research in this area. Coercive Drivers Domestic Regulation Domestic regulation is one of the few categories with a readily available relevant and complete dataset. We measured domestic regulation on a state-bystate basis using the voting results compiled by the League of Conservation Voters (LCV) in the groups annual National Environmental Scorecards.13 The LCV Scorecards are available on the groups website dating from 1971. The scorecards give each member of Congress a percent score based upon how they voted on what the LCV considers the most important environmental issues that appeared before Congress that year. To obtain one number per state per year for our model, we averaged the scores of all the Congressional officials from that state. Though LCV ratings are not a direct measure of specific environmental regulations in a particular state, they do provide an indication of the overall importance given to environmental protection in that state. An LCV rating reflects the environmental position of the states Congresspeople, which is usually a reflection of the environmental position of a majority of voters in that state. The graph below illustrates the large spread of LCV ratings across states. In addition, the graph in Appendix 6 illustrates that LCV rating is one of the
13

Available at www.lcv.org/scorecard/. The League of Conservation Voters (LCV) calls itself the independent political voice for the environment. Its website states that its mission is to advocate for sound environmental policies and to elect pro-environmental candidates who will adopt and implement such policies. In addition to producing yearly scorecards, the LCV campaigns to defeat anti-environment candidates, educates the public about environmental issues, and aims to train the next generation of environmental leaders.

25 variables that changes most from state to state. This led us to predict that LCV scores would have significant explanatory value in the model.

LCV Scores by State: 1995-2004

98
CT VT MA RI MD HI NJ WV ND WI CA IL AR DE MT NY ME MN MI SD NV NE FL VA NM OH SC WA PA GA OR IA LA CO KY AL KS NH MO IN TX MS TN NC AZ WY UT OK MA RI VT HI CT MD NJ WV WI CA DE ND IL ME AR NY MN SD MT NV MI OH NE FL VA SC PA NM WA GA IA OR LA KY CO MO IN TN NH KS TX AL AZ NC MS OK UT WY AK ID

MA CT NJ VT DE RI MD WI CA ME IL HI NY OR WV NV MN OH NH MI AR WA VA FL SD ND SC IA PA LA TN NE GA NM KY IN CO AZ NC MO MT TX UT KS MS AL AK OK WY ID

MA CT NJ VT WI RI MD HI CA IL ME DE WV OR NY MN MI NV SD OH ND AR FL WA VA NH IA SC LA PA NE GA NM IN TN NC KY CO MO MT AZ TX UT MS AL KS ID OK AK WY

VT MA RI CT NJ HI ME CA NY WI MD DE OR ND IL NV MN WV WA MI VA IN SD FL GA IA NC PA NM NE SC NH OH MT MO KS CO AR TX MS TN AZ LA UT AL KY WY OK AK ID

VT RI MA CT NJ

VT MA NJ CT

HI NY MD WI DE CA ME OR IL NV ND GA WV MN MI WA IN VA SD IA NC FL NM SC PA NE OH NH LA CO AR MT MO KS TX TN MS AZ AL KY UT AK WY OK ID

HI MD MN RI NY DE WI MI CA WA ND WV ME IL FL OR

VT MA RI NJ CT NY MN ME MD HI CA WI DE WA MI OR WV ND IL NV FL

VT RI MA NJ WA NY MD DE ME HI CT WI CA OR MI WV IL FL ND MN NH

VT HI CT MD NJ RI NY ND WA WV CA MA WI DE ME MI OR FL AR NM IL LA SC MN SD IN NV IA AZ MS TX PA NC TN NH MT MO OH NE VA KS KY GA CO AL UT OK ID WY AK

LCV Rating

LCV Rating
0

NV IA GA NC IN NM SD SC AR PA MO NH TN OH CO VA LA MT TX MS AZ UT NE KS AL OK KY AK ID WY

IA NH PA NM SD SC NC MO IN GA AR CO OH AZ TX TN VA UT MT MS LA KS AL NE AK OK KY WY ID

AR NV NM IN IA SD SC NC AZ TN PA OH CO VA TX MO MS LA GA UT KS AL NE MT OK KY AK WY ID

AK ID

1995

YEAR

2004

Year

26 Resource Drivers Socially Responsible Investing We chose socially responsible investing as an indicator because we believe it reflects investor pressure on companies to behave in more environmentally responsible ways. Our data came from the most recent version of the Social Investment Forum Industry Research Programs biennial Report on Socially Responsible Investing Trends in the United States. The 2005 version of the report contains all the information the Social Investment Forum compiled since it first published the report in 1995.14 The report contains a number of possible indicators relating to socially responsible investing (SRI), such as the dollar value of total SRI investments and the number of socially screened mutual funds. From these data, we chose to use the percentage of total investment funds identified as socially responsible as our indicator. We thought this metric gave the best measure of the

importance of SRI investing relative to the total investment market. Because the Forums report only contains data for every other year, we carried over the previous years number for the years for which we did not have data (for example, we used the same percentage for 2003 and 2004, which is the percentage for 2003). Because this is a national variable, the numbers varied by year, but were consistent across states.

14

2005 Report on Socially Responsible Investing Trends. SIF Industry Research Program. Social Investment Forum. Washington, DC. 2005

27 As the graph below indicates, after a spike in investing around 2000, the percentage of SRI investing seems to be decreasing.
SRI Investment: 1995-2004

13.25

SRI Investment
8.65 1995 YEAR 2004

SRI

Year

Buyers Buyers frequently pressure companies by requiring adherence to specific standards. The ISO 14001 standard is a measure of companies environmental management systems; buyers increasingly require it when choosing their chemical suppliers. We used QSU Publishing Companys directory of ISO 14001 registered companies to gather information regarding companies ISO 14001 certification.15 This database contains information regarding all ISO 14001 companies and the dates they initially received certification. We

conducted a search of the database to identify the number of currently certified

15

QSU Publishing Company. ISO 14001 Registered Company Directory North America. 2003, Volume 4, Number 2.

28 U.S.-based companies classified under SIC Code 28. We then reviewed each record to determine initial date of certification. In our model, we aggregated the number of companies certified; the total number of companies in the chemical manufacturing industry that were certified was our final variables for each year (see graph below). We also measured the number of companies on a national level. As you can see, there was a sharp increase in ISO 14001 certification from 1997 through 2002; the number of certified companies leveled off in the past few years.
ISO Certified Chemical Companies: 1995-2004
92

Number of ISO Certified Companies

# of ISO certified companies

1 1995 YEAR Year 2004

Market Drivers Trade Associations The American Chemistry Council (ACC) is the chemical industrys trade association in the U.S. To be a member of the ACC, a chemical company must also enroll in the Responsible Care program. Responsible Care is a global

29 initiative under which companies work together to continuously improve their health, safety and environmental performance, and to communicate with stakeholders about their products and processes.16 In the words of the industry, The Responsible Care ethic helps our industry to operate safely, profitably and with due care for future generations, and was commended by UNEP as making a significant contribution to sustainable development at the World Summit on Sustainable Development in 2002.17 We chose to use membership in Responsible Care to represent trade association pressure because we thought the number of companies that belonged indicate the support the industry gives to Responsible Cares goals. We obtained membership numbers for the years 1995 through 2001 from Michael L. Barnett, PhD, a professor at the College of Business Administration at the University of South Florida, who obtained them directly from the Chemical Manufacturing Associations member directory. We obtained

Responsible Care membership numbers from 2003 2006 from Debra Phillips, the Managing Director of Responsible Care. For the missing year of data (2002), we took the average of the 2001 and the 2003 numbers. As demonstrated by the graph on the following page, membership in Responsible Care seems to decline year over year. However, one must be careful when interpreting this data. The total number of firms in the industry is decreasing due to consolidation.18 Thus the drop in membership in Responsible Care may be simply a reflection of the shrinking number of firms, rather than an
16 17

http://www.responsiblecare.org/ http://www.responsiblecare.org/ 18 Phone conversation with Debra Phillips, October 12, 2006

30 indication of a lapse in enthusiasm for the program.

Responsible Care Membership: 1995-2004
194

Responsible Care Members

Responsible Care Members

140 1995 2004 YEAR

Year

Other Components of the Model Toxic Release Inventory Data The EPA manages the TRI and the Emergency Planning and Community Right-to-Know Act of 1986 (EPCRA) established the program. The TRI

database contains details regarding almost 650 chemicals that a variety of industries and federal facilities manage through disposal or other releases.19 We chose this variable as our Y variable because it provides a very detailed record of environmental impact, as measured by chemical waste management, for the chemical manufacturing industry. It is a comprehensive variable because of the federal requirement that all members of the industry report on their emissions. Additionally, the data are organized by facility and location, which enabled us
19

http://www.epa.gov/tri/tridata/tri03/index.htm#what

31 to gather emissions on a state level. In order to compile the appropriate data, we first selected for the 1995 list of chemicals. The EPA made significant changes to reporting requirements in 1995. The addition of new chemicals to the TRI dataset, and the removal of others, makes it difficult to compare chemical lists from pre- and post-1995 without selecting for only pre-1995 chemicals. We screened the data to select for companies classified under SIC Code 28. We then collected the data state by state for each of the ten years we examined. Finally, we normalized the data by dividing total emissions for each state by the number of factories in each state. This allowed us to control for dramatic changes in the data that may have resulted from the elimination or addition of manufacturing facilities in a state. The graph on the following page shows a surprising lack of variation in emissions between states. In addition, the graph in Appendix 6 entitled

Chemical Industry TRI Data by State shows a lack of change in emissions over time within each state. This lack of variation in our y-variable may explain some of the trouble we had creating a model with high explanatory value.

32 Chemical Industry TRI Emissions by State: 1994-2005

4.7e+06
AK

AK

AK

Normalized TRI Data

WY ID LA TX TN MS AL WV OK KY NV AR FL NC KS DE VA MO MI SC OH IA SD MD CT IN GA NE IL WA UT PA OR NY NJ MT MN CO CA WI AZ VT NM NH ND ME MA RI HI

TRI Emissions
ID ID WY LA TX TN AL MS WV OK FL KY AR NC NV KS DE VA OH SC MO IA SD MI MD NE IN IL CT WA OR NY GA NJ VT UT PA MN CO CA WI AZ NM NH ND MT ME MA RI HI

AK AK ID WY LA MS TX TN AL FL WV KS OK NV IA KY VA AR OH NC SC MO NE SD MD IL WA GA CT MI DE IN OR NJ UT PA NY NM MN MA CO CA WI VT NH ND MT ME AZ RI HI WY LA MS TX TN KS WV AL FL IN OK NV KY IA DE VA NC AR SC OH MD MO NE GA CT MI WA OR NJ PA NY MN AZ IL VT UT SD NM NH ND MT ME MA CO CA WI RI HI MS LA AL TX TN FL KS WV OK DE VA KY AR NV MD OH NE IA SC MO NC IN GA MI IL WA NY CT NJ UT PA OR NM MN ME CA AZ VT SD NH ND MT MA CO WI RI HI AK AK ID AK WY ID AK WY WY ID MS LA TX TN AL FL WV VA OK KS DE NV MD KY OH AR SC NE IA NC MO GA MI IN IL WA CT NJ VT UT PA OR NY NM ND MN SD NH MT ME MA CO CA WI AZ RI HI WY MS LA TN ID TX NE WV OK FL AR OH KS AL MD DE SC KY VA MO NC IA ND GA IN IL NV CT NJ MI WA UT PA OR NY MN CO VT SD NM NH MT ME MA CA WI AZ RI HI WY MS LA ID TN TX NE FL OK AR WV DE IA AL SC KS NH MD OH KY NV MO VA NC GA ND IN IL UT CT NJ MI WA PA OR NY NM MN VT SD MT ME MA CO CA WI AZ RI HI AK

LA WY MS TX TN AL WV FL OK KY KS DE NV AR NC VA OH MO SC IA SD MD WI NE MI IN IL WA NM GA CT NY NJ UT PA OR MN MA CO CA VT NH ND MT ME AZ RI HI

TN LA MS TX ID FL OK WV AR VA DE AL KY NV MD OH KS SC MO NE NC IA GA MI IN IL WA CT NJ UT PA NY NM ND MT MN VT SD OR NH ME MA CO CA WI AZ RI HI

1995 YEAR

2004

Year

Industry Lobbying Dollars We chose to measure industry lobbying dollars as an indicator of how much pressure corporations feel to regulate. Our hypothesis was that the greater the pressure companies felt to regulate, the more financial resources the industry would devote to lobbying the government. We obtained chemical industry lobbying dollars for the years 1998 through 2004 from the Center for Public Integrity, whose figures are based upon Senate Office of Public Record filings.20 We obtained the number for 1997 from Open Secrets.org, a campaign finance website.21 We were unable to obtain information for the years 1995 and 1996, which is likely due to the fact that the law requiring Senate Office of
20 21

http://www.publicintegrity.org/lobby/profile-pf.aspx?act=industries&year=2003&in=15 http://www.opensecrets.org/pubs/lobby00/industry.asp

33 Public Record filings to be made public was not passed until 1996. As the graph below demonstrates, there seems to be no clear trend in this variable. We therefore hypothesized that the industry lobbying dollars would not have a significant impact in our model.
Chemical Industry Lobbying Dollars: 1997-2004
27.89

Lobbying $

Lobbying $ Amount

22.6 1997 YEAR 2004

Year

Median Income for Family of Four The economic situation in each state in each year may significantly impact TRI reductions. We used median income for a family of four per state by year as a variable that would account for this economic variability. We obtained this data from the U.S. Census. As is evident in the graph on the following page, median income trends upward consistently across all states. In addition, the graph in Appendix 6 entitled Median Income for a 4-Person Family illustrates the upward trend by

34 state. This uniformity across the variable led us to hypothesize that median income would not be a significant factor in our model.
Median Income for 4-Person Family: 1995-2004
92205
NJ CT CT NJ MA MD CT CT MD NJ MA MN DE CO RI IL HI AK WA NH VA OH MI PA WI NY NC GA OR IN CA UT KS ND NV MO VT IA FL SC WY TX NE AL ME TN SD AZ ID LA KY OK MT AR MS NM WV CT NJ MD MA AK MN HI IL NH DE MI RI VA WI CA CO WA NY PA NV IN IA GA VT ME UT KS MO OH ND SC FL WY NC OR TX AZ KY AL SD OK TN NE MT LA MS ID AR WV NM NH MN DE MI RI IL VA AK CO WI HI PA NY WA CA IN OH MO GA NV VT OR IA UT ND NC KS SC ME WY AZ FL SD TN ID TX NE KY AL OK LA NM MS WV MT AR DE NH MN AK RI VA MI CO IL NY PA HI WA WI OH CA IN VT KS IA MO NC SD GA SC NV UT OR WY ME FL TX ND AZ TN NE AL KY OK LA ID WV MT AR MS NM NH MN AK DE IL CO MI HI RI VA WI WA CA NY PA OH IN VT KS IA GA OR NC UT MO NV ME ND FL WY NE AZ TX SC TN SD ID KY AL LA MT OK AR NM MS WV MD CT NJ MA NJ CT MA MD MA MD NH MN DE IL AK VA CO HI RI NY WA WI PA MI CA OH VT IN KS IA MO NC NV GA UT OR ME SD FL AZ NE ND SC WY AL TN TX ID KY LA OK MT AR MS WV NM CT NJ MD MA NH HI RI VA MN AK DE IL NY WA MI VT WI CA CO PA OH IA KS ME NE IN GA MO WY FL OR NV AZ SD ND NC UT SC TX TN LA MT KY AL OK ID AR WV NM MS

36828

Dummy variables Some of the variation in TRI emissions per state may come from the fact that some states, because of leadership or other factors, are more environmentally conscious than others. To control for variation that might be inherent to the states, we created a dummy variable for each state.

Median Income

CT NJ CT NJ MD MA AK HI DE NH MN IL MI MA AK MD HI DE NH MN MI IL RI CO CA PA VA WA IN WI NY OH NV MO VT NC KS GA ME IA SC WY TX NE OR UT ND TN AZ SD KY AL FL ID OK LA MT WV MS NM AR

CA RI WA CO OH PA NY WI VA NV GA IN IA VT NC KS OR WY MO ME UT NE FL AZ TN SC TX ND MT AL ID SD OK LA KY WV AR MS NM

NJ MD MA DE RI MN NH CO HI IL AK MI WA WI VA OH NY PA CA OR IN NE NV KS MO IA VT NC GA UT FL SC WY AL TN ME TX AZ ND SD ID LA KY OK WV MT MS NM AR

1995

Year

YEAR

2004

CREATING THE MODEL We used linear regression as the tool to determine the explanatory value of our variables. The nature of our data dictated that we use a fixed effects model,22 which we used in combination with a robust regression to correct for heteroskedasticity. Though we had concerns about the explanatory power of some of our variables, we thought it was important to test all of them in our model rather than reject them immediately. What appeared to us to be a poor variable may have had explanatory value in combination with the other variables. An initial regression of all of our variables against the normalized TRI data produced a model with variables that differed in their explanatory value. Four variables - press articles, SRI, NGO members and ISO certified companies had t-stats where 1< t <-1 and p-values 0.300 p 0.142. The seven other variables LCV rating, lobbying dollars, public opinion, academia, Responsible Care, Westlaw and median income - had t-stats where 1 > t > -1 and p-values where 0.825 p 0.376. The model had an overall r-squared of 0.018. (See Appendix 1: Model with all Variables). A correlation analysis of our model revealed high positive and negative correlations between a number of our variables (see Table 1: Correlation Among Variables, page 37). Many of these correlations ran counter to

reasonable explanations, which made us doubt the usefulness of some of our

22

We used a fixed effects regression model because running a Hausman test revealed Prob>chi2 = 0.0629, indicating that a fixed effects model is the appropriate model for these data.

35

36 variables.23 For example, it was surprising that the number of articles published in the press about the chemical industry and the environment would decrease membership in Responsible Care. Intuitively, it would make sense for the relationship to run in the opposite direction: more press attention to environmental concerns in the chemical industry should increase membership in Responsible Care. Similarly, it seemed to run counter to expectations that the number of ISO certified companies would be so strongly negatively correlated with public opinion on the environment, academic articles and Responsible Care membership. An additional consideration with the Responsible Care

variable is that it is not controlled for industry consolidation and thus may be misleading (see Discussion of Variables). Because the Responsible Care

variable also had a low t-stat (0.34) and a high p-value (0.735) we dropped it from the model. The high t-stat (-1.04) and relatively low p-value (0.300) for ISO certified companies led us to keep that variable in the model.

23

Many of our unrelated variables happen to be similar values in the same year. For example, in 2003, the amount of lobbying dollars was $23.05 million and the number of academic articles was 21. In 2001, the number of ISO certified companies was 54 and public opinion on the environment was 57%. In 1999, the number of academic articles was 66 and public opinion on the environment was 66%. Thus the variables may seem artificially highly correlated when in fact they coincidentally happen to have the same, or similar, values.

37 Table 1: Correlation Among Variables

| pressa~s lcvrating sri lobbying public~t academic ngomem~s respon~s -------------+-----------------------------------------------------------------------pressartic~s | 1.0000 lcvrating | 0.0072 1.0000 sri | -0.0004 -0.0164 1.0000 lobbying | -0.1456 0.0033 0.0250 1.0000 publicopin~t | -0.6510 -0.0002 -0.0987 0.7705 1.0000 academic | -0.7361 -0.0061 -0.1904 0.6785 0.9662 1.0000 ngomembers | 0.0583 0.2563 0.0326 -0.0415 -0.0703 -0.0775 1.0000 responsibl~s | -0.8964 -0.0031 -0.1847 0.4331 0.8832 0.9318 -0.0754 1.0000 westlaw | -0.0607 -0.0016 -0.0004 -0.0170 0.0195 0.0328 0.1857 0.0509 medianinco~s | 0.2646 0.5129 0.2243 -0.1909 -0.3370 -0.3666 0.2067 -0.3626 ofisocerti~s | 0.6623 -0.0020 0.2966 -0.7299 -0.9469 -0.9601 0.0766 -0.8997 | westlaw median~s ofisoc~s -------------+--------------------------westlaw | 1.0000 medianinco~s | 0.0070 1.0000 ofisocerti~s | -0.0182 0.3814 1.0000

Running the robust model without the Responsible Care variable yielded a higher r-squared (0.0202). However, many variables still had t-stats that were too low and p-values that were too high to indicate a sufficient confidence level in the validity of the variables. Our next step was to run the model again, eliminating the variables with the highest p-values. These were median income and LCV rating. Median incomes high p-value confirmed our suspicion that this variable may have little explanatory value (see Discussion of Variables). Re-running the model without these variables lowered the r-squared to 0.0112, but also raised most of the t-stats and lowered most of the p-values (see Appendix 2: Eight Variable Model). We then removed the variables with the highest p-values westlaw and academic - from the model. This almost cut the r-squared in half, bringing it down to 0.0103. However, it brought all but one t-stat to a level such that 1 < t < -1 and brought all but one of the p-values to a level where 0.293 > p. We

38 eliminated the one variable whose t-stat and p-value did not meet the above criteria, which was public opinion on the environment. This produced a model with an r squared of 0.0105 where all the t-stats were significant and all the pvalues were approaching significance (see Appendix 3: Model with t stats 1 < t < -1). We then experimented with replacing press articles with a new variable, presslag, which was the press articles variable with a built in time lag of one year. We hoped this new variable would capture what we surmised was the lag time between an articles publication and its impact. Our theory was correct; presslag had a t-stat of -1.09 and a p-value of 0.274, which meant it had greater explanatory value than the press articles variable. However, presslag still had the highest p-value of the remaining variables, one that, at 0.274, was too high to give us confidence in the variables explanatory power. Removing presslag from the model decreased the r-squared from 0.0121 to 0.0110, but it also decreased the p-values of the remaining variables NGO members, SRI investing and ISO certified companies to p 0.131. SRI was the variable whose p-value was highest, at 0.131. Removing it from the model gave us our final model: NormalizedTRIdata = 213,051.5 1,125.7*ISOcertifiedcompanies + 4.05* NGOMembers

The model has an r-squared of 0.0114, t-stats of 2.32 and -3.67 and p-values of 0.021 and 0.000 (see Appendix 4: Final Model).

INTERPRETATION OF THE MODEL Our model indicates that a higher number of ISO 14001 certified chemical companies contributes to lower TRI emissions, while a higher number of NGO members contributes to increased TRI emissions. Though the dramatic

difference in coefficients seems to indicate that ISO certification has a much greater influence on lowering emissions than NGO membership does on raising them, one must take into consideration the difference in units in which each dataset is reported. The number of companies that are ISO 14001 certified ranges from 1 to 92, while the number of NGO members ranges from 172 to 198,590. Thus the NGO members influence is at least ten times stronger than it appears from the coefficients. On the surface, the relationship between higher NGO membership and higher TRI emissions seems surprising. A possible explanation for the

correlation could be that NGOs attract more members when companies are polluting more (e.g. TRI emissions are high) because individuals are concerned. Conversely, they may lose membership when emissions are low and the public thinks there is less cause for alarm. The relationship therefore may be one of correlation, not causation. The finding that ISO certification and NGO membership were the only two significant variables was surprising, particularly given that other researchers have found that domestic regulation and the existence of Responsible Care influence corporate environmental behavior (see Literature Review). A

possible explanation for the discrepancy between our research and prior

39

40 research is that we used a different dataset to represent domestic regulation than did Lyon and Maxwell and that we looked at a more narrow segment of the chemical industry than did King and Lenox. Another interesting contrast

between our findings and prior findings is that we found that consumer pressure (as represented by buyers of chemical manufacturing companies products) was significant, which contradicts what Sam and Innes found. In this case, the discrepancy may be due to the fact that Sam and Innes looked at individual consumers whereas we looked at corporate consumers.

IMPLICATIONS OF THE MODEL FOR ACTORS Those inside and outside of chemical manufacturing companies can use the findings of the model to determine the channels to which they should look to predict imminent action on reducing TRI emissions, or those channels through which they should work if they want to hasten the reduction of TRI emissions. Implications for Activists Individuals or groups interested in convincing chemical manufacturing companies to reduce their TRI emissions can use the findings of our model to target their efforts. By working through the action channels that most strongly influence TRI emissions, activists can ensure that they use their limited resources to maximum effect. Activists interested in convincing chemical

manufacturing companies to mitigate their environmental impacts should therefore focus the majority of their efforts on working with corporations that source from the chemical manufacturing industry. If they can convince these companies to care about environmental performance, it seems likely that this concern will travel down the value chain to the manufacturing companies, causing lasting change. Activists should also keep an eye on NGO membership, for a decrease in membership could signify a drop in public concern about chemical manufacturing company emissions.

Implications for Corporate Executives

41

42 The findings of our model suggest that pressure from buyers is the biggest driver of corporate environmental performance. Hence, when making decisions about whether or not to decrease their companies environmental impact, the first thing chemical manufacturing executives should examine is the environmental performance of the buyers of their products. A company that commits to ISO 14001 certification provides an outward sign that it cares about environmental performance. It makes sense that this concern would apply to the performance of a companys suppliers, as well as to its own performance. Conversely, an executive at a company for which the chemical manufacturing industry is a supplier should realize that pressure from his or her company can lead to lower TRI emissions. Corporate executives should also look at NGO membership in the states in which they operate facilities, as its decrease may indicate a decrease in concern about their environmental performance by citizens.

Implications for Environmental NGOs Our findings may be alarming for environmental NGOs. We do not think that NGOs should look at our research as an indication that they should stop recruiting members. However, our model does seem to indicate that focusing on increasing membership is not the best use of NGO resources. Like activists, these organizations should focus their attention on working with companies that are buyers of chemical manufacturing company products as a lever to change environmental performance farther back in the value chain.

43 At the same time, it should be noted that the majority of the NGOs we contacted could not provide accurate membership data and were thus excluded from the data set. Several of these organizations stated that they do not track membership data by state. Tracking membership data more accurately and regularly might benefit these groups as they attempt to measure their impact over time.

LIMITATIONS OF THE MODEL Care should be taken when interpreting the model. While our findings are significant, an r-squared of 0.0114 indicates much of the variation in chemical manufacturers TRI emissions year to year is not explained by our model. There are a number of potential explanations as to why our model was unable to explain more of the variation. Data limitations As mentioned in the methodology section, many of the variables we used reflected the data that were available, rather than the optimal data. Using data that were more closely aligned with the pressure we were trying to measure may have increased the validity of our model (see Suggested Improvements to Our Model for Future Researchers for an in-depth discussion of variable limitation). Further, the majority of the eleven variables we used varied year to year on a national level instead of a state level. Only four of our variables LCV Rating, NGO Membership, Westlaw and Median Income - varied state to state. One of these remained in our final model. The lack of interstate variation in most of our variables may mean that the variables that did change on a state level gained more explanatory value than they should have. Conversely, lack of variation within some of the variables may have robbed these variables of their true significance. Finally, our y-variable (TRI emissions) did not vary as much as may be ideal for a y-variable. This could contribute to the low explanatory value of our

44

45 model. Limitations of using a statistical model The nature of using a statistical model to explain the environmental behavior of chemical manufacturing companies excludes certain influencers from the model. For example, company leadership is an important indicator that repeatedly came up in our research and in our conversations with representatives from Responsible Care and Dow Chemical. A new CEO is often the driving force for a change in the environmental focus of a company for good or for bad. We could not capture this variable in the model. Another important driver of environmental performance we were unable to measure in our model is the cost savings to individual companies of more efficient technology, a factor that may be largely responsible for reductions in TRI emissions.

SUGGESTED IMPROVEMENTS TO OUR MODEL FOR FUTURE RESEARCHERS We hope that future researchers will use our model as a starting point for continued inquiry into the question of what triggers corporations to decrease their environmental impact. We outline suggested improvements to some of the variables below, followed by more general suggestions for future research. Variable Improvements Environmental NGOs: In addition to trying to obtain membership data for more of the seventeen NGOs we contacted, those continuing this work may want to create a variable comprising the amount of money raised by environmental NGOs per state as a supplemental indicator to the membership numbers. Membership data alone may provide a distorted picture of an NGOs influence. Some NGOs with large

memberships may have little money, and vice versa. Thus fundraising data could provide a further, and perhaps more accurate, window onto the influence of an NGO. Future researchers may also want to

normalize NGO membership as a percentage of state population, which would give a better indication of the relative size of the state constituency that cares about the environment. Press: Future teams building on this research may want to survey mainstream papers in addition to the articles in the Lexis-Nexis Environmental Database. As mentioned in the Methodology section, our assumption was that these journals represented the leading edge of reporting in this industry and that their stories would be picked up by

46

47 the wider media. Further research is necessary to verify the validity of this assumption. Data from major national newspapers such as the New York Times, Washington Post, Los Angeles Times and Chicago Tribune might answer this question and provide a more accurate indicator of the influence of the popular press. Additionally, further teams may want to obtain state-by-state data on the press. These data could be obtained by using the search terms we used in the major newspaper in each of the fifty states. The Community: The Gallup Poll data records opinions instead of actions. Research has demonstrated that individuals actions often do not correlate with their stated beliefs. Future researchers should attempt to obtain a variable that better measures community action rather than just opinion. Domestic Regulation: LCV ratings measure an elected officials voting record on the environment. These scores are therefore not necessarily indicative of an officials specific views on the environmental performance of the chemical manufacturing industry. A variable that might be a good additional measure of the power of domestic regulation would be the number of environmental regulations issued in a given state in a given year against the chemical manufacturing industry. Socially Responsible Investing: Future researchers should attempt to quantify the amount of SRI investing in the chemical manufacturing industry specifically. While we could not find an existing dataset with

48 this information, we think it may be an interesting exercise to begin tracking these data for future models. Another useful dataset to obtain would be the number of shareholder resolutions filed against the chemical manufacturing industry pertaining to the environment in each of the years in our model. This would directly measure investor

pressure on chemical manufacturing companies about environmental issues. Trade Associations: Membership in Responsible Care is not limited to chemical manufacturing companies (SIC Code 28); petroleum, pharmaceutical and other chemical companies can also join.24 Thus

dividing the total number of companies in Responsible Care by the total number of chemical manufacturing companies for which we collected TRI data did not give an accurate picture of the percentage of chemical manufacturers that belong to Responsible Care. Nor would measuring the percentage of chemical companies that are members of Responsible Care, because this percentage would reflect companies in industries outside the scope of our model. Future researchers should measure the percentage of chemical manufacturing companies that were Responsible Care members in each of the years of our model. This could be done by obtaining the names of all the companies that were members of Responsible Care for each of the years in the model and sorting them by SIC Code, then comparing that the total number of companies with that
24

American Chemistry Council, Member Companies, Retrieved from: http://www.americanchemistry.com/s_acc/sec_directory.asp?CID=250&DID=616 on December 12, 2006

49 SIC Code. The percentage would give a more accurate reflection of how involved the chemical manufacturing industry is in Responsible Care. TRI Emissions: We used TRI emissions as our y-variable because it was a reliable dataset that was gathered consistently over the time frame we examined, and because it measures chemical manufacturing companies environmental emissions. The number we obtained for each state was the aggregate amount of close to 1,000 different chemicals.25 We did not control for toxicity. Future researchers may want to control for toxicity, as a small reduction in a highly toxic chemical may be more significant than a large reduction in a less toxic chemical. Model Methodology Improvements Our model looked at the chemical manufacturing industry as a whole. This constrained the datasets we could gather because we needed to obtain information for the entire industry. Future researchers may find it easier to start at a company level and work upwards towards an industry level. Working at a company-level would likely facilitate easier data gathering and would give researchers the ability to measure factors such as leadership, cost savings and internal environmental targets, which are challenging to measure on an industry level. Building a number of models for individual companies and then

identifying the commonality between those models to build an industry-wide model may be an easier way to identify variables that are truly significant.

25

The full list of chemicals can be obtained at: http://epa.gov/tri/chemical/index.htm

CONCLUSION Activists, environmentalists and corporate executives should not take our advice to mean that they should exclusively focus on convincing chemical manufacturing product buyers to be ISO certified and driving down membership in NGOs and give up focus on all other channels. We believe that all the actors in our model impact each other in some way. For example, an increase in the number of environmental court cases against the chemical manufacturing industry may influence Congresspeople to vote for more environmental regulation, which in turn causes chemical buyers to seek ISO certified suppliers as a protective measure against regulation. Similarly, an increase in the number of articles in the press may educate the public, who in turn start to only buy from companies on the basis of ISO certification, and so on and so forth. The conclusion that should be drawn from our model is that change happens when environmental waste concerns become important enough to companies to elicit a move to cleaner manufacturing and that NGO membership may be indicative of an increased concern about the environment, not of an increase in effectiveness of that NGO.

50

BIBLIOGRAPHY 2005 Report on Socially Responsible Investing Trends. SIF Industry Research Program. Social Investment Forum. Washington, DC. 2005 Ader,Christine R. A Longitudinal Study of Agenda Setting for the Issue of Environmental Pollution, Journalism & Mass Communication Quarterly 72 (Summer 1995):303. Anton, Wilma Rose, George Deltas, and Madhu Khanna. Incentives for Environmental Self-Regulation and Implications for Environmental Performance. Journal of Environmental Economics and Management 48, 1, 632-654. 2004. Delmas, Magali and Michael W. Toffel. Stakeholders and Environmental Management Practices: an Institutional Framework. 2004. Bus. Strat. Env. 13, 209-222. Hoffman, Andrew J. Competitive Environmental Strategy: A Guide to the Changing Business Landscape. 2000. Island Press, Washington, D.C. King, Andrew A. and Lenox, Michael J., Industry Self-Regulation Without Sanctions: The Chemical Industrys Responsible Care Program. Academy of Management Journal, 2000, Vol. 43, No. 4, 698-716. Maxwell, John W., Thomas P. Lyon, Steven C. Hackett. Self-Regulation and Social Welfare: The Political Economy of Corporate Environmentalism. October 2000. Journal of Law & Economics, vol. XLIII. QSU Publishing Company. ISO 14001 Registered Company Directory North America. 2003, Volume 4, Number 2. Sam, A., and R. Innes, "Voluntary Pollution Reductions and the Enforcement of Environmental Law: An Empirical Study of the 33/50 Program," Working Paper, University of Arizona, 2004

51

APPENDICES Appendix 1: Model with all Variables Fixed-effects (within) regression Group variable (i): stateid R-sq: within = 0.0517 between = 0.0455 overall = 0.0187 corr(u_i, Xb) = -0.2966 Number of obs = Number of groups = Obs per group: min = avg = 8.0 max = 8 F(11,339) = Prob > F 2.18 = 0.0149 400 50 8

-----------------------------------------------------------------------------| Robust normalized~a | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------pressartic~s | 213.3085 145.1077 1.47 0.142 -72.11645 498.7335 lcvrating | 259.5164 379.9904 0.68 0.495 -487.9196 1006.952 sri | 6393.416 5895.986 1.08 0.279 -5203.91 17990.74 lobbying | -9086.666 16011 -0.57 0.571 -40580.09 22406.76 publicopin~t | -3971.018 4566.802 -0.87 0.385 -12953.86 5011.82 academic | 822.3369 1316.994 0.62 0.533 -1768.172 3412.846 ngomembers | 2.01561 1.586348 1.27 0.205 -1.104715 5.135936 responsibl~s | 1404.143 4146.007 0.34 0.735 -6750.996 9559.282 westlaw | -6183.522 6975.614 -0.89 0.376 -19904.46 7537.415 medianinco~s | -1.106513 5.013054 -0.22 0.825 -10.96712 8.754097 ofisocerti~s | -1599.754 1539.587 -1.04 0.300 -4628.101 1428.594 _cons | 272303.2 897682 0.30 0.762 -1493425 2038031 -------------+---------------------------------------------------------------sigma_u 347322.42 sigma_e | 124694.89 rho | .88582299 (fraction of variance due to u_i) ------------------------------------------------------------------------------

52

53 APPENDIX 2: EIGHT VARIABLE MODEL

Fixed-effects (within) regression Group variable (i): stateid R-sq: within = 0.0509 between = 0.0319 overall = 0.0112 corr(u_i, Xb) = -0.2604

Number of obs = Number of groups = Obs per group: min = avg = 8.0 max = 8

400 50 8

F(8,342) = Prob > F

2.45 = 0.0138

-----------------------------------------------------------------------------| Robust normalized~a | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------pressartic~s | 188.7636 111.5377 1.69 0.091 -30.62256 408.1498 sri | 6777.162 5710.666 1.19 0.236 -4455.287 18009.61 lobbying | -15344.68 13290.77 -1.15 0.249 -41486.61 10797.26 publicopin~t | -2683.439 3034.526 -0.88 0.377 -8652.124 3285.246 academic | 927.0522 1076.896 0.86 0.390 -1191.12 3045.225 ngomembers | 1.942028 1.309168 1.48 0.139 -.6330074 4.517063 westlaw | -5926.027 6662.665 -0.89 0.374 -19030.99 7178.932 ofisocerti~s | -2170.648 1558.801 -1.39 0.165 -5236.693 895.3966 _cons | 586475.9 328492.2 1.79 0.075 -59643.4 1232595 -------------+---------------------------------------------------------------sigma_u | 345003.06 sigma_e | 124199.75 rho | .88527134 (fraction of variance due to u_i) ------------------------------------------------------------------------------

54 APPENDIX 3: MODEL WITH T STATS 1 < T < -1 Fixed-effects (within) regression Group variable (i): stateid R-sq: within = 0.0490 between = 0.0305 overall = 0.0105 corr(u_i, Xb) = -0.2538 Number of obs = Number of groups = Obs per group: min = avg = 8.0 max = 8 F(5,345) = Prob > F 3.21 = 0.0076 400 50 8

-----------------------------------------------------------------------------| Robust normalized~a | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------pressartic~s | 175.2665 119.7608 1.46 0.144 -60.28673 410.8197 sri | 5339.994 5233.784 1.02 0.308 -4954.147 15634.14 lobbying | -16081.18 13742.26 -1.17 0.243 -43110.33 10947.97 ngomembers | 1.863761 1.317967 1.41 0.158 -.7285011 4.456023 ofisocerti~s | -2118.885 1280.629 -1.65 0.099 -4637.707 399.9375 _cons | 510769.7 283446.5 1.80 0.072 -46731.07 1068270 -------------+---------------------------------------------------------------sigma_u | 344463.45 sigma_e | 123778.54 rho | .88564292 (fraction of variance due to u_i) ------------------------------------------------------------------------------

55 APPENDIX 4: FINAL MODEL Fixed-effects (within) regression Group variable (i): stateid R-sq: within = 0.0373 between = 0.0263 overall = 0.0114 corr(u_i, Xb) = -0.3479 Number of obs = Number of groups = Obs per group: min = avg = 10.0 max = 10 F(2,448) = Prob > F 500 50 10

8.47 = 0.0002

-----------------------------------------------------------------------------Robust normalized~a | Coef. Std. Err. t P>|t| [95% Conf. Interval] -------------+---------------------------------------------------------------ngomembers | 4.053103 1.748284 2.32 0.021 .6172474 7.488959 ofisocerti~s | -1125.731 306.8528 -3.67 0.000 -1728.781 -522.6819 _cons | 213051.5 12319.82 17.29 0.000 188839.6 237263.3 -------------+---------------------------------------------------------------sigma_u | 420766.55 sigma_e | 191398.43 rho | .82855799 (fraction of variance due to u_i) ------------------------------------------------------------------------------

56 APPENDIX 5: SOURCES OF DATA FOR VARIABLES

Number of Academic Degrees Awarded by Year: Snyder, Thomas D, Alexandra G. Tan and Charlene M. Hoffman. Digest of Educational Statistics, 2003. National Center of Education Statistics, December 2004. Available on http://nces.ed.gov/pubsearch/pubsinfo.asp? pubid=2005025 as of November 14, 2006 Public Opinion: The Gallup Poll Organization. http://www.galluppoll.com. Question used: With which of these statements about the environment and the economy do you most agree protection of the environment should be given priority, even at the risk of curbing economic growth (or) economic growth should be given priority, even if the environment suffers to some extent? Median Income: U.S. Census Bureau. Median Income for 4-Person Families, by State. Downloaded October 11, 2006 from http://www.census.gov/hhes/income/4person.html TRI Data: The TRI Explorer, available on http://www.epa.gov/tri/ Courts Data: Data from Westlaw, available on http://www.lawschool.westlaw.com. Go to Westlaw Directory, then topical areas, then environmental law, then search by individual state Responsible Care Membership: Years 1995 through 2002 from Michael L. Barnett, PhD, Professor, College of Business Administration, University of South Florida. Years 2003 through 2006 from Debra Phillips, Managing Director, Responsible Care League of Conservation Voter Scores: LCV Scorecards available at http://www.lcv.org/scorecard Press Data:

57

From the Lexis Nexis Environmental Database (formerly available through Kresge Library A-Z Database) Academic Articles: From Jstor, available through Kresge Library Database Socially Responsible Investing: From the 2005 Report on Socially Responsible Investing Trends in the United States. SIF Industry Research Program. Social Investment Forum, Washington, DC. 2005, available for download at http://www.socialinvest.org Industry Lobbying Dollars: Search by Chemical under Industry at: http://www.publicintegrity.org/lobby/ NGO Membership: Data sent to us by Sierra Club in response to email query sent to: Information@sierraclub.org

APPENDIX 6: ADDITIONAL DATA GRAPHS NGO Membership by State: 1995-2004

NGO Membership
Year

58

59 Westlaw Chemical Industry Cases by State: 1995-2004

Year Graphs by State

60 Chemical Industry TRI Data by State: 1995-2004

Normalized TRI Data

Year

61 Median Income for 4-Person Family: 1995-2004

Year Graphs by State