1 By LIFAN QIAN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIR EMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2016
2 2016 L ifan Qian
3 ACKNOWLEDGMENTS First, I would like to express my very deep gratitude to my supervisor, Dr Xiang Bi, for her patient guidance and critiques on my paper. I would also like to thank other two committee members, Dr Gao and Dr Conner, for their useful advise of this research. Finally, I wish to thank my parent s for th eir support during my study.
4 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 3 LIST OF TABLES ................................ ................................ ................................ ............ 5 ABSTRACT ................................ ................................ ................................ ..................... 6 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ ...... 8 2 LITERATURE REVIEW ................................ ................................ .......................... 11 3 FRAMEW ORK AND HYPOTHESES ................................ ................................ ...... 16 4 EMPIRICAL MODEL ................................ ................................ ............................... 21 5 DATA AND VARIABLE CONSTRUCTION ................................ ............................. 26 Data Construction ................................ ................................ ................................ ... 26 Dependent Variable ................................ ................................ .......................... 26 Independent Variables ................................ ................................ ..................... 27 Data Source ................................ ................................ ................................ ............ 32 6 RESULTS AND DISCUSSION ................................ ................................ ............... 37 Hurdle Logit Model ................................ ................................ ................................ .. 37 Truncated Poisson Model ................................ ................................ ....................... 40 7 CONCLUSION ................................ ................................ ................................ ........ 51 LIST OF REFERENCES ................................ ................................ ............................... 54 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 58
5 LIST OF TABLES Table page 5 1 Variable Description ................................ ................................ ........................... 34 5 2 Summary Statistics ................................ ................................ ............................. 36 6 1 Hurdle Logit Model ................................ ................................ ............................. 45 6 2 Hurdle logit model Odds ratios ................................ ................................ ........... 47 6 3 Truncated Poisson Model ................................ ................................ ................... 49
6 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science INTRA FIRM DIFFUSION OF POLLUTION PREVENTION TECHNOLOGY: THE ROLE OF ORGANIZATIONAL STRUCTURE By Lifan Qian December 2016 Chair: Xiang Bi Major: Food and Resource Economics This paper empirically examines the extent to which organization characteristics promote the diffusion of pollution prevention technologies within a firm (parent company). We use panel data on 18879 facilities reporting to the Toxics Release Inventory over the period of 1991 to 2011 to examine the number of pollution prevention technologies adopted by a facility with respect to its size, previous experience in density. We use a two part hurdle model to estimate the likelihood of adoption and the rate of adoption, while controlling for public and regulatory pressures that may have affected the adoption of pollution prevention technologies. We also provide an easy way of ide ntifying peer effects by using a form of the installed base method specially designed for use on P2 technologies, avoiding the bias that occurs in traditional analysis. This method can also be extended to any analysis having similar characteristics to thos e of P2 technologies. We find that facilities that are located in the same state as broader range of P2 technologies if they are an adopter. Having a sibling facility in the same state promotes the rate of adoptions among adopters but does not affect the
7 likelihood of adoption. The effect on adoption rate is greater if siblings in the same state also belong to the same industry. Past experience in adopting P2 technology increases technology increases the rate of adoption but has no effect on the likelihood of adoption.
8 CHAPTER 1 INTRODUCTION Pollution prevention (P2) is any practice recognized P2 as the preferred approach ahead of re use, recycling, treatment, and disposal, since prevention has the potential to improve resource use efficiency and reduce the amount of pollutants being recycled, disposed of, and treated at the end of the pipe. Pursuant to the PPA, federal and state agencies rely on non mandatory approaches to promote the diffusion of P2 technolo gies. One key approach is information disclosure through the Toxics Release Inventory (TRI) Program. Following the passage of the PPA in 1990, the TRI has been expanded to include reporting of P2 practices by each TRI facility. Previous studies on adoption of P2 technologies by TRI facilities have found that public pressure caused by information disclosure has motivated facilities to adopt P2 technologies (Harrington 2012; 2013). Additionally, the effect on P2 technology adoption differs by policy instrumen t: reporting requirements and mandatory planning policies are more effective than numerical goal policy at encouraging adoption of P2 technology (Harrington 2013). The chief incentives for companies to adopt P2 technologies are improving their reputation a nd public image, which are affected by local and community pressure, and technology diffusion, the process by which the adoption of certain technology aggregates over time. Technology diffusion can be divided into interfirm diffusion and intrafirm diffusio n; the former refers to technology diffusion between firms and the latter to technology diffusion within a single firm or between facilities belonging to the same
9 parent company. We expect intrafirm information sharing to stimulate the diffusion of P2 tech nologies, since these technologies are often tailored to specific production processes and depend on firm specific knowledge and managerial philosophies. We have chosen to study intrafirm diffusion, because environmental benefits are visible only when P2 t echnologies are widely used, and future P2 legislation will aim to increase the intensity of P2 technology adoption within firms. To the best of our knowledge, no studies have examined the extent to which luence intrafirm diffusion of P2 technologies. The existing literature has found that facilities were more likely to adopt P2 technologies if their peers from the same industry or under the auspices of the same parent companies had done so (Bi, Deltas, and Khanna 2011; Harrington 2012; 2013), but these studies do not provide a detailed discussion of how other factors affect intrafirm technology diffusion. Doshi, Dowell, and Toffel (2013) suggest that intrafirm information sharing is likely to be affected by which also influence how a firm might respond to public and regulatory pressures. Based on this, we believe that organizational characteristics will affect intrafirm P2 technology diffusion. To fill this gap in our knowledge of intrafirm P2 technology diffusion, this paper seeks to empirically examine the extent to which organizational characteristics promote the diffusion of P2 technologies within a firm (parent company). Estimating the extent to which a firm might respond to non mandatory P2 policies by analyzing intrafirm P2 technology diffusion could provide crucial information with which policymakers could improve existing P2 policies. To conduct our empirical analysis, we compiled a panel
10 dataset on 18,879 TRI reporting facilities from 1991 to 2011 using the annual TRI reports and the National Establishment Time Series (NETS) database. Following the existing technology diffusion literature, we examined the number of P2 technologies adopted by each facility in li facilities and headquarters, thus measuring the stock, rank, and epidemic effects (Karshenas and Stoneman 1993). We estima ted two stage hurdle models, controlling for facility specific effects and public and regulatory pressures that might have affected the decision to adopt P2 technologies. headquarte rs are less likely to be a P2 technology adopter but will adopt a broader range of P2 technologies if they are an adopter. Having a sibling facility in the same state promotes the rate of adoptions among adopters but does not affect the likelihood of adopt ion. The effect on adoption rate is greater if siblings in the same state also belong to the same industry. Past experience in adopting P2 technology increases both technology i ncreases the rate of adoption but has no effect on the likelihood of adoption.
11 CHAPTER 2 LITERATURE REVIEW The literature identifies two types of technology diffusion processes interfirm diffusion and intrafirm diffusion and employs numerous theoretica l models to explain technology diffusion. Karshenas and Stoneman (1993) categorize the determinants affecting interfirm diffusion as rank, stock, order, and epidemic effects, drawing on the literature on technology diffusion. The rank effect refers to the assumption that potential adopters of new technologies have different inherent characteristics and consequently obtain different returns from employing a new technology, resulting in different adoption dates. The stock effect is the assumption that the mar ginal profit over time from using a new decreases over time, this leads to endogenous changes of output and price, affecting the profit to be had through further adopti on and limiting the extent of intrafirm diffusion. mover advantages by having the first chance to employ skilled wo rkers in the labor market, realizing that waiting will make them lose this advantage. Last, the epidemic effect treats the adoption process as endogenous learning: self propagation of information with the spread of a given technology. Information transfer, also known as knowledge spillover, reduces uncertainty costs, hastening the adoption process. Battisti and Stoneman (2005) assert that the methods used to measure interfirm diffusion can also be used to measure intrafirm diffusion, using data for computer ized numerically controlled machine tools (CNCs) in the UK metalworking and engineering
12 industry in 1993 to examine the rank and epidemic effects on technology diffusion within firms. Their study looks to the percentage of new technology adopted in a firm to indicate the extent of intrafirm diffusion. Battisti and Stoneman use employment, firm age, research and development expenditure, ownership type, and production system characteristics to measure the rank effect, and they use the number of years since th e model does not include variables with which to measure the stock effect becau se of the cross sectional data. and Woerter (2008), using firm level data from 2002 on Swiss information and as the number of ICT technologies adopted by the firm before the survey, correlate with the degree of ICT use within a firm, suggesting the presence of the rank and epidemic effects. Haller and Siedschlag (2011), using firm level data from Irish manufactu ring firms from 2001 to 2004, note that relatively young foreign owned firms that have their own websites on which they can publish information have a higher rate of intrafirm ICT adoption which also supports the existence of the rank and epidemic effects. Meanwhile, some literature studies knowledge spillover but does not fall within the bounds of the diffusion model just introduced. Jaffe, Trajtenberg, and Henderson (1993), for example, look to patent citation data to study the geography of knowledge spil lovers, finding that citations are more likely to come from the same state which suggests that knowledge spillovers are strongly geographically localized. Thompson
13 and Fox different set of control groups, argue that the effects of knowledge spillover at the state and metropolitan levels are weaker than Jaffe et al. estimate. Singh (2005) also reports, using patent citation and interpersonal networks data, that intraregional and intra firm knowledge flows are stronger than those across regional or firm boundaries. Diffusion of clean technologies can be different, however. Popp, Newell, and Jaffe (2010) argue that the incentive for firms to adopt clean technologies must come from environ mental regulations if the new technologies will only reduce toxic emissions and will not reduce costs or enhance production. The literature focuses on how environmental policy has affected the diffusion of these lower emission technologies by incentivizing firms to adopt them. Milliman and Prince (1989), dividing environmental regulation into five categories and comparing their effects on firm level P2 technology diffusion, find that auctioned emissions permits, emissions taxes, and abatement subsidies have a greater effect than do freely allocated permits or mandatory controls. Later contributions to the literature also note that market based legislation provides a greater incentive to adopt new clean technologies than mandatory legislation does (Jung, Krut illa, and Boyd 1996). Empirical literature focusing on the correlation between environmental legislation and clean technology adoption notes that environmental legislation is necessary to promote clean technology adoption (Kerr and Newell 2003; Snyder, Mil ler, and Stavins 2003). The literature also focuses on how clean technologies not only reduce toxic emissions but also reduce costs, especially in the case of energy efficient technologies. Jaffe and Stavins (1995) examine the relationship between energy p rices and the
14 adoption of thermal insulation technologies, finding a positive correlation between rising energy costs and technology adoption but noting that this still has less effect than lower adoption costs do. Similarly, Anderson and Newell (2004) rep ort that the effect of up front costs on technology adoption is 40 % greater than that of energy costs. In addition to the literature focusing on technology diffusion, studies related to the adoption of P2 technologies have examined the incentives for firms or facilities to adopt P2 technologies. Khanna, Deltas, and Harrington (2009) use firm level data from 1991 environmental program on P2 technology adoption, noting that pr ogram participants are more likely to adopt P2 technologies than nonparticipants are. Florida and Davison (2001), surveying manufacturers to analyze their motivations behind P2 adoption, report that large or corporate owned plants are more likely to adopt P2 practices than smaller plants are. Harrington (2012), examining TRI reporting facilities whose parent companies were on the S&P 500 between 1991 and 2011, classifies forty three types of P2 practices into three main categories: procedural changes, input and material practices is affected, she says, by past experience and market related factors, and the significance of community pressure and knowledge transfer in technology ado ption differs according to technology type. In her 2013 study, Harrington uses the same database as from her 2012 study to examine the effectiveness of different state level s that require action plan reporting, a target pollution level, and mandatory P2 planning policy adopt more P2 practices.
15 None of these s tudies, however, examines intra firm diffusion of P2 technologies under the current environmental policies, which focus on information disclosure through the TRI. Accordingly, we will focus on the effect of knowledge spillover (epidemic effect) on P2 adoption and control for firm and facilities characteristics (rank effect) and other circumstantial factors. The effect of kn owledge spillover can be divided into two parts: knowledge stock about P2 technology. This paper contributes to the exi sting related literature in three ways: (1) by extending the scope of intrafirm diffusion studies to P2 technologies, (2) by examining the role of organizational structure and industry networks in promoting the diffusion of P2 technologies, and (3) by usin g a larger sample size and a longer period than previous studies to examine whether the results from previous intrafirm diffusion studies on other technologies also apply to P2 technologies.
16 CHAPTER 3 FRAMEWORK AND HYPOTHESES This paper chiefly aims to study the effect of knowledge spillover, measured by P2 adoption activities. Such a correlation might reveal the mechanism through which P2 technologies diffuse withi n a firm. Unlike in the previous literature on intrafirm diffusion which used firm adoption to represent the extent of intrafirm diffusion we will undertake facility level analysis of the number of P2 te forty three different types of activities that aim to reduce pollution at its source. As a result, TRI reporting facilities might adopt more than one practice or technology in any given year and might adopt a variety of different types of P2 technologies or practices over time. In cont rast, previous studies on technology adoption have typically focused on a single technology or on a small group of technologies. marginal benefits from and marginal costs of adoption in short, that a facility will likely choose to adopt a new P2 technology at the point when its marginal benefits are greater than or equal to its marginal costs. The organizational characteristics that affect a osts of adoption include ownership type, firm size, and sibling properties. We also expect that the distance between sibling facilities will influence a
17 involved in searching for and learning how to use new technologies. Previous literature shows that reducing the costs of searching for and adapting technologies increases the probability and speed of adoption (Lenox and King 2004; Mansfield 1968). Geographic proximit y can promote knowledge sharing (Szulanski 1996), for knowledge transfer becomes more difficult and costly as distance increases (Berchicci, Dowell, and King 2012). And although proximity does not mean that such knowledge transfer is inevitable, it does cr eate a convenient opportunity for knowledge transfer in facilities that prefer face to cant if its siblings are located in the same city (Jin and Leslie 2009). Doshi, Dowell, and Toffel (2013) note that facilities that have proximate siblings improve their environmental performance more frequently. Accordingly, we assume that proximate sibli ng facilities help promote P2 technology adoption. environmental performance does not necessarily result in the adoption of more P2 technologies; facilities can choose other end o f pipe pollution abatement technologies, such as recycling and treatment, to reduce their toxic emissions. Knowledge transfer between proximate sibling facilities can consist of not only P2 technologies but also other pollution abatement technologies. This is also possible if the empirical results do not show a strong positive correlation or even if they show a strong negative correlation. Hypothesis 1(a): Facilities that have siblings in close proximity are more likely to adopt P2 technologies than are fac ilities that do not have siblings in close proximity.
18 Hypothesis 1(b): Among facilities that have adopted P2 technologies, facilities that have siblings in close proximity tend to adopt a greater number of P2 technologies than do facilities that do not hav e siblings in close proximity. We expect that knowledge transfer is inversely related to the number of distinct industries that a firm encompasses. Maritan and Brush (2003) note that knowledge transfer is more difficult in establishments that have differen t operating procedures. The U.S. Environmental Protection Agency (EPA) classifies forty three different types of P2 activities into eight categories (EPA 2007); two of these eight (1) process and equipment modifications and (2) surface preparation and fini shing relate to specific techniques. To adopt these two types of practices, facilities must learn specific knowledge that could hardly be obtained from sibling organizations that have different production processes. We have assumed that facilities that bel ong to the same industry have similar production processes and thus would be more likely to exchange information about P2 technologies than would sibling organizations that do not operate within the same industry. Nonetheless, as already stated, knowledge transfer about specific technologies between sibling facilities within the same industry is not limited to P2 technologies but rather can also apply to other pollution abatement technologies. We assume that proximate siblings belonging to the same industry will promote P2 technology adoption, but we acknowledge that the empirical results might show an insignificant correlation or a significant negative correlation, for the P2 technologies and other pollution abatement technologies might be substitutes for e ach other.
19 Hypothesis 2(a): Facilities that have siblings in close proximity that are in the same industry are more likely to adopt P2 technologies than are facilities that do not have siblings in close proximity that are in the same industry. Hypothesis 2 (b): Among facilities that have adopted new P2 technologies, facilities that have siblings in close proximity that are in the same industry tend to adopt a greater number of P2 technologies than do facilities that do not have siblings in close proximity th at are in the same industry. We expect that the distance between facilities and their parent companies also Technological innovations might have been developed at headq uarters and disseminated to individual subsidiary facilities; we assume that headquarters would then have a higher level of knowledge about these new technologies than would the subsidiaries, and technical support from headquarters would reduce individual costs of learning and adapting to new technologies. Accordingly, greater distances adopting new P2 technologies. Additionally, headquarters would sustain s tronger pressure from the local community and other stakeholders if subsidiaries have poor environmental performance, spurring headquarters to supervise subsidiaries and encourage them to improve their environmental performance. A facility that is near its headquarters might receive more supervision and enjoy less freedom to make its own decisions in such a situation. Again, the same problem discussed in the other two hypotheses also exists when considering the effects of headquarter proximity: facilities c an use other pollution
20 abatement technologies to improve their environmental performance, excluding P2 technologies. We have assumed that close proximity to headquarters will promote the adoption of P2 technologies, but empirical results might show that th is effect does not exist or even that the opposite is true. Hypothesis 3(a): Facilities that are in close proximity to their headquarters are more likely to adopt P2 technologies than are those that are farther away from their headquarters. Hypot hesis 3(b): Among facilities that have adopted new P2 technologies, those that are in close proximity to their headquarters tend to adopt more P2 technologies than do those that are farther away from their headquarters.
21 CHAPTER 4 EMPIRICAL MODEL We use a two making processes, separating their adoption decisions into two steps: (1) whether they will be a new P2 technology adopter and (2) how many practices they will adopt once they have decided to becom e adopters We assume these two stages are independent of each other. T he coefficients affecting whether facilities will adopt new P2 technologies might differ from variables. Althoug h t his assumption is quite strong and evidence to support it, the only practical method to deal with the database we have at this time is based on this strong assumption. For the first part of this two stage process, we assume that a facility would choose to be an adopter when the marginal profits of adoption are greater than or equal to the marginal costs of adoption. In other words, the facility will adopt at least one P2 technology when the net profit of adoption is greater than o r equal to zero. We employ the logit model based on the property of the first stage: where is a dummy variable denoting the adoption decision of facility which belongs to parent company at time equals 1 if facility adopts at least one new P2 technology at time but equals 0 if facility does not adopt any new P2 technology at time
22 is a latent variable indicating the difference between marginal benefit and marginal cost for facility which belongs to parent company at time The vector denotes a time varying regressor of organizational structure that can affect the denotes the time invariant regressor of knowledge spillover from peer facilities that can affect the facilit adopting new P2 technologies. The vector denotes the time varying interaction valued by product of organizational structure and knowledge spillover. The vector denotes the time invariant regressor of knowledge stock for facility .The vector denotes the time for adopting new P2 technologies. The vector denotes the time varying control from adoption. The vector denotes the environmental circumstances variables that capture the local community pressure and regul atory pressure. The vectors denote the industry, state, and time dummies, respectively, that control for the industry and time specific effects and unobserved effect across time. This approach is similar to the unconditional fixed effects logit. The vec tor is the error term, which has a logistic distribution. Significant potential problems can arise when studying knowledge spillover from siblings. Indeed, the literature raises three issues that might arise when identifying peer effects: self selection of peers, correlated unobservables, and simultaneity (Manski 1993; Moffitt 2001; Soetevent 2006). In our case, the problems of self selection of peers
23 and correlated unobservables will make the mean conditional error terms of firm characteristics unequal t o zero ( ), leading to inconsistent estimation of Per the literature, we add a set of Standard Industrial Classification (SIC) code dummies and state dummies into the model as fixed effects that control for unobserved industry and state effects to help solve these problems (Manchanda, Xie, and Youn 2008; Nair, Manchanda, and Bhatia 2010). To control for unobserved effects, which might change across time, we add a full set of year dummies to capture them. The problem of simultaneity in our case arises bec adoption decision as well. That being the case, we will get an estimation of this combined effect instead of single causal peer effects from its siblings. We employ the Youn 2008; Narayanan and Nair 2013. The installed base method uses the previous adoption of siblings as explanatory variables with which to estimate results. A lagged the fixed effect by adding SIC and state dummies to eliminate the influence of self selection of peers and correlated unobservables. The li terature indicates that models using the installed base method might be estimated inconsistently when the lagged explanatory variables used for estimation correlate with the error term (Nerlove 1971; Nickell 1981). To avoid this problem, we use the exogeno us shock experienced by all TRI facilities in 1995. The EPA has expanded the list of chemicals requiring reporting since the TRI reports were first required in 1987, boosting the number of TRI chemicals for reporting from 363 to 606,
24 with 240 of these new chemicals permanently increased since 1995. We use the P2 adoption on 363 original chemicals from 1991 to 1994 as the installed base with which to measure peer effects. T explanatory variable, because the lagged dependent variable, which correlates with the error term, will make the estimation inconsistent. Much as we have done with peer installed base with which to measure the knowledge stock. We employ a truncated Poisson mo del to estimate the second stage of the making process, conditioned on adoption of at least one technology. In our dataset, we do not find an evidence for significant over dispersion. A Poisson model is preferred to a negative binomial model, as indicated by the standard deviation is less than two times mean. For each facility that has adopted at least one new P2 technology, we assume that the cumulative positive count of its P2 adoption fits a Poisson distribution. The expected number of adoption is expressed as follows: is the expected number of cumulative P2 adoptions for facility which belongs to parent company at time
25 We use the same sets of explanatory variables for the truncated Poisson model as are used in the logit model. We expect that the coefficients of these variables differ, for they have different effects on both the likelihood and the rate of adoption.
26 CHAPTER 5 DATA AND VARIABLE CONSTRUCTION Data Construction Dependent V ariable We use two different variables to represent the results of the two stages of facility decision making. Each facility is required to report its P2 adoptions for each TRI chemical for up to forty three categories. To construct these two dependent var iables, we focus on a set of 240 chemicals These 240 chemicals were included in the TRI reporting requirement in 1995, when the EPA expanded the original list of TRI chemicals, and they did not experience reporting changes from 1995 to 2011. The first dep endent variable, Adopt P2 dummy is a dichotomous variable used in the logit model. Adopt P2 dummy was coded as 1 if the sum of P2 activities adopted from the beginning year to the given year was greater than zero, meaning that the facility adopted at least one kind of P2 activity for all its 240 chemicals since the beginning year, but was coded as 0 o facility is regarded as an adopter after it adopt one P2 technology, no matter whether it will adopt new P2 practices in the f uture. To avoid a situation in which a facility has already been an adopter but has not ions exist only of facilities that have never adopted P2 technologies and of facilities that have adopted P2 technologies for the first time.
27 The second dependent variable, Cumulative P2 is the count variable used in the truncated Poisson model. Cumulativ e P2 is the aggregate of new P2 activities adopted on the 240 chemicals for each facility since 1995. Independent Variables Our variables of interests focus on the effect of knowledge spillover, which includes the effects of organizational structure and pe er effects, as well as their interaction terms. The vector represents time varying organizational structure. We The sibling in same state binary dummy variable has a value of 1 if the facility has at least one sibling located in the same state and 0 otherwise. The same industry sibling in same state is located in the same city and also b elongs to the same industry (i.e., reporting the same two digit SIC code) but is 0 otherwise. The headquarter in same state is a binary dummy variable that takes a value of 1 if the facility is located in the same state as its headquarters but that takes a value of 0 otherwise. The vector denotes the time invariant regressor of knowledge spillover from activities of other siblings belonging to the same parent comp any, suggesting that the We created the variable which takes the sum of the adoption of P2 technologies by siblings belonging to the same parent company from 1991 to 1994. We use the existing adoption data of 363 types of chemicals on siblings from
28 1991 to 1994 as the proxy of the spillover effect from the siblings on the 240 newly added chemicals after 1995. To exclude the effect of beginning year, we divide the sum by the times of reporting from 1991 to 1994. To further examine the effects of organizational structure and peer effects, we create the intera ction term s the product of organizational structure and knowledge spillover. Because we use three also includes three variables: (1) spillover headquarters the product of and headquarter in same state (2) spillover sibling the product of and sibling in same state and (3) spillover same industry sibling the product of adoption and same industry sibling in same state We built control variable ve ctor adoption decision was affected by its adoption experience. The vector includes only The variable facili takes the sum of P2 technologies adopted by the facility from 1991 to 1994. To exclude the effect of the beginning year, we divide the past experi ence over time, we create an interaction term that takes the product of a nd a time trend. The time trend is generated by the difference of 1995 and the given year plus one. The vector denotes the time varying firm characteristics that can affect the include firm size, number of siblings, number of industry categories, and ownership type.
29 might affect its financial ability and the costs it incurs searching for an mixed. Most suggest that larger firms are more likely to implement new technologies than s maller firms are (Mansfield 1968; Karshenas and Stoneman 1995). Other literature, however, notes that smaller firms tend to intensity their subsequent adoption more than larger ones do after initial adoption of new technology (Mansfield 1963; Fuentelsaz, G omez, and Polo 2003). Thus intrafirm diffusion of new technology is faster within smaller firms than within larger firms. This is likely owing to the differences in costs for adapting to a new technology between larger and smaller firms, such as the cost t o train employees in its use. Small firms are able to increase their degree of adoption to lower the average costs for each facility. We aggregate the number of sales nu mber of sales, which we use as the proxy of firm size. The variable uses the lagged one year total of sales and natural log to reduce skewness. The variable number of unique SIC codes represents the count of unique industries (defined by two number of siblings represents the number of subsidiaries that a parent company owns. La Porta, Lopez de Silanes, and Shleifer (1999) report that firms that have more than 20 % private equi ty are more likely to adopt productivity enhancing practices a result of the ownership concentration, which reduces interagency costs between managers and shareholders. Private and government companies usually have a higher degree of ownership concentratio n than do publicly traded companies. Accordingly, we expect that facilities that belong to privately held firms are more likely to adopt P2 technology.
30 However, publicly traded firms might be more motivated to reduce pollution and improve their public imag e. Konar and Cohen (1997) note that the disclosure of traded firms might be motivated to adopt P2 technologies to reduce pollution. On the other hand, the adoption of new P2 technologies might require huge investments in the reducing stock prices and returns to shareholders. Meanwhile, other types of pollution control methods, such as end of pipe abatements, might be more cost effective. In that event, the effect of public ownership on P2 technology adoption might be ambiguous. We use a binary dummy variable public ownership which takes a value of 1 if the parent company is publicly owned and 0 if the parent company is privately or government owned in a given year. We built control variable vector adoption decision is affected by facility characteristics. Two variables were created to percentage sales to represent the s as a part of total sales to approximate the level of a skewness. To control for the scope of the P2 technologies, we create the variable number of chemicals which eq uals the number of chemicals that belong to the group of 240 chemicals that were added in 1995. The last set of variables we built to control for the circumstantial characteristics of facilities. We built control variable vector to account for the probab ility that the
31 The total volume of release might serve as a proxy for the extent of specific facility regulatory pressure and, further, the cost of liabilities related to health risk and environmental governance. We create the variable lagged toxic release which takes the value of the sum of toxic release of all chemicals. We c reate the variable lagged HAP release to represent the level of regulation pressure. Hazardous a ir pollutants (HAP) are technologies for these chemicals reach the highest industry standard. The literature performance (Arora and Cason 1999; Earnhart 2004; Wolverton 2009). The local community can press a facility to adopt a new P2 technology through citizen suits or by lobbying for stricter legislation (Earnhart 2004). We also created t he variable county median income which uses the log of median household income in each county to represent county level community pressure. We created the variable LCV score Voters (LCV) National Environm ental Scorecard. The scorecard calculates the proportion of environmental bills voted on by a particular member of Congress, reporting it as a value ranging from 0 to 1, with the latter equaling 100% We used this variable to capture state level community pressure. We create the variable number of facilities in a city which takes the value of TRI facilities in the same city, and we take a natural log of the value to reduce skewness. W e also create the vectors ,and which include the industry dummies (SIC 1 digit code), state dummies, and time dummies into the model to control for the
32 specific industry and state effects. The variable description and summary statistics are listed in Tables 5 1 and 5 2, respectively. Da ta Source NETS data from 1991 to 2011. The EPA requests that facilities belonging to certain industry sectors report the toxic release of chemicals on the list published by the EPA pursuant to the Right to Know Act, enacted in 1986. Issuance of these public annual reports began in 1987. Besides data about toxic release of each regulated chemical, the IC) code, information on the parent company, and emission medium. Pursuant to the Pollution Prevention Act of 1990, the TRI reports are expanded to include the number of P2 technologies adopted for each regulated chemical since 1991. We obtained data for P 2 adoption, location, release, SIC code, and parent www.epa.gov/tri. We linked the facilities in the TRI with their correspondent establishments in NETS data using establish ment information from January 1990 to NETS data for 1991 2011. The TRI report data are from a facilities level database, whereas the NETS data are from an establishment level database. These two databases do not match perfectly: an observation in one database might not have a corresponding data in another database. A few observations in the TRI have more than one corresponding observations in the NETS database. We kept only ob servations that had a unique corresponding object in each database for a given year.
33 The information on facility location, number of employees, and parent company ownership is obtained from the NETS database. Note that the subsidiaries of a parent company that do not report to the TRI cannot be identified in our dataset. Using the reported location of the facility from the TRI, we have merged this the period 1995 2011. T he county income data are obtained from the U.S. Department of Agriculture (USDA) Economic Research Service (ERS) at http://www.ers.usda.gov. The LCV data are from http://scorecard.lcv.org The final dataset contain s 121,451 observations involving 18,879 facilities. The sample sizes used for the logit model and the logit model for facilities that have siblings are 121,451 and 56,335, respectively. The truncated Poisson models use only the observations that have a pos itive dependent variable. The sample sizes used for the whole truncated Poisson model and the truncated Poisson model for facilities that have siblings are 9,675 and 5,462, respectively.
34 Table 5 1. Variable Description
35 Table 5 1. Continued.
36 Table 5 2. Summary Statistics
37 CHAPTER 6 RESULTS AND DISCUSSION Hurdle Logit Model The regression results in Table 6 1 present the effects of different factors on the likelihood of P2 adoption, using Adopt P2 dummy as the dependent variable. The results in the fi rst and second column of Table 6 1 are for the full sample; those of the other two columns are for facilities that have at least one sibling. The results show in every column. Starting wit h the analysis of organizational structure and peer effects, the results in Table 6 1 show that only headquarter in same state likelihood of being a P2 technology adopter when taking into account factors of organizational structure. The effect is significantly negative at a 1 % significance level, which contradicts our hypothesis that facilities that are in close proximity to their headquarters are more likely to adopt P2 technologies. This phenomenon may result logies that are an equivalent substitute for P2 technology. The other two organizational structure variables, sibling in same state and same industry sibling in same state are not statistically significant in our results, contradicting our hypotheses that close proximity siblings and close proximity siblings in the same industry are more likely to be P2 technology suggestion is also supported by the results of the peer effect. The variable used to measure peer effects, is significant onl y in column 4, and that at a 10% significance level;
38 and the interaction terms of peer effects and organizatio nal structure are not significant problem described earlier, in the empirical w ork section. The sample is also different in her study. is significantly positive in all columns at a 1% significance level. The variable t P2 adoption trend is significantly negative in all columns at a 1% significance level. It suggest s experience will increase the likelihood of adoption but that the effect will decrease over time. This suggestion is not a regarding P2 technology adoption. va riable used to measure the firm size and financial condition, is significantly negati ve in all columns at a 5% significance level in the full sample and a 1% significance level in the subset of facilities that have at least one sibling. This result is contrary to the findings of the literature on diffusion of other technologies. Perhaps because information about P2 technology is sufficiently present in the market, the advantages large firms enjoy in searching for new technology might not exist for P2 technology. The variable number of siblings i s significantly negative at a 5% significanc e level in column 2 and at a 10% significance level in other columns. The variable number of unique SIC code is not significant in full sample and is sig nificantly
39 positive at a 1% significance level in the subset of facilities that have at least one sibling. The variable public ownership is significantly positive in the first two c olumns (full sample) at 1% and 5% significance level, respectively. It is i nsignificant in the column 3 and column 4, the subset of facilities that have at least one sibling. Now we turn to control variables, which include facility characteristics and circumstantial characteristics. The variable percentage sales is statistically significant in all columns except column 2. It i s significantly negative at a 1% significanc e level in column 2 and at a 10% significance level in columns 3 and 4. The variables number of chemicals and lagged toxic release are significantly positive in all columns at a 1% significance level. The variable lagged HAP release used to capture the regulation pressure, is significantly negative in all columns at a 1% significance level, suggesting that a facility might be sensitive to the regulatory threat but c hoose to adopt other pollution abatement technology in its bid to fulfill requirements. The variables county median income LCV score and number of facilities city are insignificant in all columns, perhaps because the variables of toxic release have captu red the largest part of local community pressure and regulatory pressure. odds ratios of these variables, as listed in Table 6 2 and interpret the odds ratios of some main v ariables. Holding other variables at a fixed value, difference between the probability of being a P2 adopter for a facility located in the same state as its headquarters and being a P2 adopter for a facility not located in the same state as its headquarter s is 0.774. Thus the probability of a facility that is located in the same state as its headquarters is 23% lower than for a facility that is not. The calculation of change
40 in odds ratios for one variable always keeps other variables fixed, which we will n ot r epeat hereafter. We can see a 6% increase in the odd s of being a P2 adopter for a 1% 1994 for a subset of the full sample P2 adoption during 19 91 1994 will produce about a 72% increase in the odds of being a P2 adopter. The odds of a public facil higher than those of a private or government owned facility. A 1% toxic rel ease correlates with about a 12% increase in its odds of being a P2 adopter. technology adopter, examining the implications we can draw from the results. Overall, only variable of organizational structure that has a significant effect is headquarters in same state but the effect is significantly negative. External knowledg e from siblings has contained behavior, not the effects of siblings or proximity to headquarters and resulting stricter supervision. Another implication is that more intense community pressure is the main driver of a regulatory pressure decreases the likelihood of adoption might indicate the adoption of other substitutable pollution abatement technologies. Truncated Poisson Model As with the case analyzed in the hurdle logit model, the results in the fi rst and second column of Table 6 3 are for the full sample; those of the other two columns are for facilities that have at least one sibling. T he regression results in Table 6 3 present
41 the effects of different categories of factors on the rate of P2 adoption after a facility has become an adopter, using Cumulative P2 as the d ependent variable. The results show a high degree of homogeneity in all columns, a difference evident only at the significance level. Starting with the analysis of organizational s tructure, the results in Table 6 3 show that the variables headquarter in sa me state sibling in same state and same industry sibling in same state are al l significantly positive at a 1% significance level, supporting our hypotheses that facilities that are in close proximity to their headquarters or that have proximate siblings that belong to same industry tend to adopt more P2 technologies after they have become adopters. The variable same industry sibling in same state has larger positive effect on the rate of adoption than does the variable sibling in same state which is also consistent with our hypothesis. Now we continue with the analysis of peer effects. The variable used to measure peer effects, is significa ntly positive at a 1% significance level in columns 2 and 3, representing the sample of facilitie s that have at least sibling. This result supports our analysis just now in which we judged that a facility will adopt more P2 technologies when it has lower adoption costs thanks to better knowledge transfer after having become an adopter. The variable fi is not significant in the full sample, perhaps because the full sample estimates the mixed effect of facilities with and without siblings. For a facility that does not have sibling, the variable adoption is always zero, which mig ht affect the estimation results. For the interaction terms created by peer effects and organizational structure, the variables spillover headquarter and spillover sibling ar e significantly positive at a 1% significance level.
42 The variable spillover same i ndustry sibling i s significantly negative at a 1% significance level. and are significantly positive in all columns at a 1% experience will increase the rate of adoption and that the effect will keep increasing over 1991 t preference regarding P2 technology adoption. It also suggest the existence of a complementary relationship between P2 technologies used on different kinds of chemicals. We will no variables and number of siblings are signi ficantly negative at a 1% significance level in all columns, much as with the logit model. The variable number of unique SIC code i s significantly positive at a 1% significance level in all columns. The variable public ownership is significan tly negative at a 1% significance level in the full sample, unlike with the logit model. The condition of public ownership suggests that a facility that has already been a P2 technology adopter will choose to adopt other substitutable pollution abatement technologies when it faces stronger community pressure or wants to further improve its public image. Now we turn to control variables, which include facility characteristics and circumstantial characteristics. The variable percentage sales is significantly pos itive in columns 1 and 2 at a 1% significance level and insignificant in other columns, a
43 diverge nce from the results in the previous model. The variable number of chemicals is significantly positive in all columns at a 1% significance level. The variable lagged toxic release county i s significantly negative at a 1% significance level in columns 1 and 2. It i s significantly negative at a 5% significan ce level in column 3 and at a 5% significance level in column 4. It has the opposite sign to that of the logit model, perhaps because when the facility was already a P2 technology adopter and still wanted to continue improving its reputation, it might have adopted other substitutable pollution abatement technologies. This could account for why the variable lagged toxic release county has a significant negative effect on rate of P2 adoption. The variable lag ged HAP release i s significantly negative at a 1% significance level in columns 3 and 4 and is not significant in other columns. The variable county median income i s significantly negative at a 1% significance level in columns 1 and 2 and is not significan t in other columns. The variable LCV score is not significant in all columns, and the variable number of facilities city is significan tly negative in column 3 at a 1% significance level, signific ant negative in column 4 at a 5% significance level, and not significant in other columns. adoption after it becomes a P2 technology adopter, examining the implications of our results, which suggest that organizational structu re will affect knowledge transfer from siblings and that the effects of sibling spillover might reach proximate siblings belonging to the same industry. These results also suggest that a facility might adopt substitutable pollution abatement technologies t o fulfill mandatory requirements while
44 dealing with higher levels of local community pressure, as is the case with public ownership
45 Table 6 1 Hurdle Logit Model Dependent variable: Adopt P2 dummy (1) (2) (3) (4) VARIABLES Full sample Full sample Have siblings Have siblings Head in same state (A) 0.256*** (0.084) Sibling in same state (B) 0.108 (0.105) Same industry sibling in same state (C) 0.021 (0.272) 0.035 0.067* 0.065 (0.033) (0.040) (0.042) Spillover x headquarter (A x D) 0.025 (0.021) Spillover x sibling (B x D) 0.019 (0.032) Spillover x same industry sibling (C x D) 0.010 (0.060) 0.586*** 0.578*** 0.513*** 0.513*** (0.037) (0.038) (0.051) (0.051) 0.037*** 0.036*** 0.037*** 0.037*** (0.005) (0.005) (0.007) (0.007) 0.057** 0.062** 0.179*** 0.182*** (0.025) (0.026) (0.038) (0.038) Number of siblings 0.060* 0.071** 0.091* 0.089* (0.031) (0.032) (0.049) (0.049) Number of unique SIC codes 0.006 0.008 0.048*** 0.047*** (0.014) (0.015) (0.017) (0.017) Public ownership 0.212*** 0.183** 0.158 0.155 (0.082) (0.082) (0.100) (0.100) Percentage sales 0.391*** 0.198 0.411* 0.399* (0.123) (0.156) (0.214) (0.217)
46 Table 6 1. Continued Dependent variable: Adopt P2 dummy VARIABLES (1 ) Full sample 2 Full sample 3 Have siblings 4 Have siblings Number of chemicals 0.509*** 0.510*** 0.541*** 0.541*** (0.020) (0.020) (0.027) (0.027) Lagged toxic release 0.115*** 0.114*** 0.118*** 0.117*** (0.022) (0.022) (0.029) (0.029) Lagged HAP release 0.062*** 0.061*** 0.068*** 0.068*** (0.019) (0.019) (0.024) (0.024) County median income 0.024 0.030 0.047 0.049 (0.027) (0.027) (0.036) (0.035) LCV score 0.175 0.172 0.227 0.229 (0.198) (0.197) (0.269) (0.269) Number of facilities in a city 0.019 0.020 0.045 0.045 (0.033) (0.033) (0.044) (0.044) SIC code dummies State dummies Year dummies Included Included Included Included Included Included Included Included Included Included Included Included Constant 3.937*** 3.935*** 0.604 0.578 (0.957) (0.957) (1.341) (1.342) Observations 121,451 121,451 56,335 56,335 Note: Standard errors in parentheses, *** p<0.01, ** p<0.05, p<0.1
47 Table 6 2 Hurdle logit model Odds ratios (1) (2) (3) VARIABLES Full sample Have siblings Have siblings Head in same state (A) 0.256*** (0.084) Sibling in same state (B) 0.108 (0.105) Same industry sibling in same state (C) 0.021 (0.272) 0.035 0.067* 0.065 (0.033) (0.040) (0.042) Spillover x headquarter (A x D) 0.025 (0.021) Spillover x sibling (B x D) 0.019 (0.032) Spillover x same industry sibling (C x D) 0.010 (0.060) 0.578*** 0.513*** 0.513*** (0.038) (0.051) (0.051) 0.036*** 0.037*** 0.037*** (0.005) (0.007) (0.007) 0.062** 0.179*** 0.182*** (0.026) (0.038) (0.038) Number of siblings 0.071** 0.091* 0.089* (0.032) (0.049) (0.049) Number of unique SIC codes 0.008 0.048*** 0.047*** (0.015) (0.017) (0.017) Public ownership 0.183** 0.158 0.155 (0.082) (0.100) (0.100) Percentage sales 0.198 0.411* 0.399* (0.156) (0.214) (0.217) Number of chemicals 0.510*** 0.541*** 0.541*** (0.020) (0.027) (0.027) Lagged toxic release 0.114*** 0.118*** 0.117*** (0.022) (0.029) (0.029) Lagged HAP release 0.061*** 0.068*** 0.068*** (0.019) (0.024) (0.024) County median income 0.030 0.047 0.049 (0.027) (0.036) (0.035) LCV score 0.172 0.227 0.229 (0.197) (0.269) (0.269) Number of facilities in a city 0.020 0.045 0.045 (0.033) (0.044) (0.044)
48 Table 6 2. Continued (1) (2) (3) VARIABLES Full sample Have siblings Have siblings SIC code dummies State dummies Year dummies Included Included Included Included Included Included Included Included Included Constant 3.935*** 0.604 0.578 (0.957) (1.341) (1.342) Observations 121,451 56,335 56,335 Note: Standard errors in parentheses, *** p<0.01, ** p<0.05, p<0.1
49 Table 6 3 Truncated Poisson Model Dependent Variable: Cumulative P2 (1) (2) (3) (4) VARIABLES Full sample Full sample Have siblings Have siblings Head in same state (A) 0.165*** (0.012) Sibling in same state (B) 0.108*** (0.016) Same industry sibling in same state (C) 0.588*** (0.043) 0.004 0.052*** 0.092*** (0.004) (0.006) (0.006) Spillover x headquarter (A x D) 0.011*** (0.003) Spillover x sibling (B x D) 0.035*** (0.004) Spillover x same industry sibling (C x D) 0.138*** (0.009) 0.157*** 0.152*** 0.098*** 0.101*** (0.009) (0.009) (0.012) (0.012) 0.006*** 0.006*** 0.007*** 0.007*** (0.001) (0.001) (0.001) (0.001) 0.021*** 0.023*** 0.057*** 0.062*** (0.004) (0.004) (0.006) (0.006) Number of siblings 0.065*** 0.063*** 0.107*** 0.103*** (0.004) (0.005) (0.007) (0.007) Number of unique SIC codes 0.021*** 0.019*** 0.020*** 0.022*** (0.002) (0.002) (0.003) (0.003) Public ownership 0.153*** 0.152*** 0.128*** 0.107*** (0.011) (0.011) (0.015) (0.015) Percentage sales 0.100*** 0.046** 0.021 0.046 (0.016) (0.022) (0.031) (0.031)
50 Table 6 3. Continued Dependent Variable: Cumulative P2 VARIABLES (1) Full sample (2) Full sample (3) Have siblings (4) Have siblings Number of chemicals 0.136*** 0.134*** 0.154*** 0.165*** (0.002) (0.002) (0.003) (0.003) Lagged toxic release 0.020*** 0.023*** 0.011** 0.009* (0.003) (0.003) (0.005) (0.005) Lagged HAP release 0.002 0.001 0.012*** 0.011*** (0.003) (0.003) (0.004) (0.004) County median income 0.028*** 0.036*** 0.003 0.000 (0.004) (0.004) (0.005) (0.005) LCV score 0.033 0.033 0.045 0.060 (0.027) (0.027) (0.038) (0.038) Number of facilities in a city 0.004 0.001 0.028*** 0.018** (0.005) (0.005) (0.007) (0.007) SIC code dummies State dummies Year dummies Included Included Included Included Included Included Included Included Included Included Included Included Constant 0.948*** 1.091*** 0.433* 0.304 (0.130) (0.130) (0.223) (0.223) Observations 9,675 9,675 5,462 5,462 Note : Standard errors in parentheses, *** p<0.01, ** p<0.05, p<0.1
51 CHAPTER 7 CONCLUSION This P2 technology adopter and on the number of P2 technologies that it will adopt after it becomes an adopter. Overall, our empirical results show th decision but the influence might be heterogeneous in each stage. Specifically, proximity to headquarters increases the number of adoptions but reduces the likelihood of fac belonging to the same industry both boost the number of adoptions but have no effect to a sibling belonging to same industry is stronger than the effect of mere proximity to a sibling. Our findings about other variables that control for firm characteristics, facility characteristics, and circumstantial characteristics all indicate that loc al community number of P2 technologies that it will adopt. The results also show that regulatory pressure reduces both likelihood and rate of adoption. We suggest that in such a case, facilities might be adopting substitutable pollution abatement technologies to fulfill requirements and to further improve their public after having already become a P2 technology adopter. These findings have several implications for futur e policy. First, if the government still wants P2 technology to be the main pollution reduction method, then it should weigh
52 particular, efforts to make more people aware o f the benefits of P2 technologies might spur P2 adoption. Moreover, if the government includes the percentage of toxic reduction, making it a part of the evaluation of environ mental performance, facilities might pass by other pollution abatement technology to adopt P2 technology instead. past P2 adoption has a positive effect on both the likelihood and the rate of P2 adoption. technologies, including through interviews with members of management. Such a policy should be implemented as soon as possible so that it c an have a long term positive effect on the likelihood and rate of P2 adoption. A little increase in adoption today will accumulate through the years. Third, the empirical results in the truncated Poisson model imply that the general information in the mark et is sufficient thanks to its dissemination by the government, but specific industry specific knowledge is still insufficient and does much to affect rates of P2 adoption. Further policy should also strive to create opportunities for facilities belonging to the same industry to exchange the learning they gain while adopting industry specific P2 technologies. Our study also has implications for future research into adoption of P2 technologies. First, we provide an easy way of identifying peer effects by usi ng a form of the installed base method specially designed for use on P2 technologies, avoiding the bias that occurs in traditional analysis. This method can also be extended to any analysis having similar characteristics to those of P2 technologies. Second the
53 empirical results regarding the variable also imply that a complementary relationship might exist between different P2 technologies used on different chemicals. This, in particular, provides fertile ground for future study
54 LIST OF REFERENCES Southern Economic Journal 65, no. 4: 691 716. Anderson, Soren T., and Richard G. N Technology Adoption: The Case of Energy Energy Economics 26, no. 1: 27 50. Firm Diffusion of New Process ional Journal of Industrial Organization 23, no. 1 2: 1 22. Strategic Management Journal 33, no. 9: 1053 71. The Role of Information Spillover, Mandatory Regulation, and Voluntary Program Doshi, Anil R., Glen W. S. 1209 31. ics 80, no. 3: 408 32. Management Review 43, no. 3: 64 84. Fuentelsaz, Lucio, Jaime Gomez, and Yolanda P 51. Evidence from Firm 3775 88. Stage Adoption of Different Types of Pollution 73. 78. and Intra Firm Diffusion of Technology: The Example of E Commerce: An Analysis Based on Swiss Firm no. 3: 545 64.
55 76, no. 5: 984 1001. Jaffe, Adam B., Manuel Trajtenberg, and Rebecca Henderson. Quarterly Journal of Economics: 577 98. Regulations: The Effects of Alternative Policy Instruments on Technology S43 S63. 1: 237 67. Abatement Technology at the Industry Level: An Evaluation of Policy 95 111. Karshe RAND Journal of Economics 24, no. 4: 503 28. of the Economics of Innovation and Technological Change: 265 97. Kerr, Suzi, and Richard G. Newell. 2003. Policy Induced Technology Adoption: Evidence from the U.S. Lead Phasedown. The Journal of Industrial Economics 51, no. 3: 3 17 43. Pollution Prevention Techniques: The Role of Management Systems and 106. Konar, Shameek, and Mar Economics and Management 32, no. 1: 109 24. La Porta, Rafael, Florencio Lopez de porate 517. 331 45.
56 Man 6: 961 76. New York: Norton. Quarterly Journal of Economics 77, no. 2: 290 311. of Economic Studies 60, no. 3: 531 42. Management Journal 24, no. 10: 945 59. Milliman, Scott R., a and Management 17, no. 3: 247 65. Moffitt, Terrie E. 2001. Sex Differences in Antisocial Behaviour: Conduct Disorder, Delinq uency, and Violence in the Dunedin Longitudinal Study. New York: Cambridge University Press. Jo urnal of Marketing Research 47, no. 5: 883 895. Narayanan, Sridhar, and Harikesh S. Nair. 2013. "Estimating causal installed base effects: A bias correction approach." Journal of Marketing Research 50, no. 1: 70 94. on the Estimation of Dynamic Economic Econometric Society: 359 82. Journal of the Econometric Society: 1417 26. 937. iffusion 70.
57 Environmental Regulation on Technology Diffusion: The Case of Chlorine no. 2: 431 35. 20, no. 2: 193 228. : Impediments to the Transfer of 43. Thompson, Peter, and Melanie Fox iew: 450 60. U.S. Environmental Protection Agency. EPA Definition of "Pollution Prevention", by F. Henry Habicht II, Washington, D.C., 1992. https://www.epa.gov/sites/production/files/2014 09/documents/pollprev.pdf U.S. Environmental Protection Agency. Tox ic chemical Releases Inventory Reporting Forms and Instructions. Washington, D.C., 2007 Related Factors in for Environmen tal Economics Working Paper 2009 03.
58 BIOGRAPHICAL SKETCH ajor is food and resource economics. He used his time on resource and environmental economics studies, graduating with Master of Sc ience degree in the fall of 2016