|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 GOVERNMENT INVOLVEMENT ON LIMITING CHOICE IN SCHOOLS AND ON FIRM PERFORMANCE THROUGH SUBSIDIES By MICHELLE ANDREA PHILLIPS A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2012
2 2012 Michelle Andrea Phillips
3 To my family
4 ACKNOWLEDGMENTS I would like to thank all the individuals who made this research possible. I am immensely grateful to my committee members: Dr. Lawrence Kenny Dr. Sanford Berg, Dr. David Denslow, Dr. Alfonso Flores Lagunes, and Dr Richard Scher for their mentorship and advice. Without their guidance this dissertation would not have been possible. I would also like to thank Dr. Chunrong Ai, Dr. Steven Slutsky, Dr. Jonathan Hamilton, Dr. Lynne Holt, Dr. James Dewey, Dr. David Sappington, Dr. Roger Blair, a nd the Economics Department at the University of Florida for their helpful suggestions and constructive comments. I am also thankful to Dr. Shinji Yane and Katsufumi Fukuda for assistance in obtaining and interpreting data on Japanese water systems and to Dr. Maria Corton for providing me with helpful ideas and comments. I am also grateful for comments provided by participants at the 2011 and 2012 Public Choice Society meetings, and seminars at the University of Florida and Missouri University of Science a nd Technology. I am grateful to the McKnight Doctoral Fellowship Program and to the University of Florida Economics department and PURC for providing me with financial support through out my graduate studies. I gained invaluable experience working as a res earch and teaching assistant for Dr. Sanford Berg. I am also grateful for the Lockhart Endowment, for providing me with travel support. I would also like to extend my gratitude to my family members and friends for the love and support they provide. I th ank my husband Chuck Isaacson, my parents Mark Phillips and Silvana Schaffhauser, my brothers Bryan and Nicholas Phillips Schaffhauser and my aunt Penny Phillips.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 9 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 STATE INVOLVEMENT IN LIMITING TEXTBOOK CHOICE BY SCHOOL DISTRICTS ................................ ................................ ................................ ............. 12 1.1 Introduction ................................ ................................ ................................ ....... 12 1.2 Empirical Model ................................ ................................ ................................ 15 1.3 Results ................................ ................................ ................................ .............. 25 1.4 Conclusion ................................ ................................ ................................ ........ 29 2 WHAT CHARACTERIZES SCHOOL DISTRICTS THAT CHOOSE TO OPT ................................ 42 2.1 Background ................................ ................................ ................................ ....... 42 2.2 Empirical Model ................................ ................................ ................................ 44 2.3 Results ................................ ................................ ................................ .............. 48 2.4 Conclusion ................................ ................................ ................................ ........ 49 3 INEFFICIENCY IN JAPANESE WATER UTILITY FIRMS: A STOCHASTIC FRONTIER APPROACH ................................ ................................ ........................ 56 3.1 Introduction ................................ ................................ ................................ ....... 56 3.2 Background: Japanese Water Utilities ................................ .............................. 57 3.3 Explanation of Stochastic Frontier Models and How They Compare to OLS .... 59 3.4 Empirical Model ................................ ................................ ................................ 60 Battese and Coelli Model ................................ ................................ .................. 60 Production Function ................................ ................................ ......................... 64 Ineffici ency Effects ................................ ................................ ........................... 64 3.5 Policy Implications and Conclusion ................................ ................................ ... 70 APPENDIX A MARGINAL EFFECTS FOR STATE LEVEL TEXTBOOK SELECTION TABLES .. 76 B MARGINAL EFFECTS FOR INDIANA DISTRICT LEVEL TEXTBOOK WAIVER TABLE ................................ ................................ ................................ .................... 82
6 C FUNDAMENTALIST CHURCHES ................................ ................................ .......... 83 D REGIONAL CLASSIFICATION OF COUNTIES AS DEFINED BY THE INDI ANA DEPARTMENT OF EDUCATION ................................ ................................ ........... 85 E JAPAN BACKGROUND INFORMATION ................................ ................................ 86 F EFFICIENCY ................................ ................................ ................................ .......... 88 G ALTERNATIVE MODEL SPECIFICATIONS ................................ ........................... 93 LIST OF REFERENCES ................................ ................................ ............................... 96 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 103
7 LIST OF TABLES Table page 1 1 Classification of states and number of biology textbooks ................................ ... 31 1 2 Description and sources: State level textbook selection variables. ..................... 33 1 3 Summary statistics: State level textbook selection variables. ............................. 35 1 4 Fundamentalism by state ................................ ................................ ................... 36 1 5 Ordered logit cross section regression explaining state textbook policies ................................ ................................ ..... 38 1 6 Ordered logit cross section regression explaining state textbook policies ................................ ................................ ...... 39 1 7 Choice / Non Choice logit panel explaining state textbook policies. ................... 40 2 1 Summary statistics and sources: Indiana district level study .............................. 53 2 2 Indiana district level logit model explaining textbook waiver use. ....................... 55 3 1 Production function variables ................................ ................................ .............. 73 3 2 Inefficiency effects model variables ................................ ................................ .... 73 3 3 Summary statistics ................................ ................................ ............................. 74 3 4 OLS estimates ................................ ................................ ................................ .... 74 3 5 Stochastic frontier estimates: Inefficiency effects model using regional dummies and allowing for delta 0 ................................ ................................ ....... 75 A 1 Marginal effects for T able 1 5 column 1 ................................ ............................. 77 A 2 Marginal effects for T able 1 5 column 2 ................................ ............................. 78 A 3 Marginal effects for T able 1 6 column 1 ................................ ............................. 79 A 4 Marginal effects for T able 1 6 column 2 ................................ ............................. 80 A 5 Marginal effects for T able 1 7 ................................ ................................ ............. 81 B 1 Marginal effects for T able 2 2 ................................ ................................ ............. 82 C 1 Fundamentalist churches ................................ ................................ ................... 84
8 E 1 Japanese prefectures and regions ................................ ................................ ..... 86 F 1 Summary statistics for efficiency scores and output ................................ ........... 90 G 1 Time invariant panel production function model. ................................ ................ 93 G 2 Inefficiency effects model using regional dummies and not allowing for delta 0 ................................ ................................ ................................ ......................... 94 G 3 Inefficiency effects model using regional dummies and including subsidies ....... 95
9 LIST OF FIGURES Figure page 1 1 Belief in evolution by education level ................................ ................................ .. 41 2 1 Geographical distribution of Indiana school district waivers. .............................. 51 E 1 Japanese regions ................................ ................................ ............................... 87 F 1 2004 weighted ................................ ................................ ................................ .... 88 F 2 2005 weighted ................................ ................................ ................................ .... 89 F 3 2006 weighted ................................ ................................ ................................ .... 89 F 4 2007 w eighted ................................ ................................ ................................ .... 90 F 5 Efficiency and output ................................ ................................ .......................... 91 F 6 Subsidies for bottom 40 firms ................................ ................................ ............. 92 F 7 Subsidies for top 40 firms ................................ ................................ ................... 92
10 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy GOVERNMENT INVOLVEMENT ON LIMITING CHOICE IN SCHOOLS AND ON FIRM PERFORMANCE THROUGH SUBSIDIES By Michelle Andrea Phillips August 2012 Chair: Lawrence Kenny Major: Economics I e xamine government intervention o n limiting textbook choice and o n efficiency of water utilities through the use of subsidies Who gets to decide what textbooks are used i state States can let each school district decide, provide standards that must be followed and make available an incomplete listing of books meeting those standards, or allow schools to only choose books from a list provid ed by the state. I n C hapter 1, I present a model that provides an explanation for state limits on textbook selection by school districts. My findings corroborate the extant literature in regards to interference by state governments in local affairs and extend the morality politics literature by finding a strong link between fundamentalism and state level policies. I also find that sta te book lists are less likely in more educated states, in states with smaller school districts, and in states with strong er teacher unions In C hapter 2 I analyze factors that are expected to influence petitions for the use of other textbooks at the school district level for the state of Indiana. The findings provide evidence supporting the costs of decision making as measured by gini co efficients suggesting that counties that are more unequal in income distribution have
11 a ha rder time reaching agreements. T he results also suggest that citizens who are more f undamentalist than average are more likely to request a waiver to opt out from In C hapter 3 I examine inefficiencies in Japanese water utility companies. capacity to maximize output given a fixed level of inputs. The findings suggest that the average operation rate, customer density and size variables are associated with lower levels of inefficiency (or higher levels of efficiency), while t he raw water proxy, subsidy and outsourcing ratio variables are associated with higher levels of inefficiency Since inefficienc y exists, there i s an firm s whenever possible Further examination of the variable s found to be influencing technical inefficiency particularly those that are under managerial control, is recommended
12 CHAPTER 1 STATE INVOLVEMENT IN LIMIT ING TEXTBOOK CHOICE BY SCHOOL DISTRICTS I present a model that provides an explanation for state limits on textbook selection by school districts. I examine the roles played by decision making costs, effectiveness of voters, religious composition, power of teachers, and propensity of state governments to interfere with or to help districts in textbook selection policies at the state level. There has been virtually no research on this topic. My findings corroborate the extant literatur e in regards to interference by state governments in local affairs and extend the morality politics literature by finding a strong link between fundamentalism and state level policies. I also find that state book lists are less likely 1) in more educated states, where voters are better able to select the most appropriate textbook, 2) in states with smaller school districts, where voters are more involved in the schools, and 3) in states with stronger teacher unions, giving teachers more power in textbook s election. 1.1 Introduction During recent years, several states that require textbook selection at the state level have been in the spotlight for controversies regarding what they choose to include and exclude from their book lists. Most recently, attentio n was drawn to Texas where church doctrine and downplays the influence of figures su ch as Thomas Jefferson (Strauss 2010). Some conservatives dislike Jefferson because he advocated separation between church and state, while some liberals dislike him because he held slaves.
13 controversial topics 1 For scientists, evolution constitutes an important building block necessary for the understanding of modern biology, while for some Christians it constitutes a direct attack on their beliefs. Politicians and school board members often make headlines over their stances, comments and actions regarding evolu tion and creationism. In Florida, for instance, an intense debate over teachings of evolution versus intelligent design and other theories culminated in the adoption of new state standards in 2008 explicitly requiring for evolution to be taught in schools This requirement was so controversial that state legislators took actions att empting to undermine it (Postal 2011). In March 2011, Florida Senator Steve Wise filed the most c theory of (Florida Senate 2011). The bill immediately sparked the interest of media outlets and another round of fights between both sides erupted. Specifically, scientific groups saw it as an attack on the teachings of evolut ion in schools. The bill died three months after it was filed, but based on legislative actions from the past four years, a similar form of it should be expected to resurface soon. Similarly, in 2006, the Utah Origins of Life bill sought to require teachers to issue a disclaimer to their students stating that not all scientists agree about evolution. Stephen Urquhart, the Mormon majority whip, (Johnson 2006). The bill was soon voted down. Who States can: (1) let each school district decide, (2) provide standards that must be 1 This statement is based on the number of court cases filed against schools regarding controversial topics.
14 followed and make available an incomplete listing of books meeting those standards (allowing pe titioning of waivers for the use of other textbooks ), or (3) allow schools to only choose books from a list provided by the state. Selecting good textbooks is important for student success. In a study relating to Brazilian and Indian educational producti on functions, Pritc hett and Filmer (1999) found that increases in test scores per dollar spent on learning materials including textbooks were about 19 times greater than those from increases in teacher salaries (Pritchett and Filmer 1999, quoted in Boissie re 2004). According to Stream (200 8 ) 80 90% of classroom and homework assignments in American schools are textbook driven. Furthermore, a large amount of taxpayer dollars roughly 4.3 billion are involved in the textbook mar ket each year (Finn and Ravi tch 2004). This paper provides an explanation for state limits on textbook selection and for district decisions regarding the use of textbook list waivers. Follow up research should focus on the impact of textbook choice policies on student outcomes. Wou ld a commission at the state level be more effective in knowing what textbooks the children of a remote rural district should use or would teachers and people familiar with the background of the students do a better job at choosing school materials? It is important to understand what characterizes states that decide between these options. Specifically, why is it that certain states have decided to impose limits on the materials that can be used by its teachers while others have not? Understanding what in fluences these policies should provide insight into what makes states more or less flexible regarding their delegation of authority to choose textbooks. Given the importance of textbook use, who gets to make these types of choices could have important con sequences for the performance of the educational sector.
15 In order to examine this question, the costs of decision making, effectiveness of voters, religious composition, power of teachers, and propensity of state governments to intervene (or to help inep t districts) in textbook selection are analyzed at the state level. To my knowledge this is the first study to empirically examine textbook selection policies 2 This study extends the existent literature in public choice, economics and political science. For instance, numerous studies have linked greater educational attainment to more involved electorates. This article provides some evidence for the idea that more educated people tend to demand more choice, are more involved in decision processes, and a re more confident in their decisi ons. This study also provides evidence consistent with there being less parental involvement in larger school districts. Furthermore, this paper corroborates the extant literature in regards to intervention by state gover nments in local affairs and extends the morality politics literature by finding a strong link between fundamentalism 3 and both state level and district level textbook selection policies. This is a relevant policy to examine because the benefits obtained by special interest groups are concentrated while the impacts (or costs) to the rest of society, are dispersed. 1.2 Empirical Model This study examines several factors influencing state textbook selection policies utilizing two different specifications: a cross sectional ordered logit model and a panel 2 Fuchs and (2007) examine an int ernational education production function to assess whether having textbook autonomy leads to higher test scores by looking at 15 year olds in thirty countries and find that the use of autonomy when choosing textbooks only matters if there is a state exit e xam. However, the authors do not examine the policies themselves. 3 In morality politics the term traditionalist term fundamentalist instead, to avoid confusion with the more common usage of the word traditionalist. The term traditionalist is commonly associated with churches, such as the Catholic Church, which are not part of the group encompassing religious traditionalism as defined by the morality politics literature. Please refer to the Appendix for a complete description of what the term fundamentalist entails.
16 data logit model (PLM). Data availability dictated the use of these two different approaches because there is a trade off between keeping a large r sample with fewer variables and using a specification with fewer observations, but a greater number of relevant variables. The state of Hawaii was dropped from all specifications because it only has one school district 4 Policies determining who can se lect textbooks vary by state; they can be classified into three major groups: 1. These are states that let each school district choose the textbooks it wants to use. In this case it is usually the duty of teachers, parents, and principals to choose the textbooks. 29 states correspond to this category 5 2. ommended list s These are states that have a list that is recommended, but not mandatory. These states usually have shown to meet those standards (the students are then tested on the standards). Basically, these states have recommendations from a state agency, but their districts are allowed to choose books from outside that list. Currently, 12 states correspond to this category. 3. y s These are states in which the list of books that a district can use is chosen at the state level. Presently, 8 states fit into this category. Table 1 1 provides a list of states and their corresponding classification. I classified each state by examining its specific laws. A spreadsheet with this information is available upon request. Table 1 2 defines the variables used and their sources. In order to provide an explanation of state policies regarding textbook selection, I estimated the fo llowing cross sectional ordered logit models 6 : 4 For the state of Hawaii, the re is no disparity between what the state and districts want since there is only one district in the state. 5 California has a mandatory state list for grades 1 8, but is a local choice state for high school. Since this paper focuses on high school texts, California is classified as a local choice state. 6 The estimates presented correspond to high school.
1 7 (1 1) (1 2) The o nly difference between model s (1 1) and (1 2 ) is the inclusion of the teach model (1 2 ). The two models were tested because using The d defined in the following way: 0 if a state has complete choice (local choice states) 1 if a state is in between (recommended list states) 2 if a state has a mandatory state list (mandatory states) In order to account for differences that might arise from some states having long lists while others have short lists, the following models were also estimated: (1 3) (1 4) Models (1 3) and (1 4) differ from models (1 1) and (1 2 ) in the definition of the dependent varia tate l 0 if a state has complete ch oice (local choice states) 1 if a state is in between (recommended list states)
18 2 if a state has a long mandatory state list for 9 th grade. A long list is defined as a list that has a higher amount of books than the average of 24 books 7 3 if a state has a short mandatory state list for 9 th grade (i.e. lower than average when counting books from all lists) Once again, models (1 3) and (1 4 variable. As mentioned earlier, I also estimate the following PLM fo r the years 1970, 1980, 1990, and 2000: (1 5 ) The PLM model is very similar to the cross sectional models except for two important differences: (1) the dependent because the authors who collected the data did not distinguish between man datory and The advantage of using a panel is that it increases sample size and should provide more powerful test stati stics. The disadvantage for this specific case is the lack of data for the variables mentioned above. T hroughout the years, Watts (2009 ), Tulley (1985), and Zinth (2005) kept track of the number of states deciding whether or not to give their districts a choice in regards to textbook selection. The authors made a distinction between choice and non choice states, but they did not distinguish between mandatory and recommended list states. For this reason, the dep endent variable in the panel ist 3 0 if a state has complete choice (local choice states) 7 This was also tested by setting the cut off at the median of 20 books. The empirical results using the median as a cut off rather than the mean are v ery similar and are available upon request.
19 1 if a state does not have complete choice (mandatory and recommended list states) The panel model uses time fixed effects 8 and the robust standard errors are cluste red at the state level. Once again, the state of Hawaii was excluded because it only has one district. For this reason the panel has 196 observations. Descriptions and sources for the variables used can be found in Table 1 2. Table 1 3 presents summar y statistics. One hypothesis for having a mandatory state list has to do with bureaucratic influences present in a state and ensuring that funds are used in appropriate ways (accountability). Withholding state funding unless a policy is adopted can often be used to dictate policies. States in which the state supplies a large share of school funding are expected to be more likely to dictate educational policies and, thus, textbook policies. A common reason given in favor of mandatory state lists is that s tate oversight of school books is cost and time efficient, saving the districts time by narrowing the lists they can choose from, and helping to ensure alignment with standards set by the state 9 Textbook list restrictions also provide a way for states to help local districts that they deem unfit to make wise choices for themselves (due, perhaps to a lack of expertise or resources). It is expected that states that are more bureaucratic are more likely to 10 The bureaucratic aspect can be captured b 8 Initially state fixed effects were added as well, but given the shortness of the panel the model was over parameterized. 9 For a thorough discussion of arguments for and against textbook adoption see Farr and T ulley (198 9) and Finn and Ravitch (2004). 10 State level funding may also reflect a desire to ensure basic levels of educational support where some districts are very poor.
20 of school revenue coming from the state and represents expected state involvement in local affairs, following Husted and Kenny (2000). Sc hool system revenues are obtained from a combination of federal, state, and local sources. The variable state revenue measures the importance of school system revenues coming from the state, and is defined as: State Revenue = ( State Revenue 100) / tota l revenue. Some religious groups, known in the literature as fundamentalists, have been in the spotlight for championing issues related to textbook selection and adoption policies. Groups with these beliefs are measured by the number of adherents to fund amentalist churches as a percent o classification of fundamentalist churches has been used before in the morali ty politics literature and follows Johnson (1976). The morality politics literature is a body of literature that looks at how the political culture of an area can explain political outcomes and characteristics. The Southern Baptist Convention and the Ame rican Baptist Association are examples of congregations cla ssified as fundamentalist. Appendix C contains a complete listing of congregations that are classified as being fundamentalist. The hypothesis regarding fundamentalism is that a state with a high er percentage of adherents to fundamentalist churches would be more likely to have a state mandated list 11 Citizens belonging to fundamentalist congregations are strongly vocal against topics such as the teaching of evolution in schools and gay marriage. Fundamentalist believers in creationism consequently tend to oppose having their children exposed to 11 I also tried using a variable measuring percentage of votes for Bush in 2000 and average senator Americans for Democratic Action ( ADA ) scores. The results are very similar to using fundamentalism, but fundamentalism was chosen because data is available f or a longer period of time (prior to and after the 2000 elections).
21 evolutionary theories; thus they are expected to want to influence book lists in support of their ideologies. These types of interest groups are powerfu l, concerned about this issue, well organized, and well known for their ongoing lobbying and strong political influence 12 (Delfattore, 1992; Batista Oliveira 1995). Answers in Genesis and other strong anti evolution organizations have been known to use the ir hefty budgets to undermine the teaching of evolution (Cole 2000 as cited in Moore 2004). Furthermore, this does not seem to be a large enough issue for other parties to get together and exert their influence. It is also easier for fundamentalists to focus their efforts at the state level rather than at the district level for the simp le reason that it is easier to lobby for an entire state than numerous districts. For example, if an advocacy group were to tackle the Even though the relationship between fundamentalists and public policies has been extensively studied, this paper is unique in that it examines textbook selection policies in this framework for the first time. Hutcheson and Ta ylor (1973) found fundamentalism to be strongly correlated with various political system and policy outcome variables relating to education and tax policies. Hutcheson and Taylor (1973, p p 418 dentification with and Meier (1980, p. 148) 12 See Delfattore (1992) for a documented account of the disproportionate influence these groups have had over the years.
22 studied the relationship between religion and referenda on moral issues using statewide referenda in Oklahoma, relating to issues such as repealing prohibition and authorizing atter and, in fact, is clearly as Using the fundamentalist classification, the states with the most fundamentalist groups as a percent of their population were Arkansas (37%), Miss issippi (36.6%), and Alabama (35.6%). The states with the lowest percentage were Rhode Island (0.9%), Connecticut (1.1%) and New York (1.1%). It is expected that the more fundamentalist the state, the more likely it is to prefer mandatory state lists. N evertheless, it is not clear that, all other things equal, adding more fundamentalists to states that are already highly fundamentalist would make fundamentalists more influential. It is possible that having a state with 35% fundamentalists does not trans late into more influence than having a state with 25% fundamentalists, given that the value of 25% is already very high. Perhaps what matters is whether or not a state is more fundamentalist than the rest. In order to account for this possibility, the fu ndamentalist variable is also measured as a dummy taking a value of 1 if a state has more fundamentalist church members than the average sta The regressions that provided the best fits for each specifi cation were reported The data regarding fundamentalism were obtained from the Glenmary Research database in Jones et al. (2002, p. 15 r children and the estimated number of other participants who are not considered members; for
23 tionship between types of state lists and the percentage of fundamentalist adherents for each state, which can be appreciated in Table 1 4 Several studies have linked greater educational attainment to more confident, better informed and more involved ele ctorates. Husted and Kenny (2007), for example, found that more educated voters are less likely to set limits on educational spending, given that more educated voters are generally more confident in their abilities. Similarly, Adams and Kenny (1986) foun d that more educated voters are less likely to impose te rm limits on their governors. Schmidt, Kenny and Morton (1996) used an objective measure of deviation from voter wishes and found that reactions to bad voting records were greater in more educated st ates. In addition to these studies, Dewey and Kenny (2009) found that more educated cities were better represented than those with less education when examining the impact of the surge in property values on municipal expenditures for the 2000 2006 period. It is anticipated that states with more educated people would demand or be given more choice, because they would be expected to be more involved in the decision process and to be more confident in their decisions. Furthermore, textbook list restrict ions can provide a way for states to help local districts that they deem unfit to make choices for themselves due, perhaps, to a lack of expertise by two variables: bac dvanced d professional degree.
24 A More choice would be expected in smaller districts since it is easier to reach agreements and more parental involvement is expected than in larger districts. On the othe r hand, if a district is large enough there might be less of a need to rely on the state government for textbook choice if economies of scale can be reached at the district level. One also would expect less parental involvement in larger districts, since there is a larger incentive to free ride, making a large local district less effective than a small one. Thus, for very large school districts, there would be less of a loss in going to a state decision. District size is defined as the number of students enrolled in the state divided by the number of dis value of 1 if a district is large, or a value Large districts are defined as districts t hat belong to the 8 th 10 th largest district deciles 13 when looking at the distribution of students/districts across states. This range was chosen because it provided the best fit. More heterogeneous districts might have a harder time reaching agreements and, for that reason, prefer to delegate to a state decision. A gini coefficient variable is used to control for heterogeneity between households within each state, providing a measure of household income inequality at the sta te level. The gini variable from 0 to 1. A low gini represents a state with more equal distribution (value of 0 if districts look the same, meaning that everyone has the same income), while a high gini represents a state with high inequality of income (value of 1 if districts have perfect 13 Similar results were obtained when using the average and the median district as cut off points between large and small districts. The dummy was created following the same logic that was explained for the fundamentalist dummy.
25 inequality). Utah and New Hampshire have the lowest gini coefficients (0.4104 and 0.4151 respectively) indicating that the income distribution is more equal than in other states, while New York and Connecticut have the highest (0 .4985 and 0.4809 respectively), indicating that income is distributed more unequally than in other states. e a measure of the power teachers can have on their specific states. Powerful groups of teachers would likely exert their influence and require more of a say on what happens in the classroom. o favor more choice in textbooks since teachers would prefer selecting their own texts over having someone else make that decision for them. defined as total union expenditures per student at the state l union, giving a measure of the influence and strength of teachers. Finally, following Fischel (2005), a measure of homeownership is included in the model. Home values are higher when schools are doing a good job, so one wou ld expect states with more homeowners to be better monitors and to have less of a need to restrict choice. percentage of homeowners in the state, relative to renters. 1.3 Results The first set of results is reported in Tables 1 5 through 1 7 Table 1 5 presents the ordered logit cross section results explaining textbook selection policies excluding shown in Table 1 6 while results using the 4 year panel logit model in Table 1 7 These different specifications were tested, since there is a trade off between keeping the larger
26 sample and between using specifications with fewer observations, but a better fit. It is important to note that the coefficients presented in Tables 1 5 through 1 7 only give us the direction of the effects. Due to the nonlinear nature of logit models, the partial effects (i.e. the marginal effects) for a given explanatory variable ar e not given by its coefficient and are different for each observation For this reason, two routinely followed approaches are used to compute marginal effects. The first method, known as Partial Effects at the Average (PEAs) consists on computing the ma rginal effects at the average values of each continuous explanatory variable. In the case of discrete variables, dy/dx is calculated for a discrete change of the dummy variable from 0 to 1. state in the sample. The second method, known as Average Partial Effects (APEs) consists on computing marginal effects for each state and then averaging these over the sample. Results with the marginal effects calculated as Average Partial Effects and Pa rtial Effects at the Average for each outcome can be found in Appendix A Columns (1) and (2) in Tables 1 5 and 1 6 differ in the definition of the dependent variable. For column (1) the dependent variable is defined as local choice, recommended or mandatory state list, whereas for column (2) the mandatory state list group is further subdivided into two groups: short mandatory state list and long mandatory state list. Though different measures of several variables were tested, only the results that provided the best fits were presented given that they all yielded very similar results. The coefficients of different mea sures of the fundamentalist variables are highly significant and, as hypothesized, positive in every specification except for one of the
27 panel specifications. These results indicate that the probability of having a more restricted state list increases as more people belong to fundamentalist groups. Specifically, the marginal effects from the ordered logit cross section presented in Appendix A indicate that the probability of being a local choi ce state decreases, for example by 37% for states which have m ore fundamentalist adherents when compared to those that do not. The panel results suggest that the probability of having a mandatory state list increases as more people belong to fundamentalist groups. These results are to be expected, given that fundam entalists feel more strongly about evolution than others and are, thus, more active and effective in bringing about policies they prefer. According to the estimates, the district size variable and the district size dummy variable are always highly significant and positive for the cross sectional specifications, indicating that the probability of having a more restricted state list increases as school districts become larger. District size is significant in 2 out of the 3 panel specifications. The marginal effect in Table A 1 indicates that the probability of being a local choice state decreases by 11% for states with a district size one sta ndard deviation (i.e. 7,301) above the mean This result is consistent with there being less parental involvement in larger districts. This is because there is a large incentive to free ride and a gain from the state government taking over in the sense that a state bureaucracy is seen as better than having decisions made by uninterested voters. all cross sectional specifications, as expected. However, these results are not corroborated by the panel and as such should be viewed with caution. The cross
28 sectional results suggest that the probability of having a more restricted state list falls as the residents of a state become more educated. The marginal effect in Table A 1 indicates that the probability of being a local choice state increases by 13% for a state that has a percentage of residents with advanced degrees one standard deviation above the mean. This is expected, since states with more educated people would be m ore confident in their choices and demand more choice. The model provides some evidence for the state revenue hypothesis. The coefficient for the state revenue variable is significant and positive in most regressions (except for the model including the te probability of being a more restricted state list increases as state revenue becomes a larger share of total school revenue (i.e. more intervention in local decisions). Specifically in Table A 2 the margina l effect indicates that the probability of being a local choice state decreases by 12% for a state that has a state revenue share of total revenue one standard deviation above the mean. For the panel section, the marginal effects are always positive and s ignificant, indicating that the probability of having a mandatory state list increases as state revenue becomes a larger share of total school revenue. This is consistent with the bureaucratic involvement hypothesis. As expected, states that are more bur eaucratic are more likely to have mandatory state lists since states where the state level government intervenes more in local affairs are more likely important t o emphasize that these results are not present in all the specifications. The model does not provide support for the hypothesis predicting that homeowners have less of a need to restrict choice, suggesting that homeownership
29 does not play a role in stat e level textbook selection policy restrictions. The state income gini variable is slightly significant and positive in a few unreported regressions, available upon request, indicating that the probability of having a more restricted state list increases a s states have a more heterogeneous population. Heterogeneous states are states in which there is a more unequal income distribution. A possible explanation for this is that it might be harder for districts to reach agreements in the presence of more hete rogeneous populations. Note, however, that the marginal effects for the specifications with the best ove rall fits presented in Appendix A are not statistically different from 0. the results excluding this variable and are shown in Table 1 6 significant in one specification, indicating that, as hypothesized, the probability of having a more restricted state list falls as teachers have more power (as refl ected by their expenditures). This result is consistent with the idea that more powerful teachers would exert their influence, and require more of a say on what happens in the classroom. Specifically in Table A 5 the marginal effect indicates that the pr obability of being a local choice state increases by about 9% for states that have state level union expenditures one standard deviation (i.e. 13.03) above the mean. 1.4 Conclusion Chapter 1 presented several models examining textbook selection policies. Religious composition, effectiveness of voters, costs of decision making, power of teachers, and propensity of state governments to intervene or to help inept districts are all important factors.
30 This analysis extends the extant morality politics literatur e and provides strong evidence of the link between religious fundamentalism and government policies. As expected, the probability of being a local choice state falls with greater adherence to fundamentalist groups. This is expected because these groups t end to have more interested and focused constituents relative to other groups. The results of this study are consistent with those of Husted and Kenny (2007), Adams and Kenny (1986), Schmidt, Kenny and Morton (1996) and Dewey and Kenny (2009) offering som e evidence that states with a more educated populace demand or are given more local choice, because they are expected to be more involved in the decision processes and to be more confident in their decisions. The evidence suggests that the probability of having more restricted textbook lists increases as school districts become larger, supporting the notion that less parental involvement is expected in larger districts, since there is a large incentive for parents to free ride, making a large local dis trict less effective than a small one. Thus, as expected, for very large school districts, there seems to be less of a loss in going to a state decision. Furthermore, this paper corroborates the extant literature in regards to intervention by state gover nments in local affairs, by suggesting that there is a link between how much a state government contributes to school revenues and how much it intervenes
31 Table 1 1. Classification of states and number of biology te xtbooks Type of state State 9 th grade 10 th grade 11 th grade 12 th grade Mandatory list Alabama 22 books 22 books 23 books 23 books Florida 17 books 17 books 17 books 17 books Louisiana 14 books 14 books 14 books 14 books Mississippi 53 books 56 books 60 books 60 books New Mexico 26 books 26 books 26 books 26 books N. Carolina 18 books 18 books 18 books 18 books Oklahoma 32 books 31 books 31 books 29 books Tennessee 16 books 17 books 16 books 16 books Recommended list Arkansas 17 books 17 books 17 books 17 books Georgia 22 books 21 books 18 books 18 books Idaho 28 books 28 books 30 books 30 books Indiana 40 books 40 books 40 books 40 books Kentucky 34 books 34 books 35 books 35 books Nevada 39 books 36 books 39 books 39 books Oregon 8 books 8 books 8 books 8 books S. Carolina 20 books 13 books 14 books 14 books Texas 7 books 7 books 11 books 11 books Utah 22 books 9 books 9 books 9 books Virginia 14 books 14 books 14 books 14 books W. Virginia 5 books 3 books 1 book s 1 book s Local choice Alaska ----Arizona ----Colorado ----Connecticut ----Delaware ----Illinois ----Iowa ----
32 Table 1 1 C ontinued Type of state State 9 th grade 10 th grade 11 th grade 12 th grade Kansas ----Maine ----Maryland ----Massachusetts ----Michigan ----Minnesota ----Missouri ----Montana ----Nebraska ----New Hampshire ----New Jersey ----New York ----North Dakota ----Ohio ----Pennsylvania ----Rhode Island ----South Dakota ----Vermont ----Washington ----Wisconsin ----Wyoming ----Local choice for high school California ----Notes: California has a mandatory state list only until 8 th grade. The state lists and links to relevant state laws are available upon request. The results presented for the short and long list classification in this paper correspond to the 9 th grade. The long and short list classification was established for each grade separately, with very similar results being obtained for each grade.
33 Table 1 2. Description and sources: State level textbook selection variables. Independent Variable Description Source Cross section: US Census Bureau (Census of Population). Year: 2007 Panel: US Census Bureau (Decennial Census of Population 1940 2000). Years: 1970, 1980, 1990, 2000. Advanced Degree % Advanced D egree (more than b Cross section: US Census Bureau Census of Population. Year: 2007 District Size Total students in a state / number of districts in a state. Thus, a larger number (1/1 vs 500/1) means larger districts. Cross section: Common Core. Year: 2007 Panel: Common Core. Years: 1969 70, 1979 80, 1989 90, 1999 00 District Size Dummy Dummy takes value of 1 if average district is large, value of 0 if district is small. A large district is defined as a district corresponding to the 8 th 10 th deciles when looking at the distribution of students/districts across states. Cross section: Common Core. Year 2007 Homeownership % Homeowners Cross section: US Census Bureau: Household by state: 1984 Panel: US Census Bureau: Census of Housing, 1900 2000. Years: 1970, 1980, 1990, 2000. Fundamentalist Total adherents of fundamen talist churches as a percent of total population Cross section: Glenmary Research Center. Year 2000. Panel: Glenmary Research Center. Years: 1971, 1980, 1990, 2000. Fundamentalist Dummy Dummy takes value of 1 if more fundamentalists than the average, value of 0 if otherwise. Cross section: Glenmary Research Center. Year 2000.
34 Table 1 2 continued Independent Variable Description Source State Revenue School system revenue from state *100 / total School system revenue. Cross section: Census of Governments Survey of Local Government Finances School Systems (US Census Bureau 2007). Year 2007. Panel: US Dept. of Education, NCES, Statistics of State School Systems and Common Core. Digest of Education Statistics (2002). Years: 1969 70, 1979 80, 1 989 90, 1999 00. Gini Cross section: State level household income inequality measure (Gini coefficient). Panel: State level family 14 income inequality measure (Gini coefficient). Cross section: US Census Bureau, Income Report 2007. Year 2007. Panel : US Census Bureau, Historical Income Tables for States. Years: 1969 70, 1979 80, 1989 90, 1999 00. Total union expenditures by largest population by state. 14 The State Income Gini is m easured at the family level instead of the household level because data for 1970 is only available at the family level. The Census Bureau defines household as a group of persons occupying a housing unit, thereby including families and unrelated individuals A family is defined as a group of two or more persons related by birth, marriage, or adoption and residing together.
35 Table 1 3. Summary statistics: State level textbook selection variables. Variable Cross section Excluding South Carolina N=48 Cross section Including South Carolina N= 49 Panel N=196 Mean (S.D.) Max (Min) Mean (S.D.) Max (Min) Mean (S.D.) Max (Min) 26.7563 (4.7488) 37.9 (17.3) 26.6898 (4.7221) 37.9 (17.3) 17.4767 (5.972) 33.19 (6.7) Advanced Degree 9.65 (2.4515) 16 (6.4) 9.6204 (2.4346) 16 (6.4) District Size 4,831.359 (7,363.557) 35,237.5 (337.643) 4,897.953 (7,301.34 ) 35,237.5 (337.43) 4,745.5220 (6,411.5430) 37,500 (235.211) District Size Dummy 0.2917 (0.4593) 1 (0) 0.3061 (0.4657) 1 (0) 0.3061 (0.4621) 1 (0) Homeownership 70.1917 (4.3169) 77.6 (55.9) 70.2714 (4.3080) 77.6 (55.9) 66.6827 (4.9488) 75.2 (47.3) Fundamentalist 9.8667 (10.6749) 37 (0.9) 10.2061 (10.8271) 37 (0.9) 11.5788 (11.6312) 47.0900 (0.1537) Fundamentalist Dummy 0.2708 (0.4491) 1 (0) 0.2857 (0.4564) 1 (0) 0.3010 (0.4599) 1 (0) State Revenue 49.8532 (11.9998) 87.7584 (31.7056) 49.7302 (11.9053) 87.7584 (31.7056) 46.1500 (13.5360) 73.9902 (8.2357) Gini 0.4483 (0.0194) 0.4985 (0.4104) 0.4458 (0.0192) 0.4985 (0.4104) 0.3820 (0.0325) 0.4720 (0.3170) 21.9264 (13.0259) 63.8157 (2.1544)
36 Table 1 4. Fundamentalism by state Type of state State % Fundamentalist Mandatory list Alabama 35.6 Florida 10.8 Louisiana 19.4 Mississippi 36.6 New Mexico 9.2 N. Carolina 22.5 Oklahoma 34.5 Tennessee 29.5 Average for mandatory list 24. 8 Recommended list Arkansas 37.0 Georgia 24.5 Idaho 4.7 Indiana 5.9 Kentucky 27.8 Nevada 3.6 Oregon 6.3 S. Carolina 26.5 Texas 19.5 Utah 1.3 Virginia 13 .0 W. Virginia 7.3 Average for recommended list 14.8 Local choice Alaska 7.6 Arizona 5.7 Colorado 4.8 Connecticut 1.1 Delaware 2.9 Illinois 4.7 Iowa 3.3 Kansas 7.1 Maine 1.8 Maryland 4.4 Massachusetts 1.4 Michigan 3.8 Minnesota 4.2 Missouri 17.4 Montana 6.1 Nebraska 3.7 New Hampshire 1.4 New Jersey 1.4 New York 1.1 North Dakota 4.1
37 Table 1 4. Continued Type of state State % Fundamentalist Ohio 4.6 Pennsylvania 2.3 Rhode Island 0.9 South Dakota 4.9 Vermont 1.6 Washington 5.6 Wisconsin 6.2 Wyoming 6.7 Average for local choice 4.3 Local choice for high school California 3.8
38 Table 1 5. Ordered logit cross section regression explaining state textbook policies Variable Mean (S.D.) Coef ficient (t stat) (1) Coef ficient (t stat) (2) State Revenue 49.7302 ) (11.9053) 0.0774** 3) (1.99) 388** 0.0612* ) ** (1.68) 2 ** 2* Homeownership 70.2714 ) (4.3080) 0.1405 3**) (1.33) 2 3 2** 0.1023 ) ** (1.02) *22 ** District Size 4,897.953 0 ) (7,301.346 0 ) 0.00016** ) (2.43) 222 ** District Size Dummy 0.3061 ) (0.4657) ... 2.3145*** ) (2.75) 22*** Advanced Degree 9.6204 ) (2.4346) 0.5851* )*2 ( 2.55) **222 0.5280** )* ( 2.34) 22*** Fundamentalist Dummy 0.2857 ) (0.4564) 2.7096** )2 (2.26) **222 2.5350** )* (2.14) 22*** Gini 0.4458 ) (0.0192) 56.4161 )**2 (1.55) **222 42.9408 )*** (1.30) 22*** Pseudo R 2 0.4951 )**2 0.4536 )*** Notes: (1) Dependent variable is 0 if local choice state, 1 if recommended state, and 2 if mandatory state list. High school biology textbooks (2 ) Dependent variable is 0 if local choice state, 1 if recommended state, 2 if long ma ndatory state list, and 3 if short mandatory state list (using average as cut off for long and short lists). 9 th grade biology textbooks (similar results were obtained for 10 th 11 th and 12 th grade biology textbooks). Number of observations = 49. 2 tail ed tests. = significant at 10% level, ** = significant at 5% level, *** = significant at 1% level T statistics in parenthesis.
39 Table 1 6. Ordered logit cross section regression explaining state textbook policies Variable Mean (S.D.) Coef (t stat) (1) Coef (t stat) (2) State Revenue 49.8532 (11.9998) 0.0451 ** ) (1.23) 2 ** 2 0.0587 )** (1.58) 22** Homeownership 70.1917 ) (4.3169) 0.1090 ) ** (0.95) 2 ** 2 0.0894 )** (0.83) 22** District Size 4,831.359 )2 (7,363.557) 2 0.0001** ) (2.36) **22 0.0001** ) (2.40) 22** Bachelor s 26.7563 ) (4.7488) 0.1870* )* ( 1.78) 22** 0.2148** ) ( 2.14) 22** Fundamentalist 9.8667 ) (10.6749) 0.0971* )* (1.78) 22** Fundamentalist Dummy 0.2708 ) (0.4491) 2.5790** ) (2.03) 22** Gini 0.4483 ) (0.0194) 41.8208 )* (1.23) **22 34.6331 )** (1.03) 22* 21.9264 ) (13.0259) 0.0754* )* ( 1.72) 22** 0.0440 )** ( 1.02) 22** Pseudo R 2 0.4887 )** 0.4449 )** Notes: (1) Dependent variable is 0 if local choice state, 1 if recommended, and 2 if mandatory state list. High school biology textbooks (2) Dependent variable is 0 if local choice state, 1 if recommended, 2 if short mandatory state list, and 3 if long mandator y state list (using average as cut off for long and short lists). 9th grade biology textbooks (similar results were obtained for 10th, 11th, and 12th grade biology textbooks). Number of observations = 48. 2 tailed tests. = significant at 10% level, ** = significant at 5% level, *** = significant at 1% level T statistics in parenthesis.
40 Table 1 7. Choice / Non Choice logit panel explaining state textbook policies. Variable Mean (S. D. ) Coef (t stat) (1) Coef (t stat) (2) Coef (t stat) (3) State Revenue 46.1500 (13.5360) 0.0467*** )2 (2.70) 22*2** 0.0502 *2*) (1.97) **222 0.0467* )s (1.79) 22s* Homeownership 66.6827 ) (4.9488) 0.0065 )*2** (0.13) 22***2 0.0170 )2** (0.27) **222 0.0065 )s* (0.10) 22s* District Size 4745.5220 ) (6411.5430) 0.00005 )*** (1.57) 222*** 0.00007* )* (1.71) 222** 0.00005 )* (0.94) *299 17.4767 ) (5.9272) 0.1469 *)**2 ( 1.83) 222*** 0.1354 )2** ( 1.07) 222** 0.1469 )*2 ( 1.17) *222 Fundamentalist 11.5788 ) (11.6312) 0.0855** (2.45) 222*** 0.1109** )2 (2.21) 222** 0.0855 )*2 (1.48) *222 Gini 0.3820 (0.0325) 7.2462 (0.66) 9.3478 )**2 (0.68) 222** 7.2462 )*2 (0.53) *222 Pseudo R 2 0.3999 )***2 0.3952 )2** 0.3999 )*2 Notes: N = 196. 2 tailed tests. = significant at 10% level, ** = significant at 5% level, *** = significant at 1% level. Dependent variable is 0 if local choice and 1 if nonchoice. Robust Standard Errors are used in all regressions. (1) Includes tim e fixed effects and south dummy. ( 2) Includes time fixed effects and clustering by state (3) Includes time fixed effects, south dummy, and clustering by state T statistics in parenthesis.
41 Figure 1 1. Belief in evolution, by edu cation level (Gallup poll). Source: Newport (2009)
42 CHAPTER 2 WHAT CHARACTERIZES S CHOOL DISTRICTS THAT CHOOSE TO OPT OUT OF A MMENDED TEXTBOOK LIS T? EVIDENCE FROM INDIANA As mentioned in C hapter 1, some states provide textbook lists that are highly recommended but allow for districts to deviate from those lists throu gh the use of waivers. In C hapter 2 I analyze factors that are expected to influence petitions for the use of other textbooks at the school district level for the state of Indiana. Specifically, the second section of this paper provides some insight into what characterizes districts recommended biology list during the year 2004. The resul ts provide evidence supporting the costs of decision making as measured by gini coefficients for the state of Indiana, suggesting that counties that are more unequal in income distribution have a harder time reaching agreements. In addition to this, the r esults suggest that citizens who are more fundamentalist than average are more likely to request a waiver to opt out 2.1 Background The textbook selection policies studied examine biology textbooks specifically because evolution has always been the most controversial topic. Moore, Jensen and Hatch (2003) indicate that in the 1920s, Tennessee, Arkansas, and Mississippi passed laws that prohibited the teaching of human evolution According to these authors, Mississippi was the last state to nullify its ban on the teaching of evolution in 1970. Nowadays, teachers encounter pressure from both sides. On the one hand, several authors have noted that teachers are pressured to avoi d teaching evolution (Zimmerman 1987; Kraemer 1995; an d Randak 2001 as cited in Moore 2004). On the
43 other hand, both court decisions and a variety of professional scientific societies have consistently supported t he teaching of evolution (Moore 2004). The state of Indiana, one of the thirteen recommended list states, provides an ideal setting to study the factors that characterized individual school districts that commended textbook list. The Indiana Department of Education has collected and published easily accessible data on their adoption process, laws, adoption outcomes, and other district level variables of int erest which are described in section 2.2 Indiana has 291 school districts. Figure 2 1 illustrates the distribution of school districts that received waivers f or high school science classes during the year 2004. As shown in Figure 2 1 there are districts requesting waivers in all six regions. A brief textbook adoption code for the year 2004 the last year for which data on science textbook adoptions at the school district level was available, is provided below (Indiana legis lative services agency 2004) The p rocedures for textbook adoption must include the involvement of both teachers and parents on an advisory committee for the preparation of recommendations. The majority of members must be teachers but no less than 40% of the committee must be comprised of parents. The local superintendents must forward to the state board of education a list of their selections for all subjects and grades. The board shall examine these lists, and, if the board finds a deviation from the state adopted list and that there has been no waiver granted under section 27 of the chapter, the board shall notify the local superintendent of the deviation. If the school corporation does not comply with this chapter within thirty days of receiving the notification, the board shall cancel the accreditation of the offending schools. After giving the advisory committee an opportunity to give its recommendation, the governing body of a school corporation may request a waiver from the adoption requirements if the governing body believes that t he educational needs of the students attending that school corporation can best be served by adopting no textbook or adopting a textbook that has not been adopted by the commission.
44 A request for a waiver must be submitted on a form approved by the Indian a State Board of Education before June 1 of the year preceding the first school year for which the waiver is to apply. The Indiana State Board of Education shall grant the waiver if it determines that the request is reasonable. Thus, for the year 2004, I ndiana had a system in place in which parents were permitted to influence the choice of textbooks used in the classroom, reflecting local choice. This local decision was, however subject to approval from the state, which also added difficulty to the proc ess by requiring extra steps for the waiver procedure to be students to achieve profici submit a waiver form and obtain approval from the Indiana State Department of met the state standards. It is important to note that the content of biology textbooks that are selected in this state plays a very important role. Recent studies found that even though Indiana has some of the best state standards regarding evolution 43% of its biology teachers avoid scientifi c validity (Rutledge and Warden 2000; Rutledge and Mitchell 20 02, cited in Moore 2004). 2.2 Empirical Model What characterizes school districts that opt out of a re commended state list? In order to assess this question, the following cross sectional logit model is estimated: (2 1)
45 The dependent variable waiver 0 if a school district uses a textbook from the state recommended list 1 if a school district opts out of the state rec ommended list by using a waiver adoption report by category which is available in their website at the school district level. Summary statistics and sources for the explanatory variables used can be found in Table 2 1 As mentioned earlier, the term fundamentalist refers to churches that are classified C hapter 1 the results from the state level analysis suggest that the probability of being a local choice state decreases as states have more fundamentalist adherents when compared to those that do not. These results are to be expected, given that fundamentalists feel more strongly about evolution than others and are, thus, more active and effective in bringing about polici es they prefer. These results do not, however, give us any insight in regards to what happens in states that do not have mandatory lists, perhaps because fundamentalists do not constitute a large enough group to demand one, as seems to be the case with th e state of Indiana. In Indiana, the percent fundamentalist at the state level is 5.9%, which is much lower than the percent for the average state (10.20%) Furthermore, judging by ndiana has some of the best evolution education state standards in the nation (Rutledge and Warden 2000, Rutledge and Mitchell 2002, quoted in Moore 2004). Given these circumstances, you would expect fundamentalist districts to clearly be better off by re questing a waiver. It is important to note that what matters in this case is how fundamentalist each school district is relative to the state (given that the recommended list is chosen at the state
46 f undamentalist dummy was calculated for each county using the following steps: Step 1: Calculate fundamentalist difference where: Fundamentalist Difference = % Fundamentalist in county % Fundamentalist in the state of Indiana Step 2: Create fundamen talist dummy where fundamentalist is: 0 if the county is either less fundamentalist than the state or just slightly more fundamentalist than the state (by less than 1 percentage point). 1 if the county is much more fundamentalist than the state (more fundamentalist by more than 1 percentage point). The implicit assumption here is that being just slightly more fundamentalist than the state may not cause you to go through the extra effort required to change a textbook, but being much more fundamentalist than the state would. The cut off used, suggesting support for this assumption, provided the best fit. As was the case with the state level analysis, the classification of fundamentalist churches used was taken from the morality politics literature and follows Johnson (1976). Educational attainment is also included. As mentioned in C hapter 1 several studies have linked greater educational attainment to more confident, better informed and more involved electorates. It is anticipated that states with m ore educated people would demand (or be given) a waiver if the state list does not match their beliefs, because they would be expected to be more involved in the decision process and to be more confident in their decisions. It is also expected that states with more educated residents would be more likely to prefer relatively evolutionist content, given that they
47 would be expected to have a better understanding of science. Several studies have found a link between education and evolution beliefs. As an ex ample, a recent Gallup those with p 2009). Furthermor school districts with more educated citizens to be less likely to request a waiver. The is captured by th e variable Since several steps and agreements between school district parents and teachers are required to be able to decide to opt out from a state list, apply for and obtain a waiver, the size of a school district is expected to influence the use of waivers. As me ntioned earlier, one would expect less parental involvement in larger districts, since there is a large incentive to free ride. Thus, you would expect large districts to have a harder time reaching agreements and applying for waivers. District size in th is context district size In a similar fashion, one would expect heterogeneity between households within a district to affect the use of waivers. For this reason, a gini coeff gini providing a measure of household income inequality, is used. School districts are expected to be less likely to get a waiver when they are more unequal (i.e. have a higher gini coefficient). This variable is a proxy for the cost s of reaching a decision. The higher the inequality, the harder it is to reach a decision regarding a waiver. The
48 gini coefficients were calculated from the US 2000 Census by Professor Mark Burkey from NCA&T State University (Burkey 2000). A variable me asuring rurality is also included. This is because urban inhabitants information than their rural counterparts that induces (Sass and Saurman 1993 p.163 ). In order to control for the re lative rurality of school districts, dummies using the 2004 Locale Census definitions were created. The categories used are listed in Table 2 1 In addition to regional c lassification. These regions are listed in Appendix D Given that some variables were only available at the county level, errors were clustered at the county level. 2.3 Results The logit model results for the use of waivers analysis for the state of Indi ana are available in Table 2 2 Results with the marginal effects calculated as Average Partial Effects (APEs) and Partial Effects at the Average (PEAs) for e ach outcome can be found in Appendix B The fundamentalist variable is defined as a dummy taking a value of 1 if the county is more fundamentalist than the state of Indiana by more than 1 percentage point and 0 otherwise. The coefficient of this variable is significant and, as hypothesized, positive. These resu lts indicate that the probability of having a waiver increases when a county is more fundamentalist than the state of Indiana by more than 1 percentage point when compared to those counties that are not (i.e those counties that are either less fundamentali st than the state or more fundamentalist but by less than 1 percentage point). The marginal effects from Table B 1 indicate that the probability of having a
49 waiver increases, for example, by 9% when a county is more fundamentalist than the state of Indian a by more than 1 percentage point than otherwise. These results are to be expected given that, as explained in the background section, Indiana is a mostly pro evolution state that does not have a mandatory list. Under these circumstances, one would expec t fundamentalists to be better off by requesting a waiver. The income gini variable is slightly significant and negative indicating that the probability of having a waiver decreases when a county is more unequal (has a higher gini coefficient). This picks up the costs of reaching a decision. The more unequal a county is, the harder it is to make a decision and reach agreements, so the less likely the school districts would be to get a waiver. There is lack of support for the predictions regarding educatio nal attainment and district size implying that, unlike the state level predictions, education and district size do not seem to play a role in the use of waivers for the state of Indiana. 2.4 Conclusion In C hapter 2 I examine factors associated with Indi ana districts opting out of the state list of approved textbooks. Fundamentalism seemed to play a very important role in the use of waivers for the state of Indiana, a state that has a smaller proportion of fundamentalists than other states, strong evolut ion educational standards and relatively pro evolutionist content coverage in its state recommended textbooks. The evidence suggests that, as expected, it is the fundamentalist groups that can stand to benefit the most from the use of waivers in this type of environment. The study also finds support for decision making costs being greater in more heterogeneous districts. Specifically, the presence of higher income inequality within a
50 county is found to lessen the probability that a district seeks a waiver from the Indiana state list of approved textbooks.
51 Figure 2 1. Geographical distribution of Indiana school district waivers. The shading indicates school districts that used a waiver for Science textbooks during the year 2004. Map Source: Indiana Bus iness Research Center using Tiger/Line 2006 Second Edition Files from the US Census Bureau, August 2007 Accessed 14 July 2011. Available http://www.stats.indiana.edu/maptools/scho ol_districts.asp
52 Figure 2 1. Continued.
53 Table 2 1. Summary statistics and source s: Indiana district level study Independent Variable n=291 Mean (S.D) Max (Min) Description Source 18.8084 ) (11.1516) 78.7505 ) (4.3919) % of population over 25 years old ho lding a b higher. District level variable. School District Demographic System (SDDS). Year: 2000 District Size 3,449.014 0 (4,381.217 0 ) 39,989 ) (149) Number of students enrolled in the district. District level variable. National Center for Education Statistics Common Core of Data. Year: 2004 Regional Dummies Regions determined by the Indiana Department of Education. See Appendix for details. Indiana Department of Education. Year: 2004 Fundamentalist Dummy 0.5017 ) (0.5 ) 1 ) (0) Dummy takes value of 1 if the county is more fundamentalist than the state by more than 1 percentage point and 0 otherwise. County level variable. Glenmary Research Center.Year: 2000 Rurality Dummies (omitted dummy) Rural outside CBSA/MSA (1) Rural inside CBSA/MSA (2) Small town, Large town and Urban fringe of midsize city (3) Urban fringe of large city, Midsize city, and Large city. District level variable. NCES. Common Core. Year: 2004.
54 Table 2 1. Continued Independent Variable n=291 Mean (S.D) Max (Min) Description Source Gini 0.4031 ) (0.0258) 0.4719 ) (0.3447) Gini coefficients calculated by M. Burkey using US Census 2000 data. County level variable. website. NCA&T State University, Greensboro, NC. Year: 2000
55 Table 2 2. Indiana district level logit model explaining textbook waiver use. Variable Mean (S.D.) Coef (t stat) Fundamentalist Dummy 0.5017 (0.5009) 0.9366** (2.26) Gini 0.4031 (0.0258) 13.7255* ( 1.84) Rurality_1 Dummy 0.954 (1.77) Rurality_2 Dummy 0.5405 (0.96) Rurality_3 Dummy 1.1611 (1.60) District Size 3,449.0140 (4,381.2170) 0.00004 ( 0.78) 18.8084 (11.1516) 0.03 ( 0.92) Regional_1 Dummy 0.1007 (0.17) Regional_2 Dummy 0.5916 ( 0.94) Regional_3 Dummy 1.344 ( 1.69) Regional_4 Dummy 0.1566 ( 0.31) Regional_5 Dummy 1.2932 ( 1.26) Pseudo R 2 0.08 Notes: N = 291. = significant at 10% level, ***=significant at 5% level, *** = significant at 1% level. 2 tailed tests. Dependent variable is 0 if do not use a waiver, 1 if use a waiver. Robust standard errors are clustered at the county level. T statistics in parenthesis.
56 CHAPTER 3 INEFFICIENCY IN JAPA NESE WATER UTILITY F IRMS: A STOCHASTIC F RONTIER APPROACH 3.1 Introduction Benchmarking is an important and relevant tool equipping managers and decision makers with powerful information in the form of performance based rankings. Once performance based rankings are established an obvious question arises: What is behind predicte d differences in efficiency between firms in an industry? What characterizes firms that perform better than others? What can governments do to increase the efficiency of firms in an industry? The main purpose of this paper is to identify factors influenc ing inefficiency of capacity to maximize output given a fixed level of inputs. Specifically, a firm is considered to be efficient when it is producing the maximum amount of output given its input endowment (reflecting past investment and operating decisions). This is a very important policy issue, because the need for water in day to day activities must be good. Can the supply of water be improved with existing resources by tackling the inefficiencies of firms rather than by increasing expenditures? If yes, the starting point is to figure out where these inefficiencies lie. This could be followed by the establishment of mechanisms to that motivate firms to emulate efficient firms, thereby encouraging them to obtain more output for a given set of inputs. According to Horn and Saito (2011) several countries promote performance efficiency. Examples abound: the UK imposed incentive
57 enhancing regulations (price caps), the Netherlands m erged small water utilities to form large water utilities, and France has allowed for private sector participation. To my knowledge, this is the first paper to examine factors influencing the inefficiency of Japanese water utility firms using the one step Stochastic Frontier Analysis (SFA) inefficiency effects model proposed by Battese and Coelli (1995). Other studies have examined Japanese water utilities using several Data Envelopment Analysis and Stochastic Frontier Analysis techniques. 3.2 Background : Japanese Water Utilities Water services in Japan are supplied by a very large number of suppliers, many of which are very small. The ten largest suppliers provide 28% of output, while the other 1,180 firms provide the remainder. This is a consequence of ownership structures: utilities are mainly owned by the local governments of cities, towns and villages (Ueda and Benouahi 2009). They operate under a municipality owned principal that promotes independent local monopolies. Regulation of the water and sanitation sector is the responsibility of the Ministry of Health, Labor, and Welfare, Ministry of Land Transportation and Tourism and Ministry of the Environment. In 2005, Japanese water utilities had cover age of 97.2% (Ueda and Benouahi 2009). The Mini stry of Internal Affairs and Communication is in charge of collecting data and providing performance indicators for water utilities. These performance indicators consider just one characteristic at a time, and include variables such as water losses, finan cial performance, and revenue collection. Because these indicators examine just one characteristic at a time, they do not allow for direct comparisons between firms. Furthermore, as noted by Marques, Berg, and Yane (2011) if estimates are to be used
58 by p olicy makers to reward or punish firms, they must take into account elements beyond the control of management. Recent studies analyzing Japanese water utilities use a variety of techniques. Marques, Berg, and Yane (2011) apply non parametric Data En velopment Analysis (DEA) to the same dataset used in the present paper using a new technique that takes into account exogenous variables one at a time. Aida et al. (1998) evaluate the performance of water suppliers in Japan using a range adjusted measure of efficiency (RAM) for the year 1993. Yane and Berg (forthcoming 2013) use a translog production function to examine the robustness of efficiency score rankings using different distributional assumptions in a Stochastic Frontier Analysis setting. Mizuta ni and Urakami (2001) study water utilities in Japan with an emphasis on size and scale economies. They conclude that the optimal size of a water supply system in Japan is more than 10 times greater than the average size of even the large water supply sys tems at the time, so they note that scale is an issue. Urakami and Parker (2011) examine the effects of consolidation amongst Japanese water utilities, using translog cost functions with a sample from 1996 to 2006. They find that consolidation had a smal l but beneficial impact on cost effectiveness and believe that the cost savings were offset by extra expenditures incurred by having to supply water to areas with low population density (Urakami and Parker 2011). Horn and Saito (2011), estimate cost effic iency and scale economies using SFA for a sample from 1999 2008. This paper differs from Horn and Saito (2011) in that it examines a production function (rather than a cost function) and analyzes inefficiency effects in addition to inputs. The main contr ibution of this paper to the benchmarking literature of water utilities in Japan
59 consists on identifying factors contributing to inefficiency in a SFA framework. This is followed by policy implications. 3.3 Explanation o f Stochastic Frontier Models and How They Compare t o O LS The majority of studies measuring technical efficiency use either DEA, a non parametric method or SFA, a parametric method According to Coelli (2005) some advantages of SFA over DEA are that (1) it accounts for noise 1 and (2) it can be used to conduct conventional tests of hypotheses. Some disadvantages are (1) the need to specify a distributional form for the inefficiency term and (2) the need to specify a functional form for the production function. Stochastic fro ntier models were first developed by Aigner, Lovell and Schmidt (1977) and Meeusen and van den Broeck ( 1977 ). The main difference between Stochastic Frontier Analaysis and OLS lies in the presence of a one sided error term. In SFA there are two error ter ms: the OLS random error term (representing noise) and an additional error term representing productive inefficiency. Unlike OLS, these models assume that firms can be inefficient and predict efficiency scores for each firm 2 An obvious next step in thi s regard is to try to understand the factors influencing these inefficiencies. Two approaches have been followed in the literature when trying to inefficiencies : a one step and a two step approach. Earlier studies focus on a two step appro ach consisting of a stochastic frontier model in the first stage typically followed by a regression of the inefficiency effects predicted by the first stage against exogenous variables such as firm size, age of employees, etc. in the second 1 In DEA, all deviations from the frontier are considered to be d ue to inefficiency. The data are assumed to be completely nois e free ( Kumbhakar and Bokosheva 2009). 2 If all firms were able to operate at the frontier (i.e. be fully efficient), SFA would not be required.
60 stage. These m odels attempt to provide an explanation for the predicted inefficiencies affecting the firm. For example, Pitt and Lee (1981) study firm ownership, age, and size as sources of inefficiency for the Indonesian weaving industry. This approach is no longer recommended. According to Coelli et al. (2005), failing to include environmental variables in the first stage can lead to biased estimators of the parameters and to biased predictors of efficiency. Schmidt and Wang (2002) examine one step and two step ap proaches and provide Monte Carlo evidence in favor of one step approaches. They also find that two stage approaches lead to biased estimates, and that these biases are substantial. Both authors suggest a one step approach. In a one step approach, envir onmental variables are allowed to directly affect the stochastic component of th e production frontier (Coelli et al. 2005). In these models, both the stochastic frontier and the way in which the one sided error depends on environmental variables can be es timated in a single step (Schmidt and Wang 2002). There are several studies examining efficiency effects using a one step approach. For instance, Iraizoz et al. (2005) examine the influence of age, debt, and other variables on inefficiencies for the Span ish beef sector in the 1990s. Zhu, Karagiannis, and Lansink (2011) examine the impacts of income transfers, farm size, degree of specialization, and other variables on the technical efficiency of Greek olive farms. To my knowledge, this is the first pape r applying this method to Japanese water utility firms. 3.4 Empirical Model Battese and Coelli (1995) Model This paper follows a one step stochastic frontier approach to measure inefficiency effects proposed by Battese and Coelli
61 (1995) model considers a stochastic frontier production function for panel data of the form: (3 1) or, alternatively, as presented by Coelli et al. (2005) (3 2) determ inistic noise inefficiency component composite error term Where Y it denotes the production at the t th th is a (1 x k) vector of values of known functions of inputs; is a (k x 1) vector of unknown parameters to be estimated ; the 2 v ) random errors, inde pendently distributed of the the negative random variables, associated with technical inefficiency of production, which are assumed to be independently distributed, such that is obtained by truncation (at zero) of the norm al distribution with mean, z it 2 ; z it is a (1 x m) vector of explanatory variables associated with technical inefficiency of production of firms over time; and This model uses the maximum likel ihood method for simultaneous estimation of the parameters of the stochastic frontier and the model for technical inefficiency effects
62 (Battese and Coelli 1995). In this context, technical efficiency is defined as a variable taking a value between 0 an d 1 of the i th firm relative to the output that could be produced by a fully efficient f irm using the same input vector (Coelli et al. 2005). It is important to emphasize that this model differs from other SFA models in that the it 3 This allows for the estimation of inefficiency effects for the firms being studied. We estimate the following Cobb Douglas 4 stochastic frontier production function: ( 3 3 ) where the technical inefficiency effects are assumed to be defined by: (3 4 ) where ln denotes the logarithm with base e. The focus of this paper is on the identification and policy implications of the inefficiency effects and not in identifying the most efficient producers or prov iding rankings of Japanese utility firms. We use a balanced panel consisting of 1,190 water utility firms over a 4 year period totaling 4,760 observations for the years 2004 2007. The data are from the (Chihou Kouei Kigyou Nenkan). The variables used in the production function are summarized in Table 3 1 5 The 3 The z it 4 An alternative trans log specification was tested, but the estimations did not converge. 5 For input observations that have original values of zero, we adopt the standard practice of calculating log values by adding one to these original values.
63 variables used in the inefficiency effects section of the model are summarized in Table 3 2. This paper uses the same output and inputs as Yan e and Berg (forthcoming, 2013). Summary statistics are provided in Table 3 3 The model was estimated using Frontier 4.1. software developed by Tim Coelli. The coefficients estimated for both the production function and inefficiency effect s model are p resented in Table 3 5 The first step is to test whether Stochastic Frontier Analysis is necessary. Since in stochastic frontier models the composite error is given by one can essentially test whether the part of the model is necessa ry. If it is not necessary, OLS would provide consistent estimates, because there would not be a need for an inefficiency component. The estimate for gamma 6 is 0.29 (t stat 5.09). Thus, gamma is statistically different from zero at the 1% level. This m eans that at least some of the variation in the composite error term is due to the inefficiency component, and that SFA is preferred over OLS. As noted earlier, one of the disadvantages of SFA is that different distributional assumptions can result in different predictions of technical efficiency 7 Since this model specification does not have the Battese and Coelli (1992) model as a special case and is non nested, it does not have a set of restrictions that can be defined to permit a test of one speci fication versus another (Coelli 1996). A few alternative spec ifications are available in Appendix G Readers can also refer to Yane and Berg (forthcoming, 2013) 6 Where gamma is defined as and varies from 0 to 1. A gamma value of 0 would indicate that OLS provides consistent estimates. 7 Nevertheless, when firms are ranked based on predicted technical efficiencies, the rankings are often robust to distributional choice s (Coelli 2005).
64 who use the same dataset (for a 2 year period) to test alternative SFA specifications for Japanese water utilities. It is imp ortant to note, however, that with a sufficiently large number of periods, a SFA model using panel data can help in mitigat ing the strong distributional assumptions that are necessary to disentangle the effects of inefficiency and noise and reduce the bias du e to unobserved heterogeneity (Coelli et al. (2005) and Farsi et al. (2005)) The mean efficiency score for Japanese firm s in the 4 year sample was 0.54. The least efficient firm had a score of 0.09, while the most efficient firm had a score of 1 (i.e. fully efficient). Summary statistics for predicted efficiencies and a graphical yearly representation of efficiency scores as they relat e to output are available in Appendix F Production Function As expected, the inputs have a positive impact on output. All inputs are statistically significant at conventional levels. An increase of 1% in length of pipe (K), for instance, is associated with an increase in total delivered water volume of 0.33%. The sum of the coefficients is essentially 1 (0.99), indicating constant returns to scale. Inefficiency Effects As mentioned earlier, the main focus of this paper consists on analyzi ng the factors contributing to inefficiencies in Japanese water utility companies. Several factors influence technical inefficiency according to the estimates from Table 3 5 The average operation rate, customer density and size variables are associated with lower levels of inefficiency (or higher levels of efficiency). The raw water proxy, subsidy and outsourcing ratio variables are associated with higher levels of inefficiency 8 The 8 Or, alternatively, decreases in technical efficiency
65 numbers in Table 3 5 show only the direction of the effects and shoul d not be interpreted as marginal effects. The results are discussed further below. The subsidies variable (rsubp) is positively significant, implying that firms with higher levels of subsidies are more inefficient. Currently, the subsidy schemes for inve stments and operations come from two pots of money and do not seem to be directed at rew arding strong performance. F ollow ing Scotti et al. (2012), I rule out the possible endogeneity between inefficiency and subsidies based on the decision for subsidies b eing mainly a political one (and not one that is taken on the basis of efficiency reasons). I c heck for this statistically in F igure F 7 by examining the distribution of subsidies for the top 40 and bottom 40 firms efficiency wise. There is no visible pattern, suggesting that subsidy decisions are not based on efficiency levels. Based on the literature examining stochastic frontier analy sis efficiency effects models, it is expected that subsidies will decrease technical efficiency. Rezitis, Tsiboukas, and Tsoukalas (2003) find that European Union subsidies for farmers are associated with decreases in technical efficiency. This is attribu incentives to achieve higher productivity and profitability, in turn reducing their motivation for improving efficiency in the production process (Rezitis, Tsiboukas, and Tsoukalas 2003). Zhu, Karagiannis, and Lansink (2011) study direct income transfer subsidies on greek olive farms and find a negative impact on technical efficiency. They suggest that the motivation for improving technical efficiency is lower when subsidies are available. The findings are consistent with the literature. The average operation rate variable (aveop) is a measure of capital utilization. It is expected that the more you utilize your capacity the more efficient your firm will be.
66 Similarly, low rates of capacity utilization are often time s cited as contributing to high costs in small scale water utilities (Horn and Saito 2011). This variable has a negatively significant coefficient, indicating that firms with more average water delivered per delivered water capacity tend to be less ineffi cient (more efficient). This result is expected, since these are firms that are operating at a higher capacity level and thereby making better use of their available resources. These results are consistent with the Urakami and Parker (2011) study which f inds that high capacity utilization has a positive impact on cost savings of 26 % per year The customer density variable (cusden) has a negative coefficient. According to this model, firms with more customer density tend to be less inefficient (m ore efficient). Again, this result is to be expected. Assuming a fixed network length, for instance, adding more and more customers should translate into higher levels of output given these fixed input levels. In relation to size (size), larger firms are expected to be more efficient (at least up to a point) if economies of scale are present. The results are negatively significant, implying that larger firms tend to be less inefficient (more efficient) than smaller sized firms for the sample studied. (2001) study which finds that firms in Japan are much smaller than the optimal size. More recently, Urakami and Parker (2011) also find that there are economies of scale for water utilities in Japan. Horn and Saito (2011) find economies of scale for small water utilities but suggest that scale diseconomies are likely to be present for larger utilities. It is important to note that the government of Japan has already started to address this issue: Ur akami and Parker (2011) mention that consolidation efforts have
67 taken place for political rather than economic reasons. Specifically, in 1999 under the that local governmen ts could be consolidated into large authorities. Thanks in part to this law, the number of Large Water Suppliers fell from 1,958 in the year 2000 to 1, 602 in 2005 (Urakami and Parker 2011). The results of this paper suggest that further consolidation may be beneficial. The raw water ratio proxy (rraw) is positively significant. This result is puzzling, as it implies that firms with more raw water availability (i.e. firms that are better endowed naturally) are more inefficient. It may be that depende nce on local resources requires greater treatment costs than when water is provided by a raw supplier. The finding that self (2011) study in which a high percentage of purchased water leads to an incre ase in costs of 1.6 2.2 % per year when compared to exploiting owned resources. This issue warrants further attention in further studies. Outsourcing was also examined. According to Ueda and Benouahi (2009), outsourcing in Japanese water companies takes place for (1) routine O&M of treatment plants and network pipes, (2) checking and executing repair works, (3) engineering design and construction supervision, (4) information and telecommunication technology services, and (5) metering and billing. Whole operations can be outsourced to other utility companies or to private companies. There are several hypotheses regarding both the advantages and disadvantages of outsourcing and privatization in the literature. One the on e hand, it is expected that contracting out services to private firms could increase the efficiency of water utilities by
68 allowing them to concentrate their energies on core tasks. Outsourcing arrangements may also provide access to technology, equipment, and expertise not available in the wa ter utility (Lee and Jouravlev 199 7 cited in Baumert and Bloodgood 2004). On the other hand, when water supply is outsourced, local governments face problems of asymmetry of information and in complete contracts (Willia mson 1976 cited in Perard 2009 ). This can lead to high transaction costs. Theories about ownership in the case of monopoly markets remain ambiguous and cannot completely explain the choice of privatization/delegation in th e water supply industry (Perard 2009). The analysis of water utility outsourcing functions, concessions and privatization has seen mixed results over the years. As stated by Perard (2009), privatization is considered advantageous due to benefits stemming from efficiency in a fully compe titive market where private ownership would be more efficient than public ownership. However, this is less clear for less competitive markets associated with network scale economies, like water supply 9 Empirical studies listed by Perard (2009) confirm t Kousassi (2002) find that private operators are more cost efficient. In a different study of African utilities, Kirkpatrick, Parker, and Zhang (2004) find no significant difference between public and private water utility operators. Studies of Argentinean water utilities find both positive effects and negative effects for the introduction of private sector 2003, Rais, Esquivel and Sour 2002 and Artana, Navajas, and Urbizt ondo 1999 ). Chile is often cited as a country where 9 are referred to Perard (2009) for more information on this topic and for a complete summa ry of the results of privatization in different countries.
69 privatization has worked, while Cochabamba (Bolivia) and Tucumn (Argentina) are cited as cities where it was very problematic. Outsourcing of functions of Japan ese water utilities started in 2001 with an amendment to th e Water Act (Urakami and Parker 2011) and has been increasing steadily over the years. Increases in outsourcing in Japan are attributed to natural attrition of in house career staff (Ueda and Beno uahi 2009). This is consistent with a 2004 report by the Ministry of Health, Labour, and Welfare presented in Waterwork s water companies. The outsourcing ratio variable (rout) is positively significant, implying that firms with more outsourcing are more inefficient. These results suggest that policymakers should be careful when proposing outsourcing, as it is currently s et up, as a solution to lem. Note, however, that it may be too early to assess the impact of outsourcing in Japan 10 Furthermore, these results differ slightly from the Marques, Berg, and Yane (2012) study that finds that outsourcing has mixed effects for Japanese water utilitie s. F urther studies regarding Japanese outsourcing are suggested Finally, the coefficients used to test differences based on regional location are for the most part negatively significant (though some regional dummies are insignificantly different from ze ro). Since the omitted region was region 1 (Hokkaido), these results 10 is reversed. We rule out this possible endogeneity problem since anecdotal evidence su ggests that the outso urcing decision is due to an ag ing workforce.
70 are all relative to the northernmost island 11 This result suggests that technical efficiency varies somewhat systematically across regions. This is expected given the differences in we ather, topography, geography, natural resources, etc. Since this factor is exogenous, labeling it inefficiency is somewhat problematic. More resources are required in some regions to produce the same output. So, again, the inclusion of explanatory facto rs in the model requires some care when interpreting performance for any particular utility. 3.5 Policy Implications and Conclusion As mentioned earlier, a 2004 report by the Ministry of Health, Labour, and Welfare presented in Waterworks Vision (2004) lis facing Japanese water companies. The findings from this paper suggest that outsourcing as a strategy to combat this problem should receive closer examination. If comparable data from before 2001 becomes avail able, a study comparing inefficiency during the years before and the years after outsourcing was implemented could provide policymakers with important information regarding this strategy. Another important issue that calls for further examination is that of centralization/decentralization and consolidation. Is it better to have a few large firms or many small individual firms? Another OECD island country that addressed this question is England (and Wales) where decentr until 1973 when 157 water undertakings, 1398 sewerage and sewage disposal authorities and 29 river authorities were merged into 10 regional water au thorities 2011). 11 Please refer to Appendix E for a listing of the regions.
71 Ueda and Benouahi (2009) argue that one of the items that the Japanese wa ter utilit y system is able to deliver is municipality based utilities ta ilored to unique local settings This is a point that requires closer examination, given our results regarding t be compared to the potential benefits from economies of scale. Several countries have noticed this and have been involved in attempts to consolidate utility firms into larger entities. These actions were directed at improving economic performance by ac hieving economies of scale and scope (Urakami and Parker 2011). The results of this paper in regards to these efforts should be continued in the future. Note, howe ver, that utilities that are too large could result in inefficiencies as well. As noted by Horn and Saito (2011) when utilities become larger, it is possible that their service areas also expand, which could in turn have negative consequences for efficien cy and costs. This could explain Urakami beneficial impact on cost effectiveness, given the possibility that cost savings were offset by extra expenditures incurred by having t o supply water to areas with low population density. In other words, it is beneficial for water utilities to expand but to an optimal size, rather than to the largest extent possible (Horn and Saito 2011). This can be achieved with proper incentive schem es. Benchmarking is an help reduce the information asymmetry between managers and those providing oversight 2010 p.114 ). In this paper, we have identified major sources of inefficiencies for Japanese water utilities. Since inefficiency exists, there is an opportunity to improve Japanese water
72 rther examination of the variables found to be influencing technical inefficiency, particularly those that are under managerial control, is recommended 12 Once this is carried out, firms far from the frontier could be provided with incentives to improve. For instance, incentives promoting further consolidation could be targeted at those utilities that improve cost containment. In addition to this, other policies could be implemented to raise efficiency. This is particularly important for Japan, given its declining population, increasing water costs, and current water utility industry forecasts: According to Data Monitor (2010) the Japanese water utilities industry is forecasted to have a volume of 84.8 billion cubic meters in 2014, a decrease of 3.3%. A good way to approach this grim forecast is to identity inefficiencies and tackle them. 12 Certain characteristics influencing efficiency are clearly outside of managerial control. Customer density, for instance, depends on where customers decide to live. It would be illogical to suggest improvements in efficiency based on increases in custom er density to a manager, particularly in rural towns facing emigration towards the big cities.
73 Table 3 1. Production function variables Variable Description [Output] Y Total delivered water volume in a year [1,000m 3 ] [Inputs] K Length of pipe [1,000m] L Total number of staff (including estimated staff from outsourcing 13 ) O Self produced intake water capacity (total intake water capacity minus purchased water capacity) 14 [1,000m 3 ] P Purchased water capacity [1,000m 3 ] Table 3 2. Inefficiency effects model v ariables Variable Description Regional d ummies Dummies for Japa nese regions. Please refer to Figure 2 and T able 5 in the Appendix for a map and description of these regions. Region 1 is the excluded dummy variable. Aveop Average operation rate defined by average delivered water volume per delivered water capacity [%] Rraw Proxy for raw water ratio defined by chemical expenditures per intake water volume [/1000m 3 ] Rout Outsourcing ratio defined by the ratio of number of staff based on outsourcing to the number of total staff [%] Cusden Customer density defined by the number of customers per length of pipe [person/1000m] Size Intake water volume [1000m 3 ] Rsubp Subsidy ratio on profit and loss account defined by the sum of subsidies on profit and loss account per water supply revenue 13 Estimated staff from outsourcing is calculated following Yane and Berg ( forthcoming, 2013). Virtual staff is calculated by dividing outsourcing expenditures b y payment per employee in each prefecture. 14 Total Intake Water Capacity = Self produced capacity + Purchased water capacity
74 Table 3 3 Summary s tatistics Variable Mean Standard Deviation Min Max [ Production Function in Ln form] LnY 8.41 1.23 4.32 14.3 LnK 5.52 1 0.19 10.18 LnL 3.13 1.24 0 .11 9.17 LnO 7.88 2.66 0 14.76 LnP 4.31 4.32 0 12.91 [Production Function] Y 12,588.87 55,733.75 75 1,624,646 K 462.06 1,042.52 1.21 26,312.52 L 69.9 324.88 1.11 9,557.68 O 15,247.28 85,905 0 2,579,742 P 5,930.73 21,911.43 0 403,043.7 [Inefficiency Effects] Aveope 60.24 11.79 10.3 120.6 Rraw 0.49 0.8 0 17.03 Rout 0.38 0.14 0 0.87 Cusden 159.78 250.23 12.74 11 463.64 Size 12,927.73 56,655.24 0 1,700,000 Rsubp 0.07 0.17 0 2.66 Regional dummies 0 1 Notes: N=4,760 Table 3 4. OLS e stimates Variable Coefficient t ratio intercept 4.2768*** 107.50 lnK 0.3360*** 30.37 lnL 0.6573*** 69.28 lnO 0.0172*** 7.08 lnP 0.0206*** 14.38 Notes: N = 4,760, R squared = 0.9206
75 Table 3 5 Stochastic frontier e stimates: Inefficiency effects model using regional d ummies and allowing for delta 0 Variable Coefficient t ratio [Production Function] intercept 5.0136*** 72.32 lnK 0.3653*** 34.54 lnL 0.5961*** 71.99 lnO 0.0166*** 7.29 lnP 0.0077*** 5.62 [Inefficiency Effects] intercept 1.0233*** 17.42 aveop 0.0037*** 10.24 rraw 0.0888*** 16.56 rout 0.2201*** 6.94 rsubp 0.5536*** 21.03 cusden 0.0003*** 20.12 size 10 6 x 0.9507*** 10.59 Region 2 dummy 0.0372 *** 1.52 Region 3 dummy 0.0746*** 3.4 Region 4 dummy 0.3048*** 15.41 Region 5 dummy 0.3543*** 14.36 Region 6 dummy 0.3543*** 15.15 Region 7 dummy 0.5095*** 23.22 Region 8 dummy 0.3538*** 17.12 Region 9 dummy 0.2041*** 8.8 Region 10 dummy 0.3615*** 15.25 Region 11 dummy 0.0951*** 4.6 Region 12 dummy 0.1779*** 7.36 Region 13 dummy 0.3094*** 8.69 N =4,760, T=4, cross sections: 1,190.
76 APPENDIX A MARGINAL EFFECTS FOR STATE LEVEL TEXTBOOK SELECTION TABLES
77 Table A 1. Marginal effects: average partial e ffects (APEs) and p artial e ffects a t t he a verage (PEAs) for T able 1 5 (ordered logit regression explaining state textbook policies), column 1. Variable Mean (S.D) Outcome 0 Outcome 1 Outcome 2 APE PEA APE PEA APE PEA Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) State Revenue 49.7302 ) (11.9053) 0.0073 ( 2.16) 0.0188 ( 1.98) 0.0028 (1.76) 0.0173 (1.85) 0.0044 (2.23) 0.0014 (1.14) Homeownership 70.2714 ) (4.3080) 0.0132 ( 1.35) 0.0340 ( 1.33) 0.0051 (1.17) 0.0314 (1.29) 0.0080 (1.42) 0.0026 (0.94) District Size 4,897.9530 ) (7,301.3460) 0.000015 ( 2.79) 0.000039 ( 2.38) 5.85e 06 (2.34) 0.000036 (2.17) 9.16e 06 (2.56) 2.97e 06 (1.20) Advanced Degree 9.6204 ) (2.4346) 0.0549 (2.89) 0.1418 (2.46) 0.0214 ( 2.16) 0.1309 ( 2.21) 0.0335 ( 2.91) 0.0108 ( 1.25) Fundamentalist Dummy 0.2857 ) (0.4564) 0.3746 ( 2.56) 0.5849 ( 3.14) 0.1936 (3.27) 0.4761 (3.07) 0.1810 (1.61) 0.1088 (1.07) State Income Gini 0.4458 ) (0.0192) 5.2917 ( 1.61) 13.6700 ( 1.49) 2.0625 (1.34) 12.6249 (1.44) 3.2292 (1.72) 1.0450 (1.03) Pseudo R 2 ( 0.4951 0.4951 0.4951 0.4951 0.4951 0.4951 Notes: T statistics in parenthesis. Outcome 0 = local choice states. Outcome 1 = recommended list states. Outcome 2 = mandatory state list. Number of Observations = 49 dy/dx is for a discrete change of the dummy variable from 0 to 1
78 Table A 2. Marginal effects: Average Partial Effects (APEs) and Partial Eff ects at the Average (PEAs) for T able 1 5 (ordered logit regression explaining state textbook policies), column 2. Variable Mean (S.D.) Outcome 0 Outcome 1 Outcome 2 Outcome 3 APE PEA APE PEA APE PEA APE PEA Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) State Revenue 49.7302 ) (11.9053) 0.0059 ( 1.72) 0.0148 ( 1.68) 0.0024 (1.42) 0.0136 (1.59) 0.0001 (0.22) 0.0007 (0.99) 0.0034 (1.74) 0.0004 (0.93) Homeownership 70.2714 ) (4.3080) 0.0098 ( 0.98) 0.0247 ( 1.01) 0.0040 ( 0.82) 0.0227 (0.98) 0.0001 (0.23) 0.0013 (0.80) 0.0056 (1.05) 0.0007 (0.80) District Size Dummy 4,897.9530 ) (7,301.3460) 0.2965 ( 2.80) 0.521 1 ( 3.44) 0.1447 (2.78) 0.4376 (3.02) 0.0347 (1.40) 0.0515 (1.09) 0.1171 (1.94) 0.0320 (1.03) Advanced Degree 9.6204 ) (2.4346) 0.0505 (2.43) 0.1275 (2.33) 0.0205 ( 1.71) 0.1170 ( 2.10) 0.0009 ( 0.22) 0.0066 ( 1.11) 0.0290 ( 2.52) 0.0039 ( 1.07) Fundamentalist Dummy 0.2857 ) (0.4564) 0.3537 ( 2.34) 0.5578 ( 2.84) 0.1802 (2.95) 0.4552 (2.79) 0.0534 (1.38) 0.0628 (1.00) 0.1201 (1.32) 0.0398 (0.92) State Income Gini 0.4458 ) (0.0192) 4.1051 ( 1.27) 10.3735 ( 1.26) 1.6704 (1.05) 9.5189 (1.22) 0.0783 (0.22) 0.5406 (0.87) 2.3564 (1.36) 0.3140 (0.87) Pseudo R 2 0.4536 0.4536 0.4536 0.4536 0.4536 0.4536 0.4536 0.4536 Notes: T statistics in parenthesis. Outcome 0 = Local choice states. Outcome 1 = Recommended list states. Outcome 2 = Long mandatory state list. Outcome 3 = Short mandatory state list. Number of Observations = 49. dy/dx is for a discrete change of the dummy variable from 0 to 1.
79 Table A 3. Marginal effects: Average Partial Effects (APEs) and Partial Eff ects at the Average (PEAs) f or T able 1 6 Variable Mean (S.D.) Outcome 0 Outcome 1 Outcome 2 APE PEA APE PEA APE PEA Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) 26.75 (4.74) 0.0168 (1.89) 0.0438 (1.74) 0.0054 ( 1.30) 0.0402 ( 1.63) 0.0114 ( 1.88) 0.0036 ( 1.12) District Size 4,831.35 (7,363.55) 0.00001 ( 2.76) 0.00003 ( 2.30) 3.44e 6 (1.56) 3e 5 (2.12) 7.31e 6 (2.57) 2.31e 6 (1.13) Gini of State Income 0.44 (0.01) 3.7579 ( 1.25) 9.7849 ( 1.18) 1.2026 (0.97) 8.9769 (1.14) 2.5553 (1.30) 0.808 (0.93) Homeownership 70.19 (4.31) 0.0098 ( 0.94) 0.0255 ( 0.95) 0.0031 (0.75) 0.0234 (0.94) 0.0067 (1.01) 0.0021 (0.78) State Revenue 49.85 (11.99) 0.0041 ( 1.26) 0.0105 ( 1.22) 0.0013 (1.01) 0.0097 (1.18) 0.0028 (1.28) 0.0009 (0.94) 21.92 (13.02) 0.0068 (2.01) 0.0176 (1.79) 0.0022 ( 2.10) 0.0162 ( 1.68) 0.0046 ( 1.57) 0.0015 ( 1.11) Fundamentalist 9.86 (10.67) 0.0087 ( 1.8) 0.0227 ( 1.69) 0.0028 (1.12) 0.0208 (1.61) 0.0059 (2.08) 0.0019 (1.04) Pseudo R 2 0.4887 0.4887 0.4887 0.4887 0.4887 0.4887 Notes: T statistics in parenthesis. Outcome 0 = Local choice states. Outcome 1 = Recommended list states. Outcome 2 = Mandatory state list. Number of Observations = 48.
80 Table A 4. Marginal effects: Average Partial Effects (APEs) and Partial Effects at the Ave rage (PEAs) for T able 1 6 Variable Mean ( S.D ) Outcome 0 Outcome 1 Outcome 2 Outcome 3 APE PEA APE PEA APE PEA APE PEA Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) 26.75 (4.74 ) 0.0194 (2.23) 0.0501 (2.06) 0.0059 ( 1.36) 0.0450 ( 1.88) 0.0022 ( 1.16) 0.0032 ( 1.15) 0.0113 ( 2.21) 0.0019 ( 1.11) District Size 4,831.35 ) (7,363.55 ) 0.00001 ( 2.72) 0.00003 ( 2.37) 3.40e 0 6 (1.69) 0.00003 (2.17) 1.25e 06 (1.00) 1.84e 06 (1.10) 6.51e06 (2.62) 1.08e 06 (1.08) Gini of State Income 0.44 ) (0.01 ) 3.1307 ( 1.04) 8.0819 ( 1.00) 0.9533 (0.86) 7.2650 (0.98) 0.3503 (0.83) 0.5153 (0.80) 1.8271 (1.06) 0.3016 (0.82) Homeownership 70.19 ) (4.31 ) 0.0081 ( 0.82) 0.0209 ( 0.83) 0.0025 (0.71) 0.0188 (0.83) 0.0009 (0.69) 0.0013 (0.68) 0.0047 (0.85) 0.0008 (0.69) State Revenue 49.85 (11.99 ) 0.0053 ( 1.65) 0.0137 ( 1.57) 0.0016 (1.19) 0.0123 (1.49) 0.0006 (1.03) 0.0008 (1.03) 0.0031 (1.66) 0.0005 (1.01) 21.92 ) (13.02 ) 0.0040 (1.08) 0.0103 (1.03) 0.0012 ( 1.22) 0.0092 ( 1.01) 0.0004 ( 0.70) 0.0007 ( 0.86) 0.0023 ( 0.99) 0.0004 ( 0.81) Fundamentalist Dummy 0.3570 ( 2.11) 0.5677 ( 2.68) 0.1632 (2.68) 0.4402 (2.87) 0.0628 (1.29) 0.0770 (0.94) 0.1310 (1.25) 0.0505 (0.87) Pseudo R 2 0.4449 0.4449 0.4449 0.4449 0.4449 0.4449 0.4449 0.4449 Notes: T statistics in parenthesis. Outcome 0 = Local choice states. Outcome 1 = Recommended list states. Outcome 2 = Long mandatory state list. Outcome 3 = Short mandatory state list. Number of Observations = 48. dy/dx is for a discrete change of the dummy variable from 0 to 1.
81 Table A 5. Marginal effects: Average Partial Effects (APEs) and Partial Effects at the Average (PEAs) for T able 1 7 (panel data regression explaining state textbook policies) Variable Mean (S.D.) Specification 1 Specification 2 Specification 3 APE PEA APE PEA APE PEA Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) Coeff (t stat) dy/dx (t stat) 17.47 ) (5.92 ) 0.0184 ( 1.90) 0.0347 ( 1.82) 0.0172 ( 1.13) 0.0320 ( 1.07) 0.0184 ( 1.25) 0.0347 ( 1.17) District Size 4745.52 ) (6411.54 ) 6.21e 6 (1.48) 0.00001 (1.58) 9.12e 6 (1.65) 0.00002 (1.72) 6.21e 6 (0.90) 0.00001 (0.95) Gini of State Income 0.38 (0.03 ) 0.9097 (0.56) 1.7115 (0.66) 1.1902 (0.70) 2.2068 (0.69) 0.9097 (0.54) 1.7115 (0.54) Homeownership 66.68 ) (4.94 ) 0.0008 (0.13) 0.0015 (0.13) 0.0022 (0.27) 0.0040 (0.27) 0.0008 (0.10) 0.0015 (0.10) State Revenue 46.15 ) (13.53 ) 0.0059 (2.85) 0.0110 (2.73) 0.0064 (2.06) 0.0119 (2.07) 0.0059 (1.81) 0.0110 (1.85) Fundamentalist 11.57 ) (11.63 ) 0.0107 (2.22) 0.0202 (2.33) 0.0141 (2.63) 0.0262 (1.97) 0.0107 (1.56) 0.0202 (1.41) Pseudo R 2 0.3999 0.3999 0.3952 0.3952 0.3999 0.3999 Notes: Number of Observations = 196.
82 APPENDIX B MARGINAL EFFECTS FOR INDIANA DISTRICT LEVEL TEXTBOOK WAIVER TABLE Table B 1. Marginal effects: Average Partial Effects (APEs) and Partial Eff ects at the Average (PEAs) for T able 2 2 (Indiana district level logi t model explaining textbook waiver use). Variable Mean ( S.D. ) APE PEA Coeff (t stat) dy/dx (t stat) Fundamentalist Dummy ... 0.0902 (1.72) 0.0816 (2.2 ) Gini 0.4031 ) (0.0258) 1.3315 ( 1.82) 1.1771 ( 1.81) Rurality_1 Dummy 0.1103 (1.42) 0.102 (1.53) Rurality_2 Dummy 0.0575 (0.83) 0.0518 (0.88) Rurality_3 Dummy 0.139 (1.20) 0.1273 (1.27) District Size 3,449.014 ) (4,381.217) 4.27e 6 ( 0.78) 3.77e 6 ( 0.79) 18.8084 ) (11.1516) 0.0029 ( 0.93) 0.0026 ( 0.95) Region_1 Dummy 0.01 (0.16) 0.0088 (0.16) Region_2 Dummy 0.0499 ( 1.22) 0.0435 ( 1.14) Region_3 Dummy 0.095 ( 3.50) 0.0797 ( 2.65) Region_4 Dummy 0.0146 ( 0.33) 0.0128 ( 0.33) Region_5 Dummy 0.089 ( 2.57) 0.0753 ( 2.04) Notes: T statistics in parenthesis. Dependent variable is 0 if do not use a waiver, 1 if use a waiver. Number of Observations = 291 dy/dx is for a discrete change of the dummy variable from 0 to 1.
83 APPENDIX C FUNDAMENTALIST CHURCHES owi ng Johnson (1976). In the 1980 s these were: All Baptist bodies, except American Baptist Convention Church of God Church of God in Christ Church of the Nazarene Evangelical United Brethren Church Pentecostal Churches Presbyterian Church in the USA African Methodist Episcopal Church African Methodist Episcopal Zion Church Several congregations changed names over the years (e.g., American Baptist Convention is now American Baptist Churches in the USA). In order to keep the list up to date, the internet was used to study the recent changes in the congregations from the s list. Several churches changed names, and a few merged for ideological reasons. In the case of church mergers, the websites of the new churches were accessed to see whether the merge affected their ideology. their corresponding Glenmary Research Center code is presented in T able C 1.
84 Table C 1. Fundamentalist c hurches Code Congregation 17 The American Baptist Association 53 Assemblies of God 57 Baptist General Conference 59 Baptist Missionary Association of America 89 The Christian Missionary Alliance 121 Church of God General Conference 123 Church of God (Anderson, Indiana) 127 Church of God (Cleveland, Tennessee) 143 Church of God in Christ, Mennonite 145 Church of God Prophecy 165 Church of the Nazarene 171 Churches of God, General Conference 179 Conservative Baptist Association of America 181 Conservative Congregational Christian Conference 189 Duck River and Kindred Baptists Associations 191 Enterprise Baptists Association 216 Evan Presbyterian Church 223 National Association of Free Will Baptists 263 International Church of the Foursquare Gospel 264 International Churches of Christ 265 International Pentecostal Church of Christ 266 Interstate and Foreign Landmark Missionary Baptists Association 304 National Primitive Baptist Convention, USA 306 New Testament Association of Independent Baptist Churches and other Fundamental Baptist Associations/Fellowships 313 North American Baptist Conference 320 ry Baptists Associations 349 International Pentecostal Holiness Church 360 Primitive Baptist Churches Old Line 365 Progressive Primitive Baptists 373 Reformed Church in the United States 388 General Association of Regular Baptist Churches 418 Southwide Baptist Fellowship 419 Southern Baptist Convention 469 Wisconsin Evangelical Lutheran Synod
85 APPENDIX D REGIONAL CLASSIFICATION OF COUNTIES AS DEFINED BY THE INDIANA DEPARTMENT OF EDUCATION Northwest: Benton, Carroll, Cass, Fulton, Jasper, La Porte, Lake, Marshall, Newton, Porter, Pulaski, St. Joseph, Starke, and White counties. Northeast: Adams, Allen, Blackford, De Kalb, Elkhart, Grant, Huntington, Jay, Kosciusko, Lagrange, Miami, Steuben, Wabash, Wel ls, and Whitley counties. East Central: Decatur, Delaware, Fayette, Franklin, Hamilton, Hancock, Henry, Howard, Johnson, Madison, Marion, Randolph, Rush, Shelby, Tipton, Union, and Wayne counties. West Central: Boone, Clay, Clinton, Fountain, Hendricks, Mo ntgomery, Morgan, Owen, Parke, Putnam, Tippecanoe, Vermillion, Vigo, and Warren counties. Southwest: Crawford, Daviess, Dubois, Gibson, Greene, Knox, Lawrence, Martin, Monroe, Orange, Perry, Pike Posey, Spencer, Sullivan, Vanderburgh, and Warrick counties. Southeast: Bartholomew, Brown, Clark, Dearborn, Floyd, Harrison, Jackson, Jefferson, Jennings, Ohio, Ripley, Scott, Switzerland, and Washington counties.
86 APPENDIX E JAPAN BACKGROUND INFORMATION Table E 1 Japanese prefectures and regions Prefecture Region 1 Hokkaido 1 Hokkaido 2 Aomori 2 North Tohoku 3 Iwate 4 Akita 5 Miyagi 3 South Tohoku 6 Yamagata 7 Fukushima 8 Ibaraki 4 North Kanto 9 Tochigi 10 Gunma 11 Saitama 12 Nagano 13 Chiba 5 South Kanto 14 Tokyo 15 Kanagawa 16 Yamanashi 17 Niigata 6 Hokuriku 18 Toyama 19 Ishikawa 20 Fukui 21 Gifu 7 Tokai 22 Shizuoka 23 Aichi 24 Mie 25 Shiga 8 Kinki 26 Kyoto 27 Osaka 28 Hyogo 29 Nara 30 Wakayama 31 Tottori 9 Chugoku 32 Shimane 33 Okayama 34 Hiroshima 35 Yamaguchi 36 Tokushima 10 Shikoku 37 Ehime 38 Kagawa 39 Kochi 40 Fukuoka 11 North Kyushu 41 Saga 42 Nagasaki 43 Oita
87 Table E 1 Continued Prefecture Region 44 Kumamoto 12 South Kyushu 45 Miyazaki 46 Kagoshima 47 Okinawa 13 Okinawa Figure E 1. Japanese r egion s
88 APPENDIX F EFFICIENCY Yearly and Sample Efficiency Scores and Output A firm with an efficiency score equal to 1 is fully efficient. The observations are weighted by percent of total water delivered and firms are sorted from most efficient to least efficient. Figure F 1. 2004 weigh ted
89 Figure F 2. 2005 weighted Figure F 3. 2006 weighted
90 Figure F 4. 2007 weighted Table F 1. Summary statistics for efficiency scores and output Variable Mean Standard Deviation Min Max [Year 2004] Output 12,418.14 56,034.32 222 1,624,602 Ln Output 8.4 1.22 5.4 14.3 Efficiency Score 0.55 0.13 0.1 0.98 [Year 2005] Output 12,711.47 55,915.39 226 1,615,886 Ln Output 8.42 1.23 5.42 14.3 Efficiency Score 0.54 0.13 0.1 0.98 [Year 2006] Output 12,609.08 55,528.16 75 1,606,415 Ln Output 8.41 1.24 4.32 14.29 Efficiency Score 0.54 0.13 0.09 1 [Year 2007] Output 12,575.38 55,433.07 81 1,606,804 Ln Output 8.41 1.24 4.39 14.29 Efficiency Score 0.54 0.13 0.15 1 [All Years] Output 12,578.52 55,710.86 75 1,624,602 Ln Output 8.41 1.23 4.32 14.3 Efficiency Score 0.54 0.13 0.09 1
91 Figure F 5: Efficiency and output Figure F 5 compares predicted efficiency and ln output for the year 2004 Quadrant I shows firms with above average output and above average efficiency. Quadrant II shows firms w h ich are efficient but have output that is lower than the average. Quadrant III shows firms that have low efficiency scores and low output levels. Quadrant IV shows firms that have low efficiency scores but higher than average output levels. Subsidies Figures F 6 and F 7 examine the possibility that subsidies could be directed at rewarding strong performance. There is no visible pattern regarding the distribution of subsidies for the top 40 and bottom 40 firms efficiency wise, suggesting that subsidy decisions are not based on efficiency levels.
92 2004 2005 2006 2007 Figure F 6. Subsidies for bottom 40 firms 2004 2005 2006 2007 Figure F 7. Subsidies for top 40 firms
93 APPENDIX G ALTERNATIVE MODEL SP ECIFICATIONS Below are some tests of alternative stochastic frontier model specifications. For more specifications, please refer to Yane and Berg ( forthcoming, 2 013 ) Table G 1. Time invariant panel production function model assuming a half normal distribution of the inefficiency term. This model assumes that the are independently half normally distributed and was estimated using Stata. Variable Coefficient t ratio intercept 7.2920*** 18.92 lnK 0.2450*** 27.32 lnL 0.4861*** 46.75 lnO 0.0121*** 4.36 lnP 0.3002*** 18.50
94 Table G 2. Inefficie ncy effects model using regional dummi es and not allowing for delta 0 Variable Coefficient t ratio [Production Function] intercept 4.3615*** 100.96 lnK 0.3006*** 26.43 lnL 0.6748*** 68.68 lnO 0.0197*** 8.05 lnP 0.0099*** 6.55 time 0.1771*** 19.11 [Inefficiency Effects] aveop 0.0021*** 3.78 rraw 0.1057*** 12.52 rout 0.4123*** 7.53 cusden 0.0002*** 11.32 size 10 6 x 0.1000 ** 2.67 time trend 0.1949*** 15.60 Region 2 dummy 0.0568 *** 1.64 Region 3 dummy 0.0500 *** 1.39 Region 4 dummy 0.2279*** 6.69 Region 5 dummy 0.1906*** 4.55 Region 6 dummy 0.2697*** 6.28 Region 7 dummy 0.5140*** 14.57 Region 8 dummy 0.2149*** 5.99 Region 9 dummy 0.1020** 2.44 Region 10 dummy 0.2329*** 5.19 Region 11 dummy 0.0373 *** 1.01 Region 12 dummy 0.1135*** 2.85 Region 13 dummy 0.1689** 2.52
95 Table G 3. Inefficiency effects model using regional dummies and including subsidies Variable Coefficient t ratio [Production Function] intercept 4.9576*** 103.8 0 lnK 0.3655*** 33.05 lnL 0.5952*** 59.07 lnO 0.0167*** 7.01 lnP 0.0079*** 5.49 time 0.0244 *** 1.51 [Inefficiency Effects] intercept 0.9530*** 20.39 aveop 0.0036*** 9.60 rraw 0.0890*** 15.64 rout 0.2170*** 6.71 subsidies 0.5556*** 20.95 cusden 0.0003*** 16.35 size 10 6 x 0.1000 *** 10.04 Region 2 dummy 0.0374 *** 1.60 Region 3 dummy 0.0725*** 3.29 Region 4 dummy 0.3042*** 15.39 Region 5 dummy 0.3527*** 14.97 Region 6 dummy 0.3533*** 14.90 Region 7 dummy 0.5098*** 23.39 Region 8 dummy 0.3537*** 17.56 Region 9 dummy 0.2043*** 8.91 Region 10 dummy 0.3586*** 14.04 Region 11 dummy 0.0952*** 4.42 Region 12 dummy 0.1828*** 7.61 Region 13 dummy 0.3086*** 8.55 time trend 0.0286* ** 1.69 Notes:
96 LIST OF REFERENCES Adams, J.D., and L.W. Kenny. 1986. Optimal t en ure of elected public o fficials. Journal of Law and Economics 29: 303 328. Aida, K, W. W. Cooper, J.T. Pastor, and T. Sueyoshi. 1998. Evaluating water supply s ervices i n Japan with RAM: A r ange ad justed measure of i nefficiency. Omega 26: 207 232. Aigner, D., C.A. Knox Lovell and P. Schmidt. 1977. Formulation and estimation of stochastic f ront ier production function m odels. Journal of Econometri cs 6: 21 37. Artana, D., F Navajas, and S. Urbiztondo. 1999. Governance and regulation: A tale of two concessions in Argentina In Spilled water: Institutional commitment in the provision of water services edited by P. T. Spiller. Washington, DC : Inter Am erican Development Bank, pp. 197 248 Batista Oliveira, J. 1995. Understanding textbook q uality in the USA: How governments distort the m arket. In ternational Review of Education 41: 481 500. B attese, G.E. and T.J. Coelli. 1995. A model for technical i neffici ency effects in a stochastic frontier p rod uction function for panel d ata. Empirical Economics 20: 325 332. Baumert, J. and L. Bloodgood. 2004 Private sector participation in the water and wastewater services industry U.S. International Trade C ommission Office of Industries Working Paper Accessed 1 April 2012. Available http://www.usitc.gov/publications/332/working_papers/wp_id_08.pdf Berg, S. and C. Li n. 2008. Consistency in pe rformance rankings: The Peru water s ector Applied Economics 40: 793 805. Berg, S. 2010. Water Utility Benchmarking: Measurement, Methodologies, and Performance Incentives London UK : International Wa ter Association Publishing. Boissie re, M. 2004. De terminants of primary education o u tcomes in developing countries. Background p aper for the e Primary Education, Operations Evaluation Department, World Bank Burkey, M. 2000. oefficients for www.ncat.edu/~burkeym/Gini.htm Coelli, T. 1996. A guide to FRONTIER version 4.1: A computer program for stochastic frontier p roduction and cost functio n estimation. University of New England. CEPA Working Paper 96/08. Accessed 26 March 2012. Available http://www.uq.edu.au/economics/cepa/frontier.htm
97 d G. E. Battese. 2005. An introduction to efficiency and productivity a nalysis New York NY : Springer Cole J R. 2000. Money flooding anti National Center for Science Education Reports 20: 64 65. Corton, M.L. 2011. Sector fragmentation and aggregation of service provision in the w at er i ndustry Journal of Productivity Analysis 35: 159 169. Currey, V. 1988. The politics of textbook a doption. Political Science and Politics 21: 25 30. Data M onitor. 2010 Industry profile: Water u tilities in Japan. Last retreived April 1, 2012 from Datamonitor Business Information Center DelFattore, J. 1992. What ensorship in America New Haven, CT : Yale University Press. Dewey, J. and L.W. Kenny. 2009. Are more educated electorates more effective in making governments more responsive to voter wishes? University of Florida Working Paper Estache, A. and E. Kouassi. 2002. Sector organization, governance, and the ineffici ency of African water utilities World Bank Policy Research Working Paper. Farr, R. and M. Tulley. 1989. State level adoption of basal r eaders: Goals, processes, and r ecommendations. Theory in to Practice 28: 248 253 Farsi, M., M. Filippini, and M. Kuenzle. 2005. Unobserved heterog eneity in stochastic cost frontier m odels: An a pp lication to Swiss n ursing h omes Applied Economics 37: 2127 2141 Finn, C. E. and D. Ravitch. 2004. The mad, mad world of textbook a doption Thomas B. Fordham Institute Washington, DC: District creative printing. Fischel, W.A. 2005. The homevoter hypothesis: How home values influence local government taxation, school finance, and land use p olicies Cambridge MA : Harvard University Press. Florida Senate ns truction in the public s Accessed 29 August 2011. Available http://www.flsenate.gov/Session/Bill/2011/1854/BillText/Filed/HTML Fuchs, T. and L. Wmann. 2007. What accounts for international d ifferences in student p erformance? A re examination using PISA data. Empirical Economics 32: 433 464.
98 Horn, T. and H. Saito. 2011 Cost efficiency and scale economies of Japanese water utilities. International Conference O n Applied Econom ics ICOAE Accessed 26 March 2012. Available http://kastoria.teikoz.gr/icoae2/wordpress/wp content/uploads/2011/10/025.pdf Husted, T.A. and L.W. Kenny. 2000. Evidence of the i mpact of s tate g overnment on p rimary and s econdary e ducation and t he e quity e fficiency t rade off. Journal of Law and Economics 43: 285 308. Husted, T.A and L.W. Kenny. 2007. Explanations for s tates a dopting l imits on e ducation al s pending. Public Finance Review 35: 586 605. H utcheson, J.D. and G.A. Taylor. 1973. Religious v ariables, p olitical s ystem c haracteristics, and p olicy o utputs in the Ameri can s tates. American Journal of Political Science 17: 414 421. Indiana Department of Education 2004. ategory. Dr. Suellen Reid, Superintendent http://dew4.doe.in.gov/WF/TEXTBOOK/DPITA/2004_ science.pdf July 21, 2011, from http://www.doe.in.gov/opd/textbook/docs/TextbookFAQsFina l090309.pdf t extbook a doption c http://www.in.go v/legislative/ic/2004/title20/ar10.1/ch9.html Iraizoz, B., I. Baraji, and M. Rapun. 2005. The Spanish b eef s ector in the 1990s: Impact of the BSE c risis on e fficiency and p rofitability Applied Economics 37: 473 484. Johnson, C.A. 1976. Political c ulture in American s tates: f ormulation e xamined. American Journal of Political Science 20: 491 509. Johnson, K. 2006. Anti Darwin b ill f ails in Utah. New York Times 28 February (online). Accessed 6 October 2011. Available http://www.nytimes.com/2006/02/28/national/28utah.html?pagewanted=print Jones, D.E., S. Doty, C. Grammich, J.E. Horsch, R. Houseal, Mac Lynn, J.P. Marcum, K. M. Sanchagrin, and R.H. Taylor 2002. Religious c ongregations and m embership in the United States Nashville, T N : Glenmary Research Center. Kirpatrick, C., D. Parker, D., and Y. Zhang. 2004. State versus private sector provision of water services i n Africa: An empirical analysis University of Manchester Centre on Regulation and Competition Working Paper Series no. 70.
99 Kraemer K teaching of evolution and creationism. M .A. thesis, Bemidji State Universi ty, Bemidji, MN. Kumbhakar, S. and R. Bokusheva. 2009. Modeling f arm p roduction d ecisions u nder an e xpenditure c onstraint European Review of Agricultural Economics 36: 343 367. Lee, T. and A. Jouravlev. 1997. Private p articipation in t he provision of w ater s ervices. Medio Ambiente y Desarrollo ECLAC series 2. b enchmarking a nalysis of Japanese w ater u Meeus en, W. and J. van den Broeck. 1977. Tech nical e fficiency and d imension of the f irm: Some r esults on the u se o f f rontier p roduction f unctions Empirical Economics 2: 109 122. Ministry of Health, Labor and Welfare. Water works vision s tudy March 2012. Available http://www.mhlw.go.jp/english/policy/health/water_supply/dl/3 1a.pdf Mizala, A., P. Romaguera, and D. Farren. 2002. The t echnical e fficiency of s chools in Chile Applied Economics 34: 1533 1552. Mizutani, F. and T. U rakami. 2001. Identifying n etwork d ensity and s cale e conomies for Japanese w ater s upply o rganizations. Papers in Regional Science 802: 211 230. Moore, R. 2004. How w ell d o b iology t eachers u nderstand the l e gal i ssues a ssociated w ith the t eaching of e volution? Bioscience 54: 860 865. Moore R., M. Jensen, and J. Hatch. 2003. Twenty q uestions: What h ave the c ourts s aid about the t eaching of e volution and creationism in public schools. Bioscience 53: 766 771. Morgan, D.R., and K.J. Meier. 1980. Politics and m orality: The e ffect o f r eligion on r eferenda v oting. Social Science Quarterly 61: 144. National C enter for Education Statistics. 2000 2011. Available http://nces.ed.gov/ccd/bat/index.asp National C enter for Education Statistics. 2000 2004 School d istrict d emographic s ystem (SDDS) Accessed 5 June 2011 Available http://nces.ed.gov/surveys/sdds/ed/index.asp?st=IN National Center for Education Statistics 2004 Common c ore Accessed 15 May 2011. Available http://nces.ed.gov/ccd/bat/index.asp
100 Newport, Frank. 2009. b irthday o nly 4 in 10 b elieve in e Gallup Website Accessed 31 August 2011. Available http://www.gallup.com/poll/114544/Darwin Birthday Believe Evolution.aspx Padian, K. 2008. Trickle down evolution: An a pproach to g etting m ajor e volutionary a daptive c hange s i nto t extbooks and c urricula. Inte grative and Comparative Biology 48: 175 188. Perard, E. 2009. Water s upply: p ublic or private? An approach based on cost of fund, t ransacti on c osts, efficiency and political c osts Policy and Society 27 : 193 219. Pitt, M. and L.F. Lee. 1981. The m easurement and s ources of t echnical e fficiency in the Indonesian w eaving i ndustry Journal of Development Economics 9: 43 62. Postal, L. 2011. Scienti iled in Florida Senate Orlando Sentinel 8 March (online) Accessed 29 August 2011. Available http://articles.orlandosentinel.com /2011 03 08/features/os antievolution bill florida sb1854 030811_1_evolution debate science standards florida senate Pritchett, L. and D Filmer. 1999. What education production functions really show: A positive theory of education e xpenditure. Economic s of Education Review: 18: 223 239 Rais, J. C. M.E. Esquivel, and S. Sour. 2002. La c oncesin d e l os s ervicios d e a gua p otable y a lcantarillado s anitario e n Tucumn Repblica Argentina Public Private Infrastructure Advisory Facility. Randak S. 2001. Th American Biology Teacher 63: 226, 228, 230. Rezitis, A., K. Tsiboukas, and S. Tsoukalas. 2003. Investigation of f actors i nfluencing the t echnical e fficiency of a gricultural p roducers p articipating in f arm c red it p rograms: The c ase of Greece Journal of Agricultural and Applied Economics 35: 529 541. Rutledge M L and M A. Mitchell. 2002. structure, acceptance, and teaching of evolution. American Biology Teacher 64: 21 28. Rutledge M L and W.A. Warden 2000. Evolutionary theory, the nature of science and high school biology teachers: Critical relationships. American Biology Teacher 62: 23 31. Salatiel, G. 2003. La participacion del sector privado en los servicios de agua y saneamiento en la provincia de salta en Argentina Banco Interamericano del Desarrollo, Departamento de Desarrollo Sostenible.
101 Sass, T. and D. Saurman 1993. Mandated e xclusive t erritories and e conomic e fficiency: An e mpirical a nalysis of the m alt b everage Industry. Journal of Law and Economics 36: 153 177. Schmidt, A., L.W. Kenny, and R. Morton. 1996. Evidence on e lectoral a ccountability in the U.S. s enate: Are u nfa ithful a gents r eally p unished? Economic Inquiry, 34: 545 567. Schmidt, P. and H.J. Wang. 2002. One s tep and t wo s tep e stimation of the e ffects of e xogenous v ariables on t echnical e fficiency l evels. Journal of Productivity Analysis 18: 129 144. Scotti, D., P. Maligh etti, G. Martini, and N. Volta. 2012. The i mpact of a irport c ompetition on t echnical e fficiency: A s tochastic f rontier a nal ysis a pplied to Italian a irport Journal of Air Transport Management 22: 9 15. Strauss, V. 2010. The answer sheet. Washington Post 24 July (online). Accessed 20 July 2011. Available http://www.washingtonpost.com/wp dyn/content/article/2010/05/23/AR2010052304066.html Stream, C. 200 8 Beyond the t hree Rs: Textbook a doption in Nevada. Southwestern Social Science Association Annual Meeting. Tulley, M. 1985. A d escriptive s tudy of the i ntents of s tate level t extbook a doption processes. Educational Evaluation and Policy Analysis 7: 289 308. Ueda, S. and M. Benouahi. 2009 Accountable water and s ani tation governance: n Water in the Arab world: Management perspectives and i nnovations e dited by N.V. Jahannathan, A. S. Mohamed, and A. Kremer. Accessed 20 February 2012. Available http://siteresources.worldbank.org/INTMENA/Resources/Water_Arab_World_full.p df Urakami T. and D. Parker. 2011. The effects of consolidation amongst Japanese water utilities: A hedonic cost functi on analysis. Urban Studies 48:2805 2825 Watts, M. 200 9. Research on t extbook s election p rocesses in the United States. IARTEM e journal 2:1 22. Wells, J. 2002. Icons of e volution: Science or myth? How much of what we teach about evolution is w rong Washington, DC: Regnery Publishing. Williamson, O. E. 1976. Franchise bidding for natural monopolies In gener al and with the respect to CATV Bell Journal of Economics 7: 73 104. Wooldri dge, J. 2005. Introductory econometrics: A m odern a pproach Califo rnia: Southwestern College Pub.
102 Yane, S. and S. V. Berg 2013 Sensitivity a nalysis o f efficiency rankings to distributional a ssumpt ions: Applications to Japanese water u tilities. Applied Economics ( Forthcoming ) 45: 2337 2348 Zhu, X., G. K aragiannis and A.O. Lansink. 2011. The Impact of d irect i ncome t ransfers of CAP on Greek erformance: Using a Non Monoto nic Inefficiency Effects Model. Journal of Agricultural Economics 62: 630 638. Zimmerman, M. 1987. The evolution creation controve rsy: Opinions of Ohio high school biology teachers. Ohio Journal of Science 87: 115 125. Zinth, K. 2005. State t extbook a doption. ECS s tate n otes. Accessed 8 January 2011. Available https://www.ecs.org/clearinghouse/57/75/5775.htm
103 BIOGRAPHICAL SKETCH Michelle Andrea Phillips Schaffhauser was born in 1983 in La Serena, Chile. In August of 2007 she graduated summa cum laude from the University of Central Florida with a Bachelor of Science degree in economics Sh e started graduate school at the Univer sity of Florida in August of 2007 and received a Ph D in economics in August of 2012. Her research inte rests include p ublic c hoice and i ndustrial o rganization. In August 2012, she will join the Department of Economics at Missouri University of Science and Technology.