Abstract
Many municipalities across the United States have turned to unit based pricing—also known as pay as you throw (PAYT)—as a vehicle for reducing municipal solid waste generation, increasing recycling, and promoting equity in paying for the service. In this paper, we reevaluate the standard analytical methods used to examine the impact of PAYT in the literature and illustrate that econometric shortcomings may have led to underestimation of policy effects. A two-tier analytical approach to examining PAYT program effects is proposed and is demonstrated in a case study of New Hampshire municipalities. (JEL O18, Q58)
I. Introduction
Economists have long promoted the use of economic incentives in environmental policy as providing more efficient solutions than command and control systems. At the local government level, many municipalities are attempting to increase efficiency through privatization of some services and implementing user fees for others (Girard et al. 2009; Mohr, Deller, and Halstead 2010). One example is management of municipal solid waste (MSW) via unit based pricing. Unit pricing or “pay as you throw” (PAYT) for waste disposal has the potential for jointly considering the often conflicting goals of efficiency and equity and offers consumers flexibility in their choice of the amount of service needed.
Until about 1988, only a handful of cities experimented with unit pricing for solid waste management, when cities across several states including Illinois, Pennsylvania, New York, and New Jersey implemented extensive and successful programs (Miranda et al. 1994). Since then the use of PAYT has grown. In the early 2000s approximately 10% of all cities and municipalities in the United States (more than 5,000 communities in 46 states) employed such programs (Lambert 2005; Miranda and Bynum 2002). By the mid-2000s, more than 7,000 jurisdictions had incorporated some form of user fee system to manage MSW (Skumatz 2008). Previous studies have extensively documented the implications of PAYT as a component of solid waste management (e.g., Morris and Holthausen 1994; Fullerton and Kinnaman 1996; Nestor and Podosky 1998; Callan and Thomas 1999; Hong and Adams 1999; Van Houtvan and Morris 1999; Kinnaman and Fullerton 2000; Skumatz 2000; Miranda and Bynum 2002; Kinnaman 2006). Various studies have found that implementation of PAYT resulted in reductions in MSW disposal from 15% to 74%, depending on factors such as whether recycling was provided before the program was implemented and whether diversionary means of illegal dumping, burning, and increased composting were taken into account (see, e.g., Miranda et al. 1994; Reschovsky and Stone 1994; Seguino, Criner, and Suarez 1995; Callan and Thomas 1997; Hong 1999; Van Houtvan and Morris 1999; Skumatz 2000; Fullerton and Kinnaman 1996, 2002; Miranda and Bynum 2002; Dijkgraaf and Gradus 2004).
It is reasonable to conclude that the adoption of PAYT by municipalities is not random but motivated by political and/or economic incentives. Failure to take into account what is essentially policy endogeneity can bias the estimation of policy effects. Note that the bias can go either direction, depending on the correlation of PAYT with the unobserved characteristics of towns. As discussed by Kinnaman and Fullerton (2000), towns may have different attitudes toward environmental conservation, but these attitudes might not be captured in the model (the omitted variable). Those environmentally conscious towns may be more likely to adopt PAYT, so that the effect of PAYT could be overestimated due to the omission of the environmental attitude variable. On the other hand, if towns with more garbage are more likely to adapt PAYT, failure to recognize this relationship could underestimate the effect of PAYT.
In the course of these studies, price elasticities were generally estimated to describe consumers' responsiveness to unit pricing policies. Perhaps the earliest estimate of own price elasticity of demand for residential waste disposal was by Wertz (1976) who used a San Francisco case study to generate an estimate of - 0.15, though Jenkins (1993) noted some obvious empirical difficulties in this initial effort. Jenkins (1993) found price elasticity of demand for residential solid waste services to be - 0.12. She estimated that in response to a price of $1.31 per 32-gallon price in a PAYT program (vs. a zero price), MSW would decrease by 20%, or 183 lbs (83.2 kg) annually per capita. In a survey article, Kinnaman (2006) summarized eight more recent studies (from 1996 to 2004) that covered all three types of curbside pricing programs: bag-and-tag, weight-based, and subscription of cans. The estimated demands for solid waste disposal were generally price inelastic (ε= - 0.01 to - 0.47) except for one estimation of the weight-based system using household data (ε = - 1.10).1
It is puzzling that while previous studies have documented significant reductions in solid waste generation by unit-based pricing, these studies have also found extremely inelastic own price demand for waste disposal. Although it could be the case that the pure existence of a nonzero marginal price is sufficient to influence the behavior of waste disposal regardless of the actual magnitude of the nonzero price, we suggest that this common conclusion of highly price inelastic demand for household waste disposal may be an overstatement due to less than ideal modeling strategies that fail to discern the dual meanings of zero marginal price—no unit price and no PAYT program adoption—for the non- PAYT municipalities in the cross-section demand analysis.
To investigate the impact of a PAYT program, we propose to examine it via two venues—the adoption of PAYT and the pricing for PAYT. We first examine the overall effect of adopting the PAYT policy on waste disposal, holding other policy and various socioeconomic variables constant. Potential policy endogeneity is examined. We then examine the effects of pricing in the estimation of demand for waste disposal among PAYT municipalities with treatments for the incidental truncation of PAYT adoption. Using a case study of New Hampshire municipalities, we find that the presence of a PAYT program, availability of curbside trash collection, and price to dispose of waste are all significant influences on per capita solid waste generation. Holding other things equal, a PAYT town produces at least 43% (or 0.22 tons) less solid waste per capita than a non-PAYT town. We also find that for the New Hampshire towns, the demand for solid waste disposal is price elastic ( I εl > 1). The case study demonstrates that the standard practice of combining both PAYT towns and non-PAYT towns in the estimation of demand for solid waste disposal masks the true price sensitivity of waste generation. For comparison, we also apply the proposed empirical methods to analyze the data of Kinnaman and Fullerton (2000) that consist of communities with or without PAYT in different states. Based on their data, the findings of PAYT program effects are qualitatively similar to the New Hampshire data in that endogeneity of PAYT may have biased the program effects downward. In contrast to the New Hampshire data, the results of price effects, after correcting for self-selection, are insignificant; that is, the implied price elasticity is not significantly different from zero. Combining the estimation results of the policy and price effects of the Kinnaman-Fullerton data indicates that the existence of a PAYT program (nonzero marginal price) does significantly reduce waste generation; however, the actual magnitude of the nonzero price does not seem to matter in their data. In contrast, the New Hampshire data show significant behavioral response to price changes.
II. Analytical and Econometric Issues
From the literature, PAYT's reduction of MSW waste production is evident even when diversionary means such as illegal dumping are taken into account. However, most studies also find highly inelastic demand for solid waste collection, implying that disposal behavior is generally not affected by pricing. Kinnaman (2006) concludes from a review of current literature that unit pricing fails to influence disposal behavior, and the net benefits of unit-pricing programs are very small or negative. We argue that there may be analytical and econometric issues overlooked in many existing studies to compromise the evaluation of PAYT programs.
To start, with the exception of Kinnaman and Fullerton (2000), most studies examine the effects of waste disposal policies without investigating the potential endogeneity of policy choices.2 For example, some municipalities that have high property tax rates to support expensive local public services such as disposing of solid waste may adopt PAYT as a means to reduce waste and curb increasing expenditures on waste management. On the other hand, some towns may be more environmentally conscious and adopt PAYT to further reduce waste and lessen residents' environmental footprints. These towns may have produced less solid waste per capita than others even without the user fee system. Similar to most public policies, PAYT is not likely adopted at random. Failure to account for self-selectivity may bias the estimated effects of the chosen programs; the direction of bias depends on the different adoption scenarios.
Another important issue arises when assessing the effectiveness of economic principles underlying PAYT via the estimation of price elasticities of waste generation. There are generally two types of PAYT price elasticity studies. One type examines the program implementation in a single town and computes arc price elasticity from zero to a set price before and after the implementation (e.g., Fullerton and Kinnaman 1996). There is usually insufficient price variation to trace out the whole demand curve for waste disposal in this type of policy experiment. Further, the relatively large percentage change in price (from zero to a positive price) tends to lead to small arc price elasticities (in absolute value). For example, suppose that the implementation of a PAYT program in a town increases the marginal price of waste disposal from $0 to $1 and results in 50% reduction in waste generation. Based on the standard formula, the estimated arc price elasticity is - 0.25. Without knowing the actual quantity decrease, one might interpret the estimated inelastic demand as ineffectiveness of PAYT to influence disposal behavior that can be misleading with regard to the actual policy effects.
The second type of study generates point price elasticity estimates by pooling both PAYT and non-PAYT towns in cross-section analysis of demand for waste disposal in which the observations are predominantly non-PAYT towns. The marginal price of waste disposal for towns with no PAYT program is effectively zero, so that the estimation outcome of the price elasticity can be driven by the large number of zero prices. Since an observation with a zero price indicates not only a zero marginal price for garbage disposal but also no adoption of PAYT, the estimation of demand slope is strongly influenced by the fact of the low PAYT adoption rate among towns (a spike of observations around zero prices) rather than the true price sensitivity of waste generation.
III. Proposed Analytical Strategies
We propose to investigate the impact of PAYT in two dimensions. Before examining the pricing effect on disposal behavior, we first examine the overall effect of the existence of a PAYT program on MSW generation rates:
[1]where PCMSWi is tons of MSW per capita produced by town i, PAYT is a dummy variable describing whether a town employs a PAYT program, PV are other political/pro- gram variables such as whether the town already has adopted policies or ordinances on mandatory recycling programs or curbside trash pickup, and SD are sociodemographic variables such as per capita income level and education.
The adoption of the PAYT program (or other programs) by a town is possibly nonrandom and can be motivated by, for example, attempts at controlling the increasing costs of solid waste collection or lowering total costs of public services to reduce property tax bur- dens.3 The endogeneity of program adoption can result in inconsistent estimation of program effects if not treated in the regression analysis. There are different econometric techniques to treat the potential dummy endogenous regressors (in our case the PAYT and other program dummy variables). The conventional method is to employ a Heckman-type treatment effects model with a two- step procedure that utilizes all observations in the second-step regression (Barnow, Cain, and Goldberger 1981). Angrist (2001) discusses several new econometric procedures to treat endogenous dummy regressors. One relatively simple procedure presented by Angrist (2001) is to employ a two-stage least squares (2SLS) estimation using all exogenous variables and the predicted values of the endogenous dummy regressors as instruments. Let D be the potentially endogenous dummy variable; W and X are sets of explanatory variables. Note that there can be partial overlap between W and X. The procedure is outlined as follows (Angrist 2001):
Step 1. Estimate D = g(W, X) and derive the predicted D, denoted as Dfit.
Step 2. Estimate Y = f(X, D) using 2SLS estimation with instruments = (X, Dfit).
Step 1 can be a linear probability model or a nonlinear model (such as the probit model). For comparison and robustness checks, we will present results based on alternative estimation strategies to examine the program effects in our case study.
We next examine the responsiveness of disposal behavior to the unit price for disposal.
[2]where Price is the unit cost of disposal for a bag of MSW. As discussed in the previous section, it is misleading to present price elasticity of demand for waste disposal by simply including both PAYT and non-PAYT towns in the regression estimation, since the slope of the demand curve and price elasticity can be significantly influenced by the majority of non-PAYT observations with zero prices. We argue that a more plausible modeling strategy to examine whether pricing of PAYT influences disposal behavior is to estimate price elasticity of garbage disposal demand with incidental truncation; that is, to estimate the demand equation using only the PAYT towns with a correction for self-selection into the PAYT program. For comparison, in the case study we will estimate the price effects first by including all towns (as has commonly been done), then report the estimation for towns with the PAYT programs only. We then employ the Heckman procedure to account for self-selectivity when estimating equation [2] with the PAYT towns.4
IV. Case Study: Payt in New Hampshire, 2000
In the 1990s, only about 6% of all solid waste management facilities in New Hampshire were under private ownership, but this 6% accounted for nearly 85% of all MSW disposal (Fortier et al. 2001). That combined with a more regional rather than local disposal network with longer transportation routes, a tight labor market, and unpredictable fuel prices resulted in average tipping fees of $52 per ton in 2000, higher than at any time in the previous 10 years (New Hampshire tipping fees averaged $77/ton statewide in 2009 [Reilly 2010]). The state's Solid Waste Task Force concluded that PAYT seemed to both reduce generation rates and increase recycling rates. This in turn could help alter the waste generation and waste diversion amounts in order to combat shrinking capacity, and reduce upward pressure on tipping fees.
Data
Our analysis focuses on the year 2000 to enable the use of the most recently available U.S. Census data of the town-level demographic information. Thus, all program dummy variables are defined according to adoption prior to 2000 (i.e., program dummy = 1 if PAYT adopted before or in 2000 and program dummy = 0 otherwise). The outcome variable is the per capita MSW in tons. Our analysis includes 200 towns in New Hampshire.5 Among them, 31 towns adopted PAYT by 2000. None of these PAYT towns used subscription services that required households to precommit to a certain number of bags or containers. Information on the town-level solid waste management programs and details regarding each program, including marginal cost to the homeowner in cents per gallon (typical PAYT programs use 15 gallon trash bags; these volume measures can be converted to pounds per gallon), were obtained from the New Hampshire Governor's Recycling Program and by contacting individual town halls. The state of New Hampshire Revenue Department provided municipal tax rates, while the state Department of Environmental Services and town managers provided the annual MSW disposal rates in terms of tonnage per municipality. All relevant variables and their definitions are summarized in Table 1. The summary statistics are presented for all towns, as well as separately for the PAYT and non-PAYT towns. The mean annual waste disposal rate was 0.482 tons per person in 2000. When converted to a pounds- per-day unit this works out to 2.66 lbs (1.2 kg) per person, per day. Comparing the PAYT and non-PAYT towns, the average annual per capita MSW (PCMSW) is considerably lower in the PAYT towns (0.307 tons) than in the non-PAYT towns (0.517 tons). Two other measures seem to differ between the PAYT and non-PAYT towns. The average per capita solid waste expenditure (PCSWExp) seems lower and the property tax rate (PrTax00) seems higher for the PAYT towns. The difference in mean property tax rate between PAYT and non-PAYT towns is statistically significant.
There are other programs relevant to MSW management. Kinnaman and Fullerton (2000) note the importance of coupling curbside recycling with PAYT. Under PAYT, the convenience of curbside recycling may encourage recycling to divert waste disposal and enhance effects of PAYT. A dummy variable, Curb- Cycl, which equals 1 if the town has curbside recycling and is 0 otherwise, was created. Since curbside trash pickup lowers the costs of waste disposal for households, it is expected that curbside trash pickup will result in more waste generation.6 We also create a dummy variable, CurbPkUp, which equals 1 if the town has curbside trash pickup and is 0 otherwise.7 Another dummy variable, MandRec, which equals 1 if mandatory recycling exists and is 0 otherwise, was created. The presence of a mandatory recycling program may reduce waste generation (depending on enforcement).
The two main equations to be estimated are the PAYT choice equation and the per capita MSW (PCMSW)equation. The factors we believe to motivate or influence the implementation of PAYT include (but are not limited to) factors that affect a town's finance, expenses of solid waste management, a town's education level and income level, and the size of a town. The variables we are able to construct and use include CIP (the existence of a capital improvement plan), PCSWExp (per capita costs of solid waste management), PrTax00 (property tax rate in 2000), College (% of population with a bachelor's degree), PCInc (per capita income in $1,000), and Pop00 (population in thousands).
Variable Definition and Summary Statistics
For the PCMSW equation [1], we include dummy variables to indicate presence of programs relevant to MSW management: PAYT, CurbCycl, CurbPkUp, and MandRec1. In addition, we include the socioeconomic variables College, PCInc, and AvgHSize (average household size). Other socioeconomic variables that we have tried including are age and percent of high school graduates. They are never significant, so we do not include them in the final specification. The definition and summary statistics of all variables are given in Table 1.
Program Effects on Per Capita Solid Waste Generation in New Hampshire
Following the analytical strategy outlined in the Section III, we first estimate the PAYT equation. We then employ the 2SLS procedure to estimate program effects in the PCMSW equation as a function of the endogenous PAYT, other program dummy variables, and demographics (equation [1]). Next, to examine the demand responsiveness to the PAYT pricing, equation [2] is estimated using various estimation strategies with or without correction of incidental truncation for comparison.
Estimation Results
To determine the functional form of the PCMSW equation, we first estimate the Box-Cox model, and the results clearly indicate a natural log transformation of the dependent variable of PCMSW.8
Effects of PAYT and Other Policies on Waste Generation
Table 2 presents the estimation results of the effects of the PAYT program on logged per capita waste generation. Four sets of results are presented: the standard ordinary least squares (OLS) estimation, the conventional treatment effects model to estimate the treatment of PAYT, the 2SLS estimation that allows for endogenous PAYT, and the 2SLS estimation that accounts for potential endogeneity of both PAYT and CurbCycl.9 The estimated probit models to endogenize PAYT and CurbCycl in the treatment effects and 2SLS models are reported in Appendix Table A1. An increase in a town's average per capita income level significantly decreases a town's probability of adopting PAYT. Also, the PAYT adoption is associated with higher property tax rates. There is a negative correlation between expenditure on solid waste management and the adoption of PAYT. The presence of a capital improvement plan is positively correlated with the presence of a mandatory recycling program.
We also tested for overidentifying restrictions to examine the quality of the instruments. All instruments except for the per capita expenses of solid waste management (PCSWexp) passed the test. Since we believe the costs of waste management to be a factor that influences the decision to implement PAYT but does not affect household solid waste generation, we continue to include it in the probit equations but not in the PCMSW equation.
A Breusch-Pagen test confirms that heteroskedasticity is present in the OLS estimation, and White corrected standard errors are used. The OLS results indicate that the presence of a PAYT program has a negative and significant effect on waste generation. In contrast, the coefficient for the PAYT variable in the conventional treatment effects model, while still highly significant, is noticeably larger than the coefficient estimate in the OLS model ( - 1.132 vs. - 0.566). To describe the policy effects, we employ Kennedy's (1981) formula to compute the effects of a dummy variable in a semilog model.10 Thus, the percentage change induced by implementing a policy is 
, where 
is the estimated coefficient of the policy dummy and 
is the estimated variance. The estimated percentage change in PCMSW from having a PAYT program is 100*e(- 0.566 - 0.140*0440*0.5) - 1 = - 43.77% for the OLS model and is 100*exp(- 032 - 0.459*0.459*0.5) - 1 = - 70.98% for the conventional treatment effects model.
The estimated reduction due to the PAYT program based on the treatment effects model is much higher than indicated by the OLS estimation. Note that the inverse Mills ratio is not significant in the conventional treatment effects model. To further examine the potential endogeneity of PAYT, we turn to the 2SLS estimation. The first 2SLS estimation allows endogenous PAYT. Applying the Kennedy formula, the implied effect of PAYT is a 71.67% reduction on PCMSW. The second 2SLS estimation that allows endogenous PAYT and CurbCycl implies a PAYT program effect of a 71.66% reduction in solid waste generation. Overall, after taking into account the potential self-selectivity of towns adopting the PAYT program, the estimated PAYT program effects from the conventional treatment effects model and the 2SLS models are close in magnitude and are noticeably larger than indicated by the standard OLS estimation.
Curbside recycling and curbside trash pickup do not appear to significantly affect solid waste generation, while mandatory recycling tends to lower the per capita solid waste generation.11 Across all four models, the household size coefficient is significant and negative, implying lower per capita waste generation in larger families. This may be because households with more people tend to buy in bulk, resulting in less packaging. Also, larger households may tend to share many “common” items such as light bulbs and newspapers, again resulting in smaller amounts purchased on a per capita basis.
Price Effects on Waste Generation
Table 3 presents the results of the model from equation [2] to evaluate the effect of a change in the price of a 15 gallon bag (Price15g) on per capita waste generation. Both results for all towns and for towns with PAYT only are presented. In addition, the Heckman procedure is employed for the estimation based on the PAYT towns with control for self-selection. The probit model of PAYT in the first step of the Heckman procedure to correct for selection bias is reported in Appendix Table A1.12 Results show that in all estimations (based on all towns, PAYT towns only, and PAYT towns with correction for selectivity), the coefficient for PAYTP15G is negative and highly significant, indicating that increased price for MSW disposal can lower annual per capita MSW generation rates. The estimated price coefficient in the model for all towns is – 0.901. In contrast, the estimated price coefficients based on the PAYT towns with and without correction for self-selection are – 1.905 and – 1.965, respectively. The coefficient estimate of – 1.965 implies that for every cent increase in the price of a 15-gallon garbage bag, PCMSW is estimated to reduce by approximately 2%. (Note that the measurement unit for PAYTP15G in the model is dollars.13)
Effects of Unit Pricing on Solid Waste Generation in New Hampshire
While all three models imply that increasing the cost per bag reduces waste generation, elasticities differ considerably between the models. For all towns, Ɛd = - 0.096 (evaluated at the mean price of $0.107), which is consistent with previous studies. However, when looking at the selection model for towns that have a PAYT program, εd= – 1.312 (evaluated at the mean price of $0.689). Thus, small manipulations of per bag price in towns that have PAYT lead to elastic responses in per capita waste generation. If we take the coefficient estimate based on all towns and compute price elasticity at the mean price of the PAYT towns, the resulting price elasticity of – 0.621 ( = – 0.901*0.689) still severely underestimates the price effect of the PAYT programs. The drastic difference is driven by the large number of zeros in the price variable for those non-PAYT towns that influences both the slope estimation and the average price.14
Scatter Plot and Simple Regression of the Logged Per Capita Municiple Solid Waste on Unit Price, All Towns
Scatter Plot and Simple Regression of the Logged Per Capita Municiple Solid Waste on Unit Price, PAYT Towns
Figure 1 presents a simple plot of logged PCMSW against the per bag price for all towns. The spike of zero price observations represents the non-PAYT towns. Figure 2 plots only the PAYT towns. As can be seen, the spike of the zero-price observations substantially rotates the fit of the regression line. This illustration is of great interest since it contradicts the prevailing notion that demand for solid waste disposal is unresponsive to price changes, with our results indicating that, in actuality, it is quite responsive. A comparison of the elasticities generated in this study with a selection of previous empirical estimates for various U.S. municipalities is provided in Table 4. Caution must be taken when using price elasticities to describe policy effectiveness. It is important to draw a distinction between the policy effectiveness and the responsiveness to pricing of PAYT. This point is further demonstrated in the following section.
V. The Kinnaman-Fullerton Data Revisited
For comparison, we apply the proposed empirical strategies to analyzing the data used by Kinnaman and Fullerton (2000).15 This data set was compiled from two sources in the late 1990s. First, original data were collected for communities with user fees for solid waste disposal. The data were then merged with the data of over 800 non-PAYT communities that were provided by the International City Managers Association. After screening for missing values and outliers, the number of observations employed to generate the results of the MSW regression by Kinnaman and Fullerton (2000) is 756, including 91 PAYT communities and 665 non-PAYT communities. Among the PAYT communities, 40 have a zero price, which is not reasonable since a nonzero marginal price is expected when implementing the PAYT program. We omit those observations and use the remaining 716 observations for analysis. The variable definition and summary statistics are given in Table 5. Detailed discussion of the variables is presented by Kinnaman and Fullerton (2000). The data set consists of communities nationwide, so that state policy variables such as state law to prohibit yard waste from landfill (I–YW) and state deposit/refund system for bottles (I–DR) can be included in the regression analysis. One drawback of the data set is the inclusion of nonrandom communities due to availability of data.
We follow the model specification used by Kinnaman and Fullerton (2000) and employ our estimation strategies outlined in Section III. The estimation results of the logged per capita residential solid waste (LogG) equations are reported in Table 6.16 The first two columns of Table 6 show the OLS and 2SLS estimation of the overall policy effects, where the 2SLS model allows both PAYT and curbside recycling (I–R) to be endogenous. Comparing the estimated coefficients of PAYT in the OLS and 2SLS models, the estimated effect of PAYT is larger after accounting for policy endogeneity, similar to the results for the New Hampshire data. The implied percent change in solid waste reduction is 43.47% using 2SLS and 39.33% using OLS. Thus, the endogeneity causes a slight downward bias in the estimated policy effect.
Comparison of Elasticity Estimates of Previous Studies with Current Study (HHS)
Kinnaman and Fullerton Data: Variable Definition and Summary Statistics
The next three columns report the effects of PAYT pricing. Similar to the analysis of the New Hampshire data, we present (1) the OLS results using all observations, (2) the OLS results with PAYT observations only, and (3) the Heckman estimation to address potential selfselection of PAYT implementation. It is seen that the coefficient of price (P2) is significant and negative when all observations are included in the estimation. In contrast, the price coefficient becomes insignificant when only the PAYT communities are included or when the Heckman procedure is used to account for selection bias. It is clear that the estimation of the MSW-price relationship, when both PAYT and non-PAYT observations are included, is driven by the majority of the zero-priced non- PAYT observations. These results, combined with the significant policy effects from the 2SLS model, suggest that it is effective to implement PAYT to impose a nonzero marginal price. However, the actual magnitude of the nonzero price of the PAYT program does not seem to matter here. In contrast, the New Hampshire data show both policy effectiveness and responsiveness to PAYT pricing.
Kinnaman and Fullerton Data: Program and Price Effects on Per Capita Solid Waste Generation
VI. Discussion
Our findings suggest that the solution often espoused by environmental economists—basically, effluent fees by taxing consumer discharges—leads to a significant decrease in per capita waste production. Regression analysis showed that an increase in the price of waste disposal as imposed by a PAYT program had a statistically significant effect on MSW disposal amounts in New Hampshire towns. This decrease in MSW disposal rates may translate into an overall lower cost of disposal to a specific municipality, since most towns pay per ton tipping fees to private firms. From a policy standpoint, although results show that an increase in price may reduce per capita MSW generation rates, PAYT may not be right for every town. The design and implementation of a PAYT program is not without administrative costs, which range from staff time to recordkeeping to education and outreach to residents regarding program specifics. Further, the costs of alternative means of waste management may not be trivial. Obviously, the overall savings from reduced per capita generation will be beneficial to the community only if they outweigh the costs of program implementation.
Most New Hampshire communities do not keep detailed and complete records of solid waste recycling rates, nor do they document composting or illegal dumping/burning, so that we were unable to decompose the total effect of PAYT on waste generation into diversion and source reduction, obviating a need for further investigation. Nonetheless, the case study demonstrates the proposed two- tier analytical approach for properly examining the overall impact of a PAYT program on waste reduction. The analysis can be extended to incorporate recycling and illegal dumping should the data become available in the future.
From a methodological standpoint, our proposed two-tier analysis of the impact of PAYT is of interest both in planning future research and policy efforts and in interpreting past research. Analyses that neglect the selectivity issues may not identify coefficients that may be statistically significant, or may not estimate the correct magnitude of these coefficients. We show that PAYT can have larger effects on waste generation than indicated by the standard OLS estimation in both the New Hampshire case study and the Kinnaman-Fullerton data. Regarding the effects of changes in per capita waste generation due to changes in the per unit price, results are highly sensitive to whether models are estimated using the pooled data of all towns or the two-step procedure on the PAYT towns with correction for self-selection. We argue that it is analytically and empirically inappropriate to interpret the price coefficient in the pooled analysis as the price or policy effect of the PAYT program on disposal behavior. It is demonstrated in this study that caution must be taken when estimating price elasticities to indicate policy effectiveness.
Acknowledgments
We are grateful to Tom Kinnaman for sharing his data with us, and Don Maurer of the New Hampshire Department of Environmental Services for help with New Hampshire data collection. This paper has greatly benefited from the comments of two anonymous referees, the participants of the seminar in the Department of Economics at the University of New Hampshire, and the participants of the session of Environmental Protection, Waste Management, and Recycling I at the 4th World Congress of Environmental and Resource Economists in Montreal. We also thank Aliya Sassi for the relentless data collection effort and Son Tung Nguyen for the analysis of the Kinnaman-Fullerton (2000) data. Partial financial support from the New Hampshire Agricultural Experiment Station, Project H 335, is gratefully acknowledged. This paper is Scientific Contribution Number 2325 from the New Hampshire Agricultural Experiment Station.
Appendix
Probit Models of PAYT and Curbside Recycling Programs
Footnotes
↵1 Linderhof et al. (2001) examined the effects of weight-based pricing on household waste generation in Oostzaan, Amsterdam, from July 1993 to December 1996 with the implementation of weight-based pricing in October 1993. In their panel data analysis, they separated compostable waste and nonrecyclable waste. They found that the estimated price elasticities were greater than 1 (in absolute values) for the compostable waste but less than 1 for the nonrecyclable waste. Their study shows that households' generation of different types of solid waste may respond to unit pricing differently.
↵2 Kinnaman and Fullerton (2000) allowed endogenous curbside recycling policy choice by replacing the policy dummy variable with the predicted probability of policy adoption in the analysis of solid waste generation. They also included and endogenized the unit price of waste disposal in the analysis. However, they followed the usual modeling strategy of pooling both PAYT and non-PAYT towns so that the estimated pricing effect of PAYT could be heavily influenced by the large number of observations of the non-PAYT towns, as discussed in this study.
↵3 According to a report by the New York Times in 2007, the average property tax rate is 2.21 in the state of New Hampshire, the second highest among all states. (“State-by-State Property-Tax Rates,” New York Times, April 10, 2007, available at www.nytimes.com/2007/04/10/business/11leonhardt-avgproptaxrates.html.)
↵4 One may make an attempt to include both the PAYT dummy variable and the unit price of PAYT in the same regression model to simultaneously examine the program and pricing effects. Since the implementation of PAYT comes with a positive unit price, the confounding scenario makes it implausible to examine the effect of PAYT with price as a control variable. It is also not possible to include the interaction of the PAYT dummy and price, since the interaction term will be identical to the price variable, resulting in perfect collinearity.
↵5 There are in total 234 towns in New Hampshire; 34 towns were excluded from the analysis due to unavailability of town-level data on solid waste generation.
↵6 In our New Hampshire data, the correlation between curbside pickup and curbside recycling is 0.37, not as high as one might have expected.
↵7 One may expect to see high correlation between CurbCycl and CurbPkUp, or CurbCycl=1 only when CurbPkUp = 1. It is interesting to know that the correlation coefficient between CurbCycl and CurbPkUp is 0.37 for the New Hampshire towns, and over half of the towns with curbside recycling services do not have curbside trash pickup.
↵8 The results of the Box-Cox estimation are available upon request.
↵9 We also tried to simultaneously endogenize PAYT, CurbCycl, and CurbPkUp. Results are similar. However, when MandRec1 is also endogenized, all variables' coefficients become statistically insignificant.
↵10 The coefficient of a dummy variable in a semilog model is commonly misinterpreted as the percentage effect of a dummy variable on the dependent variable (Halvorsen and Palmquist 1980). Derrick (1984) compared three alternative estimators for the percentage impact of a dummy variable in a semilog model over a range of degrees of freedom and variances. He concluded that the estimator proposed by Kennedy (1981) was the preferred choice in practical situations.
↵11 We tried to include the interaction of PAYT and CurbCycl in estimation. The coefficient of the interaction term was never statistically significant. Curbside recycling services did not appear to reinforce the effects of the PAYT program in the New Hampshire data.
↵12 Given the small number of observations for the PAYT towns, we recommend caution in interpreting these estimates due to efficiency issues in the estimation. We also attempted to endogenize the price variable, but we could not find good instruments to explain the price.
↵13 We caution that the estimated coefficient here may seem to suggest more than 100% waste reduction for $1 change in the price of PAYT. Note that the average price of the PAYT towns is only 68.9 cents, so that a dollar increase in price will mean an approximately 150% increase in price. It is more plausible to present the marginal effect of price in cents than in dollars.
↵14 One possibility in the demand analysis is to replace the zero prices of non-PAYT towns by the predicted prices based on the fitted censored price equation for the PAYT towns, as is commonly done in labor supply analysis to predict wages for those who do not work. However, the predicted price does not explain the household behavior of the non-PAYT towns, since it responds to the actual price of zero. Further, in our case (and other cross-sectional PAYT studies in general), the majority of towns do not have PAYT (169 out of 200), so that it is not reasonable to use the relatively small number of PAYT towns to predict the prices for the non-PAYT towns.
↵15 We thank an anonymous referee for the suggestion to apply our analytical strategies to the Kinnaman-Fullerton data for comparison.
↵16 To conserve space, the estimated probit models of PAYT to accompany the treatment effects, 2SLS, and Heckman model are omitted and available upon request.
