ࡱ; |uv  !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrst{ywxz}~Root Entry FǶ@ZCompObjbWordDocumentmqlObjectPool'   4@   !"#&()*+,-./0123456789:;<>ABCDEFGJLMNOPQRSTUVWXYZ[\]^_`befgjlmnopqrstuvwxyz{|} FMicrosoft Word 6.0 Document MSWordDocWord.Document.6; ࡱ; LtEDlࡱ; tEM  .1  &` & MathTypeSymbol-2 | ܥe- eql(F \,\,,]0]0c0c0c4 f f f f f"f:\f ftk@fh"0h0h0h0h7h Chjjjj*jDkDVkkTlitk0cKh 0h0hKhKhtkKh_dB`H0hfKhKhKhKh0c0h0c0hjdcbd8]6^`aKhjKhWKh revised 4/25/96 Marine Debris, Beach Quality, and Non-Market Values V. Kerry Smith, Xiaolong Zhang, and Raymond B. Palmquist* I. Introduction Marine debris is an international problem influencing beaches around the world. Over twenty years ago, the National Academy of Sciences [1975] estimated 6.4 million tons of debris were discarded in the world's oceans annually, with most of that occurring in the northern hemisphere. The problems have not improved during the two decades since that estimate was prepared. It became a front page news story in the late 1980s. With this media attention, there has been increasing regulatory activity and some effort to monitor the problem (see Cole et al. [1995]). Nonetheless, to our knowledge, this paper reports the first attempt to measure the importance of controlling marine debris to enhance the quality of beaches. Our findings are based on a contingent valuation (CV) survey designed to estimate the economic value of controlling marine debris on recreational beaches in New Jersey and North Carolina. Our research was designed to address three issues. The first of these concerns the problems posed in measuring peoples economic value for changes in the quality of coastal resources. At a general level, most people would acknowledge (from experience) that each individual may have different reactions to and preferences for coastal resources. However, beyond this very general response to a link between the aesthetic dimensions of coastal resources and peoples perceptions of their quality, identifying specific features that would underlie each individual's ratings is difficult. Moreover, there is no reason to assume that people will consider the same characteristics important. This research was designed as an initial step in addressing this issue A second objective of our work concerns the process used to identify how people form perceptions of beach quality. We sought to evaluate whether quantitative insights could be derived from the largely qualitative research associated with designing contingent valuation (CV) questions. Because our problem involved characterizing the influence of the amount and types of debris for the perceived quality of beaches, we have attempted to evaluate the consistency of the evidence across the various stages of questionnaire development and, ultimately, in the estimates provided by our contingent valuation survey. The third goal of this research called for investigating whether independent samples of individuals would react to different amounts of quality impairment as economic theory would suggest. Current policy interest in this responsiveness as a scope test makes this aspect of our findings relevant to evaluations of the CV methodology from a more general perspective. In the next section, we describe the methods used to measure economic values for debris control. Following that discussion we summarize the development of a questionnaire to investigate peoples evaluations of the role of debris as an influence to the visual dimensions of coastal resources. Section four describes our data and results. The last section discusses their implications for future research in estimating the economic values people have for debris control. II. Marine Debris as a Quality Characteristic of Beaches To estimate the economic value a person places on a commodity (or some characteristic of a commodity) we must observe a choice. Under conditions ideally suited for measuring economic values, we would observe both the object of choice and the set of circumstances relevant to each choice. Monetary measures of economic value can relate to any object of choice, and these choices can involve anything -- an apple or a beach that is free of debris. The circumstances of a choice relate to all the factors that describe the assignment of rights, timing, degree of certainty, and consequences of a choice. When both the object of choice and the circumstances are clearly defined, an individual's decision implies the object selected is at least as good (or as valuable) as what was given up to obtain it. Thus, when a choice is made in favor of the object, the consequence related to the most important foregone alternative defines a lower bound for an individuals willingness to pay (WTP). When a choice is made not to select the object, we know that the individuals WTP must be less than this most important foregone alternative. Contingent valuation analysis of the economic value of debris control applies this basic logic by describing to each surveyed individual the characteristics of beaches with special emphasis on the amount of debris present. The interview then describes a plan to modify these conditions by controlling or removing the debris. These descriptions define the object of choicethe program to control or remove marine debris. They include a characterization of how the program will be financed and the cost imposed on each individual. The cost provides a monetary measure of what will be foregone if the individual chooses the program. Each person's evaluation can be represented using an indirect utility function, V(), expressed as a function of income, y, a vector of prices for the marketed goods that individuals consume, P, and the characteristics of the coastal beaches that are presented as part of the choice. Because it is a decision about a program to control their quality, this aspect of the evaluation is described as a function C(q) with q indicating the quality level of a coastal resource. Equation (1) uses this simple model to consider how this type of choice would be evaluated. It assumes that the program has been described as involving a one time payment, t, to improve the quality of the beaches from q0 to q1.  EMBED Equation.2 )(1)If the left side of equation (1) exceeds the right, then t provides a lower bound for the WTP associated with the improvement in the quality from q0 to q1. One of the central issues raised in using contingent valuation surveys to measure economic values stems from the recognition that CV choices do not require the financial commitment that is proposed to respondents. As a result, they are describing programs that could be offered or could be on a referendum. When an individual states a choice that would be made, there are inevitable questions as to how likely these stated intentions correspond to what would actually be done if the program were offered and the implied financial obligation capable of being imposed. Debate over the reliability of stated intentions versus observable actions has been significant in this literature. There is no consensus as to what the available evidence implies. Three years ago in January 1993, a panel of distinguished social scientists co-chaired by Kenneth Arrow and Robert Solow concluded that CV estimates of nonuse values could be used as a starting point in the development of monetary measures of economic losses in natural resource damage assessments. The Arrow-Solow Panel's evaluation held that the burden of proof in establishing the reliability of a CV survey rested with the analysts. Moreover, they proposed a set of guidelines for CV practice. We will discuss some aspects of these guidelines in the next section. III. Designing the Contingent Valuation Questionnaire Four aspects of the Arrow-Solow Panels recommendations for CV design are important to evaluating our study. These include requirements for: (a) probability sampling with sufficiently high response rates to avoid selection effects; (b) detailed information confirming that the object of choice was presented to respondents clearly; (c) use of a referendum format so that the proposed object of choice and cost are presented as part of a "take it or leave it" choice; and (d) evaluation of the responses to assure those interviewed understood questions and that their stated choices recognize the consequences of their decisions. These recommendations were made after our survey was completed. In addition to these issues of direct relevance to our research, the Arrow-Solow Panel also recommended in-person interviews and the use of a seventy-percent response rate as part of their burden of proof standard for evaluating CV studies used in litigation associated with natural resource damage assessments. Most research studies (including this one) do not have the resources required to pay for all the activities required to meet these standards. As a result, it is important to document what was done to maintain as high response rates as possible and to investigate the implications of selection effects sometimes associated with lower response rates. Describing the control and cleanup of marine debris as a meaningful object of choice requires an understanding of how people think about debris on coastal beaches, as well as their responses to programs intended to meet these objectives. Because past research on measuring the importance of aesthetic dimensions of environmental resources has suggested that a visual image can be important to peoples perceptions, we hypothesized that photographs of beaches with differing amounts of debris would be important to presenting this feature of the quality of beaches. Debris accumulation in coastal areas is a highly variable process, as most of the informal literature in this area and the more systematic results found in the Cole et al. [1995] suggest. This Cole et al. [1995] study considers four years of sampling of debris at seven national parks in coastal areas. It found that debris accumulation varied considerably between years for all parks. Storms, human activities, and unforeseen events all contribute to the stochastic processes that generate variable debris accumulation patterns. There is simply not enough information to characterize this process so that specific design treatments and conditions could be identified. Because of these limitations, we based our characterization of conditions on actual records and not a simulated pattern. Developing this profile required records of the conditions of the same beaches at different times. These records would offer some characterization of the variations in the amount of marine debris that could arise within the current regulatory regime. We were able to obtain assistance from Dr. Tony Amos of the University of Texas. Dr. Amos has maintained a photographic record of debris conditions over more than ten years for one section of Mustang Island, Texas and provided complete access to his library of slides of the beach scenes under different conditions. Our development process involved three sets of activitiesfocus groups, a pretest, and a review of the proposed survey instruments by other economists with experience in the design of contingent valuation surveys. Three focus groups were conducted over a seven month development period. Table 1 describes the locations, dates, number of participants, and tasks considered in each group meeting. The objective of these meetings was to evaluate how people would describe debris on beaches and whether photographs would be effective in explaining different debris conditions. Initially, two sets of twelve 5 inch by 7 inch photos each were presented to each participant. Each person in this first focus group was asked to rank them independently. Subsequent focus group meetings narrowed the set to twelve photos for ranking. The final set of twelve scenes included beach scenes that could be interpreted as arising from the same general area with natural and man-made debris as well as people and structures. The findings from all three focus groups indicate that participants displayed very consistent processes in forming their perceptions about the beaches presented in the photos and the role of debris in their judgments about beach quality. There was a clear indication that photographs could be used to describe alternative beach conditions and that respondents seemed capable of transferring the scenes from one location (e.g. Mustang Island, Texas) to the beaches they considered to be relevant to their local area. Moving toward our conceptual formulation that assumes the existence of a function, C(q), to describe the quality of coastal resources requires a quantitative evaluation of the consistency in focus group participants' evaluations of the beach scenes to develop this type of consistency test, indexes for the attributes of each photo. These indexes are intended to measure the features that one might expect participants would notice in each scene. For example, one index scored the amount of manmade debris that seemed to be present in a scene. Another rated the amount of natural debris (kelp). These indexes serve as quantitative measures of the physical attributes that may be related to perceived quality and can be evaluated based on each photo. By decomposing the elements present in each photo we hypothesized it would be possible to measure their role in peoples overall quality ratings for each scene. These index scores were constructed by each author independently, without considering how the focus group participants ranked the photos. The participants saw only the photos. Each index was measured on a 0 to 5 scale. For attributes used to describe positive features of a scene higher values of the index are associated with more of the positive characteristic and for potentially undesirable features higher values imply more of the negative characteristic as well. Thus, a score of 5 for density of manmade debris implies the photo had the greatest density possible among the original twenty-four scenes used in the first focus group. The average rating scale across all three authors was used to describe each attribute for each photo. The attributes included: density of natural debris, density of manmade debris, clarity of the photograph, importance of dead marine life in photo, and water quality. These were combined with qualitative variables indicating the presence of people, structures, and unique debris in each photo. To evaluate whether participant rankings could be modeled. we hypothesized that these rankings reflected preference orderings and, as such would be a function of the characteristics we judged to be present in each beach scene. More formally, Beggs, Cardell, and Hausman [1981] have demonstrated under these circumstances that data composed of a ranked set of commodities, services or, indeed, even photos by each of a number of individuals can be used to recover a function describing the underlying preferences. If we assume independent extreme value distributions for the errors associated with these preference based rankings, then the simplest form of the model for these rankings is comparable to a random utility model. Moreover, because the probability of a ranking can be described in terms of the measured attributes for each photo we can: (a) use the rankings to estimate the relative importance of each photo's characteristics; and (b) test for differences in these evaluations across distinct focus groups. An indirect utility function analogous to those given in equation (1) can be specified, where we assume that in the focus group activities the cost of a plan to provide improvements is left unspecified; the prices of other goods and income of the individual do not vary across the alternatives being ranked; and the C(q) function is simply replaced by a vector, A, of attributes describing the features each participant is hypothesized to use to evaluate the quality of the scene depicted in each photo. In this context a photo describing conditions with A1, characteristics in comparison with only A2 would be ranked higher if the utility associated with it exceeds that for A2 (i.e. Vi(y,P,A1) > Vi(y, P, A2), with I designating the ith individual). If we assume Vi is composed of deterministic (vi(.)) and stochastic components, (ei), then the probability of a ranking is directly linked to the assumed preference functions evaluated for each as in (2):  EMBED Equation.2 (2) where V( y, P, Ai ) = v( y, P, Ai) + ei Table 2 presents the models for focus groups two and three, where the same set of photos participants were ranked and comparison of the signs and statistical significance of estimated parameters across models indicates that participants based their rankings on the comparable factors. Testing the hypothesis that the ranking models for groups two and three were the same with a likelihood ratio test indicates we cannot reject the null hypothesis of equality for all parameters. Thus, based on this criterion, participants in independent focus groups evaluated the set of photos in a similar way. The focus group discussions also indicated that the typical household may not know about the sources and composition of marine debris. Some members of our groups had reasonably good information while others had very little information. to address this disparity and provide background information, we adopted a telephone-mail-telephone survey format. The first step involves a telephone interview with an adult decision maker over 18 years of age. After completing a short interview collecting attitudinal and demographic information, the respondent was asked to participate in a second interview about coastal issues. For those agreeing, an information booklet about marine debris was sent, along with one of the eight different sets of photos describing different beach conditions. The booklet contained eight pages describing marine debris; explaining where debris comes from; outlining its effects; discussing laws to control debris; and explaining (for a second CV question) the national estuarine research reserves and the specific North Carolina sites considered as part of this question. After that material had been received the second interview was conducted with the same individual. The contingent valuation questions use the same photo describing the conditions with the plan to control or cleanup debris (this is identified in footnote a for Table 2 as the second photo. It was scene B in the focus group evaluation). Four photos were selected from the set used in the three focus groups to serve as the baseline or default conditions if the plan to control debris was not selected. These photos were identified in the photo sheets sent to respondents with letters A, B, C, D. The link to how they were used in the focus groups is described in footnote a for Table 2. The information booklet also helped to frame the program offered to respondents. Appendix 1 reproduces the text of the question used to ask about the debris control and cleanup program for recreational beaches. The package also included an explanatory letter and short mail back questionnaire about respondents' recreational use of beaches. Our pretest consisted of eight cognitive interviews. Using current telephone directories four were recruited from Raleigh, NC and four from Asheville, NC. These interviews consisted of a forty-five minute discussion of a draft of the information materials composed in booklet form with draft illustrations as well as the CV question. They were conducted by phone to correspond to the actual survey conditions. After these interviews, several changes were made to the booklet to enhance readability. In addition, the contingent valuation questions were shortened and some wording changes adopted to respond to areas of confusion identified by the pretest respondents. The research design for the survey sought to distinguish four factors that could influence peoples evaluations of programs to control debris including: local conditions (by selecting two distinctly different coastal areas), the type of coastal resources affected by debris, the amount and character of debris (i.e. recreational beaches versus estuarine research areas), and the means of payment. Some of these factors, such as local beach conditions, amount and type of debris, and the means used to pay for the control program, were identified as important by participants in the focus groups. Others, such as distinguishing the type of coastal resource experiencing impacts from debris, were identified based on another research objective not directly relevant to this paper, attempting to distinguish use and passive use (or non-use) values for quality dimensions of coastal resources. As we noted above, two different CV questions were asked of each respondent. One considered debris on recreational beaches and the other focused on debris that is found in estuarine reserves, that are used primarily for habitat and research. In the questions associated with the recreational beaches we also varied the payment mechanism considering additions to state taxes, independent of a respondents usage of specific beaches, and a specified increase in a beach access fee. When these variations are combined with the specified variation in payments as well as the different combinations of photos used for each situation, it is easy to appreciate that the design was in some respects too complex for the sample size available for analysis. Because of this limitation, research was intended to be exploratory. Thus, the results are best treated as arising from a pilot study. The final sample sizes in each of our design cells are too small to develop estimates of population parameters. Our two-stage (telephone-mail-telephone) design also contributed to this reduction in sample size. Our focus here is on the estimates of respondents willingness to pay for programs to control and cleanup debris on recreational beaches. IV. Results A. Sample Our research design considered the samples relevant for New Jersey and North Carolina beaches independently. Within each sample a further distinction was drawn in defining the initial selections for the random digit dialed (RDD) samples. About half of each sample was composed of RDD numbers selected to be representative of each state. The remaining observations were drawn from regions identified as providing recreationists for each states beaches, based on the on-site surveys conducted through NOAAs Public Area Recreation Visitors Surveys. The overall response rate for the Phase I sample was 52.6%. These interviews were completed in September 1992. Of the 1,773 adults interviewed in the Phase I sample, 66.4% agreed to a second interview. The callbacks to complete these interviews and second stage interviewing ended in November, 1992, yielding a sample of 693 completions for both interviews and a response rate for this second stage of 39.1%. As we noted in discussing our design, this low response rate raises the possibility of significant selection effects and limits the relevance of our findings for developing population estimates for the economic value of debris control and cleanup programs. However, it does not necessarily invalidate the use of the surveys for testing of specific hypotheses. Our findings can serve as an approximate gauge of the economic value people may hold for improving quality conditions at recreational beaches, but this role is conditional to any model used to take account of selection effects. The structure of the Phase II questionnaire was fairly complex. Our two samples were designed to consider marine debris on recreational beaches in New Jersey (NJ) and North Carolina (NC) as well as debris on beaches in areas comprising North Carolina's Estuarine Research Reserves. Respondents in each sample received one of eight sets of photos. Each set had a photo pair describing conditions on what were described as their (NJ or NC) beaches and a pair depicting scenes for the NC Estuarine Research Reserves along with the information booklet described earlier. Each photo pair had a "clean" photo described as the condition with the plan for cleanup and control, and a photo intended to depict debris conditions without the plan. This no-plan photo varied across four levels for the recreational beaches (labeled A through D and described in footnote a for Table 2) and over two levels for the Estuarine Reserves. Considering all alternatives from each group (beaches and reserves) provided the eight sets of photo sheets. Our design also sought to evaluate payment vehicle. It considered annual income taxes versus beach access fees for the cleanup and control of debris on recreational beaches. Two CV questions were asked of each respondent - one for debris control on recreational beaches and one for Estuarine Reserves. These were randomly ordered. Respondents were also assigned randomly a payment mechanism for the plan associated with recreational beaches. The effects of all possible assignments (payment plan and CV question order) was intended to be completely random. Unfortunately, a programming error in the computer assisted interview software used in the implementation of the survey by the research firm conducting the interviews confounded the effects of payment scheme and order of the CV question. The order effect of type of resources (beach versus research reserve) was maintained. The effect of income tax versus beach access fee was not. Respondents asked about plans involving recreational beaches first always received the annual income tax payment scheme. Because our analysis focuses only on debris on recreational beaches using an annual income tax, this error does not affect our assumption that random assignment between cleanup and control of debris for beaches and estuarine reserves allows a focus on the CV questions for debris on beaches independently of the other question for the NC Estuarine Reserves. Our primary objective is to consider how the amount of debris (without the cleanup plan) affects stated choices for the plan. Thus, we have focused our summary of the results on the questions involving debris conditions on recreational beaches with a state income tax used for the payment vehicle. B. Choice Analysis Table 3 summarizes the votes for and against the control program for each of the sub-groups receiving one of the four photographs (A through D). Respondents who answered with a Dont Know response were coded as against the plan. The letters in the columns identify the sub-samples according to the photo received. Three contingency tables are reportedone for the overall sample and two that split this sample into the North Carolina and New Jersey sub-samples. The sample sizes are smaller than the number completing the Phase II survey because the remaining group in each sample was randomly assigned to the beach access fee version of the CV questions (i.e., in these questions the payment was described as an increase in a beach assess fee). With these small sample sizes, the record of individual choices alone does not offer much ability to detect differences in perceptions of beach quality conditions based on the photos. Nonetheless, the evidence for discriminating among conditions using this simple test is a little better for the New Jersey sample, where arguably users may have experienced debris conditions similar to the photos. At first these results might appear to contradict the results from the ranking tasks presented to the focus groups. There are, however, important differences in the question addressed with these contingency tables and that considered by the models with the rankings of beach scenes. Here we are studying respondents stated choices with a specific financial consequences. The rankings did not have a financial consequence. Instead they asked participants to evaluate the photos with different amounts of debris. There are also a number of other factors changing in the responses presented in these four columns. The most important of these is the different tax amounts assigned to respondents. Because this process was random (see the text of the questions reported in Appendix 1), we would expect an approximately uniform distribution of values across the four sub-samples. We expect that the tax amount should have the same directional effect on choice regardless of the photo received. Nonetheless, with these small samples, it may be difficult to rely on the randomization alone. As a result, one way to control for the effects of tax amount while testing for the effect of debris conditions is the Cochran-Mantel-Haenszel statistic (see Snedecor and Cochran [1980]). In this case, using the general association alternative hypothesis, the (2 statistic (with 3 degrees of freedom) was 6.577. This has a p-value of .087, implying that at the 10 percent level we would conclude there were differences in the effects of the debris conditions for stated choices for a debris cleanup and control plan. This conclusion is reinforced if we selected the most extreme photos (i.e. A and D). Using the simple chi square test, the choices in this case would have been judged significantly different at a .075 p-value (without taking account of amount of the charge). The New Jersey sample would also have rejected the null hypothesis for the extreme photos with a .013 p-value. To our knowledge, this is the first evidence of scope effects with quality attributes of an environmental resource. Moreover, the relative support of each sub-sample for the cleanup plan is consistent with ratings of the photos used in the CV survey based on the independent focus group evaluations. This analysis used the composite rank logit model for focus groups 2 and 3 (given in the third column of Table 2) to compute the probability each photo describing conditions without the cleanup and control plan would be ranked "best". These computations use the indexes of the attributes constructed to describe photos A through D in the estimated model for the ranks. The estimated probabilities are reported below the column headings below each photo letter in Table 3. PR = .077, for example, implies that the rank logit model's estimates suggest a probability of 8 in 100 that this scene would be ranked best. Based on the focus group responses, it represents the worst of the four default (or no plan) states. The estimates of this probability and those of the other photos are generally consistent with the percentage support for the cleanup plan. That is, there are lower probabilities of high rankings associated with higher percentages of support for the plan. This pattern is most apparent with comparisons between the best default conditions, D (PR = .57), and the worst conditions, A (PR = .077). Thus, based on these findings, where few assumptions about the parametric features of preferences have been maintained, it appears that large distinctions in objects of choice can be isolated, especially if they are consistent with local conditions. Of course, it is important to emphasize that these comparisons do not take account of any selection effects induced by the telephone-mail-telephone format used to collect these data. C. Statistical Model for WTP Responding to the limitations imposed by our relatively small samples requires that we introduce a statistical model that is consistent with the economic framework for describing choice behavior to explain respondents decisions. The Hicksian definition for willingness to pay (WTP) for a change from q0 to q1 measures the t* that will make both sides of this expression equal as in equation (3) below:  EMBED Equation.2 )(3)Solving for t* requires that we invert the indirect utility function (V()) and express t* as a function of y, P, q1 and q0. Comparing t* to the proposed t for each individual provides a model describing his or her choice (i.e. if t* > t then a respondent will report a stated choice/vote for the program, otherwise the choice/vote is against it). Of course, the analyst does not know the exact form for the function to be used in describing peoples preferences and may have incomplete knowledge of other considerations involved in their choices. Thus, this framework generally introduces stochastic errors into its description of the choice behavior. The framework describes the probability an individuals unobservable WTP (i.e. t*) will exceed t. Thus t* is an economic construction to explain what is actually knownthe peoples decisions about specific objects of choice under a set of well defined circumstances for those choices. Because probit (or logit) maximum likelihood estimators for these discrete choice models have been criticized as implying some people would not support desirable programs even if they had no cost (i.e. lead to estimates of WTP that are less than zero), following the work of Carson et. al. [1992] we have adopted an alternative formulation based on a Weibull survival model that constrains t* to be a positive random variable. The survival function, S(x), can be interpreted as a reduced form description of the probability that an individuals WTP is at least as great as x (i.e.  EMBED Equation.2 , where F()= cumulative distribution function for x). To implement the model, with the censoring implied by our referendum model, S(x) is used to describe the probability of for and 1-S(x) the against votes. The hazard rate, l(x) in this function (roughly the rate of observing WTPs greater than x, which can be treated as a demand measure for the program), is specified to be a function of individual characteristics. With a Weibull function describing this process, the survival and hazard rate models are given in equations (4) and (5):  EMBED Equation.2 (4) where l, q are parameters  EMBED Equation.2 (5)where b is a 1xK parameter vector to be estimated and zi is a Kx1 vector of values of independent variables for the i-th respondent. The median WTP,  EMBED Equation.2 , can be expressed in terms of these parameters as:  EMBED Equation.2 (6) As we noted earlier, the use of the telephone-mail-telephone survey to provide respondents the information necessary to answer our CV question introduces the prospects for selection effects. Incorporating these effects requires that we modify the hazard rate model to include another hazard rate (i.e. Heckmans [1979] inverse Mills ratio, IMR) that takes account of respondents rate of participation in the second telephone interview. This relationship can be estimated from the information included in the first round, RDD telephone interviews. Following Heckmans argument for the case of linear models, this modified specification offers an approximate method to take account of selection effects. Unlike Heckman's two-step method (or Poirier's [1980] bivariate probit), our proposed strategy is ad hoc. It treats the inverse Mills ratio as a gauge of the importance of differences between respondents and nonrespondents in our Phase II sample. Because the analysis involves combining two different distributions (i.e. Weibull for stated choices and normal for the selection model), we did not attempt to adjust standard errors for the heteroscedasticity induced in the usual selection model. We did test for the extent to which the Weibull model hazard rate displayed heteroscedasticity that could be related to our estimates of the inverse Mills ratio. This composite strategy was selected because it allows the WTP assumed to underlie stated choices to be constrained to the positive domain and offered a simple way to gauge the potential influence our selection effects. D. Empirical Findings Equation (7) provides the probit estimate for our selection model. Income, demographic, and attitudinal variables influenced respondents willingness to participate in a second telephone interview described as being about the quality of beaches and coastal areas. Prob (Completing Phase II Interview) = 0.29 + 0.004 Income (in Thousands) (1.89) (3.33) + 0.17 Environmentalist (=1) - 0.01 Age (in years) - 0.28 Below High (2.17) (-6.07) (-1.34) School Education (=1) + 0.29 Attitude toward Dumping + 0.18 Attitude toward Limiting Coastal (2.94) in Ocean (=1) (2.58) Development (=1) + 0.20 Attitude toward Coastal Pollution (=1) (7) (2.60) n=1671  EMBED Equation.2 =0.057 The number below each estimated parameter is the estimated asymptotic Z-statistic for the null hypothesis of no association. The attitude variables are all qualitative variables defined from questions on the Phase I survey. To be coded as a 1 the respondent rated reducing or restricting activities related to the identified topic in each variable name related to an attitude as very important. A code of 0 corresponds to those indicating it was important, not at all important or not sure. The variable, environmentalist, is also a qualitative variable indicating whether respondents suggested that being concerned about the environment described them very well. While these results suggest that the survey method appears to have differentially attracted higher income respondents with interests in coastal issues, this does not automatically imply selection effects will be an important factor in describing respondents choices about debris control and cleanup programs. To investigate this issue, we estimated the Weibull survival model in several ways: (1) using separate sub-samples corresponding to the groups receiving each of the four different sets of photographs depicting the default beach conditions (i.e. without debris control and cleanup program); (2) using a single model for the full sample with two different specifications for the debris conditions; and (3) with a subset of these models for the New Jersey sample. Table 4 reports the first set of results from the analysis of the four sub-samples along with the estimated median WTP derived from each sub-samples model. These models do not attempt to take account of respondents' characteristics and consider only the potential impact of the selection effects as reflected through the Heckman IMR term. The estimates of median WTP are of particular interest given our objective of evaluating whether independent respondents could discriminate between debris conditions and whether these distinctions affected their stated choices. These results support the implications of the Cochran-Mantel-Haenszel test as well as the focus group analysis in that the relative size of the estimated median WTP for control and cleanup by independent samples align with the judgments of focus group respondents about the severity of the debris conditions represented in these photos (i.e. D is preferred to B and B to C and C to A). Given these differences in their evaluations of beach conditions, we would expect the WTP for the cleanup plan to be lowest with D, next lowest with B, then C and largest with A. This is what the estimated medians suggest. Moreover, using a likelihood ratio test to test whether the parameters of the hazard functions describing respondents' stated choices for each of the four default beach scenes were equal, we reject the null hypothesis with a p-value = .05 ((2(3) = 9.33). Thus, at this simple level with somewhat more structure imposed in describing respondents' stated choices for the debris control programs, we find stronger evidence of differences in these choices with different amounts of debris. Nonetheless, it is important to acknowledge the significant caveats that condition our results. First, our samples are small and our ability to take account of potential selection effects is limited. The 95% confidence intervals for the conditional medians do overlap, suggesting our estimates offer rather limited resolution. Even with these qualifications, they do offer a remarkably consistent picture of respondents distinguishing beach conditions similarly to the participants in the focus groups. Table 5 reports multivariable specifications for the Weibull hazard rate model that include the effects of debris in two different ways--using qualitative variables for each photo in comparison to photo D (the omitted category) and the probability computed from the focus group rank logit model (designated as PR). The results suggest that with further parametric assumptions, it is possible to strengthen the support for the conclusion that respondents do discriminate between debris conditions. After accounting for income, attitudes toward the environment, and selection effects, these models indicate that differences in the debris conditions were important to respondents' stated choices. The coefficients of Photo A through Photo C compare to the omitted scene (the best conditions of the four, Photo D, with only natural debris). These estimates suggest we can estimate a significant (p-value = 0.06) discrimination relevant to peoples choices for the worst versus the best conditions with these sample sizes. Income is generally not a significant determinant of the hazard rate and sometimes has a negative effect. This finding likely arises because the effect of income is already captured to some degree through the selection effect terms, indicated as the IMR effect. A comparison of equations (1), (2), and (3) in Table 5 seems to suggest this interpretation. Another method for testing the effect of differences in the baseline debris conditions on the support for the control and cleanup plan calls for use of the computed probability of ranking a beach scene as best quality (PR) using the rank logit model. In this use, high probabilities provide smaller incentives to support the plan and thus, we would expect a negative coefficient. This formulation provides a slight improvement in the ability to isolate an effect of debris conditions. The estimated coefficients should be interpreted as comparable to an elasticity - the proportionate change in median WTP with a proportionate change in the relevant independent variable. The other models in Table 5 include several different attitude measures for the importance of coastal resources and a qualitative variable indicating whether a respondent owned vacation property. This later term could reflect wealth effects as well as indicate that a respondent might have a particular reason for being impacted by the proposed situation. We also considered both types of models (i.e. with the dummy variables for debris conditions and with the estimated PR measure) for just the New Jersey sample. The last column in Table 5 reports the estimates for one such model. Here the reduced sample size limits the discriminatory power of the model, but the estimates are consistent for both treatments of debris with the full sample results and we report only the model with the PR measure. Our analysis considered both homoscedastic and heteroscedastic specifications for the errors variance in the hazard models as they might be related to the IMR. Results are reported for the heteroscedastic model if the models estimates implied the homoscedastic form would be rejected. Otherwise the homoscedastic form was reported. Even with the most minimal modeling assumptions, we find evidence of consistent economic behavior. Independent samples responded to debris conditions in ways consistent with the focus groups. Respondents were more concerned about more serious than less serious debris conditions. These findings are potentially important because they are based on analyses that add little prior information to the tests of consistent discrimination in quality and relate to small samples. The pilot nature of our survey suggests that one be prudent in interpreting them more generally. They offer systematic evidence that stated choices for aesthetic dimensions of coastal quality follow an economic framework. Specific estimates of economic values for programs intended to avoid deterioration or to improve conditions will require larger samples and simpler, more focused designs that do not attempt to consider as many issues simultaneously. V. Implications This study has three potentially important findings. First, our results illustrate that it seems possible to design qualitative research in a way that facilitates quantitative consistency checks of the insights provided by this type of development research as part of contingent valuation surveys. Our use of ranking tasks together with the Beggs, Cardell, and Hausman framework to model these types of data offered an example of how these consistency checks might proceed in a case where the objective was understanding how the features of a beach scene relate to perceptions of quality. Second, our survey offers some evidence that stated choices about some aspects of the quality of some environmental resources give rise to measures of economic values consistent with what economic theory would expect. This may seem trivial but the expectations derived from theory largely relate to cases where it is readily possible to agree on what constitutes different amounts of the object of choice. As we consider an aesthetic or quality dimension of environmental resources, such agreement across people is not as clear-cut. Our findings suggest that there may be methods for dealing with these types of problems. Finally, a program to control and cleanup marine debris yields improvements in an environmental resource, people care about and appear willing to commit resources to pay for. The results should be considered as the findings of a detailed pilot survey intended to investigate whether measures of the economic values of programs to control and cleanup marine debris could be developed. In order to be able to use the data available, we pooled respondents across areas (i.e. those relevant to NC and NJ beaches). This process assumes the differences in local beach conditions are not important to respondents abilities to deal with the particular beach scenes shown in the photos. We also assumed that analysis of this question could proceed independently of the other valuation question on our survey. With these caveats, the findings do offer reasonably clearcut evidence that independent respondents discriminate different amounts of debris consistently, in terms of identifying the extreme conditions. When considered together with the focus group results, they provide strong support for peoples concern for programs to control and cleanup marine debris as well as for the prospects for measuring these programs economic values. Table 1: Description of Focus Groups Focus GroupDateLocationNumber of ParticipantsCompositionSelection EffectGoals1Aug. 27, 1991Raleigh, N.C.96 female; age was 27-65; 8 married; 2 owned beach propertyRecruited using listed phone numbersEvaluate ranking tasks; gauge importance of beach scene; experience with debris; N.C. beaches versus other areas; use and passive use CV questions2Oct. 28, 1991Clark, N.J.127 female; age was 21-70; greater use of beach than FG1; 1 owned beach property; 8 marriedRecruited by marketing research firmEvaluate refined ranking task; refined CV questions; evaluation of payment method; questions to be addressed in booklet; Is referendum reasonable?3March 2, 1992Asheville, N.C.93 female; age was 33-65; 3 couples, i.e. husband & wife, participated; 9 married; no-one owned beach property); greater attachment to South Carolina beachesRecruited by church; requested random group of members but church used Social Concerns Committee insteadGoals comparable to FG2 on ranking and CV; used picture board and scenes with CV to mimic insert; payment method; evaluation of referendum Table 2: Rank Logit Models for Photos with Marine Debris a Independent VariablesbFocus Group TwocFocus Group ThreecCombined Focus GroupscDensity of Natural Debris (?)-0.04 (-1.81)-0.05 (-1.93)-0.05 (-2.70)Density of Manmade Debris (-)-0.24 (-8.11)-0.23 (-6.53)-0.23 (-10.31)Unique Debris (?)-1.78 (-3.81)-2.96 (-5.25)-2.19 (-6.13)Prominence of Dead Marine Life (-)-0.18 (-5.71)-0.21 (-5.32)-0.19 (-7.80)Perceived Water Quality (+)0.36 (10.04)0.56 (12.55)0.42 (15.33)Presence of Structures (?)1.02 (2.70)0.76 (1.71)0.88 (3.07)Presence of People (-)-1.14 (-2.92)-2.39 (-5.09)-1.61 (-5.36)Log (L)186.57129.51318.78 a These models were derived using the rankings of twelve color photos. Five of these photos were used in the CV questions to describe the four different baselines with debris and the clean beach that would be available with cleanup and debris control. The photos used for these focus groups were: Photo Letter or Focus GroupDescriptionRole in Final SurveyALittered beach with unique metal debris in backgroundBBClean beach with a hotel in the backgroundwith cleanup and control programCPanorama of beach covered with kelp and plastic bottlesDDense kelp with birds nearbyDEA family sitting on a beach littered with debrisFLarge amount of plastic on beachGSeaweed and a yellow barrelHScene with what appears as outline of shipwreckIManmade debris on a beach near private beach propertyJBeach with prominent dead marine specieKA mixture of manmade trash with a tentALBeach with wood from hurricaneA*Three people with binoculars standing in kelp and plastic bottles (photo from second set of photos used in first focus group; this was not used in the rank logit model given above.C b The signs in parentheses after the names of each variable indicate the hypothesized direction of effect based on the interpretation of each index. c The numbers in parentheses below the coefficients are asymptotic Z-statistics for the null hypothesis of no association. Table 3: Cross Tabulations for Choices of Plans to Control and Cleanup Marine Debris a Photo of Debris ConditionsSampleA (PR = .077)B (PR = .273)C (PR = .117)D (PR = .533)Full Sample EMBED Equation.2 (p-value=0.337)Against Control and Cleanup Program63 (56.8)47 (58.8)60 (60.0)47 (70.2)For Program48 (43.2)33 (41.2)40 (40.0)20 (29.8)North Carolina Sample EMBED Equation.2 (p-value=0.975)Against Control and Cleanup Program41 (62.1)26 (59.1)34 (63.0)21 (63.6)For Program25 (37.9)18 (40.9)20 (37.0)12 (36.4)New Jersey Sample EMBED Equation.2 (p-value=0.097)Against Control and Cleanup Program22 (48.9)21 (58.3)26 (56.5)26 (76.5)For Program23 (51.1)15 (41.7)20 (43.5)8 (23.5) a The numbers in parentheses below the sample counts in each contingency table is the column percentage (i.e. between against and for the program). The PR column heading below each designates the estimated probability each photo would be ranked best. Table 4: Estimated WTP and Survival Models for Program to Control and Cleanup Marine Debris a Hazard RateHeteroscedasticity Adjustment bEstimated Median WTPSub-SampleSample SizeInterceptIMR EMBED Equation.2  cInterceptIMR[95% Confidence Interval]Photo A1088.49 (8.32)-7.06 (-2.99)-0.52 (-1.08)2.37 (2.66)$72.18 [34.20 - 152.32]Photo B758.05 (3.25)-6.78 (-1.35)0.1011.15 (2.84)$40.97 [11.72 - 143.23]Photo C997.32 (6.88)-4.83 (-2.38)0.0540.79 (3.61)$63.21 [29.59 - 135.04]Photo D667.34 (5.09)-6.59 (-2.35)0.0700.96 (3.17)$21.38 [5.71 - 80.02] a These estimates are for a Weibull survival model specified to be a function of the Inverse Mills ratio (IMR), or hazard rate, derived from the probit model reported in equation (7). Observations from the New Jersey and North Carolina samples were pooled due to the small sample. Tests for the effects of pooling indicated that a qualitative variable for the sample was insignificant. The estimated medians are derived from the parameters of the Weibull models in each row. b The estimates of the Weibull allow for heteroscedasticity due to the estimate of the IMR. The rationale follows Heckman [1979], but implementation assumes  EMBED Equation.2 . This is incorporated when  EMBED Equation.2  is judged significantly different from zero. c Pseudo-R2. d The confidence interval for the mean is derived using a first order Taylor series expansion to estimate the asymptotic variance of the estimated median. Table 5: Multivariate Survival Models for Program to Control and Cleanup Marine Debrisa Independent VariablesFull SampleNew Jersey Onlyln (l)Intercept7.40** (7.18)3.67** (6.52)7.44** (7.33)8.49** (8.56)8.18** (5.42)Photo A (=1)1.02+ (1.86)1.21* (2.10)1.00+ (1.89)Photo B (=1)0.37 (0.63)0.67 (1.08)0.46 (0.81)Photo C (=1)0.66 (1.21)0.75 (1.30)0.70 (1.32)PR-----2.02+ (-1.91)-2.99+ (-1.76)Incomeb (in thousands)-.002 (-0.26)0.01 (1.50)-.002 (-0.27)-.002 (-0.26)0.001 (0.11)Own Vacation Home (=1)-----1.29+ (-1.75)-1.32+ (-1.80)-1.57 (-1.17)Coastal Concern Attitude (=1)----0.72 (1.35)0.72 (1.36)0.03 (0.04)Inverse Mills Ratio-6.50** (-3.65)---6.70** (-3.74)-6.71** (-3.75)-5.55* (-2.43)HeteroscedasticityIntercept0.26 (0.89)0.97** (7.21)0.24 (0.83)0.24 (0.86)0.85** (4.47)Inverse Mills Ratio1.33* (2.41)--1.26* (2.32)1.26 (2.36)--Log (L)-189.75-221.06-187.56-187.67-80.77Sample Size348348348348154 a The numbers in parentheses below the coefficients are asymptotic Z-statistics for the null hypothesis of no association. b Not all households who responded to the Phase II interview reported their incomes. We used predicted income for these missing values. These values for households with missing income were estimated using an equation based on the phase I sample. The model used to develop these estimates for the missing values is given as follows: log (income) = 9.759 + 0.236 Full Employed (= 1) (109.38) (4.79) - 0.441 Retired (= 1) + 0.610 Education below (= 1) (-5.32) (-4.28) high school + 0.521 Education with at (= 1) + 0.159 Male (= 1) (12.12) least college degree (3.90) + 0.218 White (= 1) + 0.004 Age (in years) (4.13) (2.31) R2 = 0.255 n = 1249 ** reject null hypothesis with p-value at least 0.01 * reject null hypothesis with p-value at least 0.05 + reject null hypothesis with p-value at least 0.10 APPENDIX 1 Text of CV Question BLOCK A (Questions 7 to 10) [Randomly order with Block B. Keep track of order.] In the next part of this interview, consider the beaches in the page labeled "Alternative Coastal Conditions" and how you would feel about conditions similar to these for beaches in [North Carolina, New Jersey]. 7. I would like to ask you about a proposal for controlling debris on [North Carolina, New Jersey] beaches involving greater enforcement of the laws prohibiting waste disposal in the ocean and periodic clean-ups of the beaches. Without these efforts, we can expect that ocean disposal of waste would make conditions on [North Carolina, New Jersey] beaches resemble Photo A [some, most: Alternate and keep track.] of the time. This program involves [emphasize these terms] both enforcement and cleanup activities financed by those states whose citizens are likely to use [North Carolina, New Jersey] beaches. It would improve beach conditions to those depicted in Photo B. The proposal would involve: [Alternate which is selected and keep track.] I. activities financed by an annual surcharge added to each household's state income tax. This means each household's annual taxes would increase by _______* dollars per year. *[Instructions: There are two sets of values to be used with each sample. Randomize which value is selected and keep track.] North Carolina Sample--Use for respondents in: NC, Va/MD: [5, 10, 25, 40, 60, 100, 150, 350, 450, 750, 1,000] Midwest, PA, and NJ: [5, 10, 25, 40, 60, 100, 150, 350, 450] New Jersey Sample--Use for entire sample: [5, 10, 25, 40, 60, 100, 150, 350, 450, 750, 1,000] II. activities financed by a statewide beach use fee. The beach use fee would be [$.25, $.50, $.75, $1.50, $1.75, $2.00, $2.50, $4.50, $6, $7.50, $10, $12.50, $15] per person [Randomize which value is selected and keep track.] and would be for each time a person visited a beach anywhere in [North Carolina, New Jersey]. It would be in addition to any other local charges or state park fees. The program's funds would be used exclusively for controlling debris on beaches so conditions would be improved to the situation in Photo B. Keeping in mind your household income and current expenditures together with these new costs, if this proposal were placed on the ballot this fall as a referendum, would you vote for it? 01 Yes [Go to question 8.] _863019159F==Ole PIC LMETA (Symbol-2 | )Symbol-2 (Symbol-2  )-XSymbol-2 |D(Symbol-2 |)Symbol-2 (Symbol-2 )Times New Roman+- 2 `V 2 `y 2 `tk 2 `P 2 `C 2 ` q 2 `7V 2 `y 2 `JP 2 `C 2 `9qSymbol- 2 `-Symbol- 2 or> 2 r<Times New Roman+- 2 `H,` 2 `,` 2 `z,` 2 `@,` Times New Roman,- 2  1p 2 0p & "System-on &Topࡱ;  FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; A@d @ Vy-t,CompObj ZObjInfo Equation Native _891330413J F==P,Cq 1 ()() e  e>  e< Vy,P,Cq 0 ()() Fࡱ; ࡱ; L kdࡱ;  Z  .02 No [Go to question 9.] 03 Don't Know [Go to question 10.] References Arrow, Kenneth J., 1958. Utilities, Attitudes and Choices: A Review Note. Econometrica, 26 (January): 1-23. Arrow, K., R. Solow, E. E. Leamer, R. Radner and H. Schuman, 1994. Comment on NOAA Proposed Rule on Natural Resource Damage Assessments. ANPNM, Comment No. 69, January 7. Arrow, K., R. Solow, P. R. Portney, E. E. Leamer, R. Radner and H. Schuman, 1993. Report of the NOAA Panel on Contingent Valuation. Federal Register, January 15, Vol. 58, No. 10, 4601-4614. Beggs, S., S. Cardell and J. A. Hausman, 1981. Assessing the Potential Demand for Electric Cars. Journal of Econometrics, Vol. 16 (September): 1-19. Cameron, Trudy Ann, 1988. A New Paradigm for Valuing Non-Market Goods Using Referendum Data. Journal of Environmental Economics and Management, Vol. 15 (September): 355-379. Carson, Richard T., W. Michael Hanemann, Raymond J. Kopp, Jon A. Krosnick, Robert Cameron Mitchell, Stanley Presser, Paul A. Ruud, and V. Kerry Smith with Michael Conaway and Kerry Martin, 1996. Referendum Design and Contingent Valuation: The NOAA Panel's No-Vote Recommendation. Unpublished paper, Center for Environmental and Resource Economics, Duke University, March. Carson, Richard T., W. Michael Hanemann, Raymond J. Kopp, Jon A. Krosnick, Robert C. Mitchell, Stanley Presser, Paul A. Ruud and V. Kerry Smith, 1994. Prospective Interim Lost Use Value Due to DDT and PCB Contamination in the Southern California Bight. Report to National Oceanic and Atmospheric Administration, Natural Resource Damage Assessment, Inc., La Jolla, Ca., September. Carson, R. T., R. C. Mitchell, W. M. Hanemann, R. J. Kopp, S. Presser and P. A. Ruud, 1992. A Contingent Valuation Study of Lost Passive Use Values Resulting from the Exxon Valdez Oil Spill, Report to the Attorney General of the State of Alaska, Natural Resource Damage Assessment, Inc., La Jolla, Ca., November 10. Cole, C. Andrew, William P. Gregg, Daniel V. Richards, and David A. Manski, 1992. Annual Report of National Park Marine Debris Monitoring Program. (Washington, D.C.: National Park Service, U.S. Dept. of the Interior, July). Diamond, Peter A. and Jerry A. Hausman, 1994. Contingent Valuation: Is Some Number Better Than No Number. Journal of Economic Perspectives, Vol. 8 (Fall): 45-64. Freeman, A. Myrick III, 1993. The Measurement of Environmental and Resource Values: Theory and Methods. (Washington, D.C.: Resources for the Future). Hanemann, W. Michael, 1984. Welfare Evaluations in Contingent Valuation Experiments With Discrete Responses. American Journal of Agriculture Economics, Vol. 66 (August): 332-341. , 1994. Valuing the Environment through Contingent Valuation. Journal of Economic Perspectives, Vol. 8 (Fall): 19-44. Heckman, James, 1979. Sample Bias as a Specification Error. Econometrica, Vol. 47 (January): 153-162. Kreps, David M., 1990. A Course in Microeconomic Theory. (Princeton, N.J.: Princeton University Press). McConnell, K. E., 1990. Models for Referendum Data: The Structure of Discrete Choice Models for Contingent Valuation. Journal of Environmental Economics and Management (January): 19-34. McFadden, Daniel, 1974. Conditional Logit Analysis of Qualitative Choice Behavior, Frontiers in Econometrics, edited by P. Zarembka. (New York: Academic Press): 105-142. Poirier, Dale, 1980. Partial Observability in Bivariate Probit Models. Journal of Econometrics, Vol. 12: 209-217. Portney, Paul R., 1994. The Contingent Valuation Debate: Why Economists Should Care. Journal of Economic Perspectives, Vol. 8 (Fall): 3-18. Smith, V. Kerry, 1993, NonMarket Valuation of Environmental Resources: An Interpretive Appraisal, Land Economics, Vol. 69, (February): 1-26. Snedecor, G. W. and W. G. Cochran, 1980, Statistical Methods, 7th edition (Ames, Iowa: Iowa State University Press). Zhang, Xiaolong, 1995. Integrating Resource Types, Payment Methods, and Access Conditions to Model Use and Nonuse Values: Valuing Marine Debris Control. (Raleigh, NC: North Carolina State University, unpublished Ph.D. thesis). Zhang, Xiaolong, V. Kerry Smith, and Raymond B. Palmquist, 1992a. Marine Debris Focus Group I: Summary Report I, Resource and Environmental Economics Program, North Carolina State University. Unpublished paper, January 27. , 1992b. Marine Debris Focus Group II: Summary Report, Resource and Environmental Economics Program, North Carolina State University. Unpublished paper, January 27. , 1993. Marine Debris Focus Group III: Summary Report, Resource and Environmental Economics Program, North Carolina State University. Unpublished paper, June 17. Footnotes PAGE  PAGE 31 * Arts and Sciences Professor of Environmental Economics, Duke University and Resources for the Future University Fellow; Economist, AT&T; and Professor, Department of Economics, North Carolina State University respectively. Thanks are due to Dr. Tony Amos for providing access to all of his photos, to two anonymous referees for constructive comments on an earlier draft, to Laura Osborne and Kurt Schwabe for research assistance and to Drake Paul, Brent McLamb, Danny Kim, Anna Park, and Paula Rubio for preparing several versions of this paper. Partial support for the research was provided by NOAA (award number NA90AA-D-SG847) and the UNC Sea Grant College under project R/MRD-25.  For example, the cover story of Time in 1988 (August 1, 1988), The Filthy Seas featured marine debris and its impact on beaches in New Jersey as well as elsewhere in the U.S. About the same time comparable front page stories appeared in Newsweek under the heading Dont Go Near the Water, August 1, 1988 and Business Week as Troubled Waters, October 12, 1987.  The logic underlying this approach for interpreting discrete contingent valuation questions (as well as any individual level choice data) is outlined in Carson et al.[1994]. It is not a new framework in economic analysis and an early discussion of its importance to interpreting actual and stated choices was offered in a review article by Arrow [1958]. Thirty-five years ago, he observed that: ...in both economics and in the other behavioral sciences, the choices studied may be actual or potential. In economics, the study of actual choices would be exemplified by statistical analysis on family budget data. Potential choices would perhaps be exemplified by questionnaires which seek to establish how much an individual would buy, were conditions something other than they are now. The latter type of analysis has had little actual use in empirical economics, though one might say it is the chief content of the pure theory of demand. ...The relation between actual choices and potential ones is, of course, rather tricky. One can easily see all sorts of reasons why the response to a question, always posing a potential choice, may be different from a choice made in the actual situation. Undoubtedly in-depth interviews and questionnaires can go far in eliminating inconsistencies by elaborate sets of questions. (pp 3-4)  For an overview of the issues see the special section in the 1994 Journal of Economic Perspectives with articles by Portney [1994], Diamond and Hausman [1994] and Hanemann [1994].  The NOAA Panel [1993] report included a set of guidelines for CV studies used in litigation, recommendations for research, and a general conclusion that: ...CV studies can produce estimates reliable enough to be the starting point of a judicial process of damage assessment, including lost passive use values (p. 4610).  The structure of the model used for a rank logit is the same as the categorical logit or random utility model proposed by McFadden [1974]. It relies on modeling the probability of a choice of each of K items for those applications and a ranking in this case. It also requires the independence of irrelevant alternative (IIA) assumption to simplify the specification of probabilities for rankings. In the case of a ranking, the IIA property implies that the ordering of any pair of alternatives is independent of the other possible alternatives that might be considered. The probability of orderings are specified to be functions of characteristics of the alternatives. If the characteristic does not change across alternatives, then it will not be possible to estimate its effect on the ordering. In the first focus group, the first set of photos to be ranked did not include people or structures in a prominent role.  To enhance response rates each respondent was told that he had a chance of winning one of four $50 awards if they completed the second phase interview and mailed back the questionnaire about recreational use of beaches.  Passive use value is the term used by the Court of Appeals decision in the Ohio case to refer to non-use values. For a discussion of the conceptual definitions for use and non-use values see Smith [1993] and Freeman [1993].  Prior to asking the questions about the control and cleanup program, the second telephone interview assured the interview was with the same individual (the name had been collected in the first interview), confirmed the mailing had been received and the booklet read (if not, another interview time was scheduled), and asked about experience with conditions like those described by the photos. Following the CV question follow-up questions asked for open ended responses explaining reasons for each individuals choices. In contrast with the experience with in-person interviews, the responses to these questions were not informative. Few respondents offered any explanation for their answers.  These judgments were based on the Public Area Recreation Visitors Surveys undertaken for NOAA for Island Beach State Park in New Jersey and the Cape Hatteras National Seashore in North Carolina for one recreational season. The sample distribution for the New Jersey and North Carolina samples is given as follows: Sample RegionPlanned Phase I Sample SizeRealized Phase I Sample SizeRealized Phase II Sample SizeNew Jersey Sample New Jersey575577216 Philadelphia & Rural Penn.15015157 New York City Area10010234 Delaware505018Subtotal875880325North Carolina Sample North Carolina438445194 Eastern Virginia & Maryland879339 Midwest States17517960 Pennsylvania & New Jersey17517675Subtotal875893368Total17501773693  This conclusion recognizes that the overall response rate is the product of the initial response 52.6% and the completion rate for the second survey, 39.1% or 20.9%, dramatically below the 70% rate recommended by NOAA Panel.  See Zhang [1995] for an investigation of the relationship between the responses to the two CV questions.  Carson et. al. [1996] found that treating don't know and would-not-vote responses as votes (or choices) against a CV plan could not be rejected as the explanation for how respondents evaluated these types of responses.  This framework was first proposed for describing CV responses by Hanemann [1984]. More recent papers have considered whether a specification of preference functions is necessary to develop welfare measures, see Cameron [1988] and McConnell [1990].  These estimates apply the Weibull model to each sub-sample and to the combined sample without heteroscedastic adjustment. Using models for all sub-samples with the IMR variable in a heteroscedastic adjustment for the scale parameter leads to the same conclusion. The likelihood ratio test statistic would be -2(ln LR - ln LuR) = (2(5) = 13.8 which also reflects the null hypothesis at a p-value = .05. These estimates are based on asymptotic approximations to derive estimates for the asymptotic variance for this conditional median.  Based on further comments submitted by four of the five members of the NOAA Panel, these results may well be stronger evidence of a scope effect than the statistical tests would imply. Arrow et al.[1994] wrote that in small samples no effects would be statistically significant. Instead, they note what is important is the degree to which the response of WTP is plausibly sensitive to the change in the environmental resource. They note: A survey instrument [for contingent valuation] is judged unreliable if it yields estimates which are implausibly unresponsive to the scope of the result. This, of course, is a judgment call, and cannot be tested in a context-free manner... There would be settings in which estimates made with plentiful observations are statistically significant to scope, but at the same time are implausibly insensitive. Also, if the sample size is small and the scope difference minor, the estimates may be statistically insensitive to the scope, yet plausibly sensitive. Thus, our ability to distinguish the extreme conditions is consistent with their intuition.  =/anࡱ; |Ole,1/10/1 ,1/10/11414&1/1 h0/12424$PIC  %LMETA 'hCompObj=Z1  &` & MathType-ooTimes New Roman- 2 `@Prob 2 `D( V` 2 `( y,`` 2 ` P,`` 2 `  A` 2 ` ) ` 2 ` > 2 `Q V` 2 `( y,`` 2 `Q P,`` 2 ` A`2 `@ ) ) ``` 2 `= 2 @ ` 2 <exp 2  ( v`2   ( y,``` 2 R P,`` 2   A` 2 R))2 exp( v` 2 [( y,`` 2  P,`` 2 b  A` 2  ) ) `` 2  + 2  exp` 2 ( v` 2 ( y,`` 2  P,`` 2  A` 2 -) )` Times New RomanU- 2 "i> 2  1p 2 `  8 2 i> 2 2p 2 @? 1p 2 @1p 2  8 2 Y1p 2 Y 1p 2 \  8 2 YH1p 2 Yk2p 2  8 &  "System->cCCc>ࡱ FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; ObjInfo,1/10/136D36&3/1p/1/1P46P46?Equation Native @_864465982.<FooOle HR?Rx ?R Prob( V i ( y, P, A 1 ) > V i ( y, P, A 2  ) ) = exp( v 1  ( y, P, A 1 ))exp( v 1 ( y, P, A 1 ) ) + exp( v 1 ( y, P, A 2 ) )b w wࡱ; L$ࡱ; $~G  .1  @&s & MathTypeSymbol-2  (Symbol-2 > )Symbol-2 PIC ,1/10/1(68d(68&06/1p/1-/1 6886868ILMETA KhCompObjaZObjInfoc(Symbol-2 + )Symbol-2 (Symbol-2 )Symbol-2 ?(Symbol-2 )Times New Roman0- 2 V 2 y 2 tk 2 P 2 GC 2  q 2 V 2 Xy 2 P 2 C 2 qSymbol- 2 - 2 = Times New Roman0- 2 H*pTimes New Roman- 2 ,` 2 ,` 2 ,` 2 ,` Times New Roman0- 2 G 1p 2 0p & "System-\Hࡱ FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; 21d 1 Vy-t * ,P,Cq 1 ()()=Vy,P,Cq 0 ()()Equation Nativel88 l88 d_863341770 FooOle hPIC iLࡱ; LA| ࡱ; A  ) .1  `& s & MathTypeSymbol-2 (Symbol-2 f)Symbol-2 IMETA,1/10/1:4=:4=&:/1p/1-/1ExSt 4=kCompObj~ZObjInfoEquation Native (Symbol-2 1 )Symbol-2 D(Symbol-2 )Times New RomanP.- 2 @S 2 x 2 tk 2 v x 2 BF 2 xSymbol- 2 <= 2 5  2  = 2  -Times New Roman- 2 zProb~ Times New RomanP.- 2 X*pTimes New Roman- 2 ! 1 & "System- q 2 P C 2 qࡱ;  FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; ̀44p$ Sx()=Probt * x()=1-Fx() &ࡱ; ࡱ; L ࡱ;  .9  ._863342500"Fo<Ole PIC !$LMETA 1   & n & MathType`Symbol-2 (Symbol-2 n)Symbol-2 Td(Symbol-2 T)Times New Roman~ - 2 @S 2 x 2  e Times New Roman- 2 D;xbTimes New Roman~ - 2  `` 2 ` ``Symbol- 2 "= Symbol- 2 D-{ Symbol- 2 Dl{`Symbol- 2  qS & "System-ࡱ; CompObj#%ZObjInfo&Equation Native |_863342753 )F<< FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; `4$$* Sx()  =  e -lx() q2 ࡱ; L!{hࡱ; Ole,1/10/1 ,1/10/11414&1/1 h0/12424PIC (+LMETA HCompObj*,Z!{V   .1  @`&  & MathTypePSymbol- 2 7l Symbol- 2 b{ Times New Roman - 2 'i> 2 pzWTimes New Roman- 2 e`Times New Roman - 2 ,i,Times New Roman- 2  `` 2  ``Symbol- 2 = Symbol- 2 c-{ & "System-Fࡱ;  FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ObjInfo,1/10/1(68d(68&06/1p/1-/1-6886868Equation Native |_86334312750F<<aZOlenfo ࡱ; `4$L* l i   =  e -bz i`ࡱ; Lࡱ; .H  .1  `& x &PIC /2LMETA CompObj13ZObjInfo4 MathType Times New Roman- 2 @~Times New Roman- 2 Dtk & "System-3 t =MuW ࡱ;  FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; Equation Native F<_8633431707F<i¶Ole PIC 69L 4t&l* 2t ࡱ; L ࡱ;  ^I - .1  ` &  & MathType`Symbol-2 $(Symbol-2 META CompObj8:ZObjInfo;Equation Native )-? ? Times New Roman`h- 2 @~ 2 lnkTimes New Roman- 2 Ftk 2 7e Times New Roman`h- 2 mzW`Times New Roman- 2 ,i,Times New Roman`h- 2  `` 2 w ``Symbol- 2 2= Symbol- 2 b{ 2 ( qtTimes New Roman- 2 32 Times New Roman`h- 2  1p & "System-H45&1+ࡱ;  FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; ̀2'1,)1 2t  =  e bz i ln2() 1qtk 2 7ࡱ; ࡱ; L4@,ࡱ; _864471043>Fi¶i¶Ole PIC =@LMETA 4>H  .1  & & MathType Times New Roman- 2 s~Times New Roman- 2 HR Times New Roman- 2 W2p & "System-&2H5ࡱ; CompObj0?AFZObjInfoBEquation Native <_901622758fEFi¶Ķ FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ;  4t&* 2R 2ࡱ; Lq8ࡱ; q7 Z .Ole PICObj DGLMETAfo CompObjNativeFHZ1  &4 & MathTypepSymbol- 2 Lc Times New Roman+- 2 I(I 2 )ITimes New Roman&- 2 .` Times New Roman- 2 3p 2 I2pTimes New Roman+- 2 3 2  38Symbol- 2 = & "System-FF(Fࡱ FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; @w54d 4 c (3)2 =3.38Times Neࡱ; ObjInfo,1/10/1(68d(68&06/1p/1-/1I6886868Equation Native \_901622757Q_LFĶĶaZOlenfo       !"$'*,-./01249;<=>?@ABCDEFGILMPRSTUVWXZ^_`abcdefghࡱ; LLࡱ;   Z .1  &4 & MathTypepSymbol- 2 Lc Times New Roman- 2 I(I 2 )ITiPIC KNLMETA CompObjMOZObjInfoPmes New Roman- 2 .` Times New Roman- 2 3p 2 I2pTimes New Roman- 2  0 2 22Symbol- 2 = & "System-~|ࡱ FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; @w5L444 c (3)2 =0.22-ࡱ; ࡱ; LLࡱ; F  Z .Equation Native \_901622756SFĶĶOleion Native PIC42753 RUFLMETAfo,1/10/136D36&3/1p/1 /1P46P46CompObjNativeTV#ZObjInfo2WF%Equation Native &\1  &4 & MathTypepSymbol- 2 Lc Times New Roman- 2 I(I 2 )ITimes New Roman- 2 .` Times New Roman- 2 3p 2 I2pTimes New Roman- 2  6 2 32Symbol- 2 = & "System-~|ࡱ FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; @w54,4 c (3)2 =6.32-ࡱ; _901623162ZFĶ3ŶOle (PIC Y\)LMETA +ࡱ; LWT,ࡱ; Wf@  .1   & & MathType Times New Roman%- 2 u~Times New Roman- 2 LR Times New Roman%- 2 f2p & "System-: oFVv$'ࡱ;  FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ;  4&* 2R 2ࡱ; CompObj0[]F3ZObjInfo^5Equation Native 6<_901623161CXaF3Ŷ3ŶOleEntry  F7PICObj `c8LMETAcument :hCompObjlbdHZ@ LB48 @ࡱ; B4>   .1  & & MathTypePTimes New Roman- 2 `8lnk 2 ` ` 2 `k ` 2 `u ` 2 `y ` 2 `d  ` 2 `L  `Symbol- 2 `s 2 `a 2 ` a Times New Roman- 2 i> 2 7i>Times New RomanSy- 2 ` IMR~>Symbol- 2 `7= 2 `6+ Times New RomanSy- 2 0p 2  1p & "System-2H5&ࡱ FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ; ̀4&+ ln s i  = a 0  + a 1  IMR i- 2 ࡱ; ࡱ; LW,TObjInfoeJEquation Native K_901623159tivehF3Ŷdƶ|Ole42753 FNPICion Nativel88 l88 gjOLMETA1770 FQCompObjikYZObjInfol[ࡱ; W=  .1   & & MathTypePMT Extra- 2 Py$~Symbol- 2 !a Times New Roman- 2 81p & "System-ࡱ;  FMicrosoft Equation 2.0 DS Equation Equation.2ࡱ; ࡱ;  ?>x > "a  1"Syst Oh+'0 D h   @d Equation Native36D36&3/1p/1 /1P46P46\<SummaryInformation(]_864465982FooOleHcontaineD:\WINWORD\TEMPLATE\NORMAL.DOT0The Economic Value of Controlling Marine DebrisMarjorie B. McElroyMarjorie B. McElroy@s@@L @bMicrosoft Word 6.0&13ࡱ; | PIJuv{|!',-**w<x<@@@@A A(A)A/A0A5A6A>A?AyAzAAAAAA]cc uDp3ceKuDp3avK uDcchVcuDPcPccUUcOA8B9BLBMBNBOBmBnB}B~BBBBII8M9M:MWSXS0X1X5XBXCXOXvZwZ]]hh6iKi3j4j ooooo+o,o5o6oBosss||$~%~*~+~:~;~~~~~~~~~~c uD>3ceKuD>3avKch JcVcVcUcuDPc^c]cchuDm 5ceKuDm 5avK uDccF!"#$12܆݆()*+78hiLJȇۇ܇݇އ&'()euDBu3ceKuDBu3avKuDu3ceKuDu3avKuDu3ceKuDu3avKuDu3ceKuDu3avK]cuDʌu3ceKuDʌu3avK uDcVcchuDPcchc7e}efyz{|ƜǜȜԜ՜WXnyKMNcdtu &'(~ɽuD櫽5avKU^cUchhUchUUc]c^cuDPcchJccuDȆ3ceKuDȆ3avK uDccUcCfwxyz%&#-~\^/0C譣uDy5ceKuDy5avKuDz5UceKuDz5avK uDUcUchch^cuD䫽5ceKuD䫽5avKuD嫽5ceKuD嫽5avKUcU^cc uDcuD櫽5ceK6CDEFtvvwy"$/0<=GH{|34VWef}~$%7=>X[~h]^^cUUchUcchc uDcuDw5ceKuDw5avKR~;Bhogo  BCDO2c<O/hU~Zfm #7W^cU]]VcU^c]c U]^cU]cccchchU+>DEO|&rsdl#$g"#>?}        = > _ ` u v ~          cUcV uDPPP uDPU^ccVcX   4 5 = J K P _ a b E F     kq]ucVhJch uDPUcc!OP2, ~c">%4(,02M9!6!6!6!6!6 !6 !6!6!6!6!6!6!6!6!6!6!6!6!6!6!6!6!6!6 !6 !6!6 !6!6!6 l,"&M9x>8BPBTBUBBD.I:MOZS2X3X4X5XBXCXOXg^sb h5i6iKimu{||||~~~~Y܆!!6 !6 !6!6 !6!6!6!6!6!6!6!6!6 !6!6!6!6!6!6!3!6!6!6!6!6!6!O! l,"(,01*./de}׏:bАD\.֜ ŧ=!B!6!6!6!!!!!!!!!!!!!=!!6!6! !6!6!6 !6!6!60 l,"'jɹι׹ "02KVm%&!6!6!6 !6!`o.nn`. n nl ,6"x2lX ,6" x&(6BE}ĻWXZhxz¼ͼ MNev`. n n` . n n!! GG2lX ,6" l ,6"xxž;Ӿ۾$*34FLTZbhpqۿ GG GG GG GG l t 7,"h"l4t 7," "ۿ#*/6;BCZ`hnv|GG GG GG G!!|"l4t 7," hl t 7,"! FHIKv)*+-NOPRnopr|||||||||||||||||||||||||||l(b !l(b ! # 3568WXY\&'||||||||||||||!!!!!,,l! <l(b ! l(b !9<CFMPWZabnqx{P|||XXTTTP|XTTP|||XXTTTP|||XXhL< ld!&l@d! <<!  #*+7:ADKNUX_`abcTTTP|XTTP|XTTP|||XXTTTP|||XXTTTP|hL<L<< ld!<&cdefxy #$%!"XTTP|XTTP|||XXTTTP|||XXTTT!&l@d! hL<L< ld!< "#-.6:?FLTV\22/  (  ^6  ^^8ld :qO p%1 l4 qr%1 H\dipw ,-58=DJRX]d   ^^   ^^   ^^l :qO p%1)dfm|}~\]tuJxy   2222p222!!! #ZRl(bY# 8ld :qO p%1  $+18>EFGHUZafmryz{|#ZR##ZZRR##ZZR##ZZRlb /Y#* &,34KNQX`gou}~##ZZR#ZRR##ZZRR#ZZRRlb /Y#* !"#$%89:;<=>HMT[bgnsz#ZZRR#ZZRR#ZR#ZR##ZZRRlb /Y#*z *P#ZZRR#ZR#ZR!!!!!!,lLb /Y# lb /Y#!PQ'cdJyzst!c!!!!!!!!!!!!!!!!! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! `0*$%BCOPrs56}~HI! ! ! ! ! ! ! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!! 0 0# 0# 00#0%deopxy`aCDq!!!!!!!!!!!!!!!!!!!!!!!!!6!! !`%0(qrI"#v!"=>| }          !x!!x! !!x!!x!!!!x !!x!!x!!x !!x!!x w00 w0 l4             2 6 : = > T X \ _ ` l o r u v            0 w00 w00 w00 w00 w00 w00 l4    l4                 ) - 1 4 5 > B F J K Q V [ _ w00 w00 w00 w00 w00 w0 l4    l4  _ ` a D E     d0!x!!x!!x!!x!!x!!x!!!!!! l4  K@Normala "A@"Default Paragraph Font +@ Endnote Textc *@ Endnote Referenceh@ Footnote Text &@! Footnote Referenceh @2 Footer !)@A Page Number*      !"#$%&'()*  !"#$%&'()I,w98JWP0UvWZe3g#|ԙW#b  a }E H *      !"#$%&'()*a H   !"#$%&'( )# *!!!!!!!!! ! ! ! ! !!!!!!!!!!!!!!!!!!! !!!"!#!$!%!&!'!($!) !*1 <4%b,3;;AHOCU\{ciNqx ~1ʞ (#%DHD|.'cns  q cE9tRV&<X 6! "# $%&'() '*AeC~ ]M9&ۿc"\dzPq  _ 8?L?N?{{{܃(*DŽۄ݄&(ey{y/CE::::::::::::::: #*!!pg_1U U WSchool of the Environment"C:\DOC1\MANUSCRI\MDEB.2\041596.TXTSchool of the Environment"C:\DOC1\MANUSCRI\MDEB.2\041596.TXTSchool of the Environment"C:\DOC1\MANUSCRI\MDEB.2\042596.TXTSchool of the Environment"C:\DOC1\MANUSCRI\MDEB.2\042596.TXT AUTHORIZED GATEWAY 2000 CUSTOMERGAIA/COMMON:\COMMON\042596.DOCSchool of the EnvironmentK:\COMMON\042596.DOCSchool of the Environment"C:\DOC1\MANUSCRI\MDEB.2\042596.DOCSchool of the Environment"C:\DOC1\MANUSCRI\MDEB.2\042596.TXTSchool of the Environment"C:\DOC1\MANUSCRI\MDEB.2\042596.TXTMarjorie B. McElroy&D:\NETSCAPE\PAPERS\95PAPERS\DEBRIS.TXT@HP LaserJet 5/5M PostScriptLPT1:ADOBEPSHP LaserJet 5/5M PostScript pD; odXXHP LaserJet 5/5M PostScript pD; odXXJ@Times New Roman Symbol &ArialCG Times"hk{{ L &^ =&#i/The Economic Value of Controlling Marine DebrisMarjorie B. McElroyMarjorie B. McElroyࡱ; |