Voluntary Cooperation in the Commons? Evaluating the Sea State Program with Reduced Form and Structural Models

Joshua K. Abbott and James E. Wilen

Abstract

We utilize a variety of approaches to examine the success of a voluntary conservation program for a common property resource. The availability of panel data and a nonparticipatory group lets us use quasi-experimental methods to investigate the distribution of outcome treatment effects. We supplement these methods by incorporating a difference-in-differences structure into a behavioral model of fishing location choice to disentangle the program’s incentive effects from potentially misleading temporal variations in behavioral constraints. Our findings yield insight into the factors that support cooperation and illustrate the power of the complementary use of structural and reduced form models in program evaluation. (JEL C21, Q22)

I. Introduction

The prospects for “cooperation in the commons” have been a fruitful area of research for economists, political scientists, and other social scientists for years. There are now innumerable documented examples of successful cooperative management of common property resources by resource users—often with little or no external coercion (Acheson 1987; Arnold 1998; Ostrom 1990). Furthermore, there is substantial evidence from laboratory experiments to suggest that while collective behavior may fall short of the social welfare maximizing solution, people are often far more cooperative in their behavior than predicted by narrow self-interest alone (Ostrom, Gardner, and Walker 1994; Ostrom, Walker, and Gardner 1992; Walker, Gardner, and Ostrom 1990). This experimental and anecdotal evidence has spawned extensive research on the theoretical justification for collective action and the evolution of social norms for cooperative resource use (Ostrom 2000).

In this paper we examine the success of an endogenous and voluntary program of cooperation for the conservation of a species that is explicitly allocated by the right of first possession (common property). Fishermen in the Bering Sea trawl flatfish fishery pursue a variety of primary target species; however, as in many fisheries, they also inadvertently catch a range of nontargeted bycatch species. Chief among these bycatch species for much of the Bering Sea flatfish season is Pacific halibut, a species of zero direct value to trawl fishermen due to regulations that require it to be discarded. The total take of halibut is limited and carefully monitored for the entire fleet, and yet no vessel has a secure property right to a portion of its catch. When regulators see that the halibut quota will soon be met, they dramatically curtail the retention of target species, often effectively closing the fishery. Such closures often occur with significant remaining unused quota for target species. The tendency of this system to generate suboptimal outcomes is hardly surprising, since the equilibrium noncooperative strategy is to underinvest in avoidance because the costs of avoidance are personally born while the benefits of foregone or delayed closures are shared across vessels (Abbott and Wilen 2009).

In an effort to mitigate such perverse incentives and avoid the threat of tighter regulation, a group of vessels elected in 1995 to facilitate bycatch avoidance by pooling vessel-level data into spatial displays of participants’ bycatch outcomes using a third-party contractor, Sea State. In addition to lowering the costs of information provision and enabling participants to identify and avoid bycatch “hotspots,” Sea State may have served as a catalytic framework for the organization of informal arrangements among skippers to avoid fishing in areas of high bycatch.

Importantly, a group of vessels elected not to participate in the program for a number of years and therefore operated without the benefit of the information disseminated by Sea State. By utilizing this group of nonparticipants as a control group we are able to employ established techniques from the program evaluation literature to provide the counterfactual of what bycatch outcomes Sea State members would have shown in absence of their participation in the program. In addition to these reduced form, outcome-based measures of performance, we also develop a structural model of bycatch avoidance that allows us to use the actual spatial choices of skippers to uncover their implicit willingness to pay for bycatch avoidance. By extending the difference-in-differences approach from our reduced form models to the structural context, we are able to test for effects of the Sea State program on fishermen’s behavior apart from potentially confounding changes in fishing conditions that may have differentially impacted the participating and nonparticipating groups. The results of this structural difference-in-differences model not only help to confirm the robustness of our reduced form findings but also provide a useful framework for their proper interpretation. We conclude with observations on the importance of both institutional and natural constraints in fostering cooperation among resource users and argue for the complementary use of structural and reduced form models in the evaluation of both voluntary and externally imposed policy changes.

II. Background and Data

The subjects of this study are a group of approximately 25 catcher-processor vessels operating in the federal waters of the Bering Sea/Aleutian Islands management areas of the North Pacific. The vessels vary somewhat in their degree of capitalization, ranging in length from 110 to 230 feet with horsepower of 1,200 to 3,600 BHP and utilize bottom trawl gear to pursue a wide array of demersal species (primarily flatfish such as yellowfin and rock sole but also other groundfish species as price, abundance, and regulatory factors dictate) during trips that often last a fortnight or more. This catch is minimally processed onboard (hence their classification as the “head and gut” fleet) and frozen.

As one of the world’s most closely monitored and enforced fisheries, regulatory institutions are critical in shaping fishermen’s behavior. Although fishermen are subject to a wide array of regulations, including a network of time and area closures, the defining characteristic of the system for the purpose of this analysis is the use of common pool quotas to control the mortality of both targeted and regulatory bycatch species. Prior to each season, the North Pacific Fishery Management Council (NPFMC) establishes a biologically determined total allowable catch (TAC) limit for each of the managed target species. These limits fall on the catch of each species, regardless of whether the species is landed or discarded. At frequent intervals throughout the season, regulators synthesize data from onboard observers and weekly vessel production reports to produce cumulative estimates of fleetwide catch for each species. If regulators anticipate that a TAC will be met in the near future, they close the fishery to “directed fishing”—a status that allows the species to be retained, but only in drastically reduced proportions. The intent, if not always the effect (see Ackley and Heifetz 2001), of these closures is to remove the incentive for targeting of the species in question while allowing the harvest of other species. On occasion, this “soft cap” hardens into a prohibition on retention if the threat of significantly overshooting a TAC is imminent. Since access to the TAC is restricted due to federal limited entry regulations, and property rights to the catch are designated according to the right of first possession, the TACs on target species are common pool resources.

As with targeted species, fishermen also face common quotas on prohibited species catch (PSC), so named because fishermen are required to discard 100% of the catch of these species. The rationale is to discourage fishermen from targeting species that are also the primary targets of other nontrawl fisheries administered by entities besides the NPFMC (e.g., the Alaska Department of Fish and Game or the International Pacific Halibut Commission). For our purposes, the PSC quota for Pacific halibut is of paramount importance. The annual allocation of halibut PSC is parsed across various target fisheries based upon their anticipated usage of the quota. Estimated halibut PSC for the fleet is continuously monitored over the season in an analogous manner to the catch of target species and is debited against the available common pool quota suballocations according to the dominant target species of a vessel in a particular week. The allocations that are important in our analysis are those parsed to the yellowfin sole target fishery and, to a lesser degree, the quota for rock sole and other flatfish species.

The legacy of the common quota system is reflected in Figures 1 and 2. Figure 1 shows the annual TAC and actual total catch for yellowfin sole for the years 1993– 2004, demonstrating that catch has fallen well short of quota in all but the most recent years. The primary reason for the shortfall is indicated in Figure 2. The bycatch of halibut has consistently met or exceeded its quota allocation well before the yellowfin sole TAC is near to binding. As a result, fishermen often forego tens of thousands of metric tons of annual harvest worth tens of millions of dollars. These outcomes are consistent with the accounts of industry insiders (Gauvin, Haflinger, and Nerini 1995; Gauvin and Rose 2000) and fisheries scientists (Marasco and Terry 1982; Trumble 1998; Witherell and Pautzke 1997) who have described a situation in which the lack of secure rights to a share of the bycatch quota spurs a “race for bycatch.”

Figure 1

Annual Catch, Quota, and % Quota Utilization of Yellowfin Sole in the Bering Sea/Aleutian Islands

Figure 2

Annual Catch, Quota, and % Quota Utilization of Halibut Prohibited Species Catch Quota for the Yellowfin Sole Target Category

The Voluntary Treatment: Sea State

Weary of premature bycatch-driven closures of primary target species fisheries, the majority of the head and gut fleet (an initial group of around 18 vessels) agreed in 1995 to hire a private company, Sea State, Inc., of Seattle, to rapidly assimilate and relay data from the established onboard observer program to fishermen on the grounds. Under this program, information on PSC bycatch, including its precise spatial location and estimated quantity, is electronically submitted on a daily basis to Sea State, where it is pooled to provide fleet summary maps of bycatch rates along with expert commentary on trends in bycatch hotspots and possible evasive actions for skippers to undertake. These outputs are relayed on a daily basis, or even faster, to participants via email.

An important aspect of Sea State’s implementation is that a number of active vessels (four to six, depending upon the year and season) elected not to participate in the initial years of the program, although they did finally join in the late 1990s.1 By establishing that certain criteria of similarity are met, we are able to utilize the outcomes and behavior of this group of vessels to create the counterfactual of how the behavior and bycatch outcomes of the Sea State participatory group would have differed if they had instead opted out of the program. We can then compare this scenario with what we actually observe to discern what effects, if any, Sea State had upon bycatch rates and avoidance behaviors.

Before commencing our empirical analysis of Sea State’s performance, it is useful to consider what prior insight the literature on collective action and coalition formation can contribute. In many ways, our case bears a resemblance to the problem of self-enforced environmental agreements for airborne pol- lutants. By “abating” their bycatch, fishermen contribute a nonexcludable benefit to all other fishermen (through the lengthening of the season and expanded fishing opportunities), but their individual abatement costs can be described as increasing and convex in the degree of bycatch avoidance (see Abbott and Wilen 2009). Furthermore, given the firm legal right of common property in our case, any cooperative agreement has to be selfsustaining. In a seminal paper, Barrett (1994) analyzed the formation of such groups in the context of environmental treaties, using a static game theoretic framework in which the number of participants, the amount of abatement, and the abatement decisions of free-riders are simultaneously determined. He demonstrates that in cases where the benefits to cooperation are highest (i.e., the marginal benefits and costs of abatement are similar in magnitude and large) the incentive to free-ride will be substantial, so that only a small fraction of the parties will participate, therefore degrading the credible level of abatement sustainable under the treaty. The implication for our case is that the high rate of participation in Sea State may be indicative of a low behavioral hurdle between freeriding and the standard of abatement implicit in the agreement.2 In a model with costly and imperfect monitoring and enforcement of cooperative behavior, McCarthy, Sadoulet, and de Janvry (2001) show how the presence of significant variable costs of enforcement and an inability to credibly assess a large penalty on cheaters can result in an equilibrium level of resource exploitation substantially beyond the welfare maximizing level for the group as a whole. Given the stochasticity of catch and bycatch (making the identification of noncooperative behavior costly), the limited but by no means small number of participants, and the substantial physical challenges to self-monitoring presented by the vast Bering Sea, wemay expect, at best, amarginal gain in bycatch abatement from Sea State membership.

Despite these predictions, there are equally strong arguments that suggest a substantial degree of cooperative bycatch avoidance may be sustainable. The repeated seasonal interactions between vessels create the potential for sustained cooperation as a subgame perfect equilibrium in a noncooperative game. Recent papers have shown how the threat of substantial one-time punishments (through increased harvest of the common property resource) can sustain cooperative use of the commons, even in cases where individual actions are unobservable (Polasky et al. 2006; Tarui et al. 2008). This same research has demonstrated that full cooperation may be easier to support for an intermediate number of vessels than for a smaller group—a finding that also concurs with experimental results (Mason and Phillips 1997). Nor does the presence of a nonparticipatory group necessarily erode the sustainability of cooperation. Recent work has shown how the presence of indirect benefits from cooperation (i.e., social capital) can yield evolutionary stable equilibria in which both cooperating and noncooperating agents coexist (Oses-Eraso and Viladrich-Grau 2007). Information exchange on the locations of target species and a variety of small favors on and off the water can create excludable benefits for members of the cooperative “club.”3 Ultimately the literature is somewhat ambiguous concerning the prospects for cooperation in averting the tragedy of the commons for our case.

Data

The primary data source for our analysis is the North Pacific Groundfish Observer Program (NPGOP) database.4 Data from onboard observers, along with weekly production reports from vessels, are key inputs into the monitoring and enforcement activities of fisheries regulators. Observer coverage requirements vary by size of vessel, with vessels over 124 feet required to maintain an observer presence on 100% of all fishing days, while smaller vessels have a coverage requirement of 30%. The most important duties of observers are the gathering of a complete spatial record of fishing effort and total catch and random samples of selected hauls for target and PSC species composition. The decision of which hauls to sample is determined by randomized means conditional on an observer’s anticipated workload, minimizing the crew’s ability to anticipate a sampled haul and potentially alter its catch composition.5 After selecting a haul, observers utilize statistically sound methods of subsampling to yield valid estimates of the species composition of the sampled hauls. Upon recording their observations, observers relay this information to NPGOP officials in Seattle on a daily basis. The NPGOP program is widely regarded as among the best of its type (MRAG Americas 2000), and its data have been employed in numerous scientific studies as well as in the day-to-day management of the fishery.

The data used in our reduced form models include the species composition sampled hauls for 18 of 24 long-term members of the fleet for the interval of 1992–2000. The data include 12 vessels that were initial and continuing members of Sea State (there was no defection from the program over time) and 6 vessels that initially opted out. The records for another 6 initial Sea State participants were excluded fromour sample due to the fact that their size made them subject to only partial observer coverage. Observer data from these vessels may not be representative of overall fishing behavior, given the freedom of choice of small vessel owners in choosing when to meet their 30% coverage requirements. Finally, we limit our attention to the subseason between April and November in order to focus on the fishery for yellowfin sole and the avoidance of halibut.6

These data present a random sample of the behavior and bycatch outcomes of a substantial portion (approximately threefourths of vessels and a much higher proportion of total effort and catch) of the total fishery. Although composed of larger vessels than the overall population, we argue that the effects we uncover apply broadly, given that many of the excluded vessels are only slightly below the 100% observer coverage cutoff and are therefore qualitatively similar to their fully covered colleagues and would likely behave in a similar manner if they too were subjected to full observer coverage.

III. Did Sea State Lower Bycatch Rates?

One obvious way to examine the success of Sea State in fostering bycatch avoidance is to examine whether participants exhibit reductions in measures of bycatch relative to what they would have otherwise produced. The appropriate measure of success, as is often the case with reduced form models, is unclear, with valid arguments favoring the use of both bycatch rates and levels. In the end, we base our analysis on the bycatch rate (kilograms per metric ton of targeted catch—including all targeted species) for a number of reasons. First, it is very close in definition to the bycatch rates provided by Sea State to its participants. Second, while the catch of halibut per unit time is important to the goal of lengthening the season, the ratio of bycatch to target catch ultimately determines the ratio of quota usage at the end of the season (and whether target or bycatch quota constrains the fishery). Third, the change in the bycatch rate provides a rough but useful inverse measure of the catch benefits (before the consideration of any incurred costs) arising from avoidance behavior. The level of bycatch can be misleading in this domain because it is possible to show “success” simply by fishing in less productive grounds. Finally, this measure of the “cleanness” of a vessel’s bycatch output has the virtue of being comparable across vessels with varying fishing capacities.7 We calculate vessel bycatch rates using weekly catch totals from the species-sampled observer data. Working at this level of aggregation allows us to focus on persistent trends in bycatch rates while attenuating the considerable degree of natural and measurement variability in our sample, thinning the proportion of zero bycatch rates, and dampening serial correlation between successive observations.

Summary Statistics

Table 1 presents summary statistics of bycatch rates by year and divided across the participatory and nonparticipatory groups. We should note that initial Sea State nonparticipants eventually joined the program in the 1998 or 1999 fishing season so that the monikers of “Sea State” and “non– Sea State” really have credence only for the previous years.8 The distribution of bycatch rates is skewed right with a significant mass at zero (particularly in the earlier years of our sample). These statistics suggest that the central tendency of bycatch rates is typically quite low, particularly in the early and mid-1990s, and that the mean bycatch rate is driven by a relatively small percentage of “dirty” hauls, regardless of which group a vessel belongs to. Of course, in extrapolating this aggregate result to individual vessels we risk committing the ecological fallacy. The properties of the group need not apply to the individual members; heterogeneous central tendencies in bycatch rates across vessels could produce a similar pattern. We address this possibility in the econometric estimates that follow.

Table 1

Quantiles and Other Summary Statistics for Weekly Halibut/Groundfish Bycatch Rates (KG/MT)

Turning our attention to the distributions of the initial Sea State participants and nonparticipants, we note that the distributions of bycatch rates across the two groups are practically indistinguishable in the early and mid-1990s—not only in central tendency but in many of the other quantiles as well. This similarity continues until 1997, suggesting the lack of an obvious a priori effect on the bycatch frequency distribution from Sea State membership. The strong upward trend in bycatch rates in the late 1990s (potentially due to an increase in halibut biomass) is dampened considerably for the initial nonparticipants relative to the early adopters (keeping in mind that all vessels were in the program for much of this period). These descriptive observations are confirmed by the application of the nonparametric Wilcoxon rank-sum test (Lehmann and D’Abrera 1998). The base category in this case is the nonparticipants, and so negative z-statistics are evidence of a rightward location shift of the Sea State bycatch rate distribution relative to the non–Sea State group. In the pre–Sea State years we find no significant evidence of any location differences between the two groups. However, there is overwhelming evidence for 1998–2000 that the median bycatch rate of the initial nonparticipants was actually lower than that of the original Sea State members (a finding that we return to later). Simple comparisons of summary statistics provide no evidence of improvement in the bycatch rates of Sea State members relative to those that did not participate, despite their nearly identical performance before the implementation of the program.

Difference-in-Differences Estimation

This analysis, while suggestive, does not provide a convincing case for the success or failure of Sea State. To investigate this question more rigorously, we utilize a tool from the policy evaluation literature, the difference-in-differences (DID) estimator (Angrist and Krueger 2000; Meyer 1995). This estimator analyzes the mean effect of a treatment variable on outcomes by utilizing data gathered before and after the treatment was implemented on a group that received the treatment and an untreated control group. By subtracting the average values of the “after” and “before” outcomes for the two groups and then subtracting these differences across the groups, one is able to uncover an average treatment effect from the policy change. In the two-period case this differencing is facilitated by a regression specification in dummy variables:

Embedded Image [1]

where BycatchRateit is defined in the previous section, PostSeaStatet=1 if the observation occurs in the posttreatment period (1995 onward), SeaStatei=1 if vessel i is a member of the treatment group (i.e., it joined Sea State at its outset), and (PostSeaStatet * SeaStatei) = 1 represents the overlap between a treated vessel and the treatment period. In this case β3 is the average treatment effect on the bycatch rates of treated vessels. If its estimate is small and insignificant, then this is consistent with a minimal effect of Sea State in fostering bycatch avoidance. On the other hand, if it is large and significantly negative, then this provides positive evidence of its effectiveness. Note that both pretreatment differences across the two groups (including time-invariant, group-stable factors that influence the selection process) and shared time trends in the dependent variable are controlled for in [1].

A number of assumptions are implicit in the use of this quasi-experimental estimator, and the integrity of the results is often fragile with respect to their violation (Besley and Case 2000; Meyer 1995). First, the composition of the treatment and control groups must be time-stable; otherwise, issues of sample selection bias arise if individuals selected themselves into (or out of) the sample based on the policy treatment (Blundell and MaCurdy 1999). Secondly, the control group must be as comparable to the treatment group as possible. In other words, factors that vary between the pre and posttreatment periods must impact the two groups equally or, failing this, be effectively controlled for by the addition of observable exogenous variables. Failure to achieve this comparability may lead to the confounding of the treatment effect with other time-varying factors. Finally, the treatment should be exogenous; in other words, the application of the treatment must not be correlated with unobserved drivers of the outcome in question.

The treatment in our case is being a member of the Sea State program. The control group is therefore the handful of vessels that were not part of Sea State at its inception. In terms of stability of the two groups, our quasi-experiment is valid by design, since the sample includes only those vessels that were active in the fishery before and after Sea State, and we only attempt to draw direct inferences for the population of long-term participant vessels. There are, of course, weeks in which some vessels do not fish, causing our panel to be unbalanced and irregularly spaced. This is not a problem, however, as long as the decision to fish in the Bering Sea is not itself influenced by the anticipated bycatch rate in that week.9

In terms of comparability our treatment and control groups pursue similar fishing strategies, produce similar products, face the exact same regulatory and natural restrictions, and utilize similar vessel capital and gear in fishing. Furthermore, the comparison of the empirical distributions of bycatch rates in Table 1 and the Wilcoxon rank-sum tests suggest that the pretreatment bycatch rate distributions are indistinguishable across the two groups. This lends credence to the identifying assumption that unobserved time-varying factors that drive vessel bycatch rates vary similarly across the two groups and affect their bycatch rates equivalently.

The exogeneity of the treatment is, of course, problematic given that the assignment to Sea State was a choice for vessel owners. Fortunately, matters are made somewhat easier by the fact that our objective is to examine the average effect of a voluntary program under a particular set of institutional constraints (common property management), not one for which participation is mandated or randomized over the fishery participants. In other words, our objective is to identify the average treatment effect on the treated, not the average treatment effect for the population of all anglers in the fishery.10 The differential effect of the program due to selection is therefore an integral part of the treatment itself, not a nuisance to be purged. The implication is that variables (observable or not) that affect the magnitude of the individual treatment effect but are also correlated with the likelihood of receiving (i.e., selecting into) the treatment are to be ignored in the estimation as long as they do not have an independent role in explaining bycatch rates. However, if some variables that explain participation in Sea State also play direct roles in explaining bycatch rates, then these effects must be accounted for in the estimation (Besley and Case 2000).11 We use both direct and econometric methods of control, which we detail below.

Bertrand, Duflo, and Mullainathan (2004) determined that many prior DID studies cite standard errors that are too small due to neglected serial correlation in the errors. We therefore specify a flexible error structure that allows for vessel-specific heteroskedasticity, arbitrary contemporaneous correlation between vessels, as well as AR(1) correlation within panels and estimate this error structure for Models 1– 3 in Table 2 using a Prais-Winsten regression estimator modified for panel data (Greene 2003). This yields consistent estimates if the regression is correctly specified and is also asymptotically efficient if the specified covariance structure is correct.12

Table 2

Difference-In-Differences Regressions of Halibut Bycatch Rate

Three different DID estimates of the effects of Sea State are presented in Table 2. Note that, despite the substantial number of zero bycatch rates present in the data, we have elected to estimate linear specifications of the conditional mean. We do this for the reason that the typical parametric solutions, such as the censored Tobit model, are extremely fragile to deviations from normality and homoskedasticity—assumptions that are clearly violated by our data.13 Model 1 is closest in form to [1] although the presence of multiple years of data both before and after the advent of Sea State allows us to augment this model with yearly dummies to consider separate annual treatment effects. This also serves to immunize our results to uncertainty about the suitability of the initial nonparticipants as a control group in the later years of our sample. We also include the ratio of horsepower to vessel length, since prior analysis (Abbott 2007) suggests that highly powered vessels may exhibit lower bycatch rates due to their greater mobility across the fishing grounds and may also experience differential selection pressure to join Sea State.

A glance at the treatment variables (labeled Sea State * Year) suggests that Sea State had no discernible impact on bycatch rates in the first three years of its inception. There do, however, appear to be strong effects from 1998 onward, suggesting that bycatch rates were far higher on average for the initial Sea State participants than for those who opted out.

Notwithstanding these results, this estimator does not control for intraseasonal variations in bycatch rates due to fluctuating biological conditions and recurring closures. Failing to account for such patterns could bias the estimation of the treatment effect if the groups allocate their effort differently across the season. To account for such effects, we add monthly indicators to the previous model (with April the base category) to form Model 2. Nonetheless, estimates of the treatment effects from Sea State are practically unchanged.

Neither of the previous models controls for unobservable vessel specific factors that influence bycatch rates and that potentially drive selection into the treatment. Model 3 addresses these potential sources of endogeneity by harnessing the panel nature of the data and including vessel-specific fixed effects in place of the single control group indicator, therefore absorbing any time-invariant vessel or firm-specific factors. A strong case can be made for time-invariant effects as the most likely sources of selection bias and other forms of endogeneity. First, the initial membership of Sea State was established along preexisting and durable bonds of trust and information sharing. Secondly, we observe no reversals of the participation decision, which we might expect if the decision to participate was strongly influenced by time-varying factors that varied significantly in our sample. Thirdly, the initial development of Sea State for the control of red king crab bycatch in the rock sole fishery and its apparent success in that regard (Gauvin, Haflinger, and Nerini 1995; Gilman, Dalzell, and Martin 2006) suggest that much of the participation decision was based upon Sea State’s value in this separate fishery, since the additional cost of the program for halibut was minimal.

Model 3 presents the least-squares dummy variable (LSDV) estimates of the fixed-effects model with the estimates of the fixed effects suppressed. Accounting for time-invariant endogeneity seems to have little impact on any of the estimated coefficients or their significance levels. This suggests that whatever time-invariant factors are involved in the decision to retain Sea State’s services, they do not figure significantly in the decisions that influence weekly bycatch rates. Models 4 and 5 are identical to Model 3 except for their calculation of the standard errors. To consider the sensitivity of our inferences to alternative covariance estimators, we estimate Model 4 with a covariance matrix that is identical to that of Model 3 except it presumes no serial correlation, and we use “panel robust” White standard errors for Model 5 (Bertrand, Duflo, and Mullainathan 2004). Model 5’s standard errors are robust to heteroskedasticity, cross-correlation, and serial correlation of a general form, but the assumptions necessary for its consistency are arguably lacking here.14 Model 4 confirms the findings of Bertrand, Duflo, and Mullainathan (2004) in that t-statistics that fail to account for serial correlation are a bit larger than those that do (although the distortion is far from large). Regardless of which standard errors are chosen, the inferences on the mean effects of Sea State are remarkably consistent across multiple specifications.

While useful, limiting our analysis to the average treatment effect may be inadequate for several reasons. First, the statistics from Table 1 suggest that the conditional mean may not be particularly informative as a summary of “typical” bycatch behavior. Secondly, mean regression techniques are quite sensitive to outliers, and so our estimates may be highly leveraged by a relatively small number of extreme data points. Finally, there is no a priori reason to suspect that the treatment effect of Sea State will operate on the distribution of bycatch rates by a pure location shift. It may instead function through alterations in the variance (i.e., heteroskedasticity) or other higher moments as well. Failure to acknowledge such higher-order effects can yield partial or deceptive results (Bitler, Gelbach, and Hoynes 2006).

To address these effects we employ a semiparametric estimator known as quantile regression (Buchinsky 1998; Koenker and Bassett 1978; Koenker and Hallock 2001). Rather than specify the structure of the conditional mean, we instead specify the conditional quantiles of the dependent variable so that the interpretation of the coefficients is the marginal change in the inverse cumulative distribution of the dependent variable evaluated at a particular quantile. Quantile regression has a number of desirable properties. First, as a semiparametric estimator, the assumptions underlying its consistency as an estimation method are quite limited. Secondly, since it is estimated by minimizing the weighted absolute sum of residuals, rather than the sum of squared residuals as in the case of the conditional mean, it is much more robust to the influence of outliers. Thirdly, multiple regressions can be estimated at a variety of quantiles to reveal the effects of regressors on the entire distribution of outcomes.

To account for the large quantity of zero bycatch rates in our data we estimate censored quantile regressions (Buchinsky 1998; Chay and Powell 2001; Powell 1984, 1986). The parameters of the conditional quantile estimator are obtained by iteratively minimizing the weighted sum of the absolute residuals (where the weights are determined by the quantile in question) for the uncensored model using only those observations for which the predicted quantiles are nonzero until the same observations are excluded on repeated iterations.

Table 3 presents our estimates of the full model with fixed effects (Model 3 in the conditional mean specification) for a range of quantiles of the bycatch rate distribution. We estimate regressions only from the fourth decile up, as the iterative estimation technique tends to disproportionately drop variables from particular time periods, making the estimation of the associated dummy variables quite unstable (if not impossible) for lower quantiles.

Table 3

Probit and Censored Quantile Estimates of Difference-In-Differences Model

A comparison of the marginal effects of the regressors across quantiles reveals that there is a great deal of agreement in sign and significance across the quantiles, implying that the variables often exert uniform directional impacts or are comparable in their apparent lack of impact.15 There is little evidence of upward trend in the bycatch rate distribution for the central (fourth and fifth) deciles relative to the base year of 1992, although there does appear to be some evidence of higher bycatch rates in the upper quartiles for the 60th to 90th percentiles for the late 1990s and 2000. A comparison of the estimates for the median to those for the conditional mean of Model 3 suggests that as measurements of central tendency or typical behavior, the coefficients of the conditionalmean are somewhat inadequate in that they are clearly heavily influenced by the upper quartiles and bear a close resemblance to the quantile regression estimates for the 60th and 70th percentiles.

A glance at the estimates of the treatment effects for 1995 through 1997 reveals little consistent proof of any detectable effect, notwithstanding some evidence that the Sea State vessels had higher bycatch rates around and just below the median of the distribution in 1996 relative to their non–Sea State colleagues. We also report estimates of a probit model in Table 3, where we model the probability of a week with a recorded bycatch rate of zero. Again, there is no evidence of a higher proportion of bycatchfree weeks by Sea State members relative to the control group in the immediate aftermath of the program. These results confirm the findings of the models in the conditional mean. The added value of the quantile regression and probit estimation is that we can now draw the same inference for the entire distribution of outcomes.16

One concern in interpreting the treatment effects in our DID specifications (both here and in the structural analysis) is the possibility, due to the common PSC constraint on bycatch, that nonparticipants are nevertheless “treated” in an indirect way, potentially compromising their validity as a control group.17 Fortunately, there are two arguments that allay this criticism. First, assuming that Sea State’s existence had an impact upon nonparticipants, it seems probable that it would drive them to freeride on the cooperative vessels and avoid bycatch less than before. This would lead, ceteris paribus, to an increase in the bycatch rate among non–Sea State vessels in the treatment period and therefore bias the estimated effect of Sea State in a downward direction (i.e., it would tend to overstate its downward impact on bycatch rates). However, given that we failed to detect an effect of Sea State across a wide range of specifications, this potential bias only strengthens the robustness of our results. Second, if Sea State had a significant indirect effect on the bycatch outcomes of nonparticipants, then we would expect to see this reflected in a substantial shift in the annual dummies for the control group between the pre- and post–Sea State years in our regressions. We find no evidence of such a shift, which further suggests that our estimated treatment effects are valid.18

The robust findings that initial Sea State participants actually exhibited substantially higher bycatch rates in 1998–2000 relative to the control group are puzzling. Of course we know that the initial nonparticipant group had joined Sea State by the 1999 season and therefore lacks validity as a control group for this period. It is conceivable that this worsening in the relative bycatch performance of charter Sea State members is explainable by a disproportionate positive effect of the program on the bycatch outcomes of the new members. However, our regressions demonstrate that the gap in relative performance was not caused by a drastic improvement in the bycatch rates of the initial nonparticipants but was instead driven by a rapid increase in the bycatch rates of the original membership.

This increase can be attributed to a marked change of targeting strategy by the initial Sea State participants from 1998 forward. Table 4 shows the percentage shares of species composition sampled hauls devoted to each major groundfish target allocation.19 One immediately observes a sharp movement away from the targeting of yellowfin sole by original Sea State participants, with an increased focus instead on flathead sole and arrowtooth flounder. The primary motivator in this decision was apparently the depressed price of yellowfin sole products in this time period. TheNMFS estimated price for head and gut yellowfin sole fell 30% between 1997 and 1998 from $0.27 per pound of catch to $0.19/lb. As a result, this group of fishermen diverted their effort to the catch of more valuable flatfish species. Unfortunately, the catchability of halibut associated with these alternative targets is significantly higher than in the yellowfin sole fishery (Spencer, Wilderbuer, and Zhang 2002); our own data show that arrowtooth flounder and flathead sole targeted hauls exhibit median halibut bycatch rates of 30 and 18 kg/mt, respectively, while yellowfin and rock sole have much lower rates of 3 and 6 kg/mt, respectively.

Table 4

Shares of Species Composition Sampled Hauls Devoted to Target Categories

By contrast, the late-joining Sea State vessels did not dramatically alter their targeting behavior away from yellowfin sole after 1998, thus explaining the relative stability of their bycatch rates. The cause of this differential targeting, although not the focus of this paper, appears to lie in the comparative importance of scale for these larger vessels. In order to achieve the economies for which these vessels are designed, they must fish on well-known (low search cost) and relatively thick aggregations to achieve sufficient throughput in their operation. Yellowfin sole is considerably more aggregated than flathead sole, leading the larger vessels to remain on these aggregations despite their now lower gross return.20

IV. Did Sea State Improve Incentives for Bycatch Avoidance?

While useful, these reduced form results are inherently limited in that they reveal the joint effects of fishermen’s preferences and the biological, economic, and regulatory constraints that shape them. If constraints change in ways that are poorly understood or not easily incorporated into a reduced form structure, and the treatment and control groups are differentially affected, then the identifying assumptions of the DID are suspect. Indeed, it is plausible that an apparent lack of significant outcome effects may belie the presence of an underlying behavioral treatment effect due to the differential impacts of temporally varying factors on fishermen’s abilities to alter their bycatch outcomes. An appropriate behavioral (structural) model is adapted to the explicit incorporation of such constraints and disentangling the incentive effects of a program from its outcomes. Furthermore, even if the results of a reduced form investigation are robust (as we have done our best to insure), a structural model can prove highly complementary for the interpretation of the reduced form model and vice versa (Keane 2005).

The behavioral model we employ is a variation of a random utility model of fishing location choice that we have used successfully in other contexts to analyze the causes of fishermen’s spatial behavior (Abbott 2007; Abbott and Wilen 2008).21 The utility function of a spatial alternative conditional on participation in the fishery is (suppressing subscripts for the individual decision maker)

Embedded Image [2]

where n denotes the particular site and t denotes the temporal unit over which location choices are made—the individual haul in our case. The utility of a site is a function of its expected productivity in terms of revenues per hour less the variable cost of accessing the site (which we assume is a linear function of its distance from the current fishing location) as well as a number of control variables (Xnt) to capture the effects of phenomena such as state dependence and herding behavior. Most importantly, however, the attractiveness of a site is also a function of its expected halibut bycatch rate so that λ/β is the implicit cost in dollars of a kilogram of halibut bycatch— the amount of revenue a captain would be willing to forego to “bank” a unit of bycatch for future use.

The estimation of such a model presents a number of challenges, and our extensive data set allows us to make a number of innovations in our specification that we detail elsewhere (Abbott 2007; Abbott and Wilen 2008). We allow for substantial heterogeneity in preferences by estimating vessel-specific parameters on expected revenues, distance, and a habit persistence dummy variable.22 We also allow for vessel-specific heteroskedasticity by estimating the relative scale parameter for each vessel.23

In order to specify the two expectations variables, we depart from the prior literature, which has relied upon simple specifications using rolling averages or lags of past catch and revenues pooled across vessels, by deriving vessel-specific revenue and bycatch predictions from sophisticated species-level expected catch models. These models assume that skippers possess a temporally stable “information weighting rule” that combines signals from a variety of sources (recent catch history at a fine spatial scale, broad-scale information on the seasonal distribution of species, logbook data, etc.). These signals are allowed to vary according to a vessel’s site visitation history, and we construct proxies for each of these signals by employing the randomly sampled observer data. In the case of halibut, we simulate the sharing of information via Sea State by pooling bycatch signals across member vessels and allowing nonmember vessels access to only their own bycatch history.24 We then estimate the efficient rational expectations weighting rule for each species using quasi-likelihood exponential regression methods (Gourieroux, Monfort, and Trognon 1984a, 1984b). Finally, we utilize these equations with vessel-level catch and histories to generate predictions of the catch of each species (including bycatch) at a site, which are then combined with data on prices, vesselspecific product allocations, and information on discard requirements and spatial closures to generate regulation-reflective expected revenue predictions for each vessel and spatial alternative.25

The model is well suited to our analysis for a number of reasons. First, the margin of decision making in the model is spatial at a relatively fine scale, which is the part of bycatch avoidance Sea State is aimed at influencing. Secondly, the temporal resolution (designated at the haul) is quite fine and matches nicely with the rate of information flow and temporal scale of the relevant decision making. Thirdly, we are able to explicitly incorporate the informational asymmetries between Sea State and non– Sea State members. Finally, many of the natural, biological, and economic factors that influence fishermen’s decision making are incorporated in economically consistent and sophisticated ways within the structural model. Changes in relative prices and revisions in beliefs about the spatial occurrence of target and bycatch species are incorporated in the models of expected revenues and bycatch, and the constraints imposed by spatial closures and retention restrictions are explicitly incorporated as well.

To ascertain the effects of Sea State on the relative trade-offs between short-run economic returns and halibut bycatch, we hypothesize that Sea State membership exerted a uniform “location shift” on the distribution of the marginal disutility of halibut bycatch but that avoidance behavior may have changed over the treatment horizon for reasons that are not explainable by the treatment alone. We therefore retain the use of the DID formulation in our structural model and specify λ as follows:

Embedded Image [3]

where αi is a vessel fixed effect (which we estimate), SeaStatei=1 to indicate vessels that were originally in the Sea State “treatment” and AfterSSt and After1998t=1 to indicate the periods from 1995 and 1998 forward, respectively. δ3/β represents the behavioral treatment effect of Sea State on its participants (i.e., the incremental shadow cost of halibut from participation); positive and statistically significant estimates of this parameter are consistent with a significant behavioral effect of Sea State on bycatch avoidance. The division of the posttreatment time horizon into two periods parsimoniously mimics the use of yearly dummies for temporal and treatment effects in the previous models, allowing us to investigate the presence of treatment effects in the years immediately preceding the introduction of Sea State without contamination from later years, when the groups were combined in Sea State. It also allows us to investigate the apparent break evidenced in the reduced form models to see if the gap in bycatch outcomes across the two groups from 1998 onward is associated with a divergence in their underlying behavior.

We estimate the model using haul-level observer data from 1994 to 2000 over the same season as our reduced form models.26 We adopt a conditional logit model and accommodate the large spatial choice set and number of choice occasions by estimating the model on a strategically chosen subset of 6 locations from the potential 165 using a selection-from-alternatives estimator (Bierlaire, Bolduc, and McFadden 2003; McFadden 1978; Train, McFadden and Ben Akiva 1987). The complete explanation of the results is beyond the scope of this paper, but the model predicts the in-sample behavior of fishermen quite well—giving the highest probability to the alternative that was actually selected about 85% of the time—and the parameter values are consistent in sign and magnitude to those in the literature and suggested by economic theory.

Table 5 reports the median estimated marginal effects of the covariates on the shadow cost of halibut and the standard deviation of estimates across all vessels.27 The estimate on the treatment interaction term is positive but highly insignificant. The structural modeling approach has therefore confirmed the findings of the reduced form models by failing to find a detectable average behavioral effect of Sea State membership on spatial bycatch trade-offs. The results for 1998 onward are revealing as well. There is moderate evidence (p = 0.09) for an increase in avoidance by original nonparticipants in this period. However, there is very strong evidence of a marked relative decrease in the implicit value of halibut to original Sea State participants in the late 1990s as well, so much so that their revealed preference for bycatch avoidance significantly declined in this period relative to the previous three years. Given the stability in the central tendency of 1998–2000 bycatch rates for the late-joiners that we find in the reduced form regressions, the worsening of outcomes in this same period for the original participants is likely attributable to the deterioration in their incentives for bycatch avoidance.

Table 5

Difference-In-Differences Estimates for the Shadow Cost of Halibut

The explanation for this sudden collapse in avoidance apparently lies in the aforementioned reduction in yellowfin sole prices. A substantial portion of the halibut allocated to the head and gut fleet is designated for the targeting of yellowfin sole, meaning that vessels must retain and process substantial quantities of yellowfin sole to utilize this bycatch quota. The value of halibut quota, and the collective mandate for cooperation, is therefore closely tied to the underlying value of the revenues from yellowfin sole catch. The 1998 crash in yellowfin sole prices had the effect of substantially reducing the implicit value of holding halibut quota, driving the initial Sea State participants to weight halibut to a lesser degree in their spatial decision making. However, the apparent stability in avoidance of the original nonparticipants in spite of this price decrease seems to counter this argument. Shouldn’t these vessels have had similar exposure to variation in prices? There are two explanations that suggest otherwise. First, these vessels may have had price contracts that sheltered them from the variation experienced by the rest of the fleet. Given that we are unable to obtain vessel-specific prices for this period, we are unable to test this hypothesis. Second, these vessels differ in that they are uniformly larger and must therefore repeatedly target high-density aggregations in order to benefit from economies of scale.28 Spawning aggregations of yellowfin sole in the spring and early summer are an ideal target for such vessels, since they are typically much more spatially concentrated than the relatively scattered species targeted by their competitors in deeper waters (Abbott 2007).29 We expect these vessels to exhibit a much more stable response to price declines of yellowfin sole, since failure to avoid halibut bycatch attributed to yellowfin may deprive these vessels of the opportunity to operate at a profitable scale.

V. Why did Sea State Fail to Reduce Bycatch?

The joint verdict from the reduced form and structural models is that Sea State failed to produce a measurable effect in either bycatch outcomes or the implicit trade-offs of fishermen. This leads us to the question, “Why?” Holland (2000) suggests that weak yellowfin sole prices may have undermined bycatch avoidance. We have already noted this phenomenon for 1998 onward. However, this argument fails to explain the lack of an apparent treatment effect in the period from 1995 to 1997, when prices were comparable to pre-1995 levels. Others have attributed the failure of the program to alleged growth in the biomass of halibut over the 1990s. While possible, such an increase is not supported by (admittedly flawed) stock surveys for the area and fails to explain the apparent lack of a behavioral effect in the early years of the program, when both the annual stability of the bycatch rates and the fundamentals of halibut biology (it being a relatively long-lived species with gradual rather than sudden shifts in the age/length distribution) suggest that the biological prospects for halibut avoidance must have been roughly comparable from year to year. Finally, some authors (Gauvin, Haflinger, and Nerini 1995; Gauvin and Rose 2000) have attributed the failure to predatory behavior on the part of nonparticipants. However this allegation does not reconcile well with our findings that nonmember vessels exhibited similar bycatch rates and bycatch avoidance behaviors for the first three years of the program and that the original nonparticipants actually do a better job of avoiding halibut from 1998 onward.30

A more likely culprit seems to lie in the inherent challenges to voluntary cooperative behavior in a system fundamentally driven by common property incentives. The number of vessels involved, the limited opportunities for meaningful punishment of defectors, and common pool quotas for both target and bycatch species would seem to militate against any attempts at more formalized endogenous cooperation. Nevertheless, this judgment must be tempered by the fact that the introduction of Sea State is widely viewed to have succeeded for the same fleet over the same period in reducing the bycatch of another prohibited species, namely, red king crab in the winter rock sole roe fishery (Gauvin, Haflinger, and Nerini 1995; Gilman, Dalzell, and Martin 2006).

What features made voluntary cooperation in red king crab bycatch avoidance sustainable in the rock sole roe fishery? First, the rock sole roe fishery is highly spatially concentrated in the shallows to the north of the Alaska Peninsula due to the seasonal distribution of the target species and the frequent presence of sea ice in more northerly waters. This places the fishermen within close proximity of one another so that monitoring and enforcement is facilitated. Second, the horizon over which cooperation must be sustained is quite short. At its longest, the rock sole fishery extends from late January to the first days of March when spawning has ended. Third, the benefits of cooperation in bycatch avoidance are comparatively high. Roe-in head and gut products command prices that are three to four times higher than alternative products for rock sole and well beyond the value of any other product fishermen produce throughout the year. This high value, while potentially a lure for opportunistic behavior, may also increase the appeal of cooperation, since the personal benefits of a cooperative action are greater.

Finally, the temporal and spatial covariation between rock sole and red king crab aggregations facilitate avoidance and cooperation. Red king crab often overlap with rock sole due to the nature of their mating migrations. Crab in the southern Bering Sea and Bristol Bay typically move in large and ephemeral “waves” of sexually segregated individuals from deep waters to shallows and often move quite quickly, covering up to a mile per day (Blau 1997; Dew and McConnaughey 2005). As a result, king crab bycatch often exhibits highly spatially concentrated and relatively short-lived patterns over space (Dew 1990; Dew and Austring 2007). In such a setting, bycatch can often be dramatically curtailed by relatively small and temporary movements from well-defined “hot spots” (Gauvin, Haflinger, and Nerini 1995). The costs of such movements (both in terms of foregone catch and fuel costs) are likely small, and it is likely relatively simple to identify vessels who are fishing noncooperatively, given the defined nature of hotspots, the rapidity of Sea State updates, and the clustering of fishing effort in the grounds.

By contrast, yellowfin sole fisheries in the summer and fall are highly dispersed over many thousands of square kilometers, hampering the effective identification and punishment of cooperative and noncooperative behavior. The horizon for cooperation in halibut avoidance is essentially the entire year, as it is encountered to a substantial degree for a range of target species. Furthermore, the benefits from cooperation in bycatch avoidance, although substantial, are not remotely as great as for the rock sole roe fishery, further dampening incentives for cooperation. Finally, and perhaps most importantly, the spatiotemporal distribution of halibut frustrates bycatch avoidance. Compared to red king crab, halibut are distributed far more evenly across the Bering Sea. Although broadscale patterns do emerge due to the varying seasonal migratory patterns of particular halibut age and size classes (Adlerstein and Trumble 1998a, 1998b), these patterns are often too spatially coarse to facilitate effective avoidance. In addition, attempts to seek microscale regularities in halibut abundance may be frustrated by substantial inherent randomness (Abbott 2007). Although more research is warranted, this suggests that the spatiotemporal distribution of halibut is governed by broad seasonal regularities but with little replicability at the temporal and spatial scales needed for decision making. The data pooling and expert analysis provided by Sea State can help in this regard, but they cannot uncover regularities that may not exist.

One indicator of the special difficulties of halibut avoidance is that participants in the rock sole roe fishery have failed to match their successes in crab avoidance with halibut. Indeed, halibut bycatch has increased considerably in the wake of Sea State (partially as a result of avoidance of high crab-bycatch areas) (Gauvin, Haflinger, and Nerini 1995; Gilman, Dalzell, and Martin 2006). One might suspect that the success of cooperation in avoiding crab bycatch might have yielded similar benefits 150 Land Economics February 2010 in halibut avoidance, but such an effect has not been forthcoming.

VI. Conclusion

The failure of Sea State in fostering halibut bycatch avoidance is likely attributable to common property incentives and the spatiotemporal association of the targeted and bycatch species—to both institutional and natural constraints. Cooperation may survive or even flourish under a contradictory system of management when nature lends a supportive hand but can scarcely be expected to offset the “race for bycatch” when nature is less benevolent. This finding extends well beyond bycatch management and fisheries issues to the management of a wide range of natural systems under common property access. The primary lesson is the joint importance of institutional, economic, and natural criteria for shaping resource users’ incentives. Our study also points to the possibility of substitutability between institutional and biological criteria for effective management when voluntary mitigation of incentive problems is a possibility. Resources whose natural characteristics discourage voluntary approaches by resource harvesters may warrant greater investments in specific topdown incentives for effective management, while more naturally amenable resources may be better served by endogenous (voluntary) forms of management, in which case the best role for management may be to remove institutional impediments to such cooperation.31

This paper has also demonstrated the usefulness of combining both reduced form and structural modeling approaches for understanding the impacts of policy innovations. By implementing the quasi-experimental tool of difference-in-differences estimation within a behavioral model we were able to relate the patterns in our reduced form models to their underlying behavioral origins. The structural model lends robustness to and aids in the interpretation of the reduced form models, while the reduced form models helped to validate the results from our structural model—an important role given the considerable structure that must be imposed to identify behavioral effects (Keane 2005). The end result is a more richly nuanced analysis of the effectiveness of a voluntary policy than would have resulted from the exclusive use of either modeling approach.

Footnotes

  • The authors are, respectively, assistant professor, School of Sustainability, Arizona State University; and professor, Department of Agricultural and Resource Economics, University of California, Davis. We are grateful to Ron Felthoven, Alan Haynie, and Kerry Smith for their helpful comments. We also thank seminar participants at the University of Washington and the International Institute of Fisheries Economics and Trade 2008 meetings. Dr. Abbott would like to acknowledge funding from the National Marine Fisheries Service/Sea Grant Graduate Fellowship in Marine Resource Economics (#NA07RG0320).

  • 1 Throughout this paper we use purposefully general language about the identity of vessels due to federal regulations that prevent the disclosure of information on specific vessels or companies.

  • 2 These results do depend somewhat on a large number of total agents (100 in Barrett’s simulations). Smaller numbers of agents allow for more effective punishment of free-riders such that agreements with high rates of participation can sustain relatively high degrees of abatement compared to the noncooperative alternative.

  • 3 The pooling of bycatch information through Sea State may also generate direct benefits by reducing the costs of bycatch abatement. Furthermore, although the information products provided by the program are anonymous, when combined with observations on the grounds and informal information exchange across vessels, they may serve to increase the efficiency of monitoring and enforcement activities.

  • 4 This database is not publicly available as the data are highly confidential. An overview of the database can be found at www.afsc.noaa.gov/FMA/fma_database.htm.

  • 5 The percentage of total hauls that are species composition sampled depends on a variety of factors such as the intensity of the fishing schedule, but coverage of around 50% is typical.

  • 6 Fishing in January through March is dominated by the rock sole roe fishery and primarily limited by the bycatch of red king crab.

  • 7 We also estimate all our reduced form equations with halibut catch as the dependent variable (available from the authors on request) and find no significant qualitative differences in our results.

  • 8 Efforts to establish the exact dates early nonparticipants eventually joined have not been successful. We maintain the years of full participation due to the extra information they provide (particularly for our structural model) but design our analysis so that the conclusions are robust to the uncertainty concerning “treatment status” in the late 1990s.

  • 9 Given the common property incentives and the lack of any coordination mechanism to ensure one’s competitors withdraw from the fishery as well, such behavior is unlikely.

  • 10 See Wooldridge (2002) for an explanation of the difference between these two concepts.

  • 11 For instance, if individuals select themselves into the program based on the perceived benefits in terms of personal bycatch reduction, then we might expect that vessels with skippers of high skill in locating and avoiding bycatch to opt out, since the personal benefits of information sharing may be small compared to the potential losses of revealing recent fishing patterns to competitors. Conversely, vessels with lower-skilled captains may be attracted to the program for the opposite reasons. If this were the dominant form of selection we might expect an upward bias in the estimated average treatment effect of the program.

  • 12 We have elected to specify the covariance structure over Bertrand, Duflo, and Mullainathan’s preferred solutions of the block bootstrap or panel-robust White standard errors for the reason that these estimators rely upon the number of cross sections going to infinity for consistency. Such a rationale does not seem reasonable for our sample of 18 vessels with panels that vary in length from 31 to 235 weeks (with an average length of 154). Conversely, the consistency of our standard errors relies on the number of observations in each panel going to infinity.

  • 13 The censored median regression in Table 3 provides a semiparametric alternative to the censored Tobit model that is robust to these violations.

  • 14 See note 12.

  • 15 It should be noted that these estimates are the marginal effects conditional on the quantile being nonzero. The unconditional marginal effects are not available from the semiparametric estimator without imposing further distributional structure but are simply the estimated effects multiplied by the (unknown) conditional probability that the quantile is nonzero. The small number of dropped observations (the difference between “sample size” and “final sample size” in Table 3) suggests that the difference is immaterial over much of the sample (particularly for the larger quantiles).

  • 16 We should note that the standard errors utilized for the quantile and probit regressions are subject to the critique of Bertrand, Duflo, and Mullainathan (2004) in that they fail to account for serial correlation. We do not address this issue due to the difficulties involved in accounting for serial correlation in long-panel censored quantile estimation. However, our z-statistics should exhibit an upward bias in the presence of positive serial correlation, so that failing to account for it actually strengthens our findings in that even under conditions most amenable to finding a treatment effect for 1995– 1997 we are unable to do so.

  • 17 We thank an anonymous reviewer for this insight.

  • 18 The estimates from the structural model support this conclusion as well. The post–Sea State dummy variable for the control group does not significantly differ from zero, suggesting that evidence for strong behavioral spillover effects is weak (see Table 5).

  • 19 These targeting allocations are assigned according to the methods utilized by fisheries managers and are primarily based on the dominant species in the haul.

  • 20 John R. Gauvin, personal communication, 2008.

  • 21 The use of spatial choice models in fisheries is well established, with papers including those by Eales and Wilen (1986), Dupont (1993), Larson, Sutton, and Terry (1999), Holland and Sutinen (1999, 2000), Smith and Wilen (2003), Smith (2005), Curtis and Hicks (2000), Curtis and McConnell (2004), Hicks and Schnier (2008), and Haynie (2005).

  • 22 We estimate parameters individually, a valid approach given the length of our panels (from 287 to 2,824 choice occasions per vessel), rather than use the common (but less flexible) approach of the random parameters logit model.

  • 23 Heterogeneity of scale across cross sections can give the false impression of heterogeneity in parameters and speaks to the relative predictability of the choice behavior of vessel captains from the model covariates (Swait and Louviere 1993). However, it is rarely explicitly considered in the economic literature outside of the literature on the combination of revealed and stated preference data (Adamowicz, Louviere, and Williams 1994), perhaps due to the identification challenges presented by short panels.

  • 24 Nonparticipants have full information from 1999 forward due to their certain membership at this point.

  • 25 It is important to note that catch/revenue expectations for a particular site can differ at a point in time due to the differences in vessels’ signals—this despite a shared rule for processing these signals.

  • 26 We are forced to truncate our sample at 1994 due to inconsistencies in data reporting that make the use of earlier data problematic for our structural model. However, due to the high resolution of the data we have ample pretreatment observations.

  • 27 Each vessel has its own set of shadow costs due to the heterogeneity of the marginal utility of revenues. The median effects reported in Table 5 all correspond to a single vessel. The relative magnitudes of the marginal shadow costs for the median case are preserved for all other vessels due to the fact that heterogeneity arises solely through the “scaling” effect of β.

  • 28 John R. Gauvin, personal communication, 2008.

  • 29 John R. Gauvin, personal communication, 2008.

  • 30 There is evidence that non–Sea State vessels had higher total bycatch of halibut in 1995–1997 due to the larger scale of their fishing. However, this finding no longer holds from1998 onward. Furthermore, regressions of the bycatch rates of Sea State members on the number of nonparticipant vessels active in the fishery in a particular week found no linkage between the two.

  • 31 This is, of course, a restatement of the normative implications of the Coase theorem. However, it specifically focuses on the importance of exogenous (i.e., naturally predetermined) transaction costs to cooperation in the determination of the appropriate balance between formal and informal allocations of property rights.

References