Abstract
Empirically identifying effective resource management strategies is challenging with many concurrent regulations. We focus on two common regulations in a spatial common pool resource experiment that involves extracting two different types of resources. Pooled or individual-specific limits regulate harvest of a protected resource, which co-locates with the desirable resource. The experiment design mimics the extraction of target species and protected bycatch in commercial fisheries. We find three key results without other regulations. Desirable resource harvests are lower under pooled than individual limits; information sharing increases desirable resource harvests with individual limits, but exacerbates moral hazard under pooled limits.
1. Introduction
Government regulation of common pool resources is the norm in developed economies, but there is still significant disagreement as to the “optimal” policy design. The concept of the tragedy of the commons has long been well known to economists. This issue became more widely understood outside of economics after Garret Hardin’s seminal work was published in Science (Hardin 1968). Since then, economists, philosophers, and policy makers have focused much time and energy on how to best manage common resources (and property) to avoid this overuse.
Our focus is to further analyze the dynamics of government extraction limits, information sharing, and moral hazard in a common pool resource setting. We set up a laboratory experiment in which participants interact in a stochastic environment to find and extract a desirable resource—one that earns money and has no extraction limits—while avoiding a protected resource—one with no value and binding extraction limits. This most closely resembles commercial fishing with valuable target species (desirable resource) and heavily regulated bycatch or constraining species (protected resource). Although this article concentrates on commercial fishing, we believe our findings could have implications for government intervention in other resource environments.1
Many U.S. fisheries are subject to a complicated combination of regulations governing the fish harvest. The complexity is particularly acute in mixed stock or multispecies fisheries where relatively abundant species comingle with other species that exist in “weak” stocks. Historically, regulators have used an array of indirect and direct methods to reduce the effect of fishing effort on weaker stocks while allowing harvest of relatively abundant species. Gear restrictions and time/area closures are commonly used to alter the selectivity of fishing effort while specie, or species-complex output quotas constrain harvest more directly (see Warlick, Steiner, and Guldin 2018 for a discussion the evolution of management in the multispecies Pacific groundfish fishery). Our study considers a representative management regime for multi-species fisheries where harvest is regulated through the following measures: (1) limited entry (fishers must hold a limited entry permit to participate), (2) output quotas defined for target (relatively abundant) species, and (3) output quotas for co-occurring constraining species that are assumed undesirable, not actively targeted, and harvested incidentally in the pursuit of target species.2
Output quotas for target and constraining species are generally determined by the regulator based on a total allowable catch (TAC). In limited-entry fisheries, these quotas represent a harvest right that may be allocated to individual fishers or groups of fishers. Individual transferable quota (ITQ) programs are a popular method in contemporary fisheries management for allocating species-specific harvest rights to individual fishers. An alternative to ITQ allocation schemes is allocating the harvest rights to groups of fishers or fishing sectors. Under such a sector TAC allocation scheme, fishers in the group share a TAC and group’s total output for each species is constrained accordingly.
When species-specific TACs are imposed as hard caps (Abbott, Haynie, and Reimer, 2015), all fishers in the fishery or group are “shut down” and prevented from further extraction until the end of the regulatory period if the total harvest of a regulated species exceeds the TAC. One prominent example of hard caps in multispecies fisheries is the Georges Bank Cod Fixed Gear Sector in New England, which originally divided their allocation of cod into monthly TAC governing all members of the sector in 2006, but as of 2010 now allocate individual quota to each member (Holland and Wiersma, 2010). Another example is the Bering Sea groundfish trawl fishery discussed in Abbott and Wilen (2010), which was governed by TACs for both target species (multiple flatfish species) and prohibited species (Pacific halibut) before 2008. The TAC for prohibited species frequently prevented fishers from fully using the target species TAC because fishing activity in the sector was shut down once the sector exhausted its prohibited species TAC.
When the TAC is divided into ITQs, individual fishers own the right to their allocated quantity of catch and may individually be shut down if they exceed that allocation. While ITQs reduce moral hazard, they provide no insurance to individual fishers against a single, accidental, high-volume haul of a protected species, which could be financially devastating. One example of a fishery governed by ITQs is the West Coast Groundfish Trawl Catch Shares Program.3 This program requires fishers to own an amount of ITQ equal to or greater than their actual extraction level for a variety of different species—including target species and bycatch species (protected resource). Whereas species quotas in most sectors of the West Coast groundfish fishery are tradable, empirical evidence from Holland (2016) suggests that many fishers are reluctant to part with their ITQ for bycatch and hence the market for ITQs is limited and inefficient.
Although Wilson (1990) suggested that fishers might join a cooperative to share information about the location of target species, cooperatives in West Coast groundfish fisheries have functioned more like insurance products, helping fishers mitigate the risk of unexpected bycatch events. Holland and Martin (2019) detail three prominent risk-pooling cooperatives operating in West Coast groundfish fisheries: the Pacific Whiting Conservation Cooperative, the Whiting Mothership Cooperative, and the voluntary risk pool in the shoreside whiting sector. These cooperatives can reduce the risk of an accidental large haul of bycatch, which might shut down a particular fisher for the remainder of the season. Participants in a fishing cooperative and fishers operating under a pooled sector TAC face similar incentives, and moral hazard becomes an issue. In this scenario, moral hazard involves fishers overharvesting the constrained species because their personal cost of doing so is much lower than the group’s social cost.
Holland and Jannot (2012) analyze by-catch extraction in the West Coast limited-entry groundfish fishery and look specifically at risk-pooling cooperatives in the shoreside nonwhiting sector of the Pacific groundfish fishery, including the California Groundfish Collective. They make a number of recommendations: (1) fishing cooperatives might actually reduce the incentives facing fishers to avoid bycatch, (2) fishers should be cautious about joining fishing cooperatives, (3) fishing cooperatives should be designed to reduce adverse selection and moral hazard (and existing cooperatives appear to have attempted to do so by including rules on fishing behavior), and (4) information sharing might improve bycatch avoidance. Coming to a similar conclusion, Evans and Weninger (2014) present a theoretical model of searching for fish, allowing for information sharing and fishing cooperatives. Their model indicates minimal information sharing among commercial fishers, and they conclude that fishing cooperatives may not improve outcomes. Haskell, Mamula, and Collier (2019) find positive spillover effects among commercial fishers in the Pacific Coast groundfish fishery, particularly for those operating out of the same port, which may be the result of shared information. Similar results from Felthoven, Lee, and Schnier (2014) and Lynham (2017) suggest that fishers gain benefits from peers, but the empirical techniques in these papers cannot reveal the mechanism through which the spillovers provide benefits.
The challenge in identifying the empirical effect of fishing cooperatives is that they include multiple rules (e.g., limitations on fishing location, forced information sharing) on fishing behavior and regulation of multiple different species.4 Table 1 provides an overview of the complex web of regulations existing in a few of the fisheries sectors managed by the Pacific Fisheries Management Council and the North Pacific Fisheries Management Council. This highlights the challenge facing researchers trying to isolate the impact of any one regulatory mechanism on fishing behavior. For example, the Pacific Whiting Mothership Sector Cooperative receives 100% of the whiting ITQ allocated from the sector. The cooperative allocates quota to vessels individually for target species (whiting), but the quota is pooled across vessels for bycatch (constraining) species. Vessels operating in the cooperative must share information on catch locations and could face fines for exceeding their pro rata share of the pooled bycatch quota. In addition, the cooperative agreement includes restrictions on night fishing, precautionary closures of past bycatch hotspots, and other measures to reduce the risk of exceeding allowed target and bycatch limits.5 Empirically estimating the effect of the information sharing mandate separately from the effect of the pooled bycatch quota and other restrictions is virtually impossible. It would be particularly useful to isolate the efficacy of specific mechanisms under different fishing environments and quota regulations. This article focuses specifically on identifying the extent to which shared information can benefit commercial fishers under pooled and individual quota limits. Many of the fishing cooperatives that have developed over recent years include a requirement for fishers to share information, mostly through a third-party service that provides maps with bycatch extraction locations.
Abbot and Wilen (2010) can evaluate the effect of information sharing on bycatch extraction of fishers in the Bering Sea/Aleutian Islands management areas because a subgroup of fishers voluntarily chose to share information on bycatch locations. The fishery under study was managed using common pool TACs for target and bycatch species. By comparing fishers who chose to share with those who chose not to share, they find that participants in the information-sharing program actually increased halibut bycatch extraction rates over time relative to nonparticipants. This is consistent with moral hazard but counter to the bycatch extraction findings from other information-sharing systems discussed in Gilman, Dalzell, and Martin (2006), including extraction of a different species (red king crab) by the same fleet over the same period, using the same information-sharing program. Abbot and Wilen (2010) conjecture that these conflicting findings result from different spatiotemporal distributions of the constrained species. In a separate study of fishery habitat protection, Holland and Schnier (2006) note that the performance of different regulatory measures (e.g., individual property rights versus rotating area closures) likely depends on the distribution, growth rate, and spatial migration of the relevant species. Similarly, Abbot, Haynie, and Reimer (2015) find that introducing ITQs incentivizes West Coast groundfish fishers to alter various behavioral margins (e.g., moving quickly away from by-catch areas and reducing night fishing) that beneficially change the composition of catch and bycatch in their hauls, although the costs and returns to such adjustments differ with the spatial and temporal distributions of the relevant species.
Controlled experiments offer the ability to separately identify how individuals might react to these different rules, different spatiotemporal distributions of resources, and the pooled nature of the bycatch quota (or TAC limits). In an early experiment designed to analyze behavior in a limited-entry common pool resource environment, Walker, Gardner, and Ostrom (1990) find that individuals over-appropriate from the common pool resource, and this problem increases with access to more capital. Similarly, Walker and Gardner (1992) show that subjects eventually deplete a common pool resource, even when they might temporarily choose more sustainable harvests. A plethora of other experimental studies find significant improvements in appropriation of common pool resources from allowing subjects to communicate and/or use costly punishment (see Ostrom 1990; Ostrom, Walker, and Gardner 1992; Hacket, Schlager, and Walker 1994; Isaac and Plott 1981, among others; see Ostrom 2006 for a survey of this literature).
Many of the common pool resource experiments previously discussed use a static design in which subjects are given tokens that can be invested in two markets: one providing a fixed return, the other (the common pool resource) providing a return that depends on the number of tokens invested by all participants. This design allows researchers to isolate the effect of subject decisions when subjects have complete control over the amount of the resource to appropriate. However, it does not allow researchers to understand how individuals make decisions in harvesting environments that require costly search effort to locate spatially stochastic resources.
A new line of experimental research now uses spatial and/or temporal resource dynamics to investigate resource extraction in common pool settings. Kimbrough and Vostroknutov (2015) introduce a dynamic common pool resource experiment where the resource regrows at a rate dependent on the amount of the resource remaining after extraction in the previous period. They find that a combination of a high resource regrowth rate and the presence of enough rule-followers are necessary to sustain the resource over time. Along similar lines, Janssen et al. (2010) employ a novel experiment where subjects must move around on a computer grid (spatial dynamics) to locate and collect resources in an environment where the resource regrows at rates dependent on the resource density at that location (temporal dynamics). They find that costly punishment is no longer effective in an environment with complex spatial and temporal resource locations. However, communication still increases group performance in this setting (with and without punishment). Janssen (2013) and Janssen, Tyson, and Lee (2014) add to the literature using similar experimental designs with spatial and temporal dynamics. These papers vary combinations of available information and communication among participants, finding an important role for both in maintaining sustainable resource levels.
Our experimental design builds on these complex settings, adding an additional complication: two different types of resources. One of the resources is abundant and unregulated,6 and the other resource is constrained and regulated with binding extraction limits. We attempt to examine how information sharing affects fishing behavior and how it might do so differently across individual and shared limit environments. Importantly, our design places a high density of the regulated resource in locations with high densities of the abundant resource. This is similar to what is known about the spatial correlation between some target species and bycatch species in fisheries (e.g., Wilson 2009; Ward et al. 2015). To effectively test the effect of these different rules, we must exclude other (potentially complementary) features of fisheries, such as collective action, social preferences, sanctioning, or punishment (e.g., explicit peer pressure). For example, Carpenter and Seki (2011) find that fishers in Japan with greater levels of conditional cooperation (based on experimental measures) are more productive in their professional fishing activities. Although excluding these other features might limit the applicability of our results to specific field situations, it also allows us to identify the unique effect of other institutions. Future work could extend the experimental design to allow for additional features.
Our experimental results show that significant moral hazard exists when participants share a pooled limit for extracting a protected resource compared with when the same combined quantity is equally divided into individual extraction limits. This manifests in diminished collections of the desirable resource for participants in the pooled limit scenario. Interestingly, information sharing increases productivity in the individual limit setting but has no effect on productivity in the pooled limit environment. This validates earlier empirical findings (Holland and Jannot 2012) that fishing cooperatives result in significant moral hazard and thus should be combined with additional regulations. Our findings also suggest that policy makers and common pool resource appropriators might want to better facilitate (or perhaps require) information sharing on highly constraining, regulated species locations. Additional empirical and experimental research in this area could be extremely valuable.
2. Experiment Design
Our experiment is designed to compare within-subject treatments and between-subject treatments to identify the effect of different forms of information sharing and contrast individual versus pooled resource extraction limits. A session includes participants searching and collecting resources for multiple periods under three different information-sharing regimes, followed by a postexperiment survey. Some sessions include an individual resource extraction limit, and others included a pooled resource extraction limit. Finally, we vary the order of the information-sharing regimes in an attempt to average out any effect of learning across treatments.
Game Play
Participants are placed into groups of four on a computer screen with a 20 × 20 grid of cells. Each cell contains a token. Each token contains some amount of a desirable resource and some amount of an undesirable resource.7 The participant does not know (with certainty) the amount of each type of resource contained in a token until after they collect it. The participants can move around the screen by using the arrow keys on a standard computer keyboard. They need to press the space bar to collect a token and find out how much of the desirable and undesirable resource is contained in that token. Once the token in a specific cell on the grid is collected by a participant, there will no longer be a token available in that cell for any participants to collect for the remainder of that period. The participant’s location on the screen is represented by a yellow circle (avatar). The participant’s field of vision only includes the cell on which their avatar resides and the four adjacent cells. Thus, participants only know another participant’s location if they are on adjacent cells at the same time.8 If this occurs, the participant would see a blue circle in an adjacent cell. The vision looks like the screen shot in Figure 1, panel A (although participants’ screens were in color), where the center of the circle represents the participant’s current location. The field of vision shows the presence of dark grey diamonds (green on the actual screen), representing tokens, in all four adjacent cells. However, the field of vision does not extend beyond the immediately adjacent cells. The upper right corner of the screen shows the quantities of desirable and undesirable resources the participant has collected so far in the round. The program used to run this experiment interface was modified from open-source software available at http://commons.asu.edu/, which was introduced in Janssen et al. (2010).
Experiment Screen Shots
Participants have 60 seconds in a period to collect as much of the desirable resource as possible, without exceeding the limit on collection of the undesirable resource. Collection of the desirable resource is incentivized, with every 2,500 units of desirable resource collected worth $1. Under one environment, each individual is subject to a limit of 100 units of the undesirable resource. A participant’s avatar is frozen after collecting 100 units of the undesirable resource, and they can no longer collect tokens during that period. In the other environment, each group of four participants is subject to a pooled limit of 400 units of the undesirable resource. The avatars for all group members are frozen after the cumulative collection of the undesirable resource for the group reaches a total of 400 units, and no members can collect any more tokens for the remainder of the period.
Resource Location
The desirable and undesirable resource locations are strategically chosen for this experiment. Most tokens on the grid contain at least some small quantities of the desirable and undesirable resources. However, there are two rectangular areas with tokens that contain very high quantities of the desirable resource, each of which are dimension 9 × 8 or 8 × 9 grid spaces. We randomly choose two of the four quadrants in the 20 × 20 grid to locate these high-density desirable resource areas and then randomly place them within those 10 × 10 quadrants.9 Arbitrarily located in each high-density desirable resource area is a 4 × 4 square of tokens that also contains very high quantities of the undesirable resource. Figure 2 provides a heat map of one example of a distribution of desirable and undesirable resource densities across the grid. The left panel provides information on one distribution of desirable resources across the grid, while the right panel shows the associated distribution of the undesirable resource. The token collected in a given cell contains both the quantity of desirable and undesirable resource shown in Figure 2 (e.g., the token collected in the top left corner of the grid contains 112 units of desirable and one unit of undesirable resource.)
Heat Map of Desirable and Undesirable Resource Distributions
Note: The left panel provides information on one distribution of desirable resources across the grid. The right panel shows the associated distribution of the undesirable resource across the grid. The token collected in a given cell contains both the quantity of desirable and the quantity of undesirable resource shown. For example, the token collected from the top left corner of the grid contains 112 units of the desirable and one unit of the undesirable resource. The high-density desirable and undesirable resource areas remain the same locations across all periods of an information-sharing regime, with only small changes to specific quantities. The location of these high-density desirable and undesirable resource areas do change for each of the three information-sharing regimes played during the experiment session.
These high-density resource locations remain constant (except for small deviations in value) through every period in the same information-sharing regime (described below), but they change across information-sharing regimes. Given the monetary incentive to collect the desirable resource and limit the collection of the undesirable resource, participants face a constrained optimization problem quite similar to groundfish fishers. Also similar to many types of groundfish, the undesirable resource (bycatch) that participants (fishers) want to avoid is spatially correlated with the desirable resource (target species). Holland (2010) shows that individual limits are likely to increase efficiency when bycatch is “common and fairly predictable”; however, these individual limits are less effective when bycatch is “highly uncertain and rare.” Our design does not mimic these two extremes; rather, we consider the bycatch (undesirable resource) to be common but distributed in uncertain magnitudes. This distribution is similar to that documented by Abbott, Haynie, and Reimer (2015) for Pacific halibut in Bering Sea–Aleutian Islands nonpollock groundfish trawl fishery, where they highlight that protected Pacific halibut hotspot locations are unpredictable and co-mingled with target, but the bycatch collection can be minimized by avoiding certain areas.10
A map with the quantity of the undesirable resource collected in each cell is shown to the participant after every period of the experiment. Each session of the experiment includes three different types of information-sharing regimes. The information included on the map varies across these three regimes. Under the first regime, no information sharing is allowed, and the map only includes the quantities of undesirable resource collected by the participant at each location. Figure 1, panel B, provides an example of one such map. Participants have 30 seconds to view the map between periods. The order of the second and third information-sharing regimes vary across sessions to account for possible learning across regimes. One of these regimes includes voluntary information sharing. Before each period, participants are asked if they wish to share information on the amount and location of their collection of the undesirable resource at the end of the period. A participant who chooses to share sees a map that includes their own undesirable resource collection (quantity at each location) and that of any other group members who also elected to share. A participant who chooses not to share sees only a map with their undesirable resource collection, identical to the no-information-sharing regime.11 Choosing not to share reveals no information about one’s resource collection and gains no information about undesirable resource locations from other group members. The map includes the quantity and location of the undesirable resource collected by all group members in the forced information-sharing regime. The various combinations of pooled limits with forced or voluntary information sharing are similar to the arrangements in a variety of groundfish fisheries.12
Length of Play
Participants collect tokens during multiple periods of each information-sharing regime. The number of periods in each regime is unknown to participants and governed by a random continuation rule (see Roth and Murninghan 1978; Bigoni, Camera, and Casari forthcoming). Participants are told that there is a one in six chance of any period (after the first) being the last period of that regime. To determine the number of periods, an experimenter rolls a six-sided die. If any of the numbers one through five are rolled, the experimenter rolls the die again. If the number six is rolled, the experimenter stops rolling the die. The number of periods for a given information-sharing regime is equal to the number of times the experimenter rolls the die. This process resulted in four periods of no information sharing, seven periods of optional information sharing, and four periods of forced information sharing. Although four periods may seem short relative to the number of trips taken by a commercial fisher, the random continuation rule approximates an infinitely repeated game. The participants only know that there is an 83.3% chance than another period will begin and hence do not know that the regime will end after four (or seven) periods. For consistency in comparison with other regimes, we use only the first four periods of the optional information sharing in the empirical analysis.13
Participant Recruitment and Execution
Volunteer participants were recruited from a pool of students using SONA Systems software from a large private university in the U.S. Midwest. Participants were shown digital slides with instructions on the experiment, while a researcher read aloud from a script. The slides and the script are available as Appendix C and Appendix D, respectively. Participants were paid $5 for showing up on time and they earned an average of $16.74 (including the show-up fee) for no more than 60 minutes of their time. The payment incentive scheme was disclosed during the instructions, and participants were paid cash immediately following the experiment. In all, we ran four different types of sessions, with each session containing 16–20 participants, and each type of session was conducted twice. Table 2 summarizes the treatments (limit environment and information-sharing regime) across each session for our total of 148 participants.
At the conclusion of the experiment, each participant was asked to complete a brief questionnaire to ascertain demographic information, their propensity to cooperate and trust others to do the same, and religious and political leanings that might affect behavior in the experiment. The full questionnaire is available in Appendix A.
3. Results
Descriptive Overview of Results
The experimental design allows us to identify the effects of information sharing and pooled resource limits on the quantity of desirable resource collected and the amount of time spent harvesting before reaching the undesirable resource limit. We analyze the full sample of data, controlling for resource limit environment and presence of information sharing. Subsequent analysis focuses only on outcomes in each limit environment. Table 3 provides summary statistics from the experiment. Two things are worth highlighting from this table. First, participants were able to collect, on average, a greater amount of the desirable resource under the individual limit compared with the pooled limit on the undesirable resource. This basic finding persists even with more rigorous analysis, as presented below. Second, the personal characteristics of our participants are very similar across resource limit environments (individual versus pooled). Figure 3 depicts average units of desirable resources collected under each treatment with 95% confidence interval bars. The graph provides clear illustration of three key results: (1) Desirable resource harvests are significantly higher (under any information-sharing regime) when participants face an individual limit for the undesirable resource. The average amount of desirable resource collected is 17%, 35%, and 36% lower with a pooled limit relative to an individual limit under the no information-sharing regime, the voluntary information-sharing regime, and the forced information-sharing regime, respectively. (2) Desirable resource harvests increase when information regarding the undesirable resource densities and locations is forced or voluntarily shared in the individual-limit setting (approximately 38% higher), but (3) the returns to shared information do not exist when undesirable resource limits are pooled among participants.
Average Desirable Resource by Treatment
The results are consistent with literature documenting moral hazard in common pool resource settings.14 Moral hazard describes a scenario where individuals are incentivized to over harvest the common resource before others do. In our experiment, this manifests as less concern for avoiding undesirable resources, thus using up the pooled resource limit very quickly.15 The average time spent collecting resources before reaching the limit is 35–40 seconds in the individual setting under any information-sharing regime. In contrast, the average time drops to 20 seconds in the pooled setting without information sharing. The voluntary and forced information-sharing regimes appear to exacerbate moral hazard in the pooled environment, as groups reach the undesirable resource limit in less than 15 seconds on average. Furthermore, the moral hazard appears stronger in later periods under the pooled limit.16 We see a monotonic increase in time spent harvesting over periods under the individual limit (likely a function of learning about the spatial distribution of resources), which contrasts with the monotonic decrease in harvest time under the pooled limit. These findings validate the use of additional rules often required in fishing cooperatives and TAC fisheries, such as area closures and forced information sharing, which attempt to reduce the harvest of constrained species. These results are consistent with the empirical findings by Birkenbach, Kaczan, and Smith (2017), who use a quasi-experimental approach to show that fishing effort continues throughout the season under individual catch shares, rather than being heavily concentrated at the start of the season under fishery-wide TAC regulations.
If pooled resource limits (and pooled limits with shared information) do not create the proper incentives for sustainable resource extraction, do other mechanisms achieve this goal? In particular, the policy objective (consistent with the Magnuson-Stevens Fishery Conservation and Management Act) is to restrict collection of undesirable resources without severely reducing desirable resource harvests. Our results indicate that voluntary or forced information sharing under individual resource limits produces more sustainable resource collection (higher desirable resource harvests under the undesirable resource limit). Information sharing allows participants to better isolate areas with high-density desirable and low-density undesirable resources. However, the careful resource collection only pays off when individuals control their own undesirable resource limit. It is important to note that our experimental design did not allow participants to develop systems of monitoring or punishment, which Holland (2018) shows to be effective at reducing moral hazard. These additional forms of collective action would complicate the identification of the effect of pooled resource limits and information sharing, which were the primary goal of this study. Future work could incorporate these options to provide a more complete picture of fisheries management.
Regression Analysis
We formalize these results in a random effects regression model, shown in equation [1]:
[1]
The dependent variable represents units of desirable resource collected by participant i, in period p, regime r, and experiment session s.17 The quantity of desirable resource collected is a function of the quantity of undesirable resource collected, Uiprs, as well as the limit environment and information-sharing regime. The indicator variable RPiprs is equal to one when the undesirable resource limit is pooled and zero for the individual limit. The voluntary and forced information-sharing regimes are presented by the indicator variables 
and 
, respectively, with no shared information being the omitted category. We allow the effects of information sharing to differ based on the limit environment by including interactions for each information-sharing regime with the pooled limit indicator. The regressions additionally control for the period number in a given information-sharing regime (ρp),18 the order in which participants faced the information-sharing regimes (Ors), and whether it was the first or second experiment session of a given set of treatments (Srs).19 In some specifications, we also control for a vector of individual characteristics, Xis, based on postexperiment survey responses. The error term, εiprs, allows for group-participant specific random effects. Participants are assigned to new groups of four players each information-sharing regime. We effectively treat each participant as a new player every time they change groups to allow for individual variation in the unobserved heterogeneity under different group dynamics. Appendix Tables E3 and E4 show similar results using only participant-specific random effects and participant fixed effects. We cluster the standard errors by experiment session to account for any idiosyncrasies in the classroom conditions (e.g., time of day or a question asked during instructions).
Desirable Resource Collection
Table 4 reports the main set of regression results with the amount of desirable resource collected as the dependent variable. The first two columns include all sessions, the third and fourth columns focus on sessions run in the individual limit environment, and the fifth and sixth columns report results under the pooled limit environment. In each pair of regression results, the first column provides a parsimonious model and the second adds individual-specific controls.
We find an additional unit of undesirable resource is associated with approximately five additional units of desirable resource collected (Table 4, column (1)). This is an artifact of the co-location among undesirable and desirable resources in the experiment design. The association between undesirable and desirable resource collection is slightly stronger in the pooled limit environment shown in column (5). By contrast, we find no significant relation between undesirable and desirable resources collected in the individual limit setting in column (3). Together these findings suggest that targeting areas with high-density undesirable and high-density desirable resources is beneficial, particularly in the pooled limit environment where moral hazard arises. However, the lack of a positive association in column (3) indicates that it is equally beneficial to engage in sustainable resource collection (i.e., the more difficult task of locating areas of high-density desirable resource that do not also contain high densities of the undesirable resource) when individuals control their own undesirable resource limit.20
Information sharing of any type increases average desirable resource collection by more than 500 units (approximately one-third of the sample average), and there is no significant difference between forced and voluntary information sharing. The positive effects of information sharing only occur in the individual limit environment, as indicated by the interaction terms in column (1). The loss in magnitude and statistical significance for the information-sharing coefficient estimates in column (5) further confirm that there is no benefit to sharing information in the pooled limit environment. This may be the result of moral hazard increasing over periods in the pooled limit setting and counteracting any knowledge gains.
Participants learn about the distribution of desirable and undesirable resources throughout the experiment. Average desirable resources collected increases steadily with each period in the full sample and even more strongly in the individual limit setting, illustrating the profitability of knowledge about resource density spatial distributions. However, we find no evidence of substantial learning between treatments (resources relocate between treatments), as the order effect in which participants saw the information-sharing treatments is statistically insignificant and close to zero in all regressions.
In Table 4, columns (2), (4), and (6), we control for additional individual characteristics. These include gender, year in school, academic unit, race, ethnicity, and nativity. We also control for two latent factors derived from an exploratory factor analysis of 12 survey questions regarding an individual’s propensity to cooperate and trust others to do the same, as well as religious and political leanings.21 Interestingly, we find that women collect fewer desirable resources on average, as do more punishment-oriented individuals. We consider the possibility that survey responses reflect some degree of endogeneity if participants who collect fewer desirable resources (lower payouts) are primed to want retribution after the experiment. That said, all of the key results regarding the effects of pooled resource limits and information sharing remain consistent (and in some cases stronger) after accounting for these additional individual characteristics, which validates the random nature of our experimental design and recruiting.22
These results are consistent with the findings in Abbot and Wilen (2010), where participants in a voluntary information-sharing program (Sea State) are not able to reduce their halibut bycatch extraction rates, relative to a group of fishers who chose not to participate. However, our results conflict with the conclusions in Gilman, Dalzell, and Martin (2006), who found reduced bycatch rates for a different species (red king crab, which has a different spatiotemporal distribution) using the same information sharing program. A useful extension of our experiment would be to consider the effects of pooled limits and information sharing under different spatiotemporal distributions of the resources, specifically the degree of correlation between desirable and undesirable resource locations and the extent of resource migration over time.23
Time Spent Harvesting
Table 5 reports a set of regression results using the same basic model as the results presented in Table 4, except here the dependent variable is the amount of time spent harvesting before reaching the undesirable resource limit. The first two columns include all sessions, the third and fourth columns focus on sessions run in the individual limit environment, and the fifth and sixth columns report results under the pooled limit environment. In each pair of regression results, the first column provides a parsimonious model and the second adds individual controls.
Column (1) shows that the average time spent harvesting is 15 seconds shorter in the pooled limit environment compared with the individual limit environment. In addition, forced and optional information sharing further decrease the amount of time harvesting by another six seconds in the pooled limit environment. Interestingly, information sharing does not alter the average time harvesting in the individual limit setting. This suggests that sharing information about undesirable resource locations may exacerbate the moral hazard in the pooled resource setting. Most fisheries (or fishery sectors) using pooled resource limits also include area closures or move-on rules, likely enacted to minimize this moral hazard.24 Last, the average time spent harvesting decreases monotonically over periods for the pooled limit environment, and increases monotonically over periods under individual limits, as discussed previously. Our results are consistent with Haynie, Hicks, and Schnier (2009), who find evidence that moral hazard increases over the regulatory period in the Alaskan flatfish fishery.
Columns (3) and (5) confirm these results by showing that the information sharing regimes (forced and optional) cause a six-second decline in time spent harvesting in the pooled limit environment. These coefficient estimates are statistically significant at better than the 1% level in the pooled limit environment, while they are statistically indifferent from zero in the individual limit environment. Again, the results remain consistent after controlling for additional individual characteristics and latent factors describing an individual’s propensity to cooperate and trust others to do the same in the second column of each section in Table 5.
Distribution Effects
The presence of government intervention in markets has the potential to alter not just average outcomes, as discussed in Tables 4 and 5, but also the distribution of resources. Our results suggest that the amount of desirable resources collected is greater for the average participant in the individual limit setting than it is for the average participant in a pooled limit environment. We run a two-sample Kolmogorov-Smirnov (K-S) test of the equality of the entire distribution of desirable resources collected across the individual and pooled limit environments. The null hypothesis of equality is rejected at the 1% level of significance. In addition, a K-S test of the null hypothesis that the distribution of desirable resources collected under the pooled limit is smaller than the distribution for the individual limit is also rejected at the 1% level of significance.25 Thus, it appears that the collection of desirable resources is greater throughout the distribution (and not just at the average) for participants operating under the individual limit relative to those operating under the pooled limit.26
Similarly, the distribution of desirable resources collected under the individual limit are greater when combined with optional information sharing or forced information sharing, relative to no information sharing. K-S tests reject the null of equality between forced information sharing and no information sharing and between optional information sharing and no information. However, a K-S test fails to reject the null of equality between the distributions of desirable resources collected under forced versus optional information sharing (in both the pooled and individual limit settings).27 These results suggest that an individual limit setting, in general, and an individual limit setting with information sharing (both forced and optional) allows high and low-performing participants to collect a greater amount of the desirable resource.28
We also find that those who choose to voluntarily share information tend to have performed slightly worse in the prior period, as shown by lower average desirable resources collected and less average time spent harvesting.29 This is consistent with more “skilled” or “successful” individuals preferring to keep profitable information private. Those who make an initial period decision to share are weakly more likely to be women, and weakly less likely to be punishment-oriented individuals. Other than prior period success, gender, and a punisher personality, there are no systematic average differences in the propensity to share information across participants.30 We find no difference in the average amount of voluntary information sharing between the individual and pooled limit environments as show in Appendix Table E6.
Undesirable Resource
We focus the analysis primarily on desirable resource collection because the undesirable resource quota limit binds in almost every case.31 Thus, we have very little variation in undesirable resource quantities. Instead, we highlight the fact that more desirable resources harvested under the undesirable limit indicates more sustainable resource collection behavior. We do consider the distribution of undesirable resource harvests under the pooled limit environment and different information-sharing regimes.
There is very little variation in group average undesirable resource collected. In fact, the group average does not differ by more than a standard deviation across any of our treatments. The group standard deviation of undesirable resource collected is very small in the individual environment. This is expected, given that it is optimal for all participants to continue harvesting resources until they hit the 100-unit undesirable resource limit, which almost always binds in the allotted time.32 By contrast, we see considerable variation in undesirable resource collection in the pooled limit environment. Some group members are harvesting well beyond the average 100 units per member, and others collect far fewer undesirable resources.33 This suggests that not all players fall subject to morally hazardous behavior. The findings are consistent with prior experimental literature suggesting that communication and costly punishment could help alleviate moral hazard by allowing some members to pressure others into considering group welfare (Ostrom 2006).
4. Conclusion
This stochastic, spatial, common pool resource experiment analyzes the behavioral response to two institutional rules commonly used in fisheries management. Introducing a protected resource (in addition to a desirable resource) that can be extracted from the commons allows us to examine management practices that have not previously been studied (to our knowledge) in the experimental literature and are difficult to separately identify with empirical data. We obtain three significant results.
First, desirable resource harvests are much larger when participants face an individual limit for the protected resource, as opposed to a pooled limit. This finding is robust to the inclusion of a host of control variables and is driven primarily by a reduction in the time participants are able to spend collecting resources before the pooled limit is reached. It is important to note that increasing desirable resource harvests in this environment is coveted. We are assuming an abundant quantity of the desirable resource and only want to limit the collection of the protected resource, which is co-located with high-density areas of the desirable resource. Thus, we confirm in the lab setting, consistent with field studies, pooled limits and fishing cooperatives (on their own) are likely subject to significant moral hazard. That moral hazard manifests as rapid collection of resources without regard for the binding limit on the collection of the protected resource.
Second, sharing information about protected resource densities and locations (both forced and voluntary sharing) leads to increased desirable resource harvests in the individual limit environment. This suggests an important role for information sharing in the management of common pool resources. Our results indicate that merely providing a mechanism for information sharing could be beneficial, whether implemented as collective action or by a governing body, when used in conjunction with individual quota.
Third, returns to shared information are not present when participants face resource limits that are pooled among participants. Although this practice is common among existing fishing cooperatives and as a regulatory mechanism in some fishery sectors, our results suggest that information sharing may not have much benefit when paired with pooled resource limits, absent any other regulatory restrictions. However, mandatory information sharing is paired with other regulations (e.g., area closures, spatial fishing plan preapprovals) in every fishery management arrangement in the West Coast of the United States that we are aware of. These additional regulations are included almost certainly to avoid the kind of moral hazard found in our pooled limit setting.34
Note that we obtain our results under a specific spatiotemporal distribution of resources in the experimental environment and without open communication among subjects. It could be that information sharing is more or less valuable when the constrained species is less prevalent (rare event) or less spatially correlated with the target species. In addition, communication could complement the type of information sharing explored in this analysis, or communication could provide no additional benefit. Further empirical research and management practice experimentation is warranted to better facilitate fisheries productivity while preserving protected species.
Footnotes
Appendix materials are freely available at http://le.uwpress.org and via the links in the electronic version of this article.
↵1 E.g., one could think of agricultural farmers trying to maximize their crop yield (desirable outcome) without exceeding a regulation on water pollution from pesticide use (undesirable outcome). Further research is necessary to test whether the institutions in this experiment would apply in other settings.
↵2 In many mixed stock fisheries, abundant species can be caught along with less abundant species. These less abundant species are often called constraining species. Because the allowable catch limits for these species are set very low, the quota price is often beyond the means of many fishers. These species can constrain the harvest of more abundant species as fishers who have exhausted their quota of the constraining species will generally adopt fishing strategies to mitigate the risk of encountering the constraining species. This avoidance behavior can severely constrain the fisher’s ability to fully use the quota for abundant species.
↵3 See http://www.westcoast.fisheries.noaa.gov/fisheries/groundfish_catch_shares/.
↵4 For a large part of the contemporary history of U.S. fisheries management, regulators were prevented from enacting individual fishing quota programs (Brinson and Thunberg 2016). Lacking the scope to directly introduce individual accountability, regulators addressed moral hazard by pairing common pool output constraints with time and area closures, gear restrictions, and other restrictions on fishing effort. When the moratorium expired and individual fishing quota (IFQ) programs were permitted, many regulatory bodies responded by enacting IFQ programs. In most fisheries, many of the regulations predating IFQs, such as time/area closures and gear restrictions, were retained.
↵5 See Holland and Martin (2019) and Fraser (2019) for further information on the Pacific Whiting Mothership Sector Cooperative.
↵6 Fisheries management also regulates target species in practice, but the quota limits are generally nonbinding for these species. The California Groundfish Collective notes in their 2017 annual report that collective members and the rest of the IFQ fleet used 18% and 26%, respectively, of their allowable catch of nonwhiting target species. After including the heavily targeted whiting species, the entire IFQ fleet (collective and noncollective members) still only used 62% of their allowable allocation of target species (Kauer, Rubinstein, and Oberhoff 2019, 12).
↵7 The undesirable resource represents a protected resource. We chose this terminology so as not to introduce any biases or prime participant behavior based on personal views of environmental conservation.
↵8 We believe that only showing adjacent cells is more ecologically realistic. In many fisheries, the fishers operate in an area where they would rarely see other fishers and they cannot know the abundance of fish outside their immediate vicinity. Wilson (1990) describes and models fishers’ search for fish, which is consistent with our experimental design. Jannsen (2013) uses a very similar experimental design, in which the size of the field of vision available to participants is varied. He finds that the size of the field of vision has a significant effect on participant behavior. Given these findings, we suggest that limiting vision to the adjacent cells allows us to focus on other regulatory mechanisms in an environment that is ecologically similar to actual fisheries.
↵9 The participants were given written instructions that provided information about the desirable and undesirable resource locations. These instructions are available in the Appendix and are almost exactly what is written here.
↵10 Although these hotspot locations are unpredictable, “significant gains in bycatch avoidance can be achieved by vessels moving relatively short distances after encountering such a hot spot” (Abbott, Haynie, and Reimer 2015, 182).
↵11 If only one participant in a group of four chooses to share, then each participant will only see a map with their own undesirable resource locations (identical to the no-information-sharing regime). We require the decision to share before resource collection to match existing fishery policies that exist in six of the seven fishery arrangements shown in Table 1. Although fishers can choose whether to join a fishing cooperative that requires information sharing or to join other vessels in voluntary information sharing, these decisions take place prior to fishing. We do not allow information free-riding in the voluntary information-sharing regime. This could have induced more of our participants to share than would have if information free-riding were allowed.
↵12 The California Groundfish Collective requires members to share information on bycatch locations (and follow specific fishing plans), while sharing bycatch risk by pooling resource quota (Kauer, Rubinstein, and Oberhoff 2019), which is similar to our pooled limit with forced information-sharing treatment. Although fishers voluntarily choose to join the California Groundfish Collective, the decision conflates the issue of information and risk pooling. Similar to our pooled limit with voluntary information treatment, Abbot and Wilen (2010) describe a voluntary effort by the Alaska “head and gut fleet” to avoid premature fishery closures by sharing bycatch information. This information-sharing agreement exists in the regulatory environment of a fleet-wide shared TAC for bycatch. In this program, each vessel in the fleet provides an independent entity (Sea State) with their fishing locations and outcomes. Sea State compiles the data and distributes information on the spatial distribution of bycatch encounters to the entire fleet. By contrast, the New England Groundfish Sectors operate as self-regulating entities, each with their own system for allocating quota and implementing other management practices. Many of these sectors divide a sector-wide TAC into individual quota limits but offer some risk sharing by asking members to set aside a portion of their individual quota for a group insurance pot to cover unexpectedly large bycatch encounters. Some sectors, such as the Sustainable Harvest Sector, implement formal bycatch hotspot information sharing, but most sectors are tight-knit communities that rely on informal communication and peer trust to mitigate moral hazard (Holland and Weirsma 2010; Holland et al. 2014). These are like our individual limit environment with forced or voluntary information sharing.
↵13 Results are similar when using all seven periods from the optional information-sharing regime.
↵14 See Walker and Gardner (1992) and Janssen et al. (2010) for examples in the common pool resource experimental literature. Holland and Jannot (2012) and Evans and Weninger (2014) provide evidence from fisheries cooperatives, and Abbott and Wilen (2009) develop a predictive model showing that a common-pool quota system (similar to TAC) results in reduced target species harvest.
↵15 Appendix Figure E1 illustrates the difference in average time spent harvesting before reaching the limit.
↵16 As shown in Appendix Figure E2.
↵17 Each experiment session contains three regimes. The first is always no shared information, and we vary the order of voluntary and forced information sharing across the second and third regimes. For a given experiment session, the regimes are either all individual limit or all pooled limit environments.
↵18 Although regimes varied in the number of periods played to avoid any last-period effects, we use only the first four periods of each regime for our analysis for consistent comparison across regimes. Thus, we drop periods 5–7 in the voluntary information-sharing regime. Results are similar when including these periods.
↵19 A “treatment” refers to a specific limit environment and order of information-sharing regimes.
↵20 We ran the same regressions without including the amount of undesirable resource collected as an explanatory variable. All other coefficient estimates were qualitatively similar. The results are available on request.
↵21 The survey questions are available in Appendix A. We conduct exploratory factor analysis and retain two latent factors based on the Kaiser criterion. We use the Bartlett method to predict latent factors for each individual. See Appendix B for tables of eigenvalues, the scree plot, and rotated factor loadings from the exploratory factor analysis.
↵22 We ran fixed effects panel regressions with the sample split, similar to Table 4, columns (3) and (5). The results of these fixed effects models are qualitatively similar to what is reported in Table 4 and are available in Appendix Table E3.
↵23 Our high-density desirable and undesirable resource locations remain constant across all periods in a given information-sharing regime. The high-density desirable and undesirable resource areas are randomly relocated before the start of each new information-sharing regime.
↵24 See Table 1 for more information on the combination of regulations used on the West Coast. In the table, we encompass move-on rules with time/area closures. The difference is that move-on rules effectively close specific fishing areas (sometimes to all vessels, sometimes to an individual vessel) in real time based on current bycatch extraction data.
↵25 The test statistics for the null hypothesis of equality of the two distributions and the null hypothesis that that the distribution in the pooled limit setting is smaller than the distribution in the individual limit setting are 0.1972 and −0.1972, respectively. The p-values for both tests are less than 0.001.
↵26 Figure E3 in Appendix E shows a kernel density plot of the two distributions. The distribution of desirable resources collected under the individual limit appears to be an almost continuous rightward shift of the distribution of desirable resources collected under the pooled limit.
↵27 The test statistics for the null hypothesis that the distribution in the no information sharing regime is smaller than the distribution in the optional information sharing regime and the null hypothesis that the distribution in the no-information-sharing regime is smaller than the distribution in the forced information sharing regime are 0.2072 and 0.2270, respectively. The p-values for both tests are less than 0.001. The test statistic for the null hypothesis that the distribution in the optional information sharing regime is equal to the distribution in the forced information sharing regime is 0.0625 and the corresponding p-value is 0.593.
↵28 This can also be seen visually in Appendix Figures E4, E5, and E6 where we plot cumulative density functions of the amount of desirable resource collected in the pooled limit environment versus the individual limit environment under the three information sharing regimes (no information sharing, optional information sharing, and forced information sharing, respectively).
↵29 These results can be found in Appendix Table E5, panel A.
↵30 These results can be found in Appendix Table E5, panel B. We focus only on the initial decision to share information in panel B because individual characteristics are constant and future sharing decisions may be influenced by group dynamics.
↵31 Participants hit the binding limit on undesirable resource collection in approximately 90% of all periods in the individual environment. In the pooled limit environment, there was only one period in which one group does not reach the binding undesirable resource limit before the allotted time expired.
↵32 The only variation beyond 100 units in undesirable resources harvested in the individual environment arises from randomness in how much undesirable resource the last token collected contains relative to how close the participant is to the limit (e.g., 10 units collected when the participant already has 99 units, compared with when she has 91 units of undesirable resource).
↵33 The group average undesirable resource collection across treatments is shown in the left panel for Appendix Figure E4 and the within-group standard deviation across treatments is shown in the right panel.
↵34 See Table 1 for a more complete picture of the complex regulatory environment across different fishery sectors.
References
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.