4

Sampling and Statistical Estimation for the Angler Intercept Survey

INTRODUCTION

One important component of recreational fishing surveys is the intercept survey. As noted by the National Research Council’s (NRC’s) 2006 study, the intercept survey for the Marine Recreational Fisheries Statistics Survey (MRFSS) used a stratified multistage sampling design, but the (point and variance) estimation procedures did not account for the complex design features, potentially leading to biased estimates. Also, the previous MRFSS design did not adequately cover night fishing. Finally, that design used the concept of “alternate site,” which did not support the calculation of well-defined inclusion probabilities. To address the 2006 report recommendations, the intercept survey underwent a complete redesign in terms of both sampling and estimation procedures. The current methods used in the Access Point Angler Intercept Survey (APAIS) for the Marine Recreational Information Program (MRIP) are a vast improvement over the previous sampling and estimation procedures and reflect state-of-the-art methods in survey sampling.

This chapter discusses the initiatives implemented to address the recommendations, the current committee’s evaluations of those initiatives, and recommendations for future studies and improvements.

DATA COLLECTION AND SAMPLING DESIGN

The target population for the intercept survey consists of the marine recreational angling fishing trips that are taken during a given 2-month data-collection period, or wave. For these purposes, a “trip” is generally considered to be each time an angler engages in fishing and then subsequently leaves a particular site.

Between 5 million and 20 million fishing trips are estimated to occur during a given wave. The newer APAIS is conducted for 2-month waves in 16 states bordering the Atlantic Coast and Gulf of Mexico (excluding Texas and Louisiana), as well as in Puerto Rico (Breidt and Chromy, 2016). Depending on the state, the number of waves ranges from three to six.¹ The objective of the intercept survey is to estimate the catch per unit effort (CPUE), by species, catch category (harvested or released dead or alive), and fishing mode (charter boat, party boat, private or rental boat, shore fishing) of anglers participating in marine recreational fishing in the study states.

The two main data-collection tasks of the APAIS are counts of completed angler fishing trips and angler-intercept interviews. The angler interviews are obtained by intercepting marine recreational anglers at shore (SH), private/rental boat (PR), and charter boat (CH) access points. Sampling in the party (or head) boat (HB) mode includes riding on the boats during fishing days (no overnight fishing trips are sampled). The interviewers collect demographic data for the anglers and ask anglers about their fishing day, including any fish released or already filleted. The interviewer also examines the catch for species identification and enumeration and may weigh and measure the catch.

The current APAIS sample design is a multistage stratified design. The population is first stratified based on site group (beach-bank, artificial structures, charter boat, private/rental boat, and the special offshore group), state, wave, region, month, type of day (weekday and weekend), and 6-hour blocks within a 24-hour day (2AM-8AM, 8AM-2PM, 2PM-8PM, 8PM-2AM, 11AM-5PM). The 11AM to 5PM interval, which corresponds with peak fishing activity, was added in 2014, due to lower activity in the early morning and late afternoon/evening resulting in a small number of completed interviews (Breidt and Chromy, 2016). The inclusion probabilities are adjusted to account for the overlap with the time intervals 8AM to 2PM and 2PM to 8PM to avoid double counting. This interval corresponds to peak fishing activity. Before 2014, the APAIS used the fishing mode as a stratification variable. In 2014, the APAIS transitioned from fishing mode–stratified sampling assignments to mixed boat–mode and shore-mode sampling, and in 2016 it transitioned to fully mixed-mode sampling to ensure adequate sampling in all modes of eligible fishing anglers (Breidt and Chromy, 2016). The current APAIS design uses the site group as a stratification variable to ensure sufficient sample size for all modes of eligible fishing anglers. This decision was made to improve productivity in terms of number of completed interviews.

The first-stage sampling frame is from a spatiotemporal list of site-days (defined as a combination of a fishing site or cluster of sites and a day), which is constructed from the public-access fishing site register (SR). Field observations are entered into a web application upon return from the field. If a site closes per-

___________________

¹ See http://www.st.nmfs.noaa.gov/recreational-fisheries/Surveys/coverage.

manently (e.g., out of business, destroyed by a storm), it is retired but remains on the SR. A primary sampling unit (PSU) is a site-day within a given 6-hour time slice stratum (Breidt and Chromy, 2016). The SR is a database of all known public-access sites from Maine to Mississippi and Puerto Rico with fishing activity. Each site on the SR is assigned an identification number, which remains unchanged over time. The SR is updated regularly by field observation. The site status is coded as retired, making the site ineligible for sampling.

Each PSU consists of either a single site or two sites. Each site is assigned a fishing pressure that corresponds to a prediction of the mean number of angler fishing trips that an assigned interviewer would encounter based on the site’s most common form of fishing (e.g., shore fishing, charter boat; see Table 4.1). The assigned pressure for a given site is time-interval dependent. That is, a given site may be labeled as a high-pressure site for one interval block (e.g., 11AM-5PM) and labeled as low pressure for another interval block (e.g., 8PM-2AM). Only sites with a pressure of “3” or less can be clustered with one additional site as long as the driving time between them is less than 60 minutes and they are located in the same county. These are referred to as “two-site assignments.” Undercoverage is an important issue in the current APAIS, because the first-stage sampling frame contains almost exclusively public-access sites. Thus, private sites cannot be selected in the sample as they have zero probability of inclusion. If the proportion of private sites is large and the behavior of private sites differs from that of public sites in terms of catch, estimators of total catch (see Chapter 2) may suffer from large biases (e.g., Särndal et al., 1992; Särndal and Lundström, 2005).

TABLE 4.1 Pressure Category and Corresponding Size Measure

SOURCE: Breidt et al., 2012.

At the first stage, a sample of site-days is selected from the spatiotemporal list according to a probability proportional-to-size (PPS) sampling design. That is, the inclusion probability of a given site is proportional to its fishing pressure. Table 4.1 shows the pressure and associated size measure used in the PPS procedure. The latter is expected to lead to efficient estimates if the PSU inclusion probabilities are (approximately) proportional to the PSU catch. However, PPS procedures tend to be vulnerable to the presence of influential units when this “proportional relationship” is not satisfied. In other words, influential units may be the high-pressure sites with low catch or low-pressure sites with high catch. These units tend to make the classical estimators unstable in the sense that they have a large variance. One way to check whether or not the PPS procedure is appropriate is to plot the pressure measure against the total estimated catch for each site-hour combination and to check if the relationship is linear and goes through the origin. If this plot is not approximately linear through zero, then using the pressure measure as an indication of fishing activity may result in inefficient estimates.

To meet operational constraints (e.g., interviewers are not available on the selected dates), a rejective sampling procedure (Fuller, 2009) was developed, whereby a very large number of samples is first selected through PPS sampling, and only those samples satisfying the operational constraints are retained. Then, a sample is selected through simple random sampling of the samples that satisfy the operational constraints. The inclusion probability of a PSU at the first stage is then approximated through Monte Carlo methods (Breidt and Chromy, 2016); that is, the inclusion probability of a PSU was obtained as the proportions of samples that contained that PSU among the samples satisfying the operational constraints. For two-site assignments, the inclusion probabilities are adjusted to account for the overlap so that double counting does not occur. Without additional information on the sample allocation used at the first stage, the committee was not able to assess its effect on the efficiency of the estimates.

Depending on the type of fishing (shore or boat), there are one or two additional stages of sampling. Sampling of shore fishing is based on a two-stage sampling design, where the secondary sampling unit (SSU) is an angler trip within each PSU. Sampling of boat fishing is based on a three-stage sampling design, where the second stage consists of selecting boat trips within a selected site-day and the third stage consists of selecting angler groups within each boat trip selected at the second stage. The angler groups are the tertiary sampling units (TSUs). When possible, field staff try to achieve a census within a PSU. That is, on a given site-day, all the anglers present on the site are interviewed. However, a census is generally not possible because of refusals, language barriers, and missed eligible participants (see below). A census is never possible for a two-site cluster because the interviewers, intercepted SSUs, and intercepted TSUs are treated as if they were selected by simple random sampling without replacement at the second and third stages, even if this is not the case in practice. Therefore, the validity of

the estimates in terms of bias depends on how well simple random sampling without replacement serves as an approximation of the actual but unknown design.

An assignment consists of a time interval, a cluster of fishing sites with activity in at least one mode, the order in which those sites are to be visited (in the case of a two-site cluster), and the date and time when the cluster will be visited. Assignments are sampled, and field staff are assigned to a date/time assignment.

The APAIS instrument is relatively short (NOAA, 2016b). In the previous MRFSS design, the field sampling procedures provided survey staff with considerable flexibility. In addition, interviewers could obtain interviews from alternate fishing modes and/or sites to increase productivity and minimize the survey costs (see Breidt et al., 2012). As a result, the inclusion probabilities were difficult to compute, making design-based type estimation a difficult task. As recommended in the 2006 report, the current design does not allow samplers to decide where and when to conduct an interview. Instead, each assignment corresponds to a fixed time interval. For two-site assignments, the sampler is told the order of sites—making it relatively straightforward to compute the inclusion probabilities that will be used in the estimation procedures. Following the recommendation of the 2006 report, the alternate sites have been eliminated. Finally, unlike in the previous MRFSS design, there is no upper limit on the number of interviews. Samplers attempt to obtain the largest possible number of completed interviews for a given assignment. Over the years, the number of completed interviews has varied from approximately 6,800 to 25,800 for a given wave in the states where the APAIS is conducted. Sampling in the HB mode includes riding on the boats during fishing days. The interviewers collect demographic data for the anglers and ask anglers about their fishing day, including any fish released or already filleted. The interviewer also examines the catch for species identification and enumeration and may weigh and measure the catch. For at-sea sampling on headboats, the interviewer remains on the boat throughout the trip, collecting data on the catch as long as fishing continues.

The APAIS uses face-to-face interviews, which enables the interviewer to clarify unclear questions and to gain the respondent’s confidence. There will be variations in interviewer effects due to interviewers’ experience, training, and skills. These factors may affect the nonresponse and measurement errors.

The responsibility for training interviewers is shared by the National Marine Fisheries Service (NMFS) and its data-collection partners: Atlantic Coastal Cooperative Statistics Program, Gulf States Marine Fisheries Commission, and the Atlantic and Gulf state agencies. NMFS must approve all training programs. However, it seems that NMFS has limited control on the actual implementation of interviewer training and testing as it is currently conducted. Details about the training program can be found in the APAIS Statement of Work (2016) and Chapter 5 of this report.

In addition to variation in training, various other interviewer effects and interviewer-related variance affect data quality. Face-to-face interviews, in con-

trast to mail surveys, can introduce additional sources of error and bias through interviewer and interviewer-respondent interactions. Some interviewers are more skilled than others, for example, which can impact inter-interviewer variance—a potential source of nonsampling error. Inter-interviewer differences can be related to similarities and differences (e.g., ethnicity) between interviewers and respondents. Thus, certain characteristics of the interviewer may affect the willingness of the angler to be interviewed or respond to certain questions—although likely not to a significant extent with the APAIS given the nonsensitive nature of its questions.

The interviewer effect will vary based on interviewer experience, training, and adherence to protocols. Experience will affect interviewer confidence, proper survey pace, and methods for gaining access to respondents. Also, experience can inform interviewers’ abilities to build rapport with respondents and respond to questions about the MRIP in a positive and informative manner, which is particularly important given the confusion and distrust that exists among some constituents. Furthermore, some interviewers are more skilled than others at asking questions, which influences interviewer error and variance. Interviewers may vary in their abilities to probe deeper during open-ended questioning. For example, question 12 of the North Carolina APAIS asks anglers whether most of their fishing effort on the current trip took place mostly in the Atlantic Ocean or “other.” Interviewers are expected to code “other” responses according to Department of Marine Fisheries waterbody codes, which requires both knowledge on the part of respondents and familiarity on the part of interviewers to elicit valid and reliable responses.

Positive interactions between interviewers and respondents are critical not only for collecting high-quality data, but also for promoting the program and ensuring angler participation. Anglers talk to other anglers and will share their experiences with the program via their informal social networks and social media. The goodwill arising from a positive experience will yield future benefits. The attitudes of interviewers will impact interview success in terms of unit nonresponse, item nonresponse, measurement error, and ultimately inter-interviewer variance.

Of course, interviewer behavior and how it is perceived and interpreted is important. For example, interviewers may be perceived as idle—and wasting taxpayer money—if they have extra time between interviews or are assigned to a site during bad weather when no or few anglers are present. Interviewers could mitigate possible misperceptions by engaging in alternative activities during idle periods, such as cleaning the area or conducting qualitative interviews with available anglers. In addition, the interviewer’s appearance contributes to perceptions. Dress, the outward display of government symbols (e.g., government symbols on shirt), and the display of official government identification will all influence the interview outcome. Professionalism is important, but its degree and nature should be tailored to the state and regional setting in which the interviews take place. Although not a true interviewer effect, weather (e.g., temperature) can affect a

respondent’s willingness to be interviewed; some studies have found that high temperatures are associated with higher nonresponse rates and higher measurement error (Cohen and Krueger, 2016).

MISSING DATA

APAIS estimates are subject to missing data. For example, in the Atlantic and the Gulf of Mexico, approximately 20 percent of data are missed because not all eligible anglers complete the interviews (Breidt and Chromy, 2016). Four sources that lead to missing data are (1) refusals, which occur when an angler or party refuses to participate in the survey; (2) mid-interview refusals, which occur when the angler or party does not answer some key questions; (3) language barriers that occur when the interviewer and the angler speak different languages; and (4) missed eligibles, which occur when interviewers are busy with other anglers or parties (NMFS Statement of Work, Access Point Angler Intercept Survey). NMFS believes that the most common language barrier situation arises when a Spanish-speaking angler is approached by an English-only speaking interviewer. If an agent is too busy to interview all anglers, the selection of anglers can introduce bias if anglers are selected by party type (e.g., large groups or non-English speaking anglers are avoided). Bias can be avoided if there is no selection according to party type and if all anglers that were not interviewed are counted. A different situation exists when anglers refuse an interview, because this can be indicative of other issues (e.g., illegal catches). The committee is not aware of the percentage of missing data attributable to each of these four reasons.

Interviewers should attempt to collect some paradata, which are variables about the data-collection process (e.g., Kreuter, 2013). Paradata may include variables such as the number of anglers in the party, their gender and approximate age, and the fishing mode. These variables may be incorporated into the estimation procedures, which may help reduce the potential bias due to missing data.

WEIGHTING AND ESTIMATION

Prior to 2006, the estimation procedures did not account for the features of the complex survey design. As noted in the 2006 report, the validity (in terms of bias) of the estimates relied heavily on some (implicit) model assumptions. With the new sampling design, the inclusion probabilities and the sampling weights are well defined, making the use of design-based type procedures possible. The weighting process is therefore completely new since the 2006 report. The base weights at the first stage are defined as the inverse of the cluster inclusion probabilities. For a cluster consisting of a single site, the base weight of the site is equal to the base weight of the cluster. For a cluster consisting of two sites, a weighting methodology was developed accounting for the duration of the visit of each site. When a census is possible the weight of an angler group is equal to

the weight of the cluster. When a second stage and a third stage of sampling are involved, the weights are equal to those that would have been assigned had the SSU and the TSU been selected according to simple random sampling without replacement. Thus, the validity of estimators in terms of bias depends on whether or not this assumption holds. Based on the weighting system described above, an estimate of CPUE (defined as the estimated number of fish caught or discarded per angler on a single saltwater fishing trip) is readily obtained, where the CPUE is defined as the ratio of the estimated Catch to the estimated Effort, and Effort is the total number of single-day angler trips spent saltwater fishing (see Chapter 2). The CPUE, produced by the APAIS, is computed at the state/fishing mode/wave/fishing area level.

An estimate of the variance of the point estimates is obtained through Taylor expansion procedures by assuming that the clusters at the first stage are selected with replacement. In practice, the clusters are selected without replacement within stratum. As a result, it is anticipated that the variance estimators will exhibit an upward bias (slightly larger than necessary), although the affect is likely to be small if the first-stage sampling fraction is small (i.e., the percentage of PSUs selected from the total available for the study is small [Särndal et al., 1992; Wolter, 2007]). In the APAIS, the first-stage sampling fraction is small, because the total number of PSUs in the population is very large in comparison to the number of selected PSUs. Therefore, it is expected that the variance estimator will perform well in terms of bias.

DISCARD ESTIMATION

NMFS estimates that a substantial quantity of recreational catch is discarded rather than retained and landed. Discards include fish released relatively unharmed and those that are dead or will not survive. In 2014, approximately 60 percent of the national recreational catch was discarded before landing due to regulation or angler choice (NMFS, 2015). Of that total, approximately 63 percent occurred in the Southeast region of the United States (Table 4.2).

TABLE 4.2 Estimated Total Recreational Catch and Percentage Released at Sea for the Entire United States and the Southeast Region, 2014

SOURCE: MRIP presentation to Fishery Management Council Coordination Committee, 2015.

TABLE 4.3 Estimated Recreational Harvest and Discards of Red Snapper in the Gulf of Mexico Region, 2010-2015

SOURCE: MRIP data query, July 9, 2016.

An understanding of the magnitude of the discard issue for an individual species basis can be gained using data from red snapper recreational fisheries in the Gulf of Mexico region (Table 4.3). For this fishery, the percentage of the total catch discarded at sea is estimated to range from 0 percent to 99 percent, but there is little verification of the estimated quantities of discarded fish. This leads to a somewhat illusory precision in these estimated quantities. Estimated discard mortality associated with these fisheries varies by region, depth, and release method. For recreational red snapper fisheries, estimated discard mortality rates range from 10 percent to 22 percent, meaning that mortality from discarding at sea rivals that from removals landed by recreational anglers.

From a programmatic perspective, the primary elements needed for assessment and management are the number or biomass of fish that are caught and landed and the number or biomass of those that are caught and released but which

subsequently die due to capture effects. For fisheries management purposes, the mortality of these discarded fish is either assumed or estimated via a discard mortality rate (DMR) and subsequently converted to biomass using an estimated weight for the discarded fish (usually an average weight). The cumulative sum of these elements is the total mortality required for assessment purposes; in MRIP terms, A + B1 + DMR(B2).

The MRIP estimates the number of fish released in the recreational fisheries (B2) in several stages, depending on region. In almost all regions, the primary method of estimating the number of discarded fish is through the APAIS, and the basic data are either self-reported by anglers or reported in mandatory logbooks in the Northeast Region for-hire program and in the Southeast Region charter boat logbook program. Attempts to validate angler reports of the quantities and size composition of discarded fish have been limited; thus, the estimates of discarded numbers and sizes are subject to considerable uncertainty (Benaka et al., 2014). There is also considerable uncertainty about the estimated mortality of these discarded fish. Although not uniquely responsible for determining discard mortality, the MRIP is responsible for determining discard rates. A major review of current knowledge, ongoing research, and data gaps conducted by NMFS in 2013 (Benaka et al., 2014) catalogued the mortality estimates currently used in fisheries management by species and regions. For recreational fisheries, the mortality estimates range from 0 percent to 100 percent, depending on species, region, and research basis for the mortality of released fish. These estimates can be strong and influential assumptions for both stock assessment and fisheries management. The problem of unknown or highly uncertain estimates of both the quantity of discarded fish and the discard mortality rate for many species is common throughout the United States. The committee notes that research into the correct DMRs for use in fisheries assessment and management is not the responsibility of the MRIP, but the committee also notes that close coordination between the MRIP and the agencies that are directly responsible for estimating DMRs for recreational fishery releases would be valuable to the stock assessment process.

The issue of total mortality estimation, including discard mortality, was also noted in NRC (2006). Although significant improvements in the APAIS have been made in the MRIP, uncertainty in the estimation of mortality and biomass of fish discarded in recreational fisheries remains a hurdle for management of many species. Because the quantities of discarded fish that die can be of similar magnitude to those that are landed for many species, the implications of this uncertainty for both the determination and management of Annual Catch Limits are profound. Uncertainty and the possible downward bias in the estimates of total mortality for species that is associated with the current framework will result in underestimates of the underlying productivity of stocks and misspecification of reference points for fisheries management. Management of Annual Catch Limits may also be compromised by inaccurate accounting of total removals.

Estimation of total mortality can be improved through

• Validation of angler-/logbook-reported discard quantities
• Comprehensive estimation of the size composition of discarded fish
• Additional research on and application of DMRs for released fish
• Additional research on statistical properties of discard mortality estimates

The committee understands that the MRIP is not solely responsible for all of these efforts. The MRIP should be concerned with the validation of discarded numbers and their size composition because they directly affect the ultimate estimates of total removals. The calculation of DMRs is a broader responsibility of fisheries management. Recognizing the importance of the discard mortality issue, the NMFS undertook a national process to develop an Action Plan for fish release mortality science in 2015 (Benaka et al., 2016). The broad goals of this action plan were to

1. Support the use of planning tools such as the SMART (Benaka et al., 2016) tool to help managers, scientists, and other stakeholders determine which fish species, complexes, and/or fisheries would benefit most from improved mortality rate estimates. The SMART tool assesses and scores the impact of discarding species based on several criteria: restricted or rare status, vulnerability to exploitation, economic impact, political sensitivity and stakeholder engagement, and discard ratio.
2. Facilitate the development of improved mortality rate estimates.
3. Support effective and efficient research that leads to reduced release mortality for high-priority species, complexes, and/or fisheries.
4. Ensure that improved mortality rate estimates are incorporated effectively into existing management processes.

These objectives pertain in part to the MRIP programs of discard estimation, technology evaluation, research funding, and communication/outreach with the angler community. Therefore, there is considerable opportunity within the MRIP for improvements in discard mortality estimation and its use in stock assessment.

The mortality resulting from the discard of recreationally caught fish is the ultimate metric of importance to stock assessment and management. The production of more accurate estimates depends on not only a more comprehensive understanding of discard mortality from future research initiatives, some within NMFS and some within academic institutions, but also involves MRIP activities and mandates. We consider the primary MRIP responsibility to be the production of reliable estimates of the number and size composition of discarded fish. Estimation of the mortality associated with these discards will require coordinated research on DMRs with other components of NMFS and partner agencies. However, verification of self-reported discards is an important role for the MRIP, and

additional efforts using electronic and human-based observations are required. For example, electronic monitoring could provide a cost-effective and less intrusive option for verification than human observers, and directly addresses the responsibility of the MRIP to estimate quantities of discards rather than quantities and DMR.

COMBINING THE FISHING EFFORT SURVEY (FES) AND THE APAIS

Estimates of Effort in the FES and Total Catch in the APAIS are obtained by combining both surveys. On the one hand, the FES estimates of Effort are based solely on in-state residents and do not account for out-of-state residents. On the other hand, the APAIS collects information on angler trips for both in-state and out-of-state residents. As a result, it is possible to determine an estimate of the proportion of out-of-state residents from the APAIS. This estimated proportion is then used to correct the estimated Effort produced by the FES.

An estimate of Total Catch is obtained as

Total Catch = Effort × CPUE,

where Effort is the total number of single-day recreational angler trips spent saltwater fishing and is imported from FES, and CPUE is the estimated number of fish caught or discarded per angler on a single saltwater fishing trip. As mentioned above, the CPUE, produced by the APAIS, is computed at the state/fishing mode/wave/fishing area level. Therefore, the estimator of Total Catch can be viewed as a stratified ratio estimator, with estimation of effort from the FES multiplied by CPUE (number of fish caught from APAIS divided by the APAIS estimation of effort). The estimated Effort from the FES is not expected to suffer from significant undercoverage because of the new sampling frame and adjustments for out-of-state angler trips; therefore, it is hoped that this estimator will help to ameliorate the problem caused by the absence of private sites on the first-stage sampling frame for the APAIS (e.g., Särndal and Lundström, 2005). The underlying assumption is that private sites and public-access sites share the same behavior in terms of CPUE. Although unlikely to be observed in practice, it is not currently clear whether this assumption is a critical source of bias. This assumption needs to be studied. The variance of Total Catch is estimated through Taylor expansion procedures taking into account the fact that the Effort produced by the FES is an estimate (Goodman, 1960). NMFS has developed a computer program that allows data users to obtain domain estimates for any domain of interest. However, if the domain is too small (i.e., with a very sample size), then the point estimates may not be reliable. In this case, small area estimation techniques may prove useful for finer domains. The committee raised some concern that some data users may be using this program for fine-scale domains (which are those exhibiting a small sample size), resulting in unreliable estimates.

FOR-HIRE SURVEYS

Prior to release of the 2006 NRC report, NMFS recognized difficulties in obtaining catch and effort data from the for-hire fishery. The for-hire fishery consists of several components: guide boats, small charter boats often called six-packs, and larger charter and headboats. How these components are effectively sampled depends on whether they are moored at specific ports and on their hours of egress and ingress. Smaller vessels, such as guide boats, can use public launches or docks and may not operate on rigid schedules. Their catches are intended to be captured in the APAIS. Licensed vessels are sampled in the For-Hire Survey (FHS). Unlicensed vessels are intended to be identified during the onsite surveys. Larger vessels that are located at specific sites and operate on defined schedules (e.g., leave at 6AM and return at 5PM) are also sampled in the FHS along the U.S. East Coast, The Southeast Headboat Survey, the NE Vessel Trip Report, the Large Pelagic Survey, or through state programs depending on the region of the United States (Chromy et al., 2009). Anglers on for-hire vessels often may be noncoastal residents, and the random PPS sampling can be a mismatch when the vessels follow strict departure and return schedules. Therefore, their catch and effort might not be well sampled through the APAIS. The FHS was not well sampled in the MRFSS/Coastal Household Telephone Survey, and NMFS developed a separate survey for the Gulf Coast in 2000 and the Atlantic Coast in 2005. West Coast states and Texas retained their programs, and their sampling designs have been reviewed by the MRIP (Chromy et al., 2009).

The sampling frame for the effort component of the FHS is a comprehensive directory of for-hire boats, stratified by vessel type, state, and week. The MRIP website states, “Data collection is conducted on a weekly basis.” The contact person for the vessel that has been chosen for the week is mailed a notice and a log sheet. Within a stratum, sampling is done as a systematic random sample without replacement from the stratified list frame (NMFS, 2014b). The respondent can FAX the report, use a toll-free telephone number, or wait to be called by NMFS contractors.

This sampling represents a stratified systematic random sample, which is well understood and has well-known variance properties. The sampling unit is

the vessel, and 10 percent of the sampling units are chosen from the frame each week (NMFS, 2014b). CPUE estimates are obtained from interviews of intercepted for-hire trips in the APAIS. Compared to the MRFSS, the improvements to this onsite survey address the concerns detailed in the 2006 NRC report. The onsite survey that obtains CPUE for the FHS is now a probability sample. The MRIP now recognizes that site data should be seen as a cluster, not as a simple random sample of anglers.

The FHS frame for Virginia through Maine includes vessels with highly migratory species (HMS) and large pelagic permits. The survey questionnaire for the FHS queries vessels that caught large pelagics and highly migratory species. A separate biweekly telephone survey is conducted using the FHS frame to estimate large pelagics taken by private boats with HMS permits.

The FHS overlaps with other surveys on the Atlantic Coast, The Northeast Fisheries Science Center’s Vessel Trip Report (VTR) program from Maine to Virginia, and the Southeast Regional Headboat Survey, as well as state logbook programs. The MRIP website states, “The VTR data are not used for preliminary wave-by-wave estimates, but they are included at the end of the year when the VTR data are most complete. For all federally-permitted charter boats and headboats, the total trips reported in the VTRs are used to produce an unadjusted number of angler trips. These boats are treated as a separate ‘VTR boats’ stratum within each for-hire boat mode. All FHS data obtained for those vessels are removed, and FHS estimates of the numbers of angler trips on non-VTR boats are re-run for each wave using the remaining FHS data. The resulting FHS estimates represent a second ‘non-VTR boats’ stratum for each mode.” Initially the VTR reports were evaluated on a yearly basis, but NMFS seeks to obtain these data bimonthly for inclusion in wave-by-wave estimates.

The 2006 NRC report recommended that the for-hire sector be handled as a separate commercial sector. However, the heterogeneity of the for-hire vessel types complicates a singular approach to estimating catch and effort. Boat size, passenger capacity, whether the boat has a regular docking site, species targeted, and permits needed for targeted species vary by region and state. States and the U.S. Coast Guard (for vessels carrying 10 or more passengers) require licensing to operate charter vessels, thus providing a potential list frame for sampling. In addition, charter vessels targeting highly migratory species must have a federal permit to enter the fishery. For-hire vessels that fish for reef fish or pelagics, among other species, must also obtain special federal permits. The advantage in having a for-hire license is the ability to mandate reporting of effort and catch as a provision of for-hire license renewal. However, with a list frame, catch is self-reported and not directly verifiable without onsite validation. Moreover, the value of these data relies on enforcement of the reporting provisions in a timely manner. The level of enforcement varies depending on the regulatory agency and the robustness of its laws.

In a pilot study of electronic logbooks in the Gulf of Mexico for-hire fleet, Donaldson et al. (2013) noted that captains who did not report regularly (or at all) were allowed to file a year-end report and to continue this practice thereafter. Without penalties for this practice, data quality can be compromised because of a lack of validation. Donaldson et al. (2013) recommended that year end–only reporting be penalized, that reporting be done weekly, and that late or missing reports be quickly identified and participants notified quickly of their noncompliance. They also emphasized the importance of field validation, such as by on-board observers depending on vessel size and cooperation, and through onsite intercepts.

Total catch estimation from the FHS mirrors that of the MRIP in that it is a complemented design. Effort is obtained from sampling or censusing a list frame of vessel captains’ business telephone numbers. Catch is estimated as CPUE from the APAIS. The proportion of unlisted boats is adjusted based on a ratio derived from APAIS intercepts of angler trips on for-hire vessels that are not on the FHS frame. Because the FHS telephone survey draws from businesses holding state and federal licenses, it should be reliable as a sampling or census frame.

CONCLUSIONS AND RECOMMENDATIONS

Conclusion: The new Access Point Angler Intercept Survey design reflects a substantial improvement of the MRFSS intercept survey methodologies.

Conclusion: The new Access Point Angler Intercept Survey design uses probability proportional-to-size (PPS) sampling at the first stage. This design is expected to lead to efficient estimates if the cluster inclusion probabilities are (approximately) proportional to the cluster catch. However, PPS sampling designs tend to be vulnerable to units whose pressure estimates are poor—high-pressure sites with low catch or low-pressure sites with high catch—and can potentially cause high variance estimates.

Recommendation: The appropriateness of probability proportional-to-size sampling should be evaluated, and alternative sampling designs should be considered if needed. For example, a stratified design (based on the site pressure as a stratification variable) may avoid very small selection probabilities, which, in turn, may lead to more stable estimates. Otherwise, methods dealing with influential values should be considered, including weight smoothing (Beaumont, 2008) and weight trimming procedures (Potter, 1990).

Recommendation: For data users requiring domain estimates at a fine level, design-based estimators tend to exhibit very large variances. To address this, small area estimation procedures should be investigated for obtaining estimates for small domains.

Conclusion: Private sites have zero inclusion probability in the sample. A ratio estimator is used to compensate for the undercoverage. However, the validity of the ratio estimator relies on the assumption that the behavior of private sites is similar to that of public access sites. Otherwise, the ratio estimator may be biased, especially if the proportion of private sites is appreciable.

Conclusion: An additional question on the Fishing Effort Survey about angler use of private or public access sites will enable stratification of the respondents in the Fishing Effort Survey into two strata: (1) anglers who have used private sites and (2) anglers who have used public-access sites. Selecting a sample from each stratum and asking the selected households about their catch will make it possible to assess differences between private- and public-access sites in terms of catch per unit effort. This will provide some insight about the quality of the ratio estimator used to obtain an estimate of Total Catch.

Recommendation: NMFS should conduct pilot studies to determine the optimal method for collecting accurate information on total catch differences between public- and private-access sites. For example, NMFS could add a question to the Fishing Effort Survey questionnaire about angler use of private sites or public-access sites. Geographic maps used to identify public-access sites within the state (see Chapter 3) could be used to distinguish public-access from private sites.

Conclusion: Missing data in the Access Point Angler Intercept Survey occur because of refusals (or mid-interview refusals), language barriers, or missed eligible anglers. Missing values may lead to biased estimators if the behavior of nonresponding anglers differs from that of responding anglers.

Recommendation: Interviewers who administer the Access Point Angler Intercept Survey should attempt to collect some paradata to help to reduce the potential bias due to missing interview data.

Recommendation: Anglers are expressing a growing interest in reporting their catches electronically (use of tablets and smartphones). NMFS should conduct a study to compare angler reporting of catch via an app with angler reporting via the traditional interview.

Conclusion: As concluded by the 2006 National Research Council committee, the magnitude and fate of fish discarded by recreational anglers remains highly uncertain. Although some technological changes (e.g., iSnapper) have been incorporated into MRIP data collection, lack of validation of discard estimates significantly contributes to the uncertainty in assessing the impact of discard mortality on stock productivity estimates and management of stock removals.

Conclusion: Initiatives by other branches of NMFS to address discard mortality estimation have not been integrated into MRIP design or operational procedures.

Recommendation: The MRIP should develop and then incorporate validation programs for the estimation of the numbers of fish discarded at sea by recreational anglers. These programs should integrate with other NMFS initiatives concerning estimation of discard mortality.

Conclusion: Recent pilot studies have demonstrated the value of using electronic logbooks to record catch and effort in sectors of the for-hire fisheries. Technological advances have reduced the costs of this equipment, increased the ease of use, and provided value-added benefits to sectors of the for-hire fleet.

Recommendation: The MRIP should expand the electronic logbook program to include most of the large charter and for-hire fleets, through outreach training in electronic logbook use and through implementation of software to run on standard tablets or smartphones.

Conclusion: During the past 10 years, there has been a substantial number of methodological studies/improvements/modifications to both the Fishing Effort Survey and Access Point Angler Intercept Survey. However, the available documentation is not always clear and up to date.

Recommendation: The MRIP should invest resources to provide organized and up-to-date documentation that describes in detail each step of the Fishing Effort Survey and Access Point Angler Intercept Survey methodologies and any changes made to them.

This page intentionally left blank.