North American Journal of Fisheries Management 26:727–741, 2006 Ó Copyright by the American Fisheries Society 2006 DOI: 10.1577/M04-146.1

[Article]

Comparing and Combining Effort and Catch Estimates from Aerial–Access Designs as Applied to a Large-Scale Angler Survey in the Delaware River JON H. VøLSTAD* Versar, Inc., 9200 Rumsey Road, Columbia, Maryland 21045-1936, USA

KENNETH H. POLLOCK Department of Zoology, North Carolina State University, Raleigh, North Carolina 27695-7617, USA

WILLIAM A. RICHKUS Versar, Inc., 9200 Rumsey Road, Columbia, Maryland 21045-1936, USA Abstract.—We used probability-based aerial
access surveys to estimate effort, catch, and harvest of American shad Alosa sapidissima and striped bass Morone saxatilis by recreational anglers in the Delaware River and upper estuary in 2002. Sampling of anglers at access points and flights over the river were conducted weekly from mid-March through October. Daily flight times were randomly selected; probabilities were proportional to the observed distribution of daily angler effort in a prior aerial
access survey (random count). Additional experimental flights were scheduled to occur at the time of day with expected peak effort (maximum count). Effort estimates derived from these maximum counts were more precise than estimates derived from the random flights, but the maximum-count observations caused bias except when the daily count expansions were based on effort distributions from the concurrent access survey. The aerial and access surveys produced similar estimates of boat angler effort and little evidence of bias, but shore anglers were undercounted in the aerial survey. We maximized the precision and minimized bias in total effort estimates by combining the estimates of boat angler effort and shore angler access. An estimated sevenfold increase in the access survey sampling effort (at nearly five times the cost) would be required to achieve the same precision in the total effort estimate produced by the aerial–access survey. Effective stratification and the use of efficient model-based estimators helped us to achieve the target precision of 20% in relative standard error (RSE) for estimated recreational catch of American shad (mean ¼ 26,885 fish; RSE ¼ 16%) and striped bass (mean ¼ 47,671 fish; RSE ¼ 15%). A single access survey during the American shad run would have required a 10-fold increase in sampling effort to achieve the same precision in estimated catch at six times the cost of the complemented surveys.

Precise and accurate estimates of catch and harvest from surveys are needed to effectively manage sport fish populations in large river systems (e.g., Matlock 1991; ASMFC 1999). A variety of survey techniques may be used to collect data for these estimates. Such techniques include counting anglers from airplanes or boats or interviewing anglers along the shoreline as they complete their fishing trips (roving–access survey) or while they are actively fishing (roving–roving survey) (Pollock et al. 1994; Lockwood 2000). However, few data are available to determine which technique or combination of techniques is optimal in terms of precision per unit of survey cost. We conducted aerial observations to count anglers (fishing effort), and we used survey clerks to collect catch, * Corresponding author: [email protected] Received August 27, 2004; accepted January 23, 2006 Published online August 28, 2006

harvest, and trip-length data during interviews with anglers at access sites as they completed their fishing trips. This type of complemented survey is referred to as an aerial–access angler survey (Pollock et al. 1994; Lockwood 2000). The data were used to determine the most efficient combination of the two techniques for estimating fishing variables associated with recreational angling of American shad Alosa sapidissima and striped bass Morone saxatilis in the Delaware River and estuary during 2002. The survey was probability based and was designed to achieve precise estimates of angler effort, catch (the total number of fish caught, including the ones released), and harvest (the number of fish kept). The study was conducted on behalf of the Delaware River Basin Fish and Wildlife Management Cooperative (the Cooperative), which includes the states of Delaware, New Jersey, Pennsylvania, and New York, as well as the U.S. Fish and Wildlife Service and National Marine Fisheries Service. The

727

V øLSTAD ET AL.

728

Cooperative had a goal to achieve estimates of total catch (number of fish) for each of the two target species with a relative standard error (RSE) of 20% or less. Aerial observations and complete trip interviews of anglers at access points along the shoreline each have advantages for estimating angler fishery statistics. Aerial survey observations may be used to estimate effort efficiently because an airplane can cover a large area quickly and personnel requirements are modest (Hoenig et al. 1993). However, aerial observations provide no data about angler catch or trip length. Furthermore, these surveys (1) may yield biased estimates if some anglers are not sighted and (2) are ineffective at night. Roving-access surveys (completetrip interviews) along the shoreline do not have any of these limitations, but conducting them is labor intensive, potentially resulting in a greater cost to achieve a given level of precision. Therefore, a combination of the two survey techniques may be most efficient for estimating catch and harvest by recreational anglers in large rivers. For aerial–access surveys, aerial counts provide estimates of angling effort (pressure), whereas data from complete-trip interviews typically are used to estimate catch and harvest rates by species and trip length. The catch by species is then estimated as the product of estimated effort and estimated catch rate (Pollock et al. 1994; Lockwood 2000). In this study, we developed alternative statistical estimation techniques that generally produced more accurate and precise estimates of angler effort, catch, and harvest than the application of current standard techniques for aerial–access designs. We then showed that the complemented aerial–access survey can achieve higher precision for a fixed cost than access surveys alone. Study Area The Delaware River is located on the East Coast of the USA and extends 531 river kilometers (rkm) from the confluence of its east and west branches at Hancock, New York, to the mouth of Delaware Bay. The survey area for this study included both tidal and nontidal portions of the river, extending 451 rkm from Delaware Memorial Bridge (Interstate 295) to Downsville, New York (Figure 1), and included 82 primarily public boat and shore access sites (as per a list compiled with input from the Cooperative and local experts). According to local fisheries managers, only a small proportion of anglers are believed to access the fishery from private docks or by walking to the water from parking lots along nearby roads (i.e., locations that would not be subject to interviews).

Methods Access point survey methods.—An access point survey was conducted to provide information on (1) the amount of time anglers spent fishing on each of their fishing trips (i.e., trip length), (2) catch and harvest of American shad and striped bass, (3) harvest per unit effort (fish/angler-hour) for each species, and (4) harvest size structure by gender. Anglers were interviewed at access points immediately after they completed their fishing trips. A sampling frame was constructed to allow the representative selection of these 82 access points from within the project area over time. In this spatiotemporal sampling frame, we defined primary sampling units (PSUs) as the combination of days available for sampling during the survey period and the points of access to the fishery (Pollock et al. 1994). Each day was divided into morning (0700–1300 hours) and afternoon (1300–2100 hours) shifts. These shifts within each PSU were the secondary sampling units (SSU). Individual anglers that completed their trip during the selected shift for the access site were the ultimate sampling units. Data from completed individual angler trips were collected from angler intercepts in each selected SSU. To achieve good sampling coverage of the angler trips across the study area through time, we employed a hierarchical stratification and restricted random sampling of PSUs. The 28 primary strata were defined by four spatial strata and seven time blocks (Table 1). The four spatial strata (Figure 1) were defined by river kilometer: (1) the estuary, which was the tidal portion of the river from Delaware Memorial Bridge (rkm 111) to rkm 214; (2) nontidal 1, from rkm 214 to Delaware Watergap Bridge (Route 80; rkm 341); (3) nontidal 2, from Delaware Watergap Bridge to Narrowsburg (rkm 467); and (4) nontidal 3, from Narrowsburg to Downsville, New York, East Branch (rkm 560). The seven time blocks were intended to improve precision in total catch and harvest estimates for American shad and striped bass by defining periods with relatively homogeneous effort, based on data from two previous angler surveys (Miller and Lupine 1987, 1996) and expert opinion. The second time block was specifically defined to capture the peak in American shad fishing activity. Within a particular spatiotemporal stratum, the access points were substratified into groups based on their predicted angler usage levels (high, medium, or low), and the days were substratified into weekdays and weekend days (including holidays) (Table 2). Some spatiotemporal strata had no access points with high usage levels; none were identified in spatial stratum 3. Recreational daily angler effort on the

ANGLER EFFORT AND CATCH ESTIMATION

729

FIGURE 1.—Spatial stratification and number of angler access points within each of four strata, as defined for a creel survey of American shad and striped bass angling in the Delaware River and upper estuary in 2002. The short, dark lines delineate strata, from Delaware Memorial Bridge (rkm 111) to Downsville, New York (rkm 560). River kilometer 214 is known locally as River Mile 133.

V øLSTAD ET AL.

730

TABLE 1.—Primary stratification (region, time block) employed in a 2002 access survey in the Delaware River. The regions correspond to the primary spatial strata illustrated in Figure 1. Date (by time block number)

Region

1 (Mar 17–Apr 20)

2 (Apr 21–Jun 8)

3 (Jun 9–Jul 6)

4 (Jul 7–Aug 3)

5 (Aug 4–Aug 31)

6 (Sep 1–Sep 28)

7 (Sep 29–Oct 31)

Estuary (E) Nontidal 1 (NT1) Nontidal 2 (NT2) Nontidal 3 (NT3)

E, 1 NT1, 1 NT2, 1 NT3, 1

E, 2 NT1, 2 NT2, 2 NT3, 2

E, 3 NT1, 3 NT2, 3 NT3, 3

E, 4 NT1, 4 NT2, 4 NT3, 4

E, 5 NT1, 5 NT2, 5 NT3, 5

E, 6 NT1, 6 NT2, 6 NT3, 6

E, 7 NT1, 7 NT2, 7 NT3, 7

Delaware River is typically highest on weekends. The groupings by usage level were based on historical information and expert opinion provided by the Pennsylvania Fish and Boating Commission, fisheries biologists from other state and federal agencies, and our team. The assigned usage levels for the access points remained fixed within each of the seven time blocks. Sampling intensity and allocation were adjusted so that greater effort was directed toward zones of the river where American shad were most abundant during each phase of their migration. Freshwater fishing for American shad begins when these fish move from the ocean, where they spend most of their lives, into the Delaware River to spawn. The timing of the runs last approximately 6–8 weeks between April and June but varies with river temperature, discharge, and other environmental factors. During the beginning of the American shad run, we enhanced sampling intensity in the lower nontidal river zones but shifted the higher sampling effort towards the upper river zones as the peak of the run moved upstream. Two stages of sample selection were used to provide interview data that represented the sport fishery across all access points over the study period from March 17 through October. In the first stage, we selected PSUs randomly without replacement from each substratum in a region–timeblock cell (Tables 1, 2). Sampling of access points in the medium- and high-usage classes was scheduled on one random weekday and one random weekend day during each calendar week; sampling of low-usage sites was conduced on one random weekday and one TABLE 2.—Secondary stratification employed in an access survey in the Delaware River, exemplified by a particular region–time block cell (the estuary, Mar 17–Apr 20) to show the substrata. Some primary strata did not have any access sites in the high-usage class. Usage class

Weekday (WD)

Weekend day (WE)

High (H) Medium (M) Low (L)

E, 1, H, WD E, 1, M, WD E, 1, L, WD

E, 1, H, WE E, 1, M, WE E, 1, L, WE

random weekend day in each 2-week period. Access points in the high- and medium-usage classes were sampled at a higher frequency to maximize the coverage of completed angler trips. A larger fraction of available days were sampled during weekend days and holidays than for weekdays because angler effort was significantly higher on nonworking days. In the second stage, one randomly selected interview shift would cover angler trips that were completed either in the morning from 0700 to 1300 hours (shift 1) or in the afternoon from 1300 to 2100 hours (shift 2). These shifts covered the daytime fishing as well as the time after sunset, which usually had the greatest fishing activity (1900–2100 hours). For each SSU (shift), we counted and interviewed anglers by type (shore-based and boat anglers) after they had completed their fishing trips. Data were collected on catch, harvest, effort (angler-hours), and fish size as well as on the demographic characteristics of anglers and other factors. Using the above design, we completed interviews of 2,353 anglers intercepted at 396 PSUs during the entire survey period (Table 3). Randomization resulted in an approximately even number of sampling shifts in the morning and afternoon. Angler interviews were conducted according to plan on all days selected in the access survey, including bad weather days. The extension of all afternoon shifts to 2100 hours covered a significant portion of the postsunset angler activity (Vølstad et al., 2003). No regular sampling was conducted between midnight and sunrise; periodic verification at high-usage sites indicated minimal fishing activity during this time. Aerial survey methods.—We scheduled an airplane to fly over the river on random days within each week to provide angler effort data. For logistical reasons, the four strata (Figure 1) were pooled into two regions that each could be covered in a 2-h flight (effective flying time over the river): region 1 encompassed the estuary and nontidal 1, and region 2 comprised nontidal 2 and 3. The sampling frame, consisting of all days in the study period, was stratified into seven periods (same time blocks as for access survey), and into weekdays

731

ANGLER EFFORT AND CATCH ESTIMATION

TABLE 3.—Sample sizes by spatiotemporal strata for primary sampling units in a creel survey of the Delaware River. Nontidal 3 had no high-usage access sites. Nontidal 2 and 3 were collapsed for estimating variables. Date (by time block number)

Region Estuary Nontidal 1 Nontidal 2 Nontidal 3 All

1 2 3 4 5 6 7 1–7 (Mar 17–Apr 20) (Apr 21–Jun 8) (Jun 9–Jul 6) (Jul 7–Aug 3) (Aug 4–Aug 31) (Sep 1–Sep 28) (Sep 29–Oct 31) (Mar 17–Oct 31) 16 14 5 4 39

27 37 18 10 92

14 16 15 3 48

17 11 9 6 43

and weekend days as substrata. We scheduled 32 regular flights (one per week) in region 1 and 29 flights (one per week after April 6) in region 2. These flights formed the core aerial survey. On each flight, aerial observers made instantaneous counts of shore and boat anglers successively within portions of the area swept, yielding a progressive overall survey count (Pollock et al. 1994). The counts were recorded separately for each of the spatiotemporal strata covered in the access survey (Table 1). The PSUs for the aerial survey were defined as single days (0700–1900 hours) within a region–time block and type of day, and the SSUs were the 2-h periods within the selected day. We employed a mix of systematic and simple random sampling to select PSUs. The weekly flights in each region were alternated systematically between weekdays and weekends. The flight start and direction (upstream or downstream) were randomly selected, and the specific weekdays or weekend days were randomly selected for each week. We included contingency aerial survey days, which we predetermined using the probability sampling design (one randomly selected day every 2 weeks) in case bad weather or logistical problems forced the cancellation of a flight. We restricted the flights in the nontidal river to stratum 1 during the first 3 weeks; it was determined by ground-truthing that the fishing activity occurring upstream from the Delaware Watergap Bridge was negligible at that time. Flights in nontidal 2 and 3 (region 2) commenced after 3 weeks. The orientation of flights relative to the river and shoreline in the tidal stratum was somewhat affected by flight restrictions imposed in the vicinity of the Philadelphia International Airport after the New York terrorist attack on September 11, 2001. In that stratum, the survey plane was prohibited from flying near the airport and was forced to fly near the opposite shoreline. Thus, counts of anglers on and in the vicinity of the airport shoreline had to be made from a greater height and distance, which may have reduced the accuracy of counts in that stratum.

15 18 18 9 60

13 15 15 9 52

16 18 16 12 62

118 129 96 53 396

For the core survey, the selection of SSUs (2-h periods) within a day was based on nonuniform inclusion probabilities that were proportional to the observed distribution of daily effort in the 1995 aerial– access survey conducted in Delaware River (Figure 2). In 1995, the distribution of angler effort throughout a typical weekday was bell shaped; peak effort was around midday, and the lowest effort occurred at sunset and sunrise. During weekends or holidays, the typical effort distribution in the nontidal river observed in the historical surveys was more skewed towards morning;

FIGURE 2.—Weekday versus weekend distribution of angler effort between 0700 and 1900 hours (time strata 1–3; x-axes) during the American shad run in the Delaware River in 1995 and 2002 in nontidal strata 1 and 2 (see Figure 1). The selection probabilities (p) used to schedule weekday and weekend flights in 2002 were based on the probability (pi) values from 1995. The intervals for the 2002 random flights were selected with pi, whereas all maximum-count flights were scheduled in the interval with maximum pi.

V øLSTAD ET AL.

732

peak effort occurred at around 1000 hours. The random scheduling tended to assign more flights (on average) to the middle of the day on weekdays and to approximately 1000 hours on weekends and holidays. We applied the same effort distribution to allocate sampling in the tidal portion of the river and for time strata 3 and 4. Hoenig et al. (1993) used similar count scheduling for estimating sportfishing effort in a roving survey. To provide information that could be used to identify the most effective scheduling method for future aerial surveys, we experimentally compared maximum count (Dauk 2000; Dauk and Swartz 2001; Lockwood et al. 2001) with the nonuniform random flights within each selected day. The maximum-count flights were scheduled during the time of day that coincided with expected peak angler activity (as determined from the 1995 survey; Figure 2). We scheduled maximum-count flights on 12 random days in addition to the regular random count flights in nontidal 1. The experimental flights were conducted weekly during the peak American shad run (temporal stratum 2). Estimating equations.—For the access survey, the variables used in the estimator formulas are defined below: ns ¼ the number of PSUs sampled in stratum s; ys,i,j,f ¼ catch (or harvest) in stratum s, PSU i, and SSU j for fishing mode f; Ls,i,j,f ¼ total hours of fishing in stratum s, PSU i, and SSU j for fishing mode f; and Rs,f ¼ mean catch (or harvest) per hour for a given species in stratum s for fishing mode f. For the aerial survey, we also defined es,i,j,f as the number of hours of fishing in stratum s, PSU i, and SSU j (where j is a particular 2-h period) for fishing mode f; this was estimated from the aerial survey by multiplying the instantaneous count of anglers in a flight interval by the interval length (2 h). The catch for a single angler trip is the sum of number of fish harvested (kept) and number of fish released. The original design was slightly modified for variance estimation by collapsing nontidal 2 and 3 into a single spatial pseudostratum (Lehtonen and Pahkinen 1994). An estimator for mean catch per unit effort (CPUE) by fishing mode (boat or shore) within a stratum is X ys;i;j;f ns ^ X : ð1Þ Rs;f ¼ Ls;i;j;f ns

This ratio estimator of catch rates is proper to provide separate estimates for boat and shore anglers based on

the respective interviews from completed trips (Pollock et al. 1994, 1997; Jones et al. 1995). The variance was computed from primary sample values only (see Cochran 1977:279; Pollock et al. 1994:42), that is, X ^ s;f L ^ s;i;f Þ2 ð^ys;i;f
R ^ s;f Þ ’ varðR

ns 2 ns Lˆ s;f ðns
1Þ

:

ð2Þ

Because the sampling fraction at the primary level was negligible (typically ,3%), the bias in equation (2) is small and tends to be slightly positive (Wolter 1985:34), thus providing a conservative variance estimate. Estimates of effort (angler-hours) for daytime fishing (0700–2100 hours) were first obtained separately from the aerial and the access surveys. A composite estimate that combines the two estimates was applied when appropriate. Estimates were provided for all anglers and separately for boat and shore anglers. The estimates for these strata were then pooled to provide a total estimate for the total sample period and separate estimates for the nontidal and tidal portions of the river. Estimating effort from the aerial survey.—A modelassisted p estimator (Særndal et al. 1992) for effort (angler-hours) in spatiotemporal stratum s and PSU i for fishing mode f, based on the aerial survey, is ^ s;i;f ¼ ^es;i;j;f ; ð3Þ E ps;j;k where the expansion factor ps,j,k is the proportion of daily angler effort that occurred during the 2-h interval in which the flight i in stratum s was conducted; k defines the set of expansion factors estimated from angler interview data that applies to a particular portion of the river and a particular period (Table 4). A similar method of defining p from daily angler activity distributions has been used in other studies (e.g., Parker 1956; Fraidenburg and Bargmann 1982; McNeish and Trial 1991; Dauk 2000; Dauk and Schwarz 2001; Lockwood et al. 2001). This p expansion assumes that the activity pattern is consistent across days in each period and geographic area. A standard design-based effort estimate for 0700–1900 hours based on the randomized flights was obtained by applying equation (3) with p, which was the actual selection probability of the 2-h flights as determined from the effort distributions in 1995. The mean daily effort in each stratum s for boat or shore anglers was estimated as X ^ ¼ 1 ^ s;i;f ; E E ð4Þ s;f ns ns the conservative estimate of the variance was

ANGLER EFFORT AND CATCH ESTIMATION

TABLE 4.—Weekday versus weekend distribution of angler effort in a 2002 creel survey in the Delaware River; the broad spatial (nontidal and estuary) and temporal (time blocks 1–3 and 4–7) were strata (1–7; see Table 1) based on access survey intercepts. Proportions of total effort in 2-h time intervals were used to extrapolate aerial counts to daily effort, but the last two rows were used to extrapolate access counts recorded by shift (0700–1300 or 1300–2100 hours) to daily effort. Weekdays Nontidal

Weekends

Time interval (hours)

1–3

4–7

1–3

4–7

1–3

4–7

1–3

4–7

0700–0900 0900–1100 1100–1300 1300–1500 1500–1700 1700–1900 1900–2100 0700–1300 1300–2100

0.06 0.13 0.15 0.16 0.18 0.18 0.14 0.34 0.66

0.08 0.14 0.14 0.13 0.17 0.17 0.10 0.34 0.66

0.10 0.13 0.16 0.18 0.19 0.16 0.08 0.34 0.66

0.05 0.08 0.13 0.13 0.17 0.23 0.20 0.34 0.66

0.13 0.19 0.18 0.16 0.15 0.13 0.07 0.50 0.50

0.12 0.16 0.16 0.18 0.16 0.14 0.08 0.50 0.50

0.15 0.18 0.18 0.15 0.16 0.12 0.05 0.50 0.50

0.09 0.15 0.18 0.17 0.17 0.14 0.11 0.50 0.50

varðEs;f Þ ’

Estuary

Nontidal

^ s;i;f
Es;f Þ2 1 X ðE ; ns ns ðns
1Þ

Estuary

ð5Þ

which was computed from the PSU values only. This simplification was necessary because only one 2-h interval (SSU) was selected from each PSU. The resulting positive bias in equation (5) is likely to be small because a fairly small fraction of PSUs were sampled (about 14% on average across strata). The total effort Es,f by fishing mode f in stratum s was estimated by extrapolating the mean daily effort to total days in each category, that is, ^ ; ^ ¼NE ð6Þ E s;f

s s;f

and the variance was ^ Þ: ^ s;f Þ ¼ N 2 varðE varðE s;f s

ð7Þ

Because of poor weather conditions, only one weekend or weekday flight was completed in some months. For weekends, this happened during May in the tidal portion of the river, and during July for both the tidal portion and nontidal 1. In October, only one weekday flight was completed in nontidal 1 and in the tidal portion of the river. For these months, we used the mean population variance for two neighboring months to impute the variance of the weekend or weekday effort. Estimating effort from the access survey.—Because the access survey was probability based and provided extensive spatial and temporal coverage, we also used the interview data to estimate fishing effort for boat and shore anglers. For each PSU i in the sample, we estimated the daily effort for boat and shore anglers by

733

extrapolating the total number of angler-hours across all anglers intercepted during the shift to the total daily period (0700–2100 hours), which we accomplished by using the respective effort distributions to derive representative adjustment factors. A model-assisted p estimator for total effort (angler-hours) by fishing mode for stratum s is then XL ^s;i;f ^ s;f ¼ Ns E ; ð8Þ ns ns ps;k where ps,k is the proportion of daily angler effort in the shift covered for PSU i in stratum s; the set of expansion factors, k, applies to a particular type of day, period, and portion of the river (last two rows in Table 4). A design-based estimate of effort was obtained by replacing ps,k with the actual SSU selection probability (p ¼ 0.5). A conservative estimate of the variance of equation (8) was obtained from equations (5) to (7). As for the aerial survey, we could not use the usual formula for two-stage sampling variance because only one SSU (shift) was sampled within each selected day. Final estimates of effort and respective variances for the tidal and nontidal portions of the river were obtained the usual way by summing over strata. The precision of all reported estimates were estimated by the RSE (Jessen 1978), which is the standard error divided by the estimate. Composite estimates of effort.—For boat anglers, the final effort (best estimate) was estimated by combining the aerial and access surveys. Let Eˆ1,s and Eˆ2,s be the independent estimates of effort for stratum s from the aerial and access surveys, respectively. A composite estimator to estimate the total effort across the two surveys (Rao 2003) is ^ s ¼ /E ^ 1;s þ ð1
/ÞE ^ 2;s ; E ð9Þ where / is a weight between 0 and 1. The optimal weight was obtained by minimizing the variance of equation (9) with respect to / (Rao 2003): ^ 2;s Þ varðE : ð10Þ /opt ¼ ^ 2;s Þ ^ 1;s Þ þ varðE varðE We assumed that the effort estimates from the two surveys were independent; however, this was not strictly the case because the daily effort from 2-h flights was estimated by using expansion factors from the access survey (Table 4). The expansion factors were based on a large number of interviews and were considered constants. The overall total effort and its variance, as well as the total effort in a particular portion of the river during a period, were obtained the usual way by summing effort and variance values across the respective strata.

V øLSTAD ET AL.

734

Estimating catch and harvest.—For shore anglers, the total catch (or harvest) in each stratum s was estimated from the access survey using a model-based estimator similar to that applied for daily effort, namely, X ^ s;i ; ^ s ¼ Ns C ð11Þ C ns ns where X ^ s;i ¼ 1 ys;i;j C ps;k j is the model-based estimate of total catch (or harvest) for PSU i based on the expansion factors ps,k in Table 4. A design-based estimate was obtained by replacing ps,k with the constant selection probability for the shift (p ¼ 0.5). The variance of equation (11) was estimated conservatively by the single-stage approximation, ^ ^ s;i
C X C s ^ s Þ ’ N2 ; ð12Þ varðC s n ðn
1Þ s s ns ^ is the estimated mean catch per PSU in where C s stratum s for shore anglers. Final estimates of catch and respective variances were obtained the usual way by summing over strata. To estimate catch for boat anglers, we applied the composite estimator, equation (9), to effort estimates from the aerial and access surveys for broad spatial and temporal strata (the tidal and nontidal river crossed by time blocks 1–3 and 4–7), which we labeled s 0 , in conjunction with the estimated catch rate ^ s 0 ¼ ½/E ^ 1;s 0 þ ð1
/ÞE ^ 2;s 0  3 R ^s0 ; ð13Þ C where Rˆs 0 is the overall catch rate (fish/h) in stratum s 0 . The variance of equation (13) was estimated as follows under the assumption that effort estimates from the aerial and access surveys were independent: 2

daily total effort and (2) effort recorded during one shift in the access survey to a total daily effort. The 2002 effort distributions differed from the 1995 distributions used to plan this survey (Figure 2). The two flight schedules (random count and maximum count) that were compared experimentally in the nontidal river during the peak American shad run (time stratum 2) produced nearly identical estimates of total effort when based on model-based estimators with expansion factors determined from the 2002 access survey (Table 4); however, a slight improvement in precision was realized for the maximum-count method (Figure 3). However, the maximum-count method resulted in a negative bias when the historical 1995 effort distributions were used to extrapolate effort, because the effort distributions in 2002 differed from the distribution used in planning the survey. For the probability-based survey, the use of the selection probabilities based on the 1995 survey in a designbased estimator produced an effort estimate similar to the model-based estimates for the probability-based and randomized surveys conducted at 0700–1900 hours (Figure 3). This comparison suggests that the model-based estimates with count expansion based on the current daily effort distributions (Table 4) are unbiased for both scheduling methods. The aerial and access surveys produced similar estimates of boat angler effort in all sections of the river, but the aerial survey resulted in significantly lower estimates for shore anglers than did the access survey (P , 0.05; Table 5). The difference in effort for shore anglers is significant because the 95% confidence interval of the difference does not include zero

2

^ s 0 Þ ’½R ^ 0 3 varðE ^ s 0 Þ þ E ^ 0 varðR ^s0 Þ varðC s s ^s0 E ^ s 0 ;i ; E ^ s 0 ;i Þ; ^ s 0 covðR þ 2R

ð14Þ

where ^ s 0 ;i Þ ¼ ð1
/ÞcovðR ^ s 0 ;i ; E ^ 2;s 0 ;i Þ ^ s 0 ;i ; E covðR

ð15Þ

is the covariance between catch rate and effort. The optimal weight is provided in equation (10). Results Effort Estimates The estimated daily distributions of angler effort (Table 4) were based on interviews of 2,353 anglers intercepted in the access survey. These proportions were used in model-based estimators to expand (1) instantaneous aerial angler counts (in a 2-h interval) to

FIGURE 3.—Estimated total angler effort (angler-hours; 6SE) in nontidal stratum 1 of the Delaware River (see Figure 1) during time stratum 2 (peak of the American shad run), based on a nonuniform random daily flight schedule (RS) and a Dauk and Swartz (D&S) maximum-count daily flight schedule. The p expansion factors for instantaneous daily counts were based on the original 1995 effort distribution (0700–1900 hours) and on estimated daily effort distributions for the 2002 survey (0700–1900 and 0700–2100 hours).

735

ANGLER EFFORT AND CATCH ESTIMATION

TABLE 5.—Estimated total angler effort (h) and relative standard errors (RSEs) for the Delaware River by spatial strata and fishing mode, based on aerial and access surveys. The difference in effort between aerial and access surveys is statistically significant (sig.) at a ¼ 0.05 when the 95% confidence interval does not include zero (CL ¼ confidence limit). Effort difference Aerial

95% CLs

Access

Region

Effort

RSE

Effort

Nontidal Estuary All

211,442 126,350 337,792

0.12 0.14 0.09

159,854 88,493 248,347

Nontidal Estuary All

49,407 13,271 62,678

0.09 0.20 0.08

Nontidal Estuary All

260,849 139,621 400,470

0.10 0.12 0.08

Difference 6 SE

RSE

Lower

Upper

Sig.

Boat anglers 0.17 51,588 6 36,649 0.35 37,857 6 35,659 0.16 89,445 6 50,816


20,244
32,034
10,154

123,420 107,749 189,044

No No No

90,026 57,451 147,477

Shore anglers 0.13
40,619 6 12,844 0.16
44,180 6 9,724 0.10
84,799 6 16,058


65,794
63,239
116,273


15,444
25,121
53,325

Yes Yes Yes

249,880 145,944 395,824

All anglers 0.12 10,969 6 39,051 0.22
6,323 6 36,427 0.11 4,646 6 54,011


65,571
77,719
101,215

87,509 65,073 110,507

No No No

(Schenker and Gentleman 2001). The lower effort estimate for shore anglers in the aerial survey suggests the possibility of counting errors due to visibility bias (Pollock et al. 1994). However, input from creel clerks suggested that some anglers actively avoided being observed from the air, another factor that could contribute to undercounting of shore anglers. Final (best) effort estimates (426,395 angler-hours; RSE ¼ 4%) were based on the composite of the aerial and access estimates for boat anglers and on the access survey alone for shore anglers (Table 6). We estimated a total effort of 260,976 angler-hours (78,666 trips) in nontidal waters and 165,419 angler-hours (48,694 trips) in tidal waters (one trip involves one angler fishing on a given day). The use of only an aerial survey for estimating effort would introduce bias for shore anglers in the estuary portion of the survey area

(Figure 4). When based on the aerial survey alone, boat anglers accounted for 90% of the effort in the estuary. When the best estimates for each component of the fishery were used (Table 6), in contrast, boat anglers accounted for 66% of nontidal effort and 65% of tidal effort. Catch and Harvest Estimates We achieved the Cooperative’s 20% precision target for the total catch estimates for both American shad and striped bass. The total catch of American shad was estimated at 26,885 fish (RSE ¼ 16%), and total catch of striped bass was estimated at 47,671 fish (RSE ¼ 15%; Table 7). Estimates of total harvest of American shad and striped bass were nonsignificant (Table 8), suggesting that a catch-and-release fishery is the norm for these species in the Delaware River. Initial catch

TABLE 6.—Final estimates of angler effort (h) and relative standard errors (RSEs) in the Delaware River, by fishing mode (boat and shore) for broad spatial (nontidal and estuary) and temporal (time blocks 1–3 and 4–7) strata. The best-effort estimate for boat anglers is based on a composite of the aerial and access estimates; the best-effort estimate for shore anglers is based on the access survey alone. Boat anglers (combined surveys) Region

Time block

Nontidal

1–3 4–7 1–7 1–3 4–7 1–7 1–3 4–7 1–7

Estuary

All areas

a

a

Shore anglers (access survey)

All anglers

Effort

RSE

Effort

RSE

Effort

RSE

109,036 61,914 170,950 76,398 31,570 107,968 185,434 93,484 278,918

0.10 0.07 0.07 0.16 0.09 0.12 0.09 0.06 0.06

54,765 35,261 90,026 33,840 23,612 57,451 88,604 58,873 147,477

0.17 0.22 0.13 0.23 0.21 0.16 0.14 0.15 0.10

163,801 97,175 260,976 110,238 55,182 165,419 274,038 152,357 426,395

0.06 0.08 0.05 0.07 0.09 0.06 0.04 0.06 0.04

All seven time blocks are defined in Table 1.

V øLSTAD ET AL.

736

TABLE 7.—Estimates of total angler catch and harvest of American shad and striped bass in the Delaware River for boat and shore anglers, by broad spatial (region) and temporal (time blocks 1–3 and 4–7) strata, based on best-effort estimates. Relative standard errors (RSEs) are also shown. Catch Region and time block

a

Number

Harvest RSE

Number

RSE

3,878 0 3,878

0.32

435 0 435

0.45

0.29

0.16

4,314 0 4,314

7,997 6,349 14,346

0.29 0.42 0.24

49 138 186

0.80 0.88 0.68

24,390 8,935 33,325

0.23 0.29 0.18

0 396 396

1.00 1.00

32,387 15,284 47,671

0.18 0.24 0.15

49 534 582

0.80 0.78 0.71

American shad Nontidal 1–3 4–7 1–7 Tidal 1–3 4–7 1–7 All 1–3 4–7 1–7

25,555 0 25,555

0.17

1,330 0 1,330

0.53

26,885 0 26,885

0.16

0.17

0.53

0.32

0.45

0.29

Striped bass Nontidal 1–3 4–7 1–7 Tidal 1–3 4–7 1–7 All 1–3 4–7 1–7

FIGURE 4.—Estimates of boat and shore angler effort (angler-hours) in nontidal and tidal portions of the Delaware River during the peak of the American shad run, comparing aerial, access, and best surveys (best ¼ composite aerial– access estimates for boat anglers, access estimates for shore anglers). a

estimates reported to the Cooperative, based on the standard aerial–access estimator with catch rates expanded to the effort estimates from the aerial survey only, proved to be biased for shore anglers, and the total catch estimates were also less precise. In addition to being biased, the total catch estimates based on the standard estimator did not achieve the target precision for American shad (35,281 fish; RSE ¼ 22%) or for striped bass (36,281 fish; RSE ¼ 22%) (Vølstad et al. 2003). The current estimate of total catch of striped bass based on the access survey alone (41,966 fish) was similar to the estimate from the complemented aerial–access survey and had the same level of precision (RSE ¼ 15%); the estimate of total catch for American shad, based on the access survey alone, was unreliable (40,355 fish; RSE ¼ 52%). Discussion In addition to producing data for evaluating alternative approaches to flight scheduling, our complemented aerial–access survey also provided us with a comprehensive data set suitable for investigating the benefits and detriments of a number of different alternative means of estimating total effort and for evaluating the cost-effectiveness of these different

All seven time blocks are defined in Table 1.

alternatives. Comparison of results from maximumcount flights (Dauk and Schwarz 2001) and probability-based flights showed that when both data sets were expanded via daily distribution of effort from the 2002 survey, the maximum-count method produced somewhat more precise effort estimates (Figure 3). Instantaneous counts at times of day with peak effort would minimize the count expansion and therefore reduce the errors in the daily effort estimates (Lockwood et al. 2001). However, as can also be seen from Figure 3, application of the maximum-count method with expansion factors determined by the daily distribution from the 1995 survey resulted in a biased effort estimate, whereas the estimate based on the probability-based flights (where the count expansion was taken as the actual selection probabilities) was unbiased. The bias in the maximum-count estimate occurred because the 2002 daily effort distribution differed from the 1995 distribution, from which the maximum-count flight periods were selected (Figure 2). These comparisons suggest that the probabilitybased flight scheduling approach is preferable to the maximum-count approach because it also can yield unbiased estimates (design-based), regardless of the

737

ANGLER EFFORT AND CATCH ESTIMATION

TABLE 8.—Estimates of angler catch and harvest (by number) of American shad and striped bass in the Delaware River by fishing mode (boat and shore) for broad spatial (region) and temporal (time blocks 1–3 and 4–7) strata, based on the combined aerial–access survey for boat anglers and the access survey alone for shore anglers. Boat anglers Region and time block

a

Catch

RSE

Shore anglers

Harvest

RSE

Catch

RSE

Harvest

RSE

American shad Nontidal 1–3 4–7 1–7 Estuary 1–3 4–7 1–7 All 1–3 4–7 1–7

22,598

0.21

3,731

0.34

2,957

0.47

147

0.70

22,598

0.21

3,731

0.34

2,957

0.47

147

0.70

1,170

0.60

275

0.69

160

0.00

160

0.00

1,170

0.60

275

0.69

160

0.00

160

0.00

23,768

0.20

4,007

0.32

3,117

0.45

307

0.33

23,768

0.20

4,007

0.32

3,117

0.45

307

0.33

42

0.88

42

0.88

2,518 339 2,857

0.49 0.47 0.44

7 138 145

1.00 0.88 0.84

10,319 7,179 17,498

0.27 0.34 0.21

396 396

1.00 1.00

12,837 7,518 20,355

0.24 0.32 0.19

7 534 541

1.00 0.78 0.77

Striped bass Nontidal 1–3 4–7 1–7 Estuary 1–3 4–7 1–7 All 1–3 4–7 1–7 a

5,479 6,010 11,489

0.35 0.43 0.28

14,071 1,756 15,826

0.28 0.56 0.26

19,550 7,766 27,316

0.22 0.36 0.19

42

0.88

42

0.88

All seven time blocks are defined in Table 1.

actual distribution of angler effort. The nonuniform selection probabilities ensured that flights tended to be conducted at times of peak effort based on the 1995 survey. Conversely, the accuracy of results via the maximum-count method would be uncertain unless the flights were conducted in conjunction with an access survey that provided precise and accurate effort distributions for extrapolating instantaneous counts in the flight interval to daily total effort. Because of the shift in effort distribution from 1995 to 2002, neither scheduling method achieved the maximum potential gain in precision. Another challenge we encountered in our survey was deriving total effort estimates from aerial- and accesssurvey data when the daily surveyed periods differed between the two. The aerial survey was constrained by daylight and no flights were scheduled after 1900 hours, whereas the access surveys extended to 2100 hours to cover the period of highest postsunset angler activity. To resolve this problem, we used the estimated distribution of the daily fishing activity from 0700 to 2100 hours, based on the access survey to derive count expansion factors for the aerial surveys. Thus, instantaneous counts from 2-h flights could be expanded to estimate the daily total effort from 0700 to 2100

hours. This model-based approach enabled us to estimate total fishing effort for 0700–2100 hours, based on observations from the access survey and from the daytime aerial surveys, including the experimental flights scheduled at fixed time of day. The use of the estimated effort distributions to expand the angler catch and effort, recorded in a shift to the total for the day, improved the precision in the estimates. We determined that this solution was appropriate because the designbased and model-based expansion methods produced similar catch and effort estimates in broad spatial and temporal strata for 0700–1900 hours. This suggests that the model-based expansion of catch and effort to 0700– 2100 hours has minimum bias. Our evaluation of results from this survey indicate that a rigorous probability-based access survey complemented by an aerial survey may be the most costeffective means of obtaining precise and unbiased estimates of fishing effort across all components of a fishery of the type that occurs on the Delaware River (i.e., a combination of boat and shore anglers distributed over a large geographical region). By themselves, aerial surveys are particularly effective for counting a large number of anglers over wide geographical regions (Pollock et al. 1994). However, in

738

V øLSTAD ET AL.

this study, we found that while the aerial and access surveys produced similar estimates of boat angler effort, the aerial survey had substantial visibility bias for shore anglers. In our survey, it is possible that the flight restrictions in the tidal stratum (described in Methods) may have caused undercounting because of visibility bias in that stratum and thus contributed to the less-reliable aerial shore counts. We also noted previously the possibility that some shore anglers actively avoided observation from a plane. The estimates of total effort using only aerial survey data were therefore negatively biased and had a higher level of variability. By using data from both aerial and access surveys to compute effort estimates for boat anglers and by using only the access survey for shore anglers, we improved precision and minimized bias in overall effort estimates used for estimating catch and harvest. The combination of effort estimates for boat anglers was possible because both surveys covered the same underlying fishery and produce consistent estimates of effort and the associated variance (expressed in the same units, angler-hours). Pierce and Bindman (1994) showed that instantaneous counts can accurately represent fishing effort, as measured by continuous monitoring. The nonsignificant difference between the aerial and access-based estimates of effort for boat anglers was unexpected. We anticipated that the access survey would undercount boats because we did not include private docks in the survey area in our sampling frame. Also, in the tidal portion of the river, we expected that some boats that would have been observed from the air might dock at access points outside the area covered in our access survey, also contributing to an underestimate of effort from the access survey. Flight cancellations were another factor that we believed could impact the accuracy of the aerial counts. Weather bad enough to require flight cancellations, however, was also likely to discourage angling; we expected this to result in the aerial count yielding an overestimate because the observations would have been biased toward fishing effort on good weather days. The access survey, in contrast, was conducted according to plan, regardless of weather conditions. The similarity of the effort estimates for boat anglers based on observations from the two surveys suggests that those factors had a nonsignificant effect on the accuracy of the boat angler effort estimate. Our results provide a basis for examining the costs of survey elements relative to the reliability of survey results, both in terms of precision and accuracy. The aerial survey produced significantly more precise estimates of total effort for boat anglers (aerial RSE ¼ 9%; access RSE ¼ 16%) and did so at only 57% of the field operating cost of the access survey. However,

as we noted above, the aerial survey alone resulted in an undercount of shore anglers because of visibility bias. Thus, we chose the access-survey estimate of effort for this component of the fishery. It was possible, using only the access survey, to obtain unbiased estimates of both shore and boat anglers because of the well-defined access sites. However, based on information on field survey cost and the precision of the total effort estimate achieved in this survey, we estimated that the sampling effort in the access survey would have to be increased more than sevenfold to achieve the same level of precision in the total effort estimate as was obtained from the aerial–access surveys, and at nearly five times the field cost. Compared with the aerial–access survey, the access survey alone required an estimated 10-fold increase in sampling effort during the American shad run (time blocks 1–3) to achieve the same level of precision in the total catch estimate, and with an associated sixfold increase in survey cost. The reason is that the observed total daily catch of American shad varied substantially among access sites, partly due to large differences in angler effort and partly because the observations were limited to the relatively short period for the American shad run. In contrast, our results suggest that the total catch of striped bass could be accurately estimated from the access survey alone with no loss in precision. This suggests a more homogeneous distribution of striped bass in time and space. The large number of catch observations from access sites throughout the entire study area and period contributed to the precise estimate. Thus, use of a complemented aerial–access survey proved to be the most cost-effective approach to achieving accurate and precise effort estimates for the two primary target species in the Delaware River recreational fishery. Angler surveys in large rivers with poorly defined or widely dispersed access may only allow the estimation of catch rates from the access survey, thus requiring a complemented effort survey. If aerial surveys are used to estimate angler effort, results from our study suggest the importance of correcting counting errors by quantifying the visibility bias. This can be done by coordinated aerial counts of anglers and boats with ground-level counts (see Pollock et al. 1994:198). The use of fast-moving boats instead of planes may be costeffective and produce more accurate counts of anglers in large lakes (Lockwood et al. 2001). A major consideration in the design of a survey such as ours is the type and reliability of data that are available for use in establishing the most cost-effective approach. We were able to attain and, in fact, exceed the target precision of 20% RSE set by the Cooperative for estimated total catch of American shad (RSE ¼

ANGLER EFFORT AND CATCH ESTIMATION

16%) by using a probability-based aerial–access survey with hierarchal stratification, even though we applied conservative variance estimators. The stratification method and the allocation of sampling effort between the aerial and access surveys were effective because they were based on data from a prior American shad fisheries survey and on expert opinion from agency staff members that were very familiar with that fishery. The grouping of access points served as a dynamic stratification method. By assigning more samples to access points with medium and high angler activity, we enhanced the sampling during the peak American shad season in the portion of the river with the most effort. Similar information was not available for the striped bass fishery. Some information on catch rates of striped bass was available from the National Marine Fisheries Service’s Marine Recreational Fisheries Statistics Survey (MRFSS; http://www.st.nmfs.gov/st1/ recreational/the_mrfss.html) for the tidal river. However, the MRFSS was designed to obtain estimates for large geographic areas, and the number of intercepts at the small spatial scale is limited. In the absence of prior survey data, we scheduled relatively large sampling effort in the estuary (1) because we anticipated that a significant portion of the striped bass fishery would occur in that stratum and (2) in case the variability in effort and catch rates for striped bass was greater than that of American shad. Despite the very limited information available for striped bass, we achieved a high precision (RSE ¼ 15%) because of the fortuitously wide temporal and spatial distribution of this species (i.e., catches were recorded throughout the Delaware River and throughout the total study period). Although we were fortunate to achieve survey objectives for striped bass, generally the success of a survey that is based on data from a prior survey will be affected by the extent to which the characteristics of the fishery are the same in both sampling years. For example, the use of the 1995 American shad creel survey results in establishing the statistical design of the 2002 aerial survey was effective, although the daily effort distribution of anglers changed over time and the variability in total daily effort for 2002 was greater than expected from the 1995 results. During the peak fishing season for American shad (April and May), the coefficient of variation (CV ¼ SD of the estimate divided by the estimate) of daily effort measured from the regular aerial counts were 90% during 2002 versus 76% for 1995. The shift in estimated distribution of angler effort within a day from 1995 to 2002 may be due to differences in survey methods or to a shift in angler behavior. The 2002 estimates of daily angler effort distributions were based on a large number of interviews (2,357) from a representative sample of

739

access points over time, including those with medium and low usage. The previous surveys were primarily conducted at access points with a high usage level only and thus may not be representative of all angler trips in the nontidal river. Our pragmatic survey design involved random sampling of PSUs without replacement within weeks and the selection of only one second-stage unit within each first-stage unit for the aerial and access surveys. This ensured that the survey effort was spread out evenly over time, thus improving the precision relative to simple random sampling (Rasmussen et al. 1998). Replication at the second stage is not only impractical for large-scale angler surveys (Rasmussen et al. 1998) but also would have forced a reduction in the number of first-stage units for a fixed survey cost and, hence, reduced the precision in catch and effort estimates. Replicated sampling of SSUs within selected days for the access survey would have required more creel clerks because of work requirements, and replicated flights within days would have required the use of more than one plane. From a statistical point of view, the price to pay for employing a design without replication at the second stage is that the standard variance estimators for twostage sampling cannot be applied. Instead, we calculated variances of catch and effort from the single-stage estimators as if sampling had been done with replacement, as advocated by Cochran (1977) and Særndal et al. (1992). This simplification is conservative because it overestimates the variance when sampling is conducted without replacement (Cochran 1977:279; Wolter 1985:34). When the sampling fractions in the first-stage are low, as is often the case in angler surveys, the bias from applying the singlestage approximation without finite population corrections is small (Cochran 1977:279). Although our survey was designed to achieve high levels of precision with the resources available, numerous factors in addition to the choice of survey design and statistical estimators might affect the accuracy of estimates, and in many cases the extent of bias in the estimates may be impossible to assess. However, some measures can be taken to minimize potential for bias. For example, in this survey, which was applied in a recreational fishery that proved to be primarily catch and release, the accuracy of our catch estimates was dependent on the reliability of the catch reporting by anglers. In our survey, we incorporated measures that we believed would limit the extent to which incorrect reporting affected accuracy. Interviews were conducted immediately after a completed trip to minimize recall bias. In this specific case, the number of fish caught per trip was also fairly low, which would

V øLSTAD ET AL.

740

also minimize recall bias. In addition, we selected creel clerks who were members of the Delaware River Shad Fishermen’s Association. Most, therefore, had prior experience in angler surveys on the Delaware River, and all received training in interview techniques and species identification as part of this survey. We believe that the use of such trained fishermen contributes to establishing a rapport with anglers and enhances truthfulness in their reporting. Acknowledgments The Pennsylvania Fish and Boat Commission, on behalf of the Delaware River Basin Fish and Wildlife Management Cooperative, funded this study. We thank Bob Lorantas for contract management. We are grateful to Joe Miller and Art Lupine for their expert advice on the design and implementation of the field surveys and for providing data from two prior aerial– access surveys. Miller directed the field data collection in an exemplary manner. We thank all the creel clerks that have participated in the survey. The Statistical Analysis System programming support from Jodi Dew, Chris Swan, and Phil Wirth is greatly appreciated. Millstone Flight School did an excellent job in conducting the aerial survey flights with very few flight cancellations due to weather. The input from the Pennsylvania Fish and Boat Commission and other state and federal biologists was crucial for the development of an effective survey design. Special thanks to David Arnold, Mark Boriek, Pete Himchak, and Mike Kaufmann for augmenting our list of access points to the river. We thank Sherian George and Gail Lucas for their assistance in data management and reporting and Ed Weber for editing. The thorough review comments from Ken Newman (University of St. Andrews), the Associate Editor, and from one anonymous reviewer were much appreciated. References ASMFC (Atlantic States Marine Fisheries Commission). 1999. Amendment 1 to the Interstate Fishery Management Plan for Shad & River Herring, April, 1999. ASFMC, Washington, D.C. Cochran, W. G. 1977. Sampling techniques, 3rd edition. Wiley, New York. Dauk, P. C. 2000. Estimation in creel surveys under non standard conditions. Doctoral dissertation, Simon Fraser University, Burnaby, British Columbia, Canada. Dauk, P. C., and C. J. Schwarz. 2001. Catch estimation with restricted randomization in the effort survey. Biometrics 57:461–468. Fraidenburg, M. E., and G. G. Bargmann. 1982. Estimating boat-based fishing effort in a marine recreational fishery. North American Journal of Fisheries Management 2:351–358.

Hoenig, J. M., D. S. Robson, C. M. Jones, and K. H. Pollock. 1993. Scheduling counts in the instantaneous and progressive count methods for estimating sport fishing. North American Journal of Fisheries Management 13:723–736. Jessen, R. J. 1978. Statistical survey techniques. Wiley, New York. Jones, C. M., D. S. Robsoon, and H. D. Lakkis. 1995. Properties of catch rates used in analysis of angler surveys. Transactions of the American Fisheries Society 124:911–928. Lehtonen, R., and E. J. Pahkinen. 1994. Practical methods for design and analysis of complex surveys, revised edition 1996. Wiley, New York. Lockwood, R. N., J. Peck, and J. Oelfke. 2001. Survey of angling in Lake Superior waters at Isle Royale National Park, 1998. North American Journal of Fisheries Management 21:471–481. Lockwood, Roger N. 2000. Conducting roving and access site angler surveys. Chapter 14 in J. C. Schneider, editor. Manual of fisheries survey methods II: with periodic updates. Michigan Department of Natural Resources, Fisheries Special Report 25, Ann Arbor. Matlock, G. C. 1991. Use of surveys in decision making. Pages 1–4 in D. Guthrie, J. M. Hoenig, M. Holliday, C. M. Jones, M. J. Mills, S. A. Moberly, K. H. Pollock, and D. R. Talhelm, editors. Creel and angler surveys in fisheries management. American Fisheries Society, Symposium 12, Bethesda, Maryland. McNeish, J. D., and J. G. Trial. 1991. A cost-effective method for estimating angler effort from interval counts. Pages 236–243 in D. Guthrie, J. M. Hoenig, M. Holliday, C. M. Jones, M. J. Mills, S. A. Moberly, K. H. Pollock, and D. R. Talhelm, editors. Creel and angler surveys in fisheries management. American Fisheries Society, Symposium 12, Bethesda, Maryland. Miller, J. P., and A. L. Lupine. 1987. Angler utilization survey of the American shad fishery in the Delaware River. Report for the Delaware River Shad Fishermen’s Association, Hellertown, Pennsylvania. Miller, J. P., and A. L. Lupine. 1996. Creel survey of the Delaware River American shad recreational fishery. Report for the Delaware River Shad Fishermen’s Association, Hellertown, Pennsylvania. Parker, N. A. 1956. Discussion. Pages 59–62 in K. D. Carlander, editor. Proceedings of Iowa State Creel Survey Symposium. Iowa Cooperative Fisheries Unit, Ames. Pierce, R. B., and A. G. Bindman. 1994. Comparison of absolute fishing effort and hourly instantaneous angler counts in a small lake. North American Journal of Fisheries Management 14:447–448. Pollock, K. H., C. M. Jones, and T. L. Brown. 1994. Angler surveys and their application to fisheries management. American Fisheries Society, Special Publication 25, Bethesda, Maryland. Pollock, K. H., J. M. Hoenig, C. M. Jones, D. S. Robson, and C. J. Greene. 1997. Catch rate estimation for roving and access point surveys. North American Journal of Fisheries Management 17:11–19. Rao, J. N. K. 2003. Small area estimation. Wiley, New York. Rasmussen, P. W., M. D. Staggs, T. Douglas Beard, and S. P.

ANGLER EFFORT AND CATCH ESTIMATION

Newman. 1998. Bias and confidence interval coverage of creel survey estimators evaluated by simulation. Transaction of the American Fisheries Society 127:469–480. Schenker, N., and J. F. Gentleman. 2001. On judging the significance of differences by examining the overlap between confidence intervals. American Statistician 55:182–186.

741

Særndal, C. E., B. Swensson, and J. Wretman. 1992. Model assisted survey sampling. Springer-Verlag, New York. Vølstad, J. H., W. R. Richkus, J. Miller, A. Lupine, and J. Dew. 2003. The Delaware River creel survey, 2002, volumes I and II. Prepared for Robert Lorantas, Pennsylvania Fish and Boat Commission, Belefonte. Wolter, K. M. 1985. Introduction to variance estimation. Springer-Verlag, New York.