Sie sind auf Seite 1von 10

Recall Bias and Reliability of Survey Data: Moose Hunting in Maine Author(s): Stuart M. Mazurkiewicz, Kevin J.

Boyle, Mario F. Teisl, Karen I. Morris, Alan G. Clark Reviewed work(s): Source: Wildlife Society Bulletin, Vol. 24, No. 1 (Spring, 1996), pp. 140-148 Published by: Allen Press Stable URL: http://www.jstor.org/stable/3782847 . Accessed: 17/01/2012 16:21
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Allen Press is collaborating with JSTOR to digitize, preserve and extend access to Wildlife Society Bulletin.

http://www.jstor.org

140

BIAS RECALL

Recall

bias
moose

and

reliability of
hunting
in

survey

data:

Maine

Stuart M. Mazurkiewicz;, Kevin J. Boyle, Mario F. Teisl, Karen I. Morris,

and Alan G. Clark


Abstract Moose hunting data were collected for the 1989 Maine moose (Alces alces) hunt to investigate recall bias in factual data and reliabilityof hunters'opinions and preferences. We surveyedthe 900 residentmoose hunterswho participatedin the October 1989 hunt. Hunterswere randomlyassigned to three groups: one group (n=150) was surveyed immediatelyafterthe huntand then again 4 months later;the second group(n=600) was surveyed immediatelyafterthe huntonly; the thirdgroup(n=l 50) was surveyed4 monthsafter the hunt only. A total of 804 hunters(89%)respondedto the survey. The 4-month recall for moose huntersdid not result in recall bias, which is contraryto the findings of general studies of huntingwith annual recall. Statisticalresultsindicatedthat the opinion and preferencedata were reliable.

Key words Alces alces, huntersurveys, Maine, recall bias, reliability

Wildlife managers often depend on surveys to profile characteristics and solicit opinions of users of wildlife resources. Despite widespread use of survey data by wildlife and other resource managers, there are concerns about their accuracy. User characteristics (e.g., age, sex, and education) and participation in wildlife-related activities (e.g., days and hours spent hunting) are factual data that profile users and their levels of use. Factual data are subject to recall bias, which is not likely to occur when survey participants report their characteristics (e.g., age, education, etc.), but can be a problem when wildlife users are asked to recall the extent of their participation in wildlife-related activities. A recent study investigated recall bias in the National Survey of Fishing, Hunting, and Wildlife-Associated Recreation (National Survey hereafter; Westat, Inc. 1989). Data on fishing, hunting, and wildlife-associated recreation during 1985 were collected in early 1986 (U.S. Fish and Wildl. Serv. and U.S. Bur. of Cen-

sus 1988). Westat found that annual recall results in rates relative to reoverreporting participation on based semiannual, sponses quarterly, and monthly recall periods (Chu et al. 1992). This finding was confirmed in a study (U.S. Fish and Wildl. Serv. and U.S. Bur. of Census, 1993, L. S. Cahoon, C. A. Riker, and T. F. Moors, Recall bias in the national survey of fishing, hunting, and wildlife associated recreation, unpubl. data) conducted with the 1991 National Survey. Westat, Inc. (1989) found overreporting was most profound among anglers due to the large number of fishing trips some anglers take over an extended period, resulting in anglers including effort prior to the 1-year recall period. Gems et al. (1982) compared fishing trip data collected every 2 weeks with data for a 2-month recall period. The 2-month recall period produced lower estimates than data collected every 2 weeks, providing further evidence of recall bias. Recall bias arises from problems in survey instrument

Address for Stuart M. Mazurkiewicz (deceased) and Kevin J. Boyle at the time of research: Department of Resource Economics and Policy, University of Maine, Orono, ME04469-5782, USA. Address for Mario F. Teisl: Department of Agriculturaland Resource Economics, University of Maryland, College Park, MD, USA. Address for Karen I. Morris: Maine Department of Inland Fisheries and Wildlife, Bangor, ME 04401, USA. Address for Alan G. Clark: Maine Department of Inland Fisheries and Wildlife, Augusta, ME 04333, USA. Wildlife Society Bulletin 1996, 24(1 ):140-148

Peer refereed

Recall bias * Mazurkiewicz et al. design, poor recall by survey respondents, or a combination of these problems. Opinion and preference data are subjective evaluations. For opinions, respondents evaluate observable phenomena (eg., crowding by other hunters). For preference data, respondents choose alternative wildlife-management options (e.g., changes in the hunting season). Opinion and preference data do not record factual information and, as such, are not usually subject to recall bias; these data can change over time. For example, if a hunter finds a new job that conflicts with the timing of a hunting season, his or her preference for when a hunting season should be held may differ from before obtaining the new job. However, if conditions that affect individual hunter preferences have not changed but individual hunters' responses to an opinion or preference question vary depending on the timing of the survey, the data would be deemed unreliable. Reliability focuses on the stability of individual responses to survey questions on an initial test and subsequent retest. Although the reliability of data have been investigated in other contexts (e.g., educational testing), our study is the first to investigate reliability of wildlife users' opinions and preferences (Allen and Yen 1979, Kealy et al. 1988, Loomis 1989). Unreliable survey data may be due to unstable preferences of wildlife users, respondents inability to answer survey questions, or survey design flaws. The basic insight from the Westat, Inc. (1989) study is that recall bias in survey data for wildlife activities increases as the recall period is lengthened and the frequency of participation in an activity increases. If fishing is one extreme, then moose (Alces alces) hunting in Maine is near the other end of the continuum. Fewer than 1 of every 80 individuals who apply receive a moose hunting permit, and permit holders must wait 2 years after being drawn before they can apply again. Most moose hunters make 1 trip to their assigned hunting zones during the annual 6-day season. These factors reduce the recall period and limit the frequency of participation. The research reported here uses moose hunting data collected after the 1989 hunt to investigate potential recall bias in factual data and the reliability of hunters' opinions and preferences.

141

establishing the moose hunt requires all hunters to complete a survey after their hunts (Maine Public Law, Chapter 118, 1981). The experiment was only applied to resident moose hunters. The survey was designed and administered using the Total Design Method (Dillman 1978), which includes the design and layout of the survey questionnaire in a booklet form. We used this methodology to minimize design flaws that might lead to false conclusions of recall bias or unreliable data. The survey was pretested in 1988 and expanded for the 1989 season. Hunters were notified of the survey prior to the moose hunt in a packet of moose hunting information provided with hunting permits. The survey process included 4 mailings: the survey, a reminder and thank you card sent 1 week later to all hunters, the same survey to non-respondents 2 weeks after the initial mailing, and the survey again to nonrespondents via certified mail about 4 weeks after the initial survey mailing.

Experimental

design

Methods
The survey
Since 1982 Maine has conducted an annual moose hunt. Permits to hunt are issued via a lottery to 900 Maine residents and 100 nonresidents. The 1989 hunt was held from 17-22 October. The legislation

The test for recall bias and reliability included 3 experimental groups. Respondents in group A (n = 150) received a survey immediately after the hunt (Al) and were asked to complete the survey a second time 4 months later (A2). Respondents in group A were not told of the retest when they initially responded to the survey. This double application of the survey is called test-retest. Groups Bi (n = 600) and C2 (n = 150) were control groups; Group B1 only received a survey immediately after the hunt and Group C2 only received a survey 4 months after the hunt. Group B1 was a control for nonrandom assignment to experimental groups. Group C2 was a control group that allowed for identifying potential learning or anchoring effects, but it is not possible to distinguish between these effects. The purpose of the within- and across-subject design is relatively straight forward. The ideal design is the within subject test-retest of Group A because we wanted to investigate the temporal stability of individual hunters' responses to survey questions. However, observations of Group A at time 2 can be contaminated by these individuals' initial responses at time 1 due to learning or anchoring effects. The reason for including control groups differs slightly between the tests of recall bias and reliability. In terms of recall bias, Group A may recall factual information differently in the second administration of the survey because they had already completed the survey once. Group C2 did not have this initial priming and only answered the survey at time 2. Statistical compar-

142

Wildlife Society Bulletin 1996, 24(1):140-148 before final conclusions are reached regarding random assignment, recall bias, and data reliability.

isons of Groups Al with A2 and Groups Bl with C2 allowed us to determine conditions where Group A gave a false reading regarding recall bias (i.e., sample statistics for Groups Al and A2 do not differ and sample statistics for Groups Bl and C2 differ). The same pattern of statistical findings can result in false readings regarding reliability, but for a different reason. If the interval between survey applications is short, respondents in Group A may recall their initial responses to the survey and reiterate these answers to the opinion and preference questions in the second survey application. This is an anchoring problem. Thus, the test-retest application allowed us to track individual respondents, while the control groups identified undesirable test-retest effects. The Westat, Inc. (1989) study was only conducted with an design, equivalent to our control across-subject groups. Thus, the Westat statistical finding are not as robust as results generated using within- and acrosssubject treatments. Although subtle, the difference between recall bias and reliability is important. We investigated recall bias to identify whether individuals can remember factual information. Reliability investigates the stability of opinions and preferences over time. The survey interval should be long enough that respondents can not recall (reiterate) their initial responses to an opinion and preference question, but short enough that factors influencing opinions and preferences remain unchanged. Thus, although the experimental designs are identical, recall bias investigates survey respondent's memories and reliability investigates the stability of their opinions and preferences. Tests for random assignment to experimental groups (hunter characteristics), recall bias (hunter participation), and reliability (opinions and preferences) are based on the hypothesis: H:O A, = 0^2 = O, = 2 (1)

Data
Surveys were sent to all 900 resident moose hunters who participated in the 1989 hunt. The survey requested socioeconomic characteristics of respondents and their characteristics as users of wildlife, including respondents' sex, age, education, type of city of residence, and income. Education data were grouped into 4 categories: less than high school, high school diploma to some college education, completion of an associate's degree or bachelor's degree, and completion of an advanced degree. Origin of residence included 4 categories of human population: rural areas, small towns (<5,000), small cities (5,000-49,000), and large cities (>50,000). Hunter characteristics included whether individuals hunted moose with a subpermittee and whether the subpermittee was a Maine resident (a subpermittee accompanies the holder of a moose hunting permit and carries a weapon, but only 1 moose can be taken). Other hunter characteristics included whether respondents hunt species other than moose, fish inland waters, trap furbearers, or observe wildlife. Hunters were asked to report the number and sex of moose seen and the number of black bear (Ursus americanus) and white-tailed deer (Odocoileus virginianus) seen during the hunt. Hunters were also asked to indicate if they chose not to shoot a moose during the hunt and to record, by sex, the number of moose they chose not to shoot. The Maine Department of Inland Fisheries and Wildlife uses bulls seen/100 cows seen, calves seen/100 cows seen, and moose seen/hour hunting to monitor trends in the moose herd. Because of overlaps in habitat of moose and white-tailed deer in Maine, data are also maintained on deer seen/hour hunted. If the component variables of these ratios do not exhibit recall bias, we can conclude that the ratios are not subject to recall bias. Each hunter was assigned to a specific zone in which they could hunt, and all hunters who killed a moose were required to report to an official check station for their moose to be registered. Biological data on sex, age, and physical measurements of the moose were collected at the stations. We asked respondents to report the zone in which they hunted and to report the sex and age if they killed a moose. We compared these survey responses with the check station data to validate the survey. We collected factual data to document moose

where the overbars denote population parameters (e.g., proportions and means). Failure to reject the null hypotheses for hunter characteristics supports random assignment to experimental groups, whereas failure to reject the null hypotheses for participation (opinion and preference) data supports the conclusion of no recall bias (data reliability). If the null hypothesis is rejected, further investigation is required to identify pair-wise comparisons that give rise to the difference. We focus on tests of descriptive statistics that are of interest for developing wildlife policy. Although each variable is considered individually in the analyses, the collective statistical findings for all the variables within each data grouping are considered

Recall bias * Mazurkiewicz et al.

143

hunting effort as days and hours spent hunting. every year. (Only resident moose hunters were inHunters were asked to report travel time from their cluded in these analyses). A majority hunted game home to the zone in which they hunted, number of other than moose, fished open water, ice fished, and trips they made during the hunt, moose hunting ex- observed wildlife. Only 6% of moose hunters said penditures, and percent of these expenditures they they trapped in Maine. made in Maine. No differences were found among the groups for Opinion data included hunters' rating the quality any of the socioeconomic characteristics; trapping of their moose hunts on an integer scale ranging from was the only hunter characteristic that differed poor (1) to excellent (6) and crowding by other among groups. Because few moose hunters trapped moose hunters on an integer scale ranging from not in Maine, this difference may have been a statistical crowded (1) to crowded (9). We asked for prefer- anomaly. We concluded that hunters were assigned ence data regarding the timing of the moose hunt in randomly to experimental groups and respondent two ways. Hunters first were asked their timing pref- characteristics did not change between the first and erences with options of late September, mid-Octo- second applications of the survey instrument. This ber, or early December. Then respondents were in- evidence suggested that the data were suitable for informed about the effects of this timing on the quality vestigation of recall bias and data reliability. of the hunt and were asked to answer the question again with response options of late September, mid- Hunt attributes Hunters observed an average of 1.3 calves, 2.5 October, and no preference (Boyle et al. 1993). cows, 2.0 bulls, and 1.2 unclassified moose while Tests hunting in their assigned zone (Table 2). The null hyWe used chi-square statistics to test the equiva- pothesis of no difference among groups was rejected lency of sample proportions across treatments for bi- only for unclassified moose. The pattern of renary variables and variables with multiple categories. sponses across groups did not indicate recall bias. Analysis of variance was used to test the equivalency Nearly half of the moose hunters chose not to of sample means across treatments for continuous shoot a calf or a cow, while about 25%chose not to variables (e.g., hunter age). When the null hypothe- shoot a bull (Table 2). The null hypothesis of no difsis was rejected for any variable, a Duncan multiple ference in the sample proportions was not rejected range test was applied to determine which treat- for any of these variables. Nearly all hunters (93%) ments differed. A nonparametric Kruskal-Wallis test killed a moose. Of moose taken, hunters reported was used for continuous variables that had a spike of 86% were adult and 66% were bulls. Check station observations at zero (e.g., moose sightings). All tests data recorded 92% of resident hunters killing a were conducted with a = 0.10. moose, 89%getting an adult, and 61%getting a bull. The success rate was within the 90% confidence interval for survey data (93% CI = 1.4), but the perResults cent of hunters getting an adult or a bull were outMore than 90% of deliverable surveys were com- side the 90% confidence interval for survey data pleted and returned, with 97% responding from (86% CI = 2.0; 66% CI = 2.7). The differences, howGroup Al, 82%from Group A2, 92%from Group B1, ever, were not large. No statistical differences were and 93%from Group C2. The lower response rate for identified for these data across the experimental Group A2 was expected because some respondents groups. Hunters observed an average of 3.4 deer and 0.2 simply reject the second survey or do not take the time to complete it. In the data analyses we only in- bears while moose hunting (Table 2). The null hycluded Group A responses for individuals who com- pothesis of no difference in average observations was not rejected for deer, but was rejected for bear. pleted the test and retest. Differences existed between treatment C2 and the other 3 treatments, indicating possible over-reportRespondent characteristics Survey respondents were primarilymale (82%), av- ing recall error. This finding could indicate recall eraged 40 years old, had a high school degree (74%) bias that does not occur in the test-retest group reresided in a ruralarea or a small town, and had an av- spondents who may have anchored on their initial erage household income of $29,794 (Table 1). Nearly responses when responding to the replication of the
all respondents (97%) hunted moose with a subpersurvey.

mittee; 96% of subpermittees were Maine residents. The majorityof moose hunters (72%)hunted in Maine

Recall of the zone hunted closely followed (X2 = 13.61, 15 df, P = 0.56) the allocation of hunters to

144

WildlifeSociety Bulletin1996, 24(1):140-148

Table 1. Socioeconomic and personal characteristics of resident moose hunters surveyed for recall bias and data reliability in Maine following the 1989 moose hunt. Variables Socioeconomic characteristics: Sex (%) Male
n

A1

A2

B1

C2

All groups

Test statistics

85 113 38 1.5 113 25 51 17 5 113 36 42 22 0 103 27,742 1,502 103 99 114 95 96 75 115 92 115 75 115 64 115 10 115 60 115

83 113 38 1.5 113 21 56 17 6 113 28 47 22 3 103 30,266 1,956 103 96 114 95 96 73 115 94 115 78 115 59 115 12 115 59 115

81 547 40 0.7 547 26 60 10 4 547 35 41 18 6 505 30,094 868 505 96 547 96 507 72 551 89 552 78 552 61 552 5 552 57 552

81 136 41 1.4 136 26 61 11 4 136 35 39 23 3 125 29,880 1,544 125 97 133 96 127 74 137 92 137 83 137 65 137 4 137 60 137

82 909 40 0.5 909 26 58 13 3 909 34 42 20 4 836 29,794 649 836 97 908 96 826 72 918 90 919 79 919 62 919 6 919 58 919

2 = 1.26 P = 0.74 F = 0.79 P = 0.50

Age (years)
x

SE
n

Education (%) No high school degree High school degree College degree Advanced degree n Residence (%) Rural Small town Small city Large city
n

X2 = 26.7 P= 0.18

X2 = 18.4 P = 0.24

Income ($)
x

SE
n

F = 0.48 P 0.70 X2 = 3.46 P= 0.32


X = 0.95

Hunter characteristics: Hunter moose with subpermittee (%) n Subpermittee is Maine resident (%) n Hunt in Maine every year (%) n Hunt other than moose (%) n Open water fish (%) n Ice fish (%) n Trap (%) n Observe wildlife (%)
n

P = 0.81 X2 = 5.66 P = 0.93 X2 = 3.38 P = 0.34 X2= 2.92 P= 0.40 2 = 1.10 P= 0.78 X2 =11.59 P = 0.01 X2 = 0.64 P = 0.89

zones. Except for the southeast zone, numbers fell within 90%confidence intervals for hunters' recall of the zones in which they hunted. Hunters traveled an average of 3.3 hours to their assigned hunting zones, took an average of 1.8 trips from their homes, and averaged 2.4 days (15.7 hours) hunting (Table 3). No differences were observed for these variables. Moose hunters spent an average of $423 for their hunts, with 97% being spent in Maine (Table 3). The null hypothesis of no differences in the proportions of expenditures made in Maine was rejected. Differences occurred between Groups Al and A2 and between

Groups Al and C2. This pattern of differences did not indicate a recall bias (i.e., the null hypothesis of no difference was not rejected for Groups B1 and C2).

Opinion and preference


The majority of respondents rated their hunts as excellent and did not feel crowded by other moose hunters (Table 4). The null hypothesis of no difference was not rejected for rating of the hunt, but was rejected for crowding. Reported crowding was for A than higher Group Groups B1 and C2, but this did not indicate unreliable data because pattern

Recall bias * Mazurkiewicz et al.

145

moose hunters and observations Table2. Hunter of big gamewhilemoose hunting as reported by resident surveyed selectivity,harvest,
after the 1989 moose season in Maine. Variables Al A2 B1 C2 All groups Test statistics

Moose observed: Calves SE


n

1.3 0.2 109 2.4 0.3 109 1.9 0.2 111 1.5 0.4 109 47 113 44 112 26 113 92 113 84 103 64 104 3.6 0.8 111 0.2 0.06 113

1.4 0.2 109 2.5 0.3 109 1.9 0.2 111 1.8 0.4 109 37 113 36 112 28 113 92 113 91 103 65 104 3.4 0.6 111 0.2 0.05 113

1.3 0.1 548 2.5 0.2 551 2.1 0.1 552 1.0 0.1 545 44 544 42 545 22 544 93 548 85 507 67 510 3.3 0.2 553 0.1 0.02 551

1.4 0.1 135 2.5 0.2 135 1.9 0.1 135 1.2 0.2 132
47

1.3 0.1 901 2.5 0.1 904 2.0 0.1 909 1.2 0.1 895 44 907 41 906 24 907 93 910 86 842 66 848 3.4 0.2 909 0.2 0.02 912

X2 = 2.15

P = 0.54

Cows
x

X2 = 3.72

SE
n

P= 0.29

Bulls
x

SE
n

X2= 1.98
P= 0.58

Unidentified
x

X2 = 8.22

SE
n

P = 0.04

Hunters thatchose not to shoot: Calves(%)


n

Cows (%)
n

Bulls(%)
n

135 41 134 24 136


96

X2 = 3.21
P= 0.36 X= 1.86 P= 0.60 X = 2.26 P= 0.52 X= 1.98 P= 0.58

Killeda moose (%)


11

Adult(%)
n

Bull(%)
17

136 87 129 63 130 3.7 0.4 134 0.3 0.08 135

X2= 3.69
P= 0.30 X = 0.77 P= 0.86

Otherbig game observed: Deer SE


n F = 0.27 P= 0.84

Bear SE
n

X2 = 8.26 P - 0.04

--

Groups Al and A2, and Groups B1 and C2, did not differ. The majority of moose hunters preferred mid-October for the moose hunt, regardless of whether they were simply asked for their preference without information or were told how timing influences the quality of a hunt (Table 4). These sample proportions differed among groups for the question without information and did not differ for the question with information. Given that the proportion of hunters favoring mid-October exceeds 70% for all groups for the question without information, the difference identified is of no practical importance for setting hunting policy.

Discussion
The null hypothesis of no difference among groups was not rejected for 12 out of the 13 variables in the hunter characteristics data. This weight of evidence led us to conclude that respondents were assigned randomly to experimental groups and that the data satisfactorily investigate recall bias and reliability. The null hypothesis of no difference was not rejected for 3 of 4 variables measuring hunter observations of moose and for the 3 hunter selectivity variables. All 4 harvest variables and 5 of 6 hunter effort variables did not differ among groups. The

146

Wildlife Society Bulletin 1996, 24(1):140-148

Table 3. Travel time, effort and hunting expenditures reported by resident moose hunters surveyed after the 1989 Maine moose season.

Variables
Travel time to zone (hours) x SE n Number trips x SE n Days hunted x SE n Hours hunted x SE n Total expenditures ($) x SE n Expenditures in Maine (% of Total) x SE n

A1
3.1 0.2 112 1.8 0.1 107 2.4 0.1 113 17.1 1.5 112 378 30 110 95 1.8 106

A2
3.0 0.2 112 1.9 0.1 107 2.5 0.1 113 17.3 1.5 112 410 39 110 99 0.9 106
i

B1
3.4 0.2 545 1.7 0.1 534 2.4 0.1 548 15.8 0.6 546 435 15 548 98 0.5 539
iiiii

C2
3.3 0.2 134 2.0 0.3 133 2.3 0.1 135 14.0 1.1 132 425 28 136 97 1.1 132

All groups
3.3 0.1 903 1.8 0.1 881 2.4 0.05 909 15.7 0.5 902 423 12 904 97 0.4 883
iiii

Test statistics
F= 1.69 P= 0.17

F= 1.18 P= 0.32

F = 0.36 P= 0.78

F= 1.35 P= 0.26

F = 0.85 P = 0.47

F= 2.10 P= 0.10

iii~~~~.

preponderance of evidence supports the conclusion that recall bias was not present. Any time a large number of statistical tests are conducted,

there is a probability that a small number of hypotheses will be rejected purely by chance. More importantly, only 1 of the rejections resulted in a

Table 4. Opinion about the hunt and preference for hunt timing reported by resident moose hunters surveyed after the 1989 Maine moose hunt.

Variables
Rating of hunta 5x SE n Crowdingb x SE n Timing of hunt without information (%)c Late September Mid-October Early December n Timing of hunt with information (%)c Late September Mid-October No Preference n
a

A1

A2

B1

C2

All groups
4.7 0.1 912 2.5 0.1 916 19 76 4 911 19 76 5 910

Test statistics 2= 19.52


P 0.19

4.6 0.1 114 2.8 0.2 113 23 73 4 114 20 74 6 113

4.6 0.1 114 2.8 0.2 113 25 70 4 114 26 68 6 113

4.7 0.1 549 2.4 0.1 553 16 79 5 549 16 80 4 548

4.8 0.1 135 2.4 0.1 137 25 73 2 134 21 73 6 136

x2 = 36.96 P = 0.04

x2 =10.84 P = 0.09

2 = 8.04 P= 0.23

Rated on integer scale from poor = 1 to excellent = 6. Rated on integer scale of not crowded = 1 to crowded = 9. c Without information represents responses from hunters who were not informed about implications of season timing to quality to hunt. With information was response after briefing on implications.
b

Recall bias * Mazurkiewicz et al. pattern of pair-wise comparisons that might support recall bias. The bear observation data indicated possible recall decay, suggesting that this problem might occur for low incidence events on hunting trips that are incidental to the primary activity. Other differences appeared to be statistical anomalies. These tests for recall bias implicitly assume, based on the results of the Westat, Inc. (1989) study, that shorter periods of recall reduce recall bias. The Westat study simply compared sample statistics and did not have actual observations of hunter and angler activities to confirm this supposition. In our study, comparisons with unpublished field data maintained by the Maine Department of Inland Fisheries and Wildlife provided evidence that hunter recall for a 4-month period, although not perfect, approximated actual behavior in the field. For opinion-preference data, the difference among groups for crowding did not indicate unreliable data because average crowding was not different for Groups Al and A2 and for Groups Bi and C2. It simply appeared that respondents in Group A experienced higher crowding than did hunters in Groups BI and C2. Our findings indicate that reliable hunter opinion and preference data can be collected that are useful for developing wildlife management policies and regulations.

147

trol groups and investigate all data collected by the survey instrument. For any individual survey where recall bias is of concern, it is possible to stratify the sample as we have done to test for recall bias. This possibility also holds for concerns regarding the reliability of opinion and preference data. The next step in investigating reliability is to consider opinion and preference data where the issue of concern is more complex and less familiar to survey respondents (e.g., habitat management issues), and the period between the test and retest is extended. Acknowledgments. The co-authors submitted the manuscript and addressed the reviewers' comments in the memory of the lead author, Stuart Mazurkiewicz, who died in an automobile accident. This research was financed by the Maine Agricultural and Forest Experiment Station and the Maine Department of Inland Fisheries and Wildlife. This publication is Maine Agriculturaland Forest Experiment Station Publication No. 1892.

Literature cited
M. J., ANDW. M. YEN. 1979. Introduction to measurement ALLEN, theory. Brooks Cole Publishing, Monterey, Calif. 310pp.
BOYLE, K. J., R. L. DRESSLER,A. G. CLARK, AND M. F. TEISL. 1993.

Moose hunter preferences and setting season timings. Wildl. Soc. Bull. 21:498-504.
CHU, A., D. EISENHOWER,M. HAY, D. MORGANSTEIN,J. NETER, AND J. WAKSBERG. 1992. Measuring the recall error in self-reported

Implications
The message from our research is clear: 4-month recall for a unique 1-week hunt does not result in recall bias. In contrast, Westat, Inc. (1989) found evidence of recall bias in a general survey of hunting requiring annual recall. These studies begin to establish conditions regarding recall period and survey applications where recall bias is or is not a concern. Other researchers should lengthen the recall period beyond 4 months and investigate applications to a common activity with frequent participation over an extended time (e.g., deer hunting) to identify the conditions where recall bias begins to
occur.

fishing and hunting activities. J. Offic. Stat. 5:13-39. D. 1978. Mail and telephone surveys: the total design DILLMAN, method. John Wiley and Sons, New York, N.Y. 325pp.
GEMS, B., D. GOSH,ANDR. HITLIN. 1982. A recall experiment: im-

pact of time on recall of recreation fishing trips. Am. Stat. Assoc. 1982 Proc. on survey res. methods:372-375.
M. J., M. MONTGOMERY, ANDJ. F. DORIDIO. 1988. KEALY, Reliabil-

ity and predictive validity of contingent valuation: Does the nature of the good matter? J. Envir. Econ. Manag. 19:244263. LOOMIS, J. B. 1989. Test-retest reliability of the contingent valuation method: a comparison of general population and visitor responses. Am. J. Agric. Econ. 71:76-84.
U.S. FISH ANDWILDLIFE ANDU.S. BUREAU SERVICE OFTHE CENSUS.1988.

1985 national survey of fishing, hunting, and wildlife-associated recreation. U.S. Gov. Printing Off., Washington D.C. 137pp.
U.S. FISH AND WILDLIFESERVICE AND U.S. BUREAUOF THE CENSUS. 1993.

Our study was robust in use of control groups that allowed for identification of significant effects that may simply be statistical anomalies. The large number of variables provided extensive evidence to consider in judging real differences among study groups. Further investigations should include con-

1991 national survey of fishing, hunting, and wildlife-associated recreation. U.S. Gov. Printing Off., Washington D.C. 124pp. Westat, Inc. 1989. Investigation of possible recall/reference period bias in national surveys of fishing, hunting, and wildlife-associated recreation. Rep. to the U.S. Dep. Inter., U.S. Fish Wildl. Serv., Off. Fed. Aid. 154pp.

148

Wildlife Society Bulletin 1996, 24(1):140-148


Stuart Mazurkiewicz (photo) received his B.S. from the University of New Hampshire. He was a graduate student in the Department of Resource Economics and Policy at the University of Maine at the time this research was conducted. Stuart died in an automobile accident prior to submitting this paper to the Wildlife Society Bulletin. We at the Bulletin extend our sympathy to Stuart's family, colleagues and friends and want to express our gratitude for the opportunity to publish this interesting and informative research.

Associate Editor: Peyton

Das könnte Ihnen auch gefallen