You are on page 1of 6

A Technique for Analysis of Utilization-Availability Data

Clyde W. Neu; C. Randall Byers; James M. Peek

The Journal of Wildlife Management, Vol. 38, No. 3. (Jul., 1974), pp. 541-545.

Stable URL:
http://links.jstor.org/sici?sici=0022-541X%28197407%2938%3A3%3C541%3AATFAOU%3E2.0.CO%3B2-B

The Journal of Wildlife Management is currently published by Alliance Communications Group.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained
prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in
the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/journals/acg.html.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

JSTOR is an independent not-for-profit organization dedicated to and preserving a digital archive of scholarly journals. For
more information regarding JSTOR, please contact support@jstor.org.

http://www.jstor.org
Mon Jun 18 19:51:56 2007
A TECHNIQUE FOR ANALYSIS OF UTILIZATION-
AVAILABILITY DATA1
CLYDE W . NEU, College of Business Administration, University of Minnesota, Minneapolis 55415'
C. RANDALL EYERS, College of Business Administration, University o f Minnesota, Minneapolis 5 5 4 1 5 ~
JAMES M . P E E K , Department o f Entomology, Fisheries a n d Wildlife, University o f Minnesota, St. Paul 55101'

Abstract: A statistical technique evaluating preference or avoidance of a given habitat or forage spe-
cies is presented, using moose (Alces alces) distribution patterns in an area including the Little Sioux
Burn of northeastern Minnesota as an example. The technique is used in conjunction with a chi-square
analysis, after the chi-square has led to the rejection of the null hypothesis that a set of observations
follows an "expected" occurrence pattern. The technique involves the use of a Bonferroni z statistic
which may be used in estimating whether a specific observation occurs rnore or less frequently than
expected. The technique provides a refinement of quantitative methods which heretofore have not
been used to test a multinomial distribution applicable to the example.
I . WILDL. MANAGE. 38(31:541-545

The necessity of determining preference lated chi-square value, and whether a spe-
or avoidance of a given habitat or plant cific type is preferred or avoided.
species in terms of its availability has long The purpose of this paper is to present a
been recognized (Glading et al. 1940, Bell- method of evaluating preference or avoid-
rose and .Anderson 1943). A useful means ance of specific habitat or forage species
of determining the statistical validity of which map be used in conjunction with the
utilization-availability data is a chi-square chi-square analysis. Data on use of dif-
test of the hypothesis that animals use habi- ferent areas on the Little Sioux Burn in
tat or forage species in proportioil to their northeastern hlinnesota by nloose will be
availability to the animal, a multinomial used to present the example. U7e thank R.
distributioll ( Ostle 1963:124). A problem D. Cook, Departluent of Applied Statistics,
in using this test arises when the null hy- University of Minnesota, and C. P. Hatch.
pothesis is rejected and a significant dif- College of Forestry, TVildlife and Range
ference between the "expected and ob- Sciences, University of Idaho, for criticism
served frequency of use is observed. The of this report. D. E. Glaser, Minnesota De-
chi-square does not determine preference partment of Natural Resources, skillfully
or avoidailce of individual categories,
- so the piloted aircraft used in obtaining field data.
data must be inspected to determine which
observations contribute most to the calcu- DATA CoLLECTiON
A 33,200-acre (13,446-ha) study area cen-
Scientific Journal Series No. 8070, Agricultural tered on the I,ittle Sioux Burn which oc-
Experiment Station, University of hlinnesota, St.
Paul 55101.
curred during the week of 14-19 May 1971
Present address: University of Montana, AFIT on the Superior National Forest in north-
hlinutemail School, hlalmstrom Air Force Base, eastern Jlinnesota was established to ob-
Great Falls 59405.
PreSent address: College of Business and Eco- serve Inoose distributioils relative to the
nornics, Uni~ersitvof Idaho, hloscow 83843. burn (Table 1) . The four areas established
"resent address: College of Forestry, Wildlife for determilling llloose locations included:
and Range Sciences, University of Idaho, hloscow
83843. j 1) center of burn, ( 2 ) exterior 0.25-mile

J. Wildl. Manage. 38 ( 3 ) : 1974 541


542 UTILIZATION-AVAILABILITY
ANALYSISNeu et al.

Table 1. Locations of moose or moose tracks on the Little vations of animals by aerial search, could
Sioux Burn study area in relation to burned, unburned, and
be used with the technique described in
0.25 mile radius distance from the burn perimeter, 24
November 1971 through 12 March 1972. this paper, as long as the two assumptions
are met: ( 1 ) that the animal has an oppor-
Number of groups observed
tunity to select any of the habitat which is
Bum perimeter
deemed available, and (2) that observations
(0.25- (0.25-
mile nlile are collected in a random,unbiased manner.
Un- out- 111-
Date of flight burned side) side) Bum Total

1
DATA ANALYSIS
24 November
21 December 0 Once the proportion of each habitat
11 January 2
23 January 5 category within the study area and the
3 February 5 number of moose tracks within each habitat
22 February 4 category have been determined (Table 2 ) ,
12 March 5
the following hypothesis may be tested by
Total 22 the chi-square technique: moose utilize
each habitat category in exact proportion
to its occurrence within the study area. The
(0.40-km) width of bum periphery, ( 3 ) observed occurrence of moose tracks is com-
0.25-mile (0.40-km) width of unbumt area pared with the "expected occurrence of
adjacent to periphery of burn, and ( 4 ) un- moose tracks for each habitat category in
burnt area (Table 1 ) . Areas were deter- calculating a test statistic which is approxi-
mined by planimeter. An intensive aerial mately chi-square distributed. In order to
search of this area was made seven times use this approximation, the number of habi-
during the 1971-72 winter, between 24 No- tat categories is generally selected so that:
vember and 12 March. Each flight was (1) there is at least one expected observa-
made immediately following a snow to re- tion in each category and ( 2 ) no more than
duce chances of recording old tracks. Loca- 20 percent of all categories contain less than
tions of moose and moose tracks were five expected observations (Dixon and
plotted on a 1:24000-scale topographic map, Massey 1969:238). Recent findings indi-
and were considered accurate to within 10 cate that these conditions may be consen7a-
acres (4.05 ha). Each group of moose or tive and that the approximation can be used
moose tracks was considered as one obser- as long as the average (over all categories)
vation. Fresh moose tracks were recorded expected observation is six or more (for the
as an observation only after a search of the 0.01 level of significance of the test) ( Ros-
area did not locate the animals. Care was coe and Byars 1971).
taken not to record tracks which were close In the example used, 117 different obser-
enough together to have been made by the vations of moose or moose tracks were re-
same group, in order to eliminate possible corded. Goodness-of-fit comparisons showed
duplications of the same group. A Cessna that the expected number of moose tracks
172 airplane was used to obtain these and moose locations in each habitat cate-
observations. gory differed significantly from the occur-
Any set of observations relative to use rence of habitat categories within the study
and occurrence of habitat, including those area ( chi-square calculated = 44.14; tabu-
obtained by radiotelemetry or direct obser- lar value, 3 df, 0.10 level of probability =

J. Wildl. Manage. 3 8 ( 3 ):1974


ASALYSIS Weu et al.
UTILI~.~TION-AVAIL~BILITY 543

Table 2. Occurrence of moose tracks on burned, unburned, and peripheral portions of o 33,200-acre area surrounding
the Little Sioux Burn i n northeastern Minnesota, Winter 1971-72.

Proportion Confidence interval on


Proportions h'umber Expected" observed proportion of
of total of number of in occurrence ( p i )
Total acreage moose lnoose ~acharea ( 90% family
Location acreage ( p i 0) observed obsrned ( pi ) confidence coefficient)

In burn 0.340 25 40 0.214 0.129 < pl < 0.299


In burn, 0.25 mile fro111
periphery or less 3,345
Out of burn, 0.25 mile
from periphery or less 3,442
Out of bum, further 15,130
Total 33,200 117 117
Proportions of total acreage represent expected moose obsen-ation \ d u e $ as if moose occurred in each habitat in
exact proportion to availability.
Calculated by multiplying proportion pi0 X n; i.e., 0.455 X 117 = 53.
p i represents theoretical proportion of occurrence and is compared to corresponding p i to determine if h>potbesis of
proportional use is accepted or rejected, i.e., p , =pie.

6.25). The null hypothesis was therefore The one necessary adjustment concerns
rejected, implying that observed moose the "level of significance" of the intervals.
were not distributed proportionately to TVhen estimating a single parameter, the
occurrence of habitat categories. In addi- probability of obtaining an incorrect inter-
tion, a chi-square test of independence on val estimate is controlled at a "level of
data in Table 1 determined that moose dis- significance" a. The resulting interval esti-
tributions did not change appreciably mate is termed a (1- (Y)100 percent confi-
through the winter (chi-square calculated dence interval. When estimating two or
= 19.02; tabular value, 18 df, 0.10 level of more parameters simultaneously, however,
probability = 25.99). the probability that any one interval esti-
It is now desirable to examine several mate is incorrect increases beyond a and is
comparisons between estimated and ex- partially dependent upon the number of
pected occurrence in order to detect prefer- simultaneous estimates being made (Hop-
ence or avoidance of individual habitat or kins and Gross 1970). In order to bound
forage species. A conclusion may be dra\vn the probability error rate at a, (i.e., the
as to which habitats or forage species are probability that at most one interval esti-
creating statistical significance in the chi- mate is incorrect), a scaling don11 of the
square statistic. For this purpose, individ- significance level of each estimate is re-
ual confidence intervals may be constructed quired. The resulting interval estimate~are
for each theoretical proportion of occur- then ternled a (1- a ) 100 percent "family"
rence p, of the multinomial distribution in of confidence intervals with a (1- a ) confi-
order to determine whether expected values dence coefficient.
p,, lie within the magnitude of the signifi- In estilnating several parameters simul-
cant effects. One important adjustment in taneously, as long as the number of param-
the appropriate z statistic to use in con- eters is small (for example, 4 ) , the
structing these confideilce intervals is re- appropriate tabulated statistic to use in con-
quired, however. structing each interval estimate is:

J. Wildl. Manage. 38 ( 3 ): 1974


544 ANALYSIS h7eu et al.
UTILIZATIOX-AVAILABILITY

1 - 2 , instead of z(l-a,,21 moose one or more times on different


flights. I t can still be assumed that each
where k is the number of simultaneous esti-
observation is independent of all others for
mates being made. These are the Bonfer-
the following reasons: ( 1 ) flights were
roni normal statistics ( Miller 1966:67-69 ) .
made far enough apart in time and after
At the a level of significance, the resulting
snows. which minimized or eliminated
confidence intervals will be slightly wider
recording a set of tracks more than once;
for each of the multiple estimates than for
( 2 ) the home ranges of nloose are large
an estimate of only one parameter. The
enough to cover all four areas (Van Ballen-
form of the confidence interval will be:
berghe and Peek 1971 ) thereby providin2
an individual moose with the opportunity
to select from ally of them.
The most difficult bias in gathering this
where fit is the proportion of lnoose or type of data is related to visibility of ani-
moose tracks in the ith habitat category and mals relative to cover. Moose move into
n is the sample size, in this case the number denser conifer-dominated cover as the win-
of observed moose. ter progresses and snow depths and densi-
This statistic is a nonnal approximation ties increase ( Peek 1971 ) . One would ex-
for a variable which follows a binomial dis- pect that moose using the most open
tribution; therefore if p is close to 1or 0, n severely burned portions of the study area
should be larger to maintain a good ap- would be more readily observed than those
proximation. A conservative rule of thumb using more dense, conifer-dominated areas.
is if np and n ( 1 - p ) 3 5, then sample size and that a shift from the nlore open burn
is sufficiently large (Hayes and Winkler to the unburned areas would occur through
1970:225 ) . the winter. However, moose distribution
These intervals are conservative and the did not change through the winter. Much
confidence coefficient is bounded below by of the burned-over area contained abuil-
1 - a. Individual confidence intervals using dant forage for moose, but the adjacent un-
the z statistic are indicated in Table 2 using burned areas would provide most of the
a: = 0.10. This table indicates a significant available cover (which probably helps ex-
difference between estimated and hypothe- plain use of the area the first winter follo~v-
sized occurrence for all categories. The ing the fire). The burn periphery has ap-
burn periphery areas were used signifi- parently served as an attractant to 1noostX.
cantly more than expected according to judged from the habitat selection data.
availability, while the center bun1 and un- Since the unburned periphery coiitained
burned areas were used significantly less, cover similar to that further away fro111the
i.e., expected use of the unburned area was burn, and moose were observed more often
0.455 while confidence intervals on ob- than expected in the periphery, it may be
served use were 0.244 < p4 < 0.440. concluded that observational biases related
to cover were probably not of a magnitude
DISCUSSION which obscured habitat use in this area.
The seven aerial flights which accounted The technique reported here can be used
for 117 observations of lnoose or moose to deternline preference or avoidance of a
tracks undoubtedly counted individual given area or forage species from data
AKALYSISNeu et al.
UTILIZ..~TION-AVAILABILITY 545

gathered by aerial search, radiotelemetry, Vol. 1. Holt, Rinehart and Winston, New
air photogrammetry, trapping and marking, York. 650pp.
HOPKIXS,C. E., AND A. J. GROSS. 1970. Siguifi-
and feeding site examinations. I t provides cance levels in nlultiple comparison tests.
the investigator with a more quantitative Health Service Res. 5 ( 2 ) :132-140.
measurement of the relative value of a habi- MILLER,R. G. 1966. Simultaneous statistical in-
ferences. McGraw-Hill, New York. 272pp.
tat or food plant to an animal than has OSTLE, B. 1963. Statistics in research. Iowa
heretofore been reported. State University Press, Ames, Iowa. 487pp.
PEEK, J. M. 1971. Moose-snow relationships in
northeastern Minnesota. Pages 39-49 tn A. 0 .
LITERATURE CITED Haugen, ed. Proc. Snow and Ice Symp. in
Relation to Wildl. and Recreation, Iowa State
BELLROSE,F. C., AND H. G. A ~ E R S O N1943. . Univ.
Preferential ratings of duck food plants. Illi- ROSCOE,J . T., ASD J. A. BYARS. 1971. .\n in-
nois Nat. Hist. Surv. Bull. 22(5):417-433. vestigation of the restraints nith respect to
DIXON,W. J., AND F. J. MASSEY. 1969. Intro- sample size commonly imposed on the use of
duction to statistical analysis. LfcGraw-Hill, the chi-square statistic. J. Am. Stat. ASSOC.
New York. 370pp. 66(336)3755-759.
GLADING,B., H. H. BISWELL,AKD C. F. SMITH. \TAN BALLENBERGHE, \'., ~ i mJ. hl. PEEK. 1971.
1940. Studies on the food of the California Radiotelemetry studies of moose in north-
quail in 1937. J. Wildl. hlanage. 4 ( 2 )3128- eastern hlinnesota. J. \Vildl. hlanage. 33( 1) :
144. 63-71.
HAYES,W. L., ASD R. L. WINKLER. 1970. Sta-
tistics; probability, inference and decision. Accepted 17 December 1973.

J. Wildl. Manage. 38 (3): 1974

You might also like