Abstract.-Discriminant function analysis corn par- Habitat Structure, Forest ina 36 occu~iedand 18 unoccu~iedsites revealed Composition and Landscape thGt structural variables discriminked between sample groups better than compositional variables, Dimensions as Com~onents and the latter discriminated better than landscape variables. These results are encouraging that habitat of Habitat suitability for the structure will provide a reliable basis for a predictive classification model of habitat suitability. Such a Delmarva Fox model would be useful both for pre-screeningthe biological suitability of potential release sites and for planning, implementing and monitoring prescriptive Raymond D. Dueser,' James 1. Dooley, JL,~ habitat management. and Gary J. Taylor"

The Delmarva fox (Sciurus of habitat requirements will be essen- Methods niger cinereus) was placed on the fed- tial for both initiatives (Dueser and eral list in 1967 Terwilliger 1988). Data Base (32 FR 4001; US. Department of Inte- Habitat requirements might be rior 1970). Remnant populations expressed through any of three sepa- During a 12-mo search for remnant were restricted to four counties in rate but related components of habi- populations of the Delmarva fox eastern Maryland (Taylor and Flyger tat suitability: forest habitat struc- squirrel on the Maryland Eastern 19731, representing less than 10% of ture, forest tree species composition, Shore, Taylor (1976) located 36 "fox the historic range of the subspecies and surrounding landscape struc- squirrel present" (Present) sites with on the Delmarva Peninsula. Forest ture. Both habitat structure and for- extant populations and 18 "fox squir- clearing and habitat fragmentation est composition have been shown to rel absent" (Absent) sites. The gray throughout the range undoubtedly influence the distribution and abun- squirrel (Sciurus carolinensis) was contributed significantly to the pres- dance of fox in heterogene- present on all 54 sites. Taylor then ent endangerment (Taylor 1973). ous landscapes (Nixon and Hansen sampled the forest habitat of each The U.S. Fish and Wildlife Service 1987). site, to compare Present forest stands Recovery Plan for the restoration of Recent research has demonstrated with Absent stands. He established a the Delmarva to secure the potential influence of landscape representative 4 m x 200 m belt status emphasizes both the reintro- composition and structure on popu- transect on each site, on which he re- duction of this subspecies to suitable lations of woodland occu- corded the number of trees by spe- habitats throughout the former range pying farmland mosaics (Wegner cies per diameter-breast-height and prescriptive habitat management and Merriam 1979, Middleton and (DBH) size class (5-20 cm, 20.1-30 for established populations (Taylor et Merriam 1983, Fahrig and Merriam cm, 30.1-50 cm, and 50.1+ cm), per- al. 1983). A thorough understanding 1985). Furthermore, changes in the cent crown cover, percent understory landscape of the Delmarva Peninsula cover, understory density, and 'Paper presented at symposium, Man- almost certainly played a major role understory species composition. All agement of Amphibians, Reptiles and Small on the decline of the fox squirrel habitat measurements were taken Mammals in North America. (Flagstaff.AZ, (Taylor 1973). from June through September 1972 July 19-2 1, 1988.) Given this background, the objec- and 1973. These data formed the ini- 2RaymondD. Dueser is an Associate tive of this study was to compare the tial data base for this study. Professor in the Department of Environ- mental Sciences, University of Virginia, apparent effects of habitat structure, Taylor (1976) reported the number Charlottesville,VA 22903. forest composition and landscape di- of trees measured in each of two size JJamesI. Dooley, Jr., is a graduate re- mensions on the presence and ab- classes: "small" trees (5-30 cm DBH) search assistant in the Department of Envi- sence of the Delmarva fox squirrel on and "large" trees (> 30 cm DBH). We ronmental Sciences, University of Virginia. 54 study sites in eastern Maryland. assigned each tree to one of five taxo- Charlo~esville,VA 22903. This analysis is the first step in the nomic groups: loblolly (Pinus 4GaryJ. Taylor is Associate Director of Wildlife. Department of Natural Resources, development of a predictive classifi- taeda), combined oak species (Quercus Maryland Forest, Park and Wildlife Service, cation model of habitat suitability for spp.), American beech (Fagus gran- Tawes State Office Building, Annapolis, MD this subspecies (cf. Houston et al. difolia), combined hickory species 27401. 1986). (Ca ya spp.), and combined mixed hardwoods. We estimated the ap- planimetry of 1:I000 black-and-white ground?"), (2) tree species composi- proximate total basal area for each photographs (dated 1978) obtained tion ('Which tree species predomi- size-taxonomic class by assuming an from Eastern Shore offices of @e nate in this forest and give it its char- average DBH of 17.5 cm for small USDA Agricultural Stabilization and acter?"), and (3) landscape dimen- trees and 40.0 cm for large trees. We Conservation Service. We initially sions ("What are the land use and then estimated total basal area for all measured each landscape variable cover dimensions of the landscape trees 2 5 cm DBH and the fraction of for a 2-km2 circular sample unit cen- mosaic in which this forest is embed- that total basal area represented by tered on the sample woodland. This ded?"). Conceptually, these compo- each taxonomic group. These basal unit was chosen as a first-approxima- nents represent a gradient of scales area estimates provide a basis for tion of "minimal population area" on from "microscopic" habitat structure comparing forest "composition" in- the basis of home range size and ac- to "macroscopic" composition to dependently of forest "structure" as tivity (Flyger and Smith 1980). Based "rnegasc~pic'~ con text. reflected, for example, in the raw on the results of analyses for the 2- Two-group discriminant function percentage of trees counted in each km2unit, both smaller (1-km2)and analysis was used to compare the taxonomic group. larger (4-km2) sample units subse- Present and Absent forest stands Original data were collected on quently were described in the same identified by Taylor (1976). Each land use and cover composition of way. analysis (1) computed the univariate the landscape surrounding a random F-ratio comparing -Presentand Ab- subset of Taylor's (1976) study areas sent sites for each habitat variable, (fig. 1). Landscape variables included Statistical Analyses (2) tested the centroids of Present area of open fields, percentage of and Absent sites for equality on the area forested, internal forest perime- This comparison of habitat compo- basis of a linear combination of the ter ("edge") within the sample unit, nents is based on multivariate statis- habitat variables (i.e., a linear dis- forest shape (Blouin and Connor tical analyses of three separate but criminant function), using multivari- 1985), and distance to next nearest related components of forest habitat ate analysis of variance (MANOVA), woodland. These variables are re- suitability: (I) habitat structure (3) indicated the relative contribution ferred to below as landscape "dimen- ('What does the forest 'look like' to of each habitat variable to any ob- sions." They were measured by an observer passing through on the served difference between centroids, based on the correlation between the variable and the discriminant func- tion, (4) tested the sample variance- covariance matrices of Present and Absent sites for homogeneity using a Box's M test statistic, and (5) indi- cated the percentage of the variation in group membership (Present or Absent) explained by the discrimi- nant function, based on the correla- tion between the membership vari- able and the discriminant function. (Dueser and Shugart 1978). All analy- ses were computed both with and without arcsin-square root transfor- mations of percentage variables. Re- sults of the parallel analyses were qualitatively similar in each case. For purposes of interpretability, only the results for untransformed variables are presented here. All analyses used the MANOVA and DISCRIMINANT routines of the Statistical Package for the Social Sciences (SPSS, Nie et al. 1975). Figure 1 .-Schematic diagram of 1 -, 2- and 4-km2 sample units for measuring landscape dimensions of "fox squirrel present" and "fox squirrel absent" study areas on the Eastern As an unbiased test of the ability Shore of Maryland..Eachsample unit was centered on one of Taylor's (1976) study areas. of each set of habitat variables to classify the group membership of the centage shrub-ground cover (0.564) Forest Tree Species Composition study sites (i.e., Present or Absent), a are particularly important in dis- jackknife procedure (Efron 1979) was criminating between sites. Conceptu- All 54 study areas supported a mix used to classify each of Taylor's ally, Present sites have larger trees, of hard- and soft-mast tree species. study areas. Each site was deleted in less shrub-ground cover vegetation, Although Present sites had some- sequence, DISCRIMINANT was run and less understory than Absent sites what greater basal areas for Ameri- for data from the remaining 53 sites, (fig. 2). Present sites were correctly can beech (p > 0.07) and mixed hard- and a classification function was classified 79% of the time in the jack- woods (p > 0.051, there were no clear- computed from these data. The de- knifing procedure, and Absent sites cut univariate differences between leted site was then classified on the were correctly classified 48% of the sites in forest composition (table 2). basis of this independent classifica- time. There also was no difference in total tion function. The probabilistic ("pre- dicted) classification was then com- pared with the actual ("observed") classification for each site. Brennan et al. (1986) have pro- posed an alternative solution to the problem of habitat analysis. Logistic regression analysis is superior to multivariate analysis of variance when one or more of the predictor (e.g., habitat) variables is categorical (i.e., non-continuous), when the vari- ance-covariance matrices are non- homogeneous and/or when the data violate the assumption of multivari- ate normality (Press and Wilson 1978). Parallel analyses demonstrate that logistic regression analysis offers no inherent advantage over discrimi- nant function analysis in the present case (Dooley, unpublished). Present Absent centtoid centroid

Results

Habitat Structure

Present sites had a greater percent- age of trees larger than 30 cm DBH, lower percentage shrub-ground cover, and slightly lower understory vegetation density than Absent sites I I I I I I I II I I (table 1, p 5 0.05). Present and Absent -2 -1 0 +1 +2 sites differ structurally on the aver- Discriminant Score age (MANOVA Chi-square (5) = 14.825, p 5 0.01 1). The linear combi- nation of structure variables ac- Smaller trees counted for 26% of the variation in Oense understory group membership. The variance- covariance matrices were marginally Dense groundcover homogeneous (Box's M = 20.056, p 2 0.06). Percentage of trees greater than Figure 2.4nterpretation of discrimination between average "fox squirrel present" and "fox squirrel absent" study areas on the Eastern Shore of Maryland, based on analysis of forest 30 cm DBH (r = -0.7351, understory habitat structure. The horizontal dashed lines indicate the range of observations for a vegetation density (0.564), and per- sample group (Present or Absent). basal area (F (1,52) = 2.300, p > 0.13). Landscape Dimensions ous (Box's M = 19.926, p > 0.39). As The two types of sites were similar in with forest composition, there was composition for both small and large Five landscape variables were meas- no consistent difference in landscape trees (fig. 3). Reflecting this similar- ured for the 2-km2 circular sample dimensions between Present and Ab- ity, there was only a marginally sig- unit centered on the target woodland sent sites. Present sites were correctly nificant multivariate difference in of 38 of the Taylor's (1976) study ar- classified 78% of the time, and Ab- forest composition (MANOVA Chi- eas. Present sites were somewhat sent sites were correctly classified square (5) = 10.584, p > 0.06). closer to the next nearest forest tract 40% of the time. The linear combination of compo- than Absent sites (table 3, p < 0.03). To evaluate the possibility that the sition variables accounted for 19%of Despite this modest difference, there negative result in the test for equality the variation in group membership. was no significant multivariate dif- of group centroids came about be- The variance-covariance matrices ference in landscape dimensions be- cause we were measuring landscape were conspicuously non-homogene- tween sites (MANOVA Chi-square variables on an "incorrect" spatial ous (Box's M = 61.549, p < 0.001). (5) = 8.574, p > 0.127). scale, we repeated the landscape Present sites were correctly classified Present and Absent woodlands analysis for both smaller (1-km2)and 79% of the time, and Absent sites also were similar in area, averaging larger (4-km2)circular sample units, were correctly classified 48% of the 9.4 and 10.0 ha, respectively, as pho- still centered on the woodland of time. tographed in 1978. The linear combi- interest. Again, there were no consis- Although the correct classification nation of landscape variables ac- tent group differences on either scale rates were the same as for structural counted for 23% of the variation in (p > 0.40, table 4). variables, the two sets of variables group membership. The variance-co- Either the landscapes surrounding misclassified different sites. variance matrices were homogene- the sample Present and Absent woodlands do not differ consistently, or they differ on a scale of measure- ment or in a way not revealed by the present analyses.

Figure 3.-Average forest tree species composition of "fox squirrel present" and "fox squirrel absentwstudy areas on the Eastern Shore of Maryland. "Other" cate- gory includes a variety of small trees such as cherry (Pmnus spp.) and flowetfng dogwood (Cornus florida). Discussion ity of alternative food sources and Habitat Suitability perhaps stabilized the food supply. Present Habitat Indeed, Allen (1943) and Nixon and It is assumed that Present (i.e., occu- Hansen (1987) indicate that settle- pied) sites are more "suitable" on the The present habitat of the Delmarva ment and agriculture have worked to average than are Absent (i.e., unoc- fox squirrel consists primarily of the advantage of the fox squirrel cupied) sites. Present sites are re- relatively small stands of mature throughout the midwestern United garded here as the "standard of ex- mixed hardwoods and having States, resulting in increased abun- cellence" by which to judge the habi- relatively closed canopies, relatively dance and an expanded geographic tat requirements of the Delmarva fox open understory, and a high propor- range. squirrel. Given that a number of un- tion of forest edge. Occupied tracts Why has this not occurred with known (and unknowable) ecological, include both groves of trees along the Delmarva subspecies? Why has biogeographical, and/or historical streams and bays and small woo- the abundance of this fox squirrel factors may actually be responsible dlots located near agricultural fields. continued to decline throughout the for the absence of this subspecies In some areas, particularly in south- period of the recorded literature from any particular site within the ern Dorchester County, Maryland, (since approximately l85O)? historic range, this assumption is cor- occupied habitat includes tracts Taylor (1976) attributes the contin- rect only as a first approximation dominated by mature loblolly pine ued decline of the Delmarva fox (ref. Hanski 1982). It clearly would be located adjacent to marshes and tidal squirrel to . While unwarranted if the distribution of streams. The woodland habitats now many of the landscape changes re- squirrels among these 54 study sites occupied by the Delmarva fox squir- sulting from settlement might have were highly variable through time. rel are consistent with those occu- benefited the fox squirrel, others Nevertheless, the chance presence of pied by other subspecies of fox squir- have been detrimental. Taylor be- the squirrel on "unsuitable" sites and rel (Bakken 1952; Brown and Yeager lieves that extensive timber harvest its absence from "suitable" sites be- 1945; Weigl et al., in press). has been particularly detrimental. cause of factors other than habitat The picture of the Delmarva fox The removal of mature hardwoods suitability per se can only make it squirrel that emerges from the litera- has reduced the availability of suit- more difficult to distinguish between ture is one of a species relatively able den trees, removed reliable Present and Absent sites. These adept at utilizing a dissected, hetero- sources of concentrated hard mast, analyses based on presence-or-ab- geneous landscape dominated by ag- promoted the luxuriant growth of sence population information thus riculture and woodlot forestry. Fox understory vegetation, and perhaps circumvent many of the potential pit- squirrels are more cursorial than altered the competitive relationship falls associated with the use of popu- gray squirrels, and often are found between fox and gray squirrels to lation density as an indicator of habi- on the ground several hundred me- favor grays. Furthermore, coastal tat suitability (Van Horne 1983). ters from the nearest woodlot. They plain woodland management typi- Given its present habitat, it occupy larger home ranges than gray cally has involved both short timber seemed reasonable to propose that squirrels (30 ha vs. 3 ha), travel far- rotations (i.e., frequent harvests) and the capacity of a woodland to sup- ther between captures (307 m vs. 119 reforestation with pure stands of lob- port a population of the Delmarva m), and thus are generally more mo- lolly pine. Finally, gradual urbaniza- fox squirrel could be determined by bile (Flyger and Smith 1980). Fox tion has added yet another detrimen- habitat structure, forest composition squirrels more readily exploit agri- tal land-use practice. and/or the land use and cover com- cultural crops such corn, oats, soy- beans and fruit. They more fre- quently utilize forest edges. Fox squirrels would thus appear to be relatively well-adapted to exploit the landscape created by settlement of the coastal plain. One might conclude that man's activities on the Delmarva Peninsula should have been to the benefit of the fox squirrel. Land clearing has cre- ated woodlots. Grazing and burning have opened up the understory. Ag- riculture has increased the availabil- position of the surrounding land- variation, they classify sites almost as discriminate meaningfully scape. We anticipated originally that reliably as the structural variables, between occupied and unoc- each of these components of habitat and their dispersion matrices are cupied forest stands. Present suitability would prove to be impor- homogeneous. They do not, how- sites have larger trees, less tant in its own way, and that each ever, discriminate meaningfully be- ground cover, and less would have a perceptible influence tween Present and Absent sites. understory than Absent sites. on fox squirrel presence or absence Given the suggested importance of Significantly, these results in Maryland woodlots today. Our landscape changes in bringing about corroborate the general habi- results indicate, however, that habi- the decline of the fox squirrel, this tat descriptions reported by tat structure is the component most result was somewhat unexpected. Flyger and Lustig (1976). likely to contribute meaningfully to The correct interpretation probably the formulation of a predictive model requires recognition that most of the 2. In addition to this clear-cut of habitat suitability. Only the struc- Eastern Shore landscape has been discrimination, the structure ture variables discriminate strongly altered, fragmented and homoge- variables exhibit the most between Present and Absent sites: nized. Most of the remaining wood- desirable combination of sta- Present sites have larger trees, less lands are mere remnants of forest in tistical properties, including ground cover and less understory a mosaic of agricultural fields, wet- the highest variance explana- (fig. 2). These variables account for lands and suburban development. tion, homogeneity of disper- the greatest fraction of the explained There may simply be little important sions, and high correct classi- variation, their dispersion matrices variation remaining among these for- fica tion rates. These proper- are effectively homogeneous, and est patches. At the same time, it must ties will simplify the formu- they classify sites to the correct be recognized that a number of po- lation of a predictive classifi- group (i.e., Present or Absent) at tentially important landscape vari- cation model of habitat suita- least as well as any of the variable ables--e.g., proximity to streams and bility. sets examined. ponds (McComb and Noble 1981) Forest composition is highly vari- and proximity to roadways (Flyger 3. Although the absence of able among locations in eastern and Lustig 1976)-were not consid- meaningful discriminating Maryland, but this variation seems to ered in this analysis. information in forest compo- exert only a marginal influence on sition and landscape dimen- the likelihood of occurrence of fox sions was somewhat surpris- squirrels on any given site. The com- Management Implications ing, these results have the position variables classify sites as re- effect of simplifying the ef- liably as the structural variables, and The Recovery Plan for the Delmarva fort to quantify habitat suita- they account for only a slightly lower fox squirrel calls for both the translo- bility for the Delmarva fox fraction of the explained variation. cation of squirrels to suitable habitats squirrel. It would be impru- They do not, however, discriminate throughout the historic range and the dent to disregard forest com- strongly between Present and Absent maintenance of occupied habitat position and landscape at- sites and their dispersion matrices (Taylor et al. 1983).Will objective, tributes in the evaluation of are strongly non-homogeneous. Of quantitative habitat analysis be help- potential release sites; these course, this conclusion is based on a ful in evaluating potential release components of habitat suita- comparison of two groups of sites, all sites and planning prescriptive habi- bility must be important at of which are known to be "squirrel tat management? Results of the some level (Flyger and Lus- woods." Had there been a "tree analyses presented here provide tig 1976). There appears to be squirrel absent" category of study some basis for optimism. A number little potential, however, for area, forest composition might well of management implications follow the variables analyzed here have appeared to be more significant from these results: to contribute to a predictive (cf. Nixon et al. 1978, Sanderson et al. model of habitat suitability. 1976). 1. Of the variable sets exam- Landscape composition also varies ined, habitat structure is the 4. The discriminating structure among locations, but this variation best indicator of biological variables are easy and rela- seems not to be important on the av- habitat suitability for the Del- tively inexpensive to meas- erage in discriminating between oc- marva fox squirrel at the ure. Including site reconnais- cupied and unoccupied sites today. present time. Even this mini- sance, approximately one- The landscape variables account for a mal list of structure variables half day of field time is re- comparable fraction of the explained (table 1) has the power to quired for a team of two ex- perienced observers to col- Acknowledgments restrial Vertebrates. U. Wisconsin lect a Taylor-type data set. Press, Madison. 470 p. This study would not have been pos- Brown, Louis G., and Lee E. Yeager. 5. It should therefore be practi- sible without the cooperation of 1945. Fox squirrels and gray squir- cal to pre-screen potential those who assisted G. J. Taylor in the rels in Illinois. Illinois Natural His- release sites for habitat suita- identification of remnant populations tory Survey Bulletin 23:449-536. bility relative to Present sites. of the Delmarva fox squirrel on the Dueser, Raymond D., and H. H. Objective pre-screening has Maryland Eastern Shore. G. D. Ther- Shugart, Jr. 1978. Microhabitats in not always been possible be- res and G. W. Willey, Sr., of the a forest-floor small cause no "standard of excel- Maryland Department of Natural fauna. Ecology 59239-98. lence" has been available. Resources assisted with locating sites Dueser, Raymond D., and Karen Ter- for the landscape analysis. J. H. Por- williger. 1988. Status of the Del- 6. It also should be practical to ter assisted with air photo interpreta- marva fox squirrel (Sciunrs niger plan, implement and evalu- tion and with the statistical analyses. cinereus) in Virginia. Virginia Jour- ate prescriptive habitat man- J. Peatross prepared the figures, and nal of Science 38(4):380-388. agement for the benefit of the L. M. McCain assisted with prepar- Efron, Bradley. 1979. Computers and Delmarva fox squirrel on oc- ing the manuscript. V. Flyger and K. the theory of statistics: thinking cupied sites or potential re- E. Severson provided thoughtful re- the unthinkable. SIAM Review lease sites. The important views of the manuscript. Finally, the 21:460-480. measures of habitat structure owners of the Maryland study sites Fahrig, Lenore, and G. Merriam. (e.g., understory vegetation generously provided access to their 1985. Habitat patch connectivity density) tend to be variables property for purposes of habitat and population survival. Ecology which are amenable to characterization. This study was sup- 661762-1768. silvicultural manipulation ported by funding from the Flyger, Vagn, and Loren W. Lustig. (Nixon et al. 1980). Nongame and Endangered Species 1976. The potential for re-estab- Program of the Virginia Department lishing fox squirrels in portions of of Game and Inland Fisheries and by their former range in the north- Conclusions the Virginia Coast Reserve Long- eastern states. Transactions of the Term Ecological Research Program Northeast Section of the Wildlife We anticipated at the outset that each (NSF Grant BSR-8702333). Society 33:13-17. of three potentially important com- Flyger, Vagn, and Deborah A. Smith. ponents of habitat suitability-forest 1980. A comparison of Delmarva habitat structure, forest tree species Literature Cited fox squirrel and gray squirrel composition, and surrounding land- habitats and home range. Transac- scape dimensions-would influence Allen, D. L. 1943. Michigan Fox tions of the Northeast Section of the present occurrence of the Del- Squirrel Management. Game Divi- the Wildlife Society 37:19-22. marva fox squirrel in forest stands on sion Publication 100. Lansing, Hanski, I. 1982. Dynamics of regional the Eastern Shore of Maryland. The Michigan. 404 p. distribution: the core and satellite analyses reported here produced a Bakken, Arnold. 1952. Interrelation- species hypothesis. Oi kos 38:210- number of surprises. ships of Sciurus carolinensis (Gme- 221. Habitat structure is the only com- lin) and Sciurus niger (Linnaeus) in Houston, B. R., Tim W. Clark, and S. ponent that both discriminates be- mixed populations. Ph.D Thesis, C. Minta. 1986. Habitat suitability tween occupied and unoccupied sites University of Wisconsin. 21 2 p. index model for the black-footed in a meaningful way and exhibits a Blouin, Michael S., and Edward F. ferret: a method to locate trans- combination of statistical proper ties Connor. 1985. Is there a best shape plant sites. Great Basin Naturalist favorable for the formulation-of a for nature reserves? Biological Memoirs 8:99-114. predictive classification model of Conservation 32: 277-288. McComb, William C., and Robert E. habitat suitability. Brennan, Leonard A., William M. Noble. 1981. Nest-box and natu- The analysis of habitat structure Block, and R. J. Gutierrez. 1986. ral-cavity use in three mid-South provides a basis for optimism that The use of multivariate statistics forest habitats. Journal of Wildlife such a model would prove useful for developing habitat suitability Management 45:93-lOl. both for pre-screening potential re- index models. p. 177-182. In J. Middleton, J., and G. Merriam. 1981. lease sites and for planning, imple- Verner, M. I. Morrison and C. J. Woodland mice in a farmland mo- menting and monitoring prescriptive Ralph, eds. Wildlife 2000: Model- saic. Journal of Applied Ecology habitat management. ing Habitat Relationships of Ter- l8:703-7lO. Nie, Norman H., C. Hadlai Hull, Jean marva fox squirrel recovery plan, G. Jenkins, Karen Steinbrenner, First revision. U.S. Fish and Wild- and Dale H. Brent. 1975. SPSS: Sta- life Service. Washington, D.C. tistical Package for the Social Sci- 49 p. ences, 2nd ed., McGraw-Hill, N.Y. US. Department of Interior, Bureau 675 p. of sport Fisheries and Wildlife. Nixon, C. M., and L. P. Hansen. 1987. 1970. Endangered Species of the Managing forests to maintain United States. Washington, D.C. populations of gray and fox squir- P. rels. Illinois Dept. of Conservation Van Horne, B. 1983. Density as a mis- Technical Bulletin 5.35 p. leading indicator of habitat qual- Nixon, C. M., S. P. Havera, and R. E. ity. Journal of Wildlife Manage- Greenberg. 1978. Distribution and ment 47:893-9Ol. abundance of the grey squirrel in Wegner, J.F., and G. Merriam. 1979. Illinois. Illinois Natural History Movements by birds and small Survey Biological Notes 105.55 p. mammals between a wood and Nixon, C. M., S. P. Havera, and L. P. adjoining farmland habitats. Jour- Hansen. 1980. Initial response of nal of Applied Ecology 16:349-357. squirrels to forest changes associ- Weigl, P. D., M. A. Steele, L. J. Sher- ated with selection cutting. Wild- man, J. C. Ha, and T. S. Sharpe. In life Society Bulletin 8:298-306. press. The ecology of the fox Press, S. James, and Sandra Wilson. squirrel (Sciurus niger) in North 1978. Choosing between logistic Carolina: Implications for survival regression and discriminant analy- in the southeast. Tall Timbers Re- sis. Journal of the American Statis- search Series. tical Associa tion 73:699-705. Sanderson, H. Reed, William M. Healy, James C. Pack, John D. Gill, and Jack Ward Thomas. 1976. Gray squirrel habitat and nest-tree preference. Proceedings of the Annual Conference of the South- eastern Association of Game and Fisheries Commissions 29:609-616. Taylor, Gary J. 1973. Present status

and habitat survevJ of the Del- marva fox squirrel (Sciurus niger cinereus) with a discussion of rea- sons for its decline. Proceedings of the Annual Conference of the Southeastern Association of Game and Fisheries Commissions 27:278- 289. Taylor, Gary J. 1976. Range determi- nation and habitat description of the Delmarva fox squirrel in Maryland. M.S. Thesis, University of Maryland, College Park. 76 p. Taylor, Gary J., and Vagn Flyger. 1973. Distribution of the Delmarva fox squirrel (Sciurus niger cinereus) in Maryland. Chesapeake Science 1459-60. Taylor, Gary J., Vagn Flyger, N. Swink, and G. Willey. 1983. Del-