m; -.L< .. .n. ... £47523. - V- .0 an... . ’2‘ :3- ‘ '2 "5 r. V r1. 4f ”5%..%v " 4' rat . 3» ~ This is to certify that the thesis entitled Factors Influencing Cowbird Distributions in Forested Landscapes of Northern Michigan presented by John Matthew Stribley has been accepted towards fulfillment of the requirements for Master of Science degree inFisheries and Wildlife Major professor Date 12/31/93 0-7639 MS U is an Affirmative Action/Equal Opportunity Institution illiillll iiiliillillli Hilliilil 3 1293 01025 9079 LIBRARY Michigan State University PLACE ll RETURN BOXtoromavottflnehocthom ywrrocord. To AVOID FINES return on or baton dd. duo. DATE DUE DATE DUE DATE DUE 719.11; MSU lsAn Afflrmdivo Action/Equal Opportunity Intuition W FACTORS INFLUENCING COWBIRD DISTRIBUTIONS IN FORESTED LANDSCAPES OF NORTHERN MICHIGAN By John Matthew Stribley Dr. Jonathan B. Haufler AN ABSTRACT OF A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE Department of Fisheries and Wildlife 1993 ABSTRACT FACTORS INFLUENCING COWBIRD DISTRIBUTIONS IN FORESTED LANDSCAPES OF NORTHERN MICHIGAN BY John Matthew Stribley Forest fragmentation has been implicated in the population declines of many neotropical migratory bird species. More specifically, forest- interior bird species may be particularly vulnerable to brood parasitism by the brown-headed cowbird (Mglggh;g§_§£gz) in fragmented forest ecosystems. This study evaluated cowbird distributions relative to landscape-level patterns as well as site-specific habitat conditions occurring in northern Michigan hardwood forest stands. The Indice Ponctuel d'Abondance (IPA) method was used to census the avifauna at the approximate center of each forest stand surveyed. Vegetation sampling occurred in the immediate vicinity of each census point using various mensuration techniques. Surrounding land-use patterns were quantified from aerial photos and vegetation maps using dot—gridding techniques. The probability that cowbirds would occur at any given site was 3 - 3.5 times greater when agricultural lands were present within 3 km of a study site. Results of logistic regression analysis indicated that intra-stand structural diversity and surrounding habitat heterogeneity were important predictors of cowbirds once agriculture was present in the landscape. Thus, the removal of agriculture or other potential foraging bases within a 3 km buffer is recommended for reducing the risk of cowbird occurrence in forested habitats of neotropical migrants. Acknowledgements Funding for this project was provided by McIntire-Stennis. Special thanks are extended to the Huron Mountain Wildlife Foundation for providing access to the research facilities of the Huron Mountain Club. I would like to thank my major adviser, Dr. Jonathan Haufler, for his guidance and for effectively providing the necessary resources used in the development and completion of this research project. I would also like to thank Dr. Scott Winterstein and Dr. Donald Beaver of my graduate committee for their helpful suggestions and insight and Dr. Allen Kurta for inspiring me to pursue an academic career in the biological sciences. Special thanks go to all of my fellow graduate students for helping me through the graduate education process and to my field assistant, Tom Gibson, for helping make the 1993 field season a bearable one. It has truly been an honor and a pleasure to attend classes and work with so many bright and intelligent individuals. I would especially like to thank Gary Roloff, Kelly Millenbah, Steve Negri, Meg Clark, Katherine Braun, Tim Vandeelen, Teresa Mackey, Matt Beirne, Ly Furrow, Nan Kelly, Julie Tsatsaros, Christine Hanaburgh, Gina Karasek, Jennifer Dorset, Maya Hamady, Lisa Grise, Cathy Cook, Lou Bender, Rich and Donna Minnis, and John Niewoonder for all of your help, support and encouragement. Best of luck to all of you in the future. I would also like to thank Joe Jarecki (MDNR), Dave Riegle (Harrisville District, U.S.D.A. Forest Service), Peggy Anderson (Delta District, U.S.D.A. Forest Service), and Dr. David C. L. Gosling (Huron Mountain Wildlife Foundation) for providing essential land-use information and much appreciated assistance. iii To my friends and family, I owe more than just thanks. Their support and belief in me has never wavered. Thank you Dad, Irene, and Lyn for helping me to believe in myself. To my mother, I owe the greatest thanks of all. It was through your strength, determination, and love of life that I am compelled to learn all I can and to strive to achieve my fullest potential. To my best friend and fiancee, Sue, I dedicate this thesis to you and my family, for all of your tireless love, support, encouragement, and endurance of my moodiness throughout this whole process, I thank you. iv TABLE OF CONTENTS LIST OF TABLES LIST OF FIGURES . INTRODUCTION Objectives STUDY AREA METHODS . Avian Censusing Vegetation Sampling Landscape Quantification Statistical Methods RESULTS . Habitat Analysis Avian Census Results Cowbird Predictor Model Discussron Conclusion Management Recommendations APPENDIX Appendix A Appendix B LITERATURE CITED vi viii 14 14 16 17 18 22 22 37 37 49 55 SS 59. 59 63 76 A-2 B-l B-2 B-3 13-4 B-S LIST OF TABLES EASE Results of the chi-square test of independence for cowbird occurrence (Present vs. Absent) in stands meeting distance to nearest edge (DNE) and canopy cover criteria . Results of the chi-square test of independence for cowbird occurrence (Present vs. Absent) in stands meeting the respective distance to nearest opening (DNO), basal area (BA), and closed stand criteria . Results of correlation analysis for the relative cowbird abundance data for each variable measured . Mean and standard error for all variables measured at sites having cowbirds (Present) versus those sites lacking cowbirds (Absent) Results of avian censuses conducted from May 15 - July 7, 1992, and May 12 - July 10, 1993 Final Logistic regression model for predicting cowbird. occurrence in northern Michigan hardwood forest stands Deciles of risk analysis for observed (OBS) and expected (EXP) frequencies for the presence (P) and absence (A) of cowbirds Habitat variables used in the logistic regression analysis Common and scientific names of censused birds Univariate logistic regression analysis to determine which covariates should be used in the final cowbird predictor model . Results of logistic regression for the covariates used in the saturated cowbird predictor model Results of quantile (Q) analysis used for determining the appropriate covariate scale to be used in the logistic regression analysis Main effects logistic regression model of the covariates used in developing the final cowbird predictor model Log-likelihood and likelihood ratio test (G) results for interaction terms used in developing the final cowbird predictor model . vi 28 28 30 32 38 42 48 59 60 64 65 67 7O 71 B-6 B-7 B-8 B-9 Results of Wald statistic for interaction terms included in the main effects model Log-likelihood ratio test (G) results for deletion of non-' significant variables from the main effects model Main effects logistic regression model with non-significant variables removed with estimated coefficients (3), standard error of coefficients (SE), and Wald chi-square results Regression diagnostics for observations 21 and 47 vii 72 72 73 75 10 ll 12 LIST OF FIGURES Location of cowbird study areas in northern Michigan. Mean percent agriculture as a function of increasing distance from a study site for the 1992 and 1993 cowbird data collected in northern Michigan hardwood forest stands. Mean percent mixed openings as a function of increasing distance from a study.site for the 1992 and 1993 cowbird data collected in northern Michigan hardwood forest stands Mean percent of grassland openings as a function of increasing distance from a study site for the 1992 and 1993 cowbird data collected in northern Michigan hardwood forest stands Mean relative cowbird abundance (relative to the number of host pairs) in northern Michigan hardwood forest stands as compared across study regions Proportion of study sites having cowbirds as compared across regions for northern Michigan hardwood forest stands Mean probability of finding cowbirds (Cowbird Probability) as a function of agricultural occurrence in the vicinity of a study site for northern Michigan hardwood forest stands The probability that cowbirds would occur at a study site given the sole occurrence of agriculture within each of 4 concentric bands surrounding the study site Response of mean canopy height to predicting cowbird occurrence with (W) and without (W0) the influence of agriculture present in the model. Shaded symbols represent extrapolations beyond recorded observations . . . . . Cowbird probability as a function of the percent vertical cover in the 0-1 m stratum with (W) and without (W0) agriculture present in the model . . . . Cowbird probability as a function of the percent vertical cover in the 1-7 m stratum with (W) and without (W0) agriculture present in the model . Cowbird probability as a function of total mixed (non- agricultural) openings within 0.5 km of a study site with (W) and without (WO) agriculture present in the model viii 10 23 24 25 26 27 35 36 43 44 45 46 INTRODUCTION Historically, the Great Lakes region once supported high percentages of late successional forest ecosystems. Intensive logging activities conducted over 130 years ago claimed more than 90% of Michigan's virgin forests (Gates and Gysel 1978). Many logged areas grew back as secondary growth forests; however, agriculture, urban, and industrial developments have fragmented much of this forested landscape (Wilcove and Whitcomb 1983). JP (X: Recent studies have indicated that fragmentation of breeding habitat in temperate forests coupled with the destruction of winter habitat in the tropics has led to severe population declines for many neotropical migratory bird species (Whitcomb et a1. 1981, Wilcove 1985, Askins et a1. 1990)(i/Forest fragmentation may be detrimental to neotropical migrants due to increased nest predation and brood parasitism pressures along forest edges (Wilcove 1985). :>: Studies by Terborgh (1989) and Wilcove (1985) suggested that nest predation and brood parasitism may best explain the precipitous decline noted in some neotropical migratory bird populations. Terborgh (1989) reasoned that increased urbanization and agricultural development in and around forested regions has provided nest predators and brood parasites with optimal habitat conditions, including ample year-round food resources (e.g., bird feeders, garbage cans, and crop grains). As a result, populations of certain nest predators (e.g., raccoons [2199193 1252;], blue jays [Cyanocittg cristata], crows [goggug brachyrhynghgg], and ravens [Qgrxgg 99:53]) and brood parasites (e.g., cowbirds [Mgloghrug spp.]) have soared in these areas, applying ever increasing {—3 2 pressures upon the populations of neotropical migrants nesting nearby. Some researchers have purported that neotropical migratory birds nesting within interior core areas of large forest tracts, at least 100 m from an edge, find refugia from predation and cowbird parasitism (Reese and Ratti 1988). Cowbird parasitism in particular can devastate the nesting success of susceptible neotropical migrants, in some instances reducing their fecundity by as much as 761 (Mayfield 1965, Payne 1977, Airola 1986). Wilcove (1985) observed that nest predation rates approach 100% in small (3.8 - 13.3 ha) suburban woodlots versus 2% in the Great Smokey Mountain National Park. Similarly, other studies have indicated that the intensity of cowbird parasitism is inversely proportional to the distance from forest openings (Gates and Gysel 1978, Brittingham and Temple 1983). Brittingham and Temple (1983) noted that cowbird parasitism in Wisconsin decreased from 652 to 182 from within 100 m to 300 m of an opening (defined as areas 2 0.2 ha where the forest canopy was > 50% open). As a result, 90% of the avifauna nesting in large forest tracts have been reported to be area-sensitive (interior- forest dwelling), neotropical migratory birds, most of whom are susceptible to cowbird parasitism (Friedmann et a1. 1977, Wilcove and Whitcomb 1983). The brown-headed cowbird (Mglgthzgg aggr) is an opportunistic and obligate brood parasite, laying its eggs in the nests of over 217 known avian species throughout the continental United States (Southern 1958, Hayfield 1965, Friedmann et a1. 1977). Cowbird populations are not dependent upon the density of a particular host, given that they parasitize so many different hosts. Indeed, cowbird parasitism rates 3 are inversely density dependent, that is, as a host species' population/[q declines, the intensity of cowbird parasitism tends to increase (Fretwell 1977). The endangered Kirtland's warbler (pgggrgigg kirtlaggii) is a case in point. The population of male Kirtland's warblers declined from 502 in 1961 to 201 in 1971 (Walkinshaw 1972). Seventy five percent of the nests examined between 1957 and 1971 were parasitized. The mean clutch size was 2.35 and 0.8 young were fledged per pair per year (Walkinshaw 1972, Decapita 1993). With the implementation of a cowbird trapping program, the number of male Kirtland's warblers increased to 488, parasitism decreased to 6.32, the mean clutch size rose to 4.46, and, on average, 3.11 young were fledged per nest (Decapita 1993). Terborgh (1989) predicted that without intensive cowbird trapping efforts, the Rirtland's warbler could go extinct within 5 years. Many neotropical migrants, including the Kirtland's warbler, exhibit a k-selected reproductive strategy (Temple 1977). These birds tend to raise only 1 brood per year from an average clutch size of 2 to 4 eggs, depending on the species (Scott and Ankney 1980, Wilcove and Whitcomb 1983). One episode of cowbird parasitism can devastate the reproductive effort of these birds, again leading to population declines (Mayfield 1965, Payne 1977, Wilcove and Whitcomb 1983, Burgham and Picman 1989). A study conducted by Robinson (1990) in southern Illinois revealed that 76% of the nests located in 14 - 68 ha woodlots contained, on average, 3.3 cowbird eggs per nest. Subsequently, population levels of 7 neotropical migrants declined by 50% over a 5-year period (Robinson 1990). 4 /)K’ Cowbirds can diminish a host species' reproductive effort in a variety of ways. Female cowbirds have the unique ability of ){lsynchronizing their egg laying effort with that of their hosts' at times when the nest is unattended (Payne 1965, Mayfield 1977). Upon encountering an unattended nest, the female cowbird removes as many eggs from the nest as she lays, usually one egg per day per nest. (Mayfield, 1977). Cowbird eggs are generally larger than their hosts', garnering a disproportionate amount of the incubation heat at the expense of the smaller host eggs (Mayfield 1977). Mayfield (1977) noted that the presence of 2 or more cowbird eggs in a nest are almost always lethal to the host's brood. Cowbird eggs hatch after 10 or 11 days, 2 - 3 days ahead of the host's eggs, giving the young cowbirds a distinct size and strength advantage over their foster siblings (Payne 1965, Mayfield 1977). As a result, young cowbirds have been known to either purposely or inadvertently expel any remaining host eggs or young from the nest (Gill 1990). Young cowbirds tend to beg more often, receive more food, and develop at a faster rate relative to their foster siblings, fledging after 10 or 11 days (Southern 1958, Mayfield 1977, Payne 1977, Woodward and Woodward 1979). The reproductive effort of many neotropical migrants may be further compromised by the cowbird's extraordinary prolific abilities. A female cowbird can potentially lay up to 80 eggs in her 15-month lifespan (Scott and Ankney 1980). Amazingly, female cowbirds rely little on body reserves for their tremendous egg production. Instead, they meet their reproductive energy demands entirely through their diet of insects, rice, or various other crop grains (Ankney and Scott 1980). These X a; 5 dietary resources are normally associated with areas having high levels of human disturbance such as residential areas with bird feeders, campgrounds, livestock and agricultural areas, and garbage dumps (Rothstein et al. 1984). Cowbirds tend to maintain separate foraging and breeding areas in conjunction with the prevalence of energy-rich food resources in disturbed areas and the relatively high densities of birds nesting along forest edges, respectively (Mayfield 1965, Rothstein et a1. 1984). Many avian species are attracted to forest-field edges for a variety of reasons, including: optimal nesting sites, singing and observation perches, and cover and food availability (Gates and Gysel 1978). Similarly, forest-interior bird species tend to aggregate their nests near forest-stream edges (Gates and Ciffen 1991). These edges serve to concentrate nests into "ecological traps" whereby density- dependant mortality factors (e.g., nest predation and brood parasitism) tend to increase (Gates and Gysel 1978). There is evidence that cowbirds exhibit increased nest searching activity near habitat discontinuities (e.g., forest edges), cuing in on host nest building activity and aggressive behavior displays (Norman and Robertson 1975, Robertson and Norman 1976). Observations of cowbirds_‘4f. sitting on high perches near forest edges intently watching the nest building activity of other birds prior to parasitism bears an uncanny resemblance to predatory stalking behavior (Payne 1965, Gates and Gysel 1978). Cowbirds, like predators, may be absent from forest interiors-l- due to the tremendous difficulties attributed to locating nests in obscured vegetation. Consequently, female cowbirds may find it more 6 economical to concentrate their search efforts in edge areas where nest densities are likely to be high (Gates and Gysel 1978, Wolf 1987). Many recent studies have indicated that cowbirds are tolerant of forest interiors (Brittingham and Temple 1983, Wilcove et al. 1986, Robbins et a1. 1989, Robinson 1990, Gates and Ciffen 1991, Thompson et al. 1992). Gates and Ciffen (1991) found increased rates of nest predation and brood parasitism along forest stream edges as small as a second order stream, 3.4 m wide. Brittingham and Temple (1983) and Robbins et a1. (1989) noted that an abundance of snags near a stream or tree fall gaps in a forest canopy may serve to attract cowbirds into a forest setting. Robinson (1990) reported that cowbird eggs were prevalent throughout the 104,000 ha Shawnee National Forest of southern Illinois and that woodthrush (flylggighla_gu§§gliga) nests were just as heavily parasitized (901 parasitism) at 400 m from an edge as at the edge. Thompson et al. (1992) found no statistical difference when cowbird numbers were compared between non-harvest forest sites and clearcut management sites (~200 ha in size) in the Mark Twain National Forest of south-central Missouri. Robbins et a1. (1989) studied fragmented forest tracts of various sizes in eastern Maryland and found that tree species abundance and foliage density were significant predictors of cowbird abundance; however, forest area was not. What specific landscape arrangement or habitat component(s), upon manipulation, will tend to exclude cowbirds from forest interiors of northern Michigan is unknown. Currently, the only effective method of controlling cowbird populations is to capture the birds in baited live traps and dispatch them (Kelly and DeCapita 1982). This method has 7 proved highly effective in controlling cowbird parasitism of Kirtland's warblers; however, it requires a considerable investment in terms of time, personnel, and money (Walkinshaw 1972). A more appropriate solution for land managers would include prior knowledge of the specific habitat variables that are known to be significant predictors of cowbird occurrence and to manage for these. Numerous cowbird studies have been conducted in small woodlots surrounded by agricultural and/or urban development; however, little is known about cowbird distributions in relation to forest fragmentation in areas supporting more continuous forest cover (Haufler 1990, Thompson et a1. 1992). Given this lack of knowledge coupled with the understanding that cowbird parasitism can potentially devastate populations of many neotropical migratory birds, research was conducted to ascertain which specific habitat and land-use characteristics, if any, tend to attract cowbirds to forested areas of northern Michigan. The purpose of this study was to design a cowbird predictor model which could be used by natural resource managers to better predict which forest habitats of northern Michigan were susceptible to cowbird occupation so that land-use management plans may ultimately be developed which will exclude cowbirds from the nesting habitats of area-sensitive neotropical, migratory birds. 8 Objectives Specific objectives of this study include the following: (1) assess habitat factors influencing cowbird distributions in forested landscapes, (2) analyze current forest landscape patterns in terms of cowbird habitat affinities, (3) develop a cowbird predictor model, and (4) based on the model, make recommendations for forest management practices that are compatible with maintaining habitat for neotropical migratory birds. STUDY-AREA Research was conducted in the Huron National Forest of northeastern lower Michigan (Alcona County), Pigeon River Country State Forest of north-central lower Michigan, Hiawatha National Forest (Delta County) in the central upper-peninsula of Michigan, and the Huron Mountain Club in the northwestern upper-peninsula of Michigan (Fig. 1). Multiple land- use activities such as timber harvesting, farming, prescribed burning, and recreation are variously implemented throughout these regions (Lantz 1976, Beyer 1987). The Huron National Forest (HuNF) encompasses portions of Alcona, Crawford, Iosco, Ogemaw, and Oscoda counties and lies within the Highplains and Presque Isle physiographic districts (Albert et al. 1986). The HuNF is characterized by hilly moraines and outwash plains with well drained and excessively well drained sandy loam soils in the Highplains district and poorly drained soils of the sand lake plains along Lake Michigan in the eastern Presque Isle district (Albert et a1. 1986). Plant communities associated with the soil types are as follows; northern pin oak (ngxggs ellinggigglig) and Jack pine (Pings hankgigng) occur on the excessively well drained soil types, northern hardwoods such as beech (Eagus grangifglia), sugar maple (Age; gagghargm), hemlock (Isaac analgesia). basswood (Tina M). red oak (Oneness mare). and white pine (Pings sgrgbus) occur on the well drained soils, and northern white cedar (Thule gggiggntglig), balsam poplar (flogging balsamifgzg), trembling aspen (Populus trgmulgiggg), and paper birch (Begglg papyrifera) grow on the poorly drained soils (Albert et al. 1986). The climate of the HuNF is temperate with average temperatures 10 Huron Mountain Club I L I l: Flg. 1. Location of cowbird study areas In northern Michigan. PRCSFaPIgeon River Country State Forest ll ranging from -6.6°C in the winter to 21'C in the summer and annual precipitation ranging from 71 - 81 cm. (U.S. Department of Commerce 1979). The Pigeon River Country State Forest (PRCSF) is a 33,590 ha state forest encompassing portions of Cheboygan, Montmorency, and Otsego counties east of Vanderbilt, Michigan. Physiographic and vegetation associations are similar to those described for the Highplains district of the HuNF. Climatic conditions are influenced by the Great Lakes and tend to fluctuate between continental and semi-marine conditions throughout the year (Moran 1973). The average annual temperature (1940- 1969) is 5.6°C and the mean annual precipitation is 74.9 cm (Michigan Weather Service 1974). Managed wildlife openings occur throughout the PRCSF and agricultural properties adjoin its boundaries (Bender 1992). 'Hiawatha National Forest (HiNF) encompasses approximately 344,000 ha of predominately northern hardwood, aspen, and coniferous forests within portions of Alger, Delta, and Schoolcraft counties (800 Line Railroad Company 1964). The HiNF lies in the Escanaba, Luce, and Dickinson physiographic districts. Site conditions range from poorly drained sand lake plain, sandy end moraine, shoreline, and outwash plains in the east to drumlins and ground moraines in the west (Albert et al. 1986). Soil types in the HiNF include excessively well drained and well drained sandy loams in the upland areas and poorly drained soils in the low lying areas (Albert et a1. 1986). Excessively well drained soils are characterized by red pine (Pings resinosa), white pine, and Jack pine stands. Well drained soils are associated with beech (Eagus grandifolia), sugar maple, hemlock, basswood, and yellow 12 birch (figtgla allgghanigggig) communities. Poorly drained soils include white cedar, tamarack (Lara; larigina), black ash (Egaxingg nigra), red maple (Age; rubrum), balsam poplar, balsam fir (Abigs balsamga), trembling aspen, and paper birch plant communities (Albert et a1. 1986). The climate in the HiNF is temperate with an average May - September temperature of 15.7’C and a mean annual precipitation of 80 cm (Albert et a1. 1986). The Huron Mountain Club (HMC) is a privately owned reservation situated in the Huron mountain region of north-western Marquette County, Michigan. Since its inception in 1889, the HMC has grown to include over 7200 ha of mature - old growth forest along the shore of Lake Superior. With the exception of a 202 selective cut made for white pines in the 1890's and some peripheral clearcuts of hemlock, sugar maple and yellow birch from 1939 - 1950's, the area has received little silvicultural treatment (Simpson et a1. 1990). As in most areas of Michigan, glaciers significantly affected the geomorphology of the Huron Mountain area. The region is topographically diverse including terraces, river flats, mountain slopes, cliffs, and swamps. The relief of the area varies from 312 m (1030 ft) at the lake shore to 495 m (1632 ft) above Sea level atop Ives Hill (Westover 1971). The south facing slopes of the Huron Mountains were heavily scoured by glaciers, exposing precambrian metamorphic bedrock. These areas are sparsely vegetated with crustose and foliose lichens and juniper (Juniperus ggggugig). Excluding the 10 inland lakes on the HMC property, nearly 901 of the remaining area is heavily forested (Simpson et al. 1990). The HMC lies vwithin the Michigamme physiographic district and is dominated by a 13 Hemlock-White Pine-Northern Hardwoods vegetation association (Braun 1950, Albert et a1. 1986). Level areas having well drained medium and coarse sands are dominated by hemlocks and sugar maples with lesser amounts of yellow birch, red maple and basswood. These tree species are similarly associated with the sandy stream terraces and alluvial fan areas composed of well - moderately well drained sands and sandstone. The moderately well drained sands and sandy loans of the floodplains are dominated by sugar maple and with lesser amounts of yellow birch, basswood, and American elm (Ulmus angriggng). The moderately - well drained loamy sands of the valleys, ravines, mountain slopes, and ridges are dominated by hemlock-northern hardwoods and the crystalline bedrock slopes include mountain maple (Age; spigatum) and striped maple (Age; pgnngylganiggm). Lake margins and low-lying swampy areas with peat and muck support alder (Alnug spp.), northern white cedar, hemlock, black spruce (Riga; mariang), yellow birch, red maple, balsam fir, and black ash communities. The well drained medium sands deposited along Lake Superior's beach ridges are home to Jack pine, red pine, white pine, hemlock, and lesser amounts of red oak (Simpson et a1. 1990). The climatic conditions are moderated by the Great Lakes with a mean annual temperature (1932-1971) of 5.8°C and total annual precipitation of 78.7 cm evenly distributed throughout the year (Simpson et a1. 1990). METHODS Research was conducted at a total of 113 study sites during the spring seasons of 1992 and 1993, and of these, 57 were sampled both years. These sites were selected based on current Arc/Info, geographic information system (6.1.8.) coverage as well as information provided by U.S.D.A. Forest Service vegetation maps, Michigan Department of Natural Resources (MDNR) vegetation maps, and ground truthing. Selected study sites were located in mature, northern hardwood forest stands, no less than 4 ha in size, at least 50 years old and having a variety of ecological land-types and land-use activities surrounding them. Those stands containing mean tree sizes < 8 cm diameter at breast height (dbh), streams > 3 m wide, clearings (z 1.0 ha in area and having < 50% canopy cover), human dwellings, or similar disturbance regimes within 100 m of a study site were rejected. Avian Censusing In addition to their parasitic lifestyle, cowbirds have variously been reported to be monogamous, polygamous, promiscuous, or exhibit some combination of these mating systems (Ankney and Scott 1982). Thus, cowbirds are not known to form traditional pair bonds nor have established nesting territories (Darley 1968). Therefore, the Indice Ponctuel d'Abondance (IPA) method for determining relative avian abundance was used in this investigation (Blondel et a1. 1981). The IPA method measures the relative abundance of birds from a fixed point in a stand and has many advantages over other bird censusing techniques. The IPA method can be used by one or two persons to survey more sites in less time relative to standard plot mapping techniques 14 15 (Blondel et a1. 1981, Whitcomb et a1. 1981). Robbins et a1. (1989) determined from extended coverage data that 901 of the bird species present in a stand would be detected during 3, 20-minute counts, using the IPA method. The IPA is touted as an exceptional bird censusing technique in that complex habitat relationships can be compared between sample sites using multivariate analysis (Blondel et a1. 1981, Robbins et a1. 1989). The avifauna within each forest stand was censused following the IPA protocol established by Blondel et a1. (1981) in conjunction with the cowbird's recorded breeding season (early May - mid July) for northern Michigan (Payne 1965, 1977). A male either seen or heard was recorded as 1 pair and a female was noted as 0.5 pair. The approximate center of each forest stand was located in the field using either G.I.S., U.S.D.A Forest Service, or Michigan D.N.R forest vegetation maps. These center points served as IPA census stations. Forest stands that were > 500 ha in extent had 2 or more IPA census stations placed systematically (> 100 m from an edge) near the forest center and at least 200 m apart from one another. These points were assumed to be independent and treated as such in the analysis. Censusing was not conducted on mornings with fog, steady drizzle, prolonged rain, precipitation dripping from the leaves, extreme temperature deviations from the mean, or winds greater than beaufort 3 (13 - 19 kph) (Blondel et a1. 1981, Robbins 1981). Each point was censused 3 times during the breeding season; once in mid-May, once in early - mid June, and once in late June, early July. The order of censusing varied such that each point was censused once at dawn (0530 - 0730 hrs E.D.T), once at mid- 16 morning (0730 - 0930 hrs E.D.T), and once in the late morning (0930 - 1100 hrs E.D.T.). Vegetation Sampling The vegetation structure and composition of each forest stand was sampled in the immediate vicinity of the census station(s). Sapling density, defined as woody stems less than 8 cm dbh and at least 1.5 m tall, was measured in 2 perpendicular belt transects (2 m X 40 m) centered upon each census point and oriented to the 4 cardinal directions. The line intercept method (Gysel and Lyon 1980) was used to determine the percent vertical cover within each of 3 height strata (0 - 1 m, l - 7 m, and > 7 m) along a 20 m transect that extended 270° (randomly selected) from each census point. Horizontal cover was measured using a profile board ruled in 10 cm squares and erected to a height of 3 m at each census station. Measurements were taken in each of the 4 cardinal directions at a distance of 15 m from the profile board. The mean number of squares that were at least 502 obscured within each of 4 height intervals above the ground (0 - 0.3 m, 0.3 m - 1.0 m, 1 - 2 m, and 2 - 3 m, respectively) were recorded and a percent calculated per stratum. The mean forest canopy height was estimated by measuring the height of the tree (> 8 cm dbh) nearest the end points of each sapling belt transect and the tree nearest the census station (a total of 5 measurements) using a Haga altimeter. Similarly, the mean basal area was estimated by using a tubular gauge (Gysel and Lyon 1980). The point-centered-quarter method (Cox 1990) was used to estimate the tree species composition and density within each surveyed forest at 5 points positioned every 50 m along a 200 m transect (oriented east - 17 west) and centered upon each census station. Landscape Quantification Current landscape patterns were quantified for this study, using collages of U.S.D.A Forest Service vegetation maps, MDNR forest vegetation maps, 1987 black and white aerial photos, or 1992 infrared aerial photos, depending on location. Distances from each census point to the nearest edge and the nearest opening were measured (in meters) from the collage base maps using a metric ruler. Edges were defined as early successional stages at least 12 m wide and having no trees 2 8 cm dbh. Edges also included primary and secondary roads, transmission line corridors, shorelines, and openings. Openings were defined as areas 2 0.4 ha and having > 501 open canopy which included grasslands, agricultural fields, and mixed openings (defined as upland and lowland brush areas, seedling—sapling stage forests, and/or selectively cut forests with > 501 of the overstory removed). Spatial quantification of surrounding land-use activities did not include an assessment of forest interior core area for this study. Many researchers have traditionally included this parameter in their research (Whitcomb 1977, Faaborg 1979, Robbins et a1. 1989); however, it is important to note that in northern Michigan, forests are not isolated as ”islands", but extend across the region as a patchy mosaic occasionally interrupted by areas of human activity. To define an interior core area would be an arbitrary decision. Instead, a measure of the percent of the area surrounding each study site occurring as openings was used in the analysis per the conventions established by Robbins et a1. (1989). An acetate dot-grid overlay was used in tandem with the base maps to 18 quantify the area of agricultural, grassland, and mixed opening vegetation types occurring within a 0.5 km, 1.0 km, 2.0 km, and 3 km radius of each study site, respectively. Similarly, a Bryant transparency (Mosby 1980) was used to quantify the relative percentage of a 4-section area, centered on each study site, containing closed forest stands (> 501 canopy closure and trees > 8 cm dbh). Statistical Methods The chi-square test of independence (Ott 1988) was used to analyze the occurence of cowbirds at sites possessing certain a priori conditions. These included; mature hardwood forest stands (>50 years old) at least 400 m from an edge with z 902 canopy cover, forested regions (four sections in size) where at least 901 of the area had closed stands (>501 canopy closure and trees >8 cm dbh), and selectively harvested hardwood stands having a mean basal area z 16 nF/ha. Logistic regression was used to assess which habitat variables (or covariates) were important predictors of cowbird occurrence in northern Michigan hardwood forest stands. Based on these results a cowbird predictor model was developed using logistic regression. The ability to predict the occurrence of cowbirds in a forest stand, especially females, has more practical land management applications than knowing how many cowbirds were at the site. To this end, the logistic regression procedure was employed to develop a reasonably accurate and biologically interpretable cowbird predictor model with few, easily quantifiable covariates. Logistic regression describes the relationship of a dichotomous (e.g., cowbird absence-0, presence-l) dependent variable as a function Lil 19 of a set of independent explanatory variables (Hosmer and Lemeshow 1989). Logistic regression has fewer assumptions relative to linear discriminate function analysis and is equally adept at predicting the outcome (Harrell and Lee 1985). Unlike linear discriminate function analysis, logistic regression does not assume multivariate normality and thus is not dependent upon complex data transformations. Instead, logistic regression assumes that the explanatory variables (covariates) are independent and binomially distributed with a conditional mean regression equation bounded between 0 and 1 (Hosmer and Lemeshow 1989). Stepwise logistic regression procedures were not used in this investigation given that these methods can produce biologically meaningless results (Hosmer and Lemeshow 1989). Instead, the variable selection method was used based on the conventions established by Hosmer and Lemeshow (1989). Data sets from those sites that were sampled in 1992 and 1993 were randomly selected on a case by case basis and used in the analysis except for those sites which underwent silvicultural treatment between years. In these situations, the observations for both years were assumed to be independent and both data sets were included in the analysis. In an attempt to conform to the independence assumption of logistic regression, the land-use variables were divided into distinct, concentric bands corresponding to their respective radii descriptions (i.e. 0.5 km represents the area within 0.5 km radius of a study site, 1.0 km represents the area between the 0.5 km and 1.0 km radii, etc..). Land-use percentages measured within these concentric bands were used in the logistic regression analysis presented in Appendix B. 20 Two-sample t-tests and F-tests were performed on all variables measured in sites having cowbirds versus those lacking cowbirds. A variable was considered for inclusion in the logistic regression model if its t-test, p-value was < 0.25. This liberal p-value served to include potentially important interaction terms. The F-tests were used in selecting the appropriate t-test for equal or unequal variances. Correlation analysis of the variables provided insight as to which variables met the independence assumption in logistic regression. Those variables which correlated significantly (o—0.05) with relative cowbird abundance (RCBA) and were independent or potentially biologically significant, and had a significant t-test result, were used in the saturated logistic regression model. An examination of the Wald statistic as well as each variable's estimated coefficient within the saturated model and the model with only that variable were compared. Those variables exhibiting significance (P<0.10) were noted and remained in the model. Quantile analyses on variables remaining in the model was implemented to ascertain the correct scaling for the covariates (see Hosmer and Lemeshow [1989], for details). Estimated quantile coefficients (B) exhibiting a quadratic or U-shaped pattern would require an appropriate design variable to be used in the model given that these functions tend to exhibit a non-significant zero slope. Upon completion of this analysis, any variable having a non-significant wald statistic was removed except for potential biologically relevant parameters (based on prior investigations from the literature and a priori selected land-use variables). These remaining covariates 21 represented the main effects model. Interaction terms were generated by taking the cross product of 2 covariates. In this phase of the model development, those covariates which weakly contributed to the main effects model based on their Wald statistic, were intently scrutinized in terms of their interaction with other variables. The log-likelihood ratio was calculated for each of the interaction terms included in the model and a G statistic was used to determine if these interactions contributed significantly to the model (Hosmer and Lemeshow 1989). Finally, an examination of the odds ratios for the proposed model indicated which variable(s) were the most significant contributors to the model when controlling for one or more covariates. Regression diagnostics and the assessment of the model's fit based on the deciles of risk, followed the protocol noted by Hosmer and Lemeshow (1989). A detailed description of the logistic regression procedure used in this analysis can be found in Hosmer and Lemeshow (1989). RESULTS Habitat Analysis The mean percent of agricultural openings was relatively high in the Huron National Forest of Alcona County regardless of distance from study site (Fig. 2). The Huron Mountain Club (HMC) had no agricultural lands present within 3 km of any study site and the Pigeon River Country State Forest (PRCSF) had relatively little agricultural influence as compared to the Hiawatha National Forest of Delta County. The mean percent of mixed openings (Fig. 3) and the mean percent of grassland openings (Fig. 4) were greater in those regions undergoing intensive silvicultural management (e.g., Huron, Pigeon, and Hiawatha) and were the least in the HMC where little or no timber harvesting occurred. The mean relative cowbird abundance was highest in Alcona County (0.506), moderately low in both the Pigeon and Delta County (0.007), and non-existent in the HMC (Fig. 5). Alcona County had the highest proportion (0.701) of cowbird occupied study sites relative to the other regions (Fig. 6). Results of the chi-square test for independence indicated that cowbirds avoided forest stands that were 2 300 m from an edge and had 2 702 canopy cover (Table 1). A comparison of the distance to the nearest edge (DNE) with the distance to the nearest grassland or agricultural opening (DNO) for those stands meeting the 400 m and 902 canopy closure requirements indicated that DNO was not as significant as DNE (Table 2). 22 23 Openings Mean Percent of Agricultural .... a. ..... O. l 0.5km 2.0km 3.0 km Distance From Study Site I Huron @ Pigeon Hiawatha I WC Fig. 2. Mean percent agriculture as a function of increasing distance from a study site for the 1992 and 1993 cowbird data collected in northern Michigan hardwood forest stands. 24 0.14 O .1 m 0.1 D O D D 01 a: Mean Percent of Mixed Openings 0 a A 0.5 km 1.0km 2.0km 3.0 km Distance From Study Site I'M" Huron E Pigeon Hiawatha I i-MC Fig. 3. Mean percent mixed openings as a function of increasing distance from a study site for the 1992 and 1993 cowbird data collected in northern Michigan hardwood forest stands. 25 0.1 3.3.?“ vvvv. . . O O O O O O O O 0 0.06 0.04 0.02 chO Ucm_mmoLO mo uchLma cow} 1.0km 2.0km 3.0 km 0.5 km Distance From Study Site I Huron a Pigeon I Hiawatha I i-MC Fig. 4. Mean percent of grassland openings as a function of increasing distance from a study site for the 1992 and 1993 cowbird data collected in northern Michigan hardwood forest stands. Fig . Pairs study 26 0.056 M. l Huron Pigeon Hiawatha HMC Region Mean Relative Cowbird Abundance l l l Fig. 5. Mean relative cowbird abundance (relative to the number of host pairs) in northern Michigan hardwood forest stands as compared across study regions. 27 Proportion of Sites Having Cowbirds Huron Pigeon Hiawatha HMC Region Fig. 6. Proportion of study sites having cowbirds as compared across regions for northern Michigan hardwood forest stands. 28 Table 1. Results of the chi-square test of independence for cowbird occurrence (Present vs. Absent) in stands meeting distance to nearest edge (DNE) and canopy cover criteria. a cov 891 ggver ZQZ cgyg: DNE 400m 300m 200m 300m 200m 300m N Y N Y N Y N Y’ N Y N Y Absent 73 6 72 7 ‘44 - 35 60 19 59 20 58 21 Present 39 0 39 0 21. 18 35 4 34 5 34 5 x3 3.121 3.674 0.036 3.166 2.442 2.878 Prob 0.077' 0.055‘ 0.849 0.075‘ 0.118 0.090‘ N-the number of sites not meeting the DNE and canopy cover criteria Yhthe number of sites meeting the DNE and canopy cover criteria ‘Significant at a-0.10 level Table 2. Results of the chi-square test of independence for cowbird occurrence (Present vs. Absent) in stands meeting the respective distance to nearest opening (DNO), basal area (BA), and closed stand criteria. DNQ :4OOM BA e t on ea 2901 Cover (nF/ha) 901 801 70% Closed. Closed Closed N Y sl6m2 >1 6m2 N Y N Y N Y Absent 57 22 6 73 36 43 12 67 3 76 Present 29 10 9 30 36 3 13 26 4 35 x3 0.064 5.640 23.979 5.147 1.952 Prob 0.800 0.018‘ <0.001‘ 0.023l| 0.162 N-number of sites not meeting the DN0 and/or canopy cover criteria Y-number of sites meeting the DN0 and/or canopy cover criteria ‘Significant at the a-0.10 level 29 An analysis of the chi-square test of independence for the basal area data indicated that sites with basal areas less than lémfi/ha were more likely to have cowbirds present than otherwise; however, this result should be viewed with caution given the small number of sites meeting this criterion. Mean tree basal area correlated significantly with many covariates (Table 3) but not as significantly with relative cowbird abundance (p-0.06). As a result, forest stands with lower basal areas may possess other physical attributes (e.g, low mean canopy heights or high percentages of ground cover [VCl] and mid-story cover [VC2]) which correlate more significantly with relative cowbird abundance and which may be a better predictor of cowbird occurrence. Results of the chi-square test for the four-section forested areas indicated that as much as 201 of the area can occur in open stands and still exclude cowbirds; however, t-test analysis indicated that canopy cover (vertical cover in the > 7 m stratum) was not significantly different between forest stands occupied by cowbirds versus those stands devoid of cowbirds and did not correlate significantly with relative cowbird abundance (Tables 3 and 4). This may indicate that cowbirds are attracted by surrounding land-use patterns more readily than site specific habitat conditions. In general, those sites which had cowbirds present within them, had relatively larger grassland/agricultural openings (47.3 ha) or edge areas (35.7 ha) nearest (291 m and 186 m respectively) to them as compared to those sites where cowbirds were absent (5.04 ha openings within 409 m, 4.62 ha edge areas within 214 m). 30 Table 3. Results of correlation analysis for the relative cowbird abundance data for each variable measured. Variable . Significant Correlates r P TDEN 0.0086 0.9266 TSPP 0.1412 0.1272 BA -0.l710 0.0641 HT' -0.3088 0.0007 SAPL‘ 0.2371 0.0097 HOST 0.0511 0.5823 DNO‘ -0.2873 0.0016 0812 0.1063 0.2518 DNE -0.1539 0.0961 ESIZ 0.1211 0.1915 HCl 0.1801 0.0511 H02 0.0541 0.5604 HC3 0.1062 0.2523 HC4 0.0494 0.5953 VCl‘ 0.3360 0.0002 V02a 0.3070 0.0007 VC3 -0.0944 0.3091 61' 0.2783 0.0023 62‘ 0.1949 0.0344 03‘ 0.1839 0.0462 G4 0.1400 0.1304 Al‘ 0.3258 0.0003 A2‘ 0.4673 0.0001 A3‘I 0.6032 0.0001 A4‘ 0.5497 0.0001 M01II 0.3274 0.0003 MOZ‘ 0.2670 0.0035 Table 3 (Cont'd). Variable Significant Correlates r P M03‘ 0.2040 0.0267 MO4‘ 0.2150 0.0194 Bolded variable- represents those variables used in the saturated logistic regression model (Appendix B). r-correlation coefficient p-probability ‘Significant at p50.05 Variable descriptions are as follows: TDENH tree density (stems/ha), TSPP- tree species abundance, BAP mean tree basal area (nP/ha), HT- mean tree canopy height (m), SAPLP mean sapling density (stems/ha), HOST- cowbird host abundance, DNO- distance nearest grassland or agricultural opening, 0812- opening size (ha), DNE- distance to nearest edge (upland and lowland shrubs, stands with < 501 canopy closure, primary and secondary roads, 5 20 year-old clearcuts), ESIZ— edge size (ha), HCl— percent horizontal cover in the 0-0.3 m stratum, HC2- percent horizontal cover in 0.3 - l m stratum, HC3- percent horizontal cover in the l -2 m stratum, HC4- percent horizontal cover in the 2 - 3 m stratum, VCl- percent vertical cover in the 0 - l m stratum, VCZ- percent vertical cover in the l - 7 m stratum, VC3- percent vertical cover in the > 7 m stratum, 01- percent grassland openings within 0.5 km of study site, CZ- percent grassland openings within 1 km radius of study site, 03- percent of grassland openings within 2 km radius of study site, 04- percent of grassland openings within 3 km radius of study site, Al- percent of agricultural openings within 0.5 km of study site, A2- percent of agricultural openings within 1.0 km radius of study site, A3- percent of agricultural openings within 2.0 km radius of study site, A4- percent of agricultural openings within 3 km radius of study site, MOl- percent mixed openings (upland and lowland shrub, stands with < 501 canopy closure, 5 20 year-old clearcuts, primary and secondary roads) within 0.5 km of study site, MOZ- percent of mixed openings within 1 km radius of site, MOS- percent of mixed openings within 2 km radius of site, MO4- percent of mixed openings within 3 km radius of study site. 32 Table 4. Mean and standard error for all variables measured at sites having cowbirds (Present) versus those sites lacking cowbirds (Absent). Absent Present Variable xiSE(n-79) xiSE(n-39) TDEN 515i24.7 533i21.7 TSPP 4.4510.16 4.77i0.30 BA‘ 24.3i0.75 21.3i0.79 HT. 22.210.35 19.510.41 SAPLP 12501149 2127i252 HOST 18.510.46 19.7i4.46 DNO' 409i25.0 29li24.3 OSIZa 5.04il.56 47.3122.5 DNE 214114.2 136i9.90 ESIZ'l 4.6210.93 35.7i18.7 HCI‘ 0.58810.031 0.714i0.042 HCZ 0.271i0.024 0.30510.038 HC3 0.248i0.022 0.30410.033 HC4 0.279i0.228 0.33310.030 VCI‘ 0.126i0.018 0.287i0.035 VCZ' 0.38910.026 0.55010.041 VC3 0.906t0.015 0.904i0.023 Gl' 0.043i0.008 0.08810.017 62‘ 0.044i0.007 0.08410.014 ‘ caa 0.034i0.005 0.06610.010 64‘ 0.03310.004 0.054i0.007 A1. 0.005i0.003 0.032i0.014 AZ‘ 0.005i0.003 0.03010.008 A3' 0.004i0.001 0.02610.006 A4“ 0.00410.001 0.02510.006 MOI‘ 0.046i0.008 0.13010.020 MDZ‘ 0.07110.008 0.12310.014 Table 4 (cont'd). Absent Present Variable §i8E(n-79) xiSE(ne39) MO3‘ 0.076i0.008 0.10910.008 MO4‘ 0.073i0.007 0.104i0.007 iemean, SE-standard error Bolded Variables- represents those variables that were significant at the p50.25 level and used in the saturated logistic regression model (Appendix B). ‘Significant at p50.10 Variable descriptions are as follows: TDEN-tree density (stems/ha), TSPP- tree species abundance, BAP mean tree basal area.(uF/ha), HT- mean tree canopy height (m), SAPLP mean Sapling density (stems/ha), HOST- cowbird host abundance, DNO- distance nearest grassland or agricultural opening, 0812- opening size (ha), DNE- distance to nearest edge (upland and lowland shrubs, stands with < 50% canopy closure, primary and secondary roads, 5 20 year-old clearcuts), ESIZ- edge size (ha), HCl- percent horizontal cover in the 0-0.3 m stratum, HCZ- percent horizontal cover in 0.3 - 1 m stratum, HC3- percent horizontal cover in the l -2 m stratum, HC4- percent horizontal cover in the 2 - 3 m stratum, VCl- percent vertical cover in the 0 - 1 m stratum, VC2- percent vertical cover in the 1 - 7 m stratum, VC3- percent vertical cover in the > 7 m stratum, 01- percent grassland openings within 0.5 km of study site, 02- percent grassland openings within 1 km radius of study site, 03- percent of grassland openings within 2 km radius of study site, 64- percent of grassland openings within 3 km radius of study site, Al- percent of agricultural openings within 0.5 km of study site, A2- percent of agricultural openings within 1.0 km radius of study site, A3- percent of agricultural openings within 2.0 km radius of study site, A4- percent of agricultural openings within 3 km radius of study site, MOl- percent mixed openings (upland and lowland shrub, stands with < 502 canopy closure, 5 20 year-old clearcuts, primary and secondary roads) within 0.5 km of study site, MOZ- percent of mixed openings within 1 km radius of site, M03- percent of mixed openings within 2 km radius of site, M04- percent of mixed openings within 3 km radius of study site. 34 The t-test analysis of the grassland and mixed opening variables indicated that the relative percent of the area surrounding a forest stand where cowbirds were present was nearly twice as high relative to those stands where cowbirds were absent. The percent of agricultural openings within the surrounding landscape was the most significant factor with nearly 5 times as much agricultural lands occurring within the vicinity of cowbird occupied forest stands as compared to those stands devoid of cowbirds. Further analysis indicated that cowbirds were 3 - 3.5 times more likely to occur in a forest stand when agriculture was present in the surrounding landscape, regardless of its size or distance from the study area (Fig. 7). This was consistent with the correlation analysis results in that relative cowbird abundance correlated significantly with all 4 (Al-A4) agricultural land-use variables (Table 3). When the presence of agriculture surrounding a study site was compared between 4 concentric bands (0.5 km, 0.5 - 1.0 km, 1.0 - 2.0 km, and 2.0 - 3.0 km radii), wherein agriculture occurred no closer than each band's respective minimum distance, the probability of finding cowbirds dropped noticeably in the 2 - 3 km radii band (Fig. 8). Additional analysis of the t-test results indicated that both tree species abundance and host density were not significantly different between forest stands occupied by cowbirds and those stands lacking cowbirds and neither variable correlated significantly with relative cowbird abundance. Sapling density, horizontal cover and vertical cover within the 0 - 1 m and l - 7 m strata tended to be 1.5 - 2 times greater in those stands occupied by cowbirds relative to those stands where 35 0.5 Cowbird Probability D I 0.2 0,5 km 1.0km 2.0km 3.0km Distance from Study Site I Absent mum Present Fig. 7. Mean probability of finding cowbirds (Cowbird Probability) as a function of agricultural occurrence in the vicinity of a study site for northern Michigan hardwood forest stands. 37 cowbirds were absent. Sapling density, vertical ground cover (0 - 1 m stratum), and vertical mid-story cover (1 - 7 m stratum) were also significant correlates of relative cowbird abundance; however, none of the horizontal cover variables (HCl-HC4) proved to be significant correlates of relative cowbird abundance. Avian Census Results An examination of the avian census data indicated that at least ten neotropical migratory bird species were relatively widespread and abundant (species are underlined in Table 5). These species tended to be either ground nesters (0 - l m stratum) or mid-story nesters (1 - 7 m stratum) and were most abundant in those forest stands where ground cover and mid-story cover was relatively prevalent. Cowbirds and several avian predators (e.g., crows, blue jays, and ravens) were widespread and abundant in these stands as well. Cowbird Predictor Mbdel Variables selected for the logistic regression model based on the correlation and t-test results were noted in Tables 3 and 4 respectively. Those variables which correlated significantly with relative cowbird abundance were included in the model with the exception of sapling density given its significant correlation with other inclusive vegetation cover parameters (e.g., vertical cover within 0 - l m and l - 7 m strata). All of the land-use variables, basal area, and distance to nearest edge parameters were included based on the earlier chi-square test results. This represented the first examination of the variables used in developing the cowbird predictor model. A more detailed description of the cowbird predictor model development is 38 Table 5. Results of avian censuses conducted from May 15 - July 7, 1992, and May 12 - July 10, 1993. Species R. O of lurin- Average e points at alder of alder of g drich pairs pairs in 1 birds detected stands risers 6 were detected :1 noted as as” as as as leotropicel Migr-Its Wc-d 1-4 as u 4.5 4.0 1.67 1.61 Bey-breasted warbler 2 1 1 1.0 1.0 1.0 1.0 Blackburnian warbler 2-4 0 8 .0 2.0 0.0 1.19 - b w 1-4 53 97 5.0 5.0 1.96 2.0 l - e warb 1-4 0 79 0.0 4.0 0.0 1.63 Black and white warbler 2 2 0 1.0 0.0 1.0 0.0 Bobolink 3 l 0 1.0 0.0 1 0 0.0 Cerulean warbler 2 0 l 0.0 1.0 0.0 1.0 Chestnut-sided warbler 1-4 4 14 1.0 2.0 1.0 1.29 Chipping sparrow 1-3 3 l 1.0 1.0 1.0 1.0 W 1-4 45 52 4.0 3.0 1.53 1.33 Eastern phoebe l-4 0 3 0.0 1.0 0.0 1.0 Field sparrow 1-3 0 4 0.0 2.0 0.0 1.25 Golden-winged warbler 2 1 0 1.0 0.0 1.0 0.0 Great crested flycatcher l 0 l 0.0 1.0 0.0 1.0 Hermit thrush 1-4 0 44 0.0 4.0 0.0 1.25 Indigo bunting 2-3 3 7 1.0 2.0 1.0 1.14 W 1-4 37 61 9.0 7.0 2.8 3.16 Lark sparrow l l 0 1.0 0.0 1.0 0.0 Northern parula l 0 2 0.0 1.0 0.0 1.0 Orchard oriole l l 0 1.0 0.0 1.0 0.0 W 1-4 60 108 8.0 7.0 4.17 3.48 Pine siskin 3-4 0 11 0 0 4.0 0.0 1.45 s - e o b 1'4 51 94 4.0 5.0 2.04 2.15 W 1-4 60 109 6.0 8.0 3.55 2.90 Ruby-crowned kinglet 4 0 1 0.0 1.0 0.0 1.0 Scarlet tenager 1-4 7 14 2.0 1.0 1.29 1.0 Solitary vireo 1,2,4 0 13 0.0 3.0 0.0 1.38 Sweinson's thrush 4 0 12 0.0 3.0 0.0 1.33 39 Table 5 (cont'd). Species R‘ I of Phi- Me e points at me: of we: of g flich pairs pairs in 1 birds detected studs Idlers 0 were detected n mted ea m? u: es a: as W 1-4 - 41 0.0 2.0 0.0 1.17 M 1-4 - 57 - 3.0 - 1.42 Veery/Hood thrush. 1-3 42 - 6.0 - 1.79 - limits-throated sparrow 1-3 6 10 2.0 2.0 1.17 1.40 Yellow-bellied flycatcher 2 l 0 2.0 0.0 1.0 0.0 Yellow-bellied sepsucker 2 2 0 1.0 0.0 1.0 0.0 Yellow ruaped warbler l-4 0 49 0.0 4 0 0.0 1.51 Yellow warbler 2 0 l 0.0 1 0 0.0 l 0 Winter wren 2-4 0 18 0.0 2.0 0.0 1.22 Short dist-see Migrate Black-backed 3-toed woodpecker 2 0 1 0.0 1.0 0.0 1.0 Canada goose 1,2,4 0 8 0.0 2.0 0.0 1.13 Killdeer 1 0 1 0.0 1.0 0.0 1.0 Cannon icon 4 0 8 0.0 1.0 0.0 1.0 Ring-billed gull 3 0 l 0.0 1 0 0.0 1 0 Sandhill crane 3 0 2 0.0 1 0 0.0 1.0 Blue Jay 1-4 35 62 3.0 4.0 1.74 1.69 Brown creeper 3 6 0 1.0 0.0 1.25 0.0 Grey cetbird 1-3 3 3 1.0 l 0 1.0 1 0 Brown-heeded cowbird 1-3 29 33 3.5 3.0 1.31 1.15 American crow 1-4 39 58 4.0 5.0 1.51 1.43 Evening grosbeek 2 3 0 1.0 0.0 1.0 0.0 Northern Flicker 2-3 6 0 2.0 0.0 1.13 0.0 Comon grsckle 1-3 8 6 2.0 1.0 1.13 1.0 Mourning dove 1-2 2 1 1.0 1 O 1.0 1.0 Red-breasted nuthetch 4 0 8 0.0 3.0 0.0 1.25 White-breasted nuthetch l-4 20 43 2.0 3.0 1.18 1.12 Broad-winged hewk 4 0 2 0.0 1.0 0.0 1.0 Red-shouldered hawk 1-2 3 4 1.0 1.0 0.83 1.0 Rufous-sided towhee 2 3 0 l. 0 O . 0 1 . 0 0 .0 40 Table 5 (cont'd). Species R‘ f of w-n— Average a points at radar of under of g Idlicb pairs pairs in 1 birds detected stands there 0 were detected n noted an sfl' In 93 an as American robin 1-4 37 59 3.0 3.0 1.38 1.27 Red-winged black bird 1-3 4 2 2.0 1.0 1.25 1.0 Pam fluid-Its Barred owl 2 1 0 1.0 0.0 1.0 0.0 Black-capped chickedee 1-4 5 43 1.0 4.0 1.0 1.5 Rutted grouse 1-2 2 2 1.0 1.0 1.0 1.0 Spruce grouse 4 0 1 0.0 1.0 0.0 1.0 Downywoodpecker 2-4 0 10 0.0 2.0 0.0 1.10 Hairy woodpecker 1-4 35 51 3.0 3.0 1.26 1.16 Pileeted woodpecker 1-4 1 7 1.0 1.0 1.0 1.0 Northern raven 2-4 24 24 3.0 2.0 1.46 1.13 European sterling 1-2 3 0 2.0 0.0 1.33 0.0 Tufted titmouse 1-2 1 3 1.0 1.0 1.0 1.0 Wild turkey 1-3 9 6 1.0 1.0 1.0 1.0 ‘Region 1: Huron National Forest (Alcona County) Region 2: Pigeon River Country State Forest Region 3: Hiawatha National Forest (Delta County) Region 4: Huron Mountain Club bStudy years wherein 60 sites were censused in 1992 and 110 sites were censused in 1993 °flnggxlingg birds were the most common neotropical migrants censused. dScientific names are provided in Appendix A, Table A-2 ‘Identification error 41 presented in Appendix B. The final logistic regression model for predicting cowbird occurrence achieved a 91.61 concordance level with just 5 covariates (Table 6). Basal area and distance to nearest edge were not significant contributors to the model contrary to prior chi-square test results. In an effort to determine which of the 5 remaining variables were significant predictors of cowbirds with and without the influence of agriculture, cowbird probabilities were calculated for each covariate using its respective minimum, 25th percentile, median, 75th percentile, and maximum values, while controlling for the occurrence of agriculture in the landscape. The mean canopy height covariate exhibited an inverse relationship with cowbird probability with or without the presence of agriculture (Fig 9). This is consistent with the negative value of its estimated coefficient or slope. Cowbird probability was greater for low canopy heights (5 19 m), especially when agricultural lands were present. Extrapolation beyond the minimum recorded canopy height observation using the logistic regression procedure‘indicated that a 50% cowbird probability would be expected for a study site with a zero tree canopy height value when controlling for the presence of agriculture. The remaining covariates displayed a positive relationship with cowbird probability and tended to have the highest probabilities when agricultural lands were present (Figures 10 - 12). Each of these covariates, by itself, was a significant predictor of cowbird occurrence in the absence of agricultural lands. Percent ground cover (VCl) proved to be the most significant predictor of cowbird abundance when the influence of agriculture was controlled for in the model. 42 Table 6. Final Logistic regression model for predicting cowbird occurrence in northern Michigan hardwood forest stands. Variable 8 SE weld x3 Prob> x3 Intercept 4.182 2.435 2.950 0.0859 HT -0.4009 0.1226 10.70 0.0011 VCl 5.654 1.583 12.76 0.0004 VC2 2.591 1.159 4.997 0.0254 AD4 1.564 2.360 2.880 0.0172 T01 4.005 0.657 5.671 0.0897 Concordant- 91.61 Tied- 0.0% Discordant- 8.41 B— estimated coefficient 88- standard error of estimated coefficient HT- mean canopy height VCl- percent vertical cover in the 0 - 1 m stratum VCZ- percent vertical cover in the 1 - 7 m stratum AD4- occurrence of agriculture within 3.0 km of a study site TOl- total percent of non-agricultural openings (primary and secondary roads, 5 20 year-old clearcuts, selective cut areas with < 501 canopy closure, grassland areas, upland and lowland brush) within 0.5 km of a study site 43 1.0 - ————— wo > if W 1) <5 13 9 CL 9 '5 s C) Q 13 l 80 4O Mean Canopy Height Fig. 9. Response of mean canopy height to predicting cowbird occurrence with (V) and without (90) the influence of agriculture present in the model. Shaded symbols represent extrapolations beyond recorded observations. 44 > ;—--3 /‘ W cm 0.6 — / D / 9 4s Q I E i g 0.4 ~—,' 0 l O .' .' 0.2 J; } r i 0.0 ' l 1 l 1 J 00 0.2 0.4 0.6 0.8 1.0 V01 Fig 10. Cowbird probability as a function of the percent vertical cover in the 0-1 m stratum with (V) and without (90) agriculture present in the model . 3‘ f? 0.6 — 13 9 D. E g 0.4 — C) o 0.2 — 45 Fig. 11. Cowbird probability as a function of the percent vertical cover in the 1-7 m stratum with (V) and without (90) agriculture present in the model . 46 ----- WO >i / z; W é 1/ i If E I g 0.4 —‘I O I Q I I l 0.2 'T I' I l 0.0 I L I I I I 00 0.2 0.4 0.6 0.8 IO T01 Fig. 12. Cowbird probability as a function of total mired (non- agricultural) openings within 0.5 km of a study site with (V) and without (90) agriculture present in the model. 47 The Hosmer and Lemeshow chi-square goodness-of-fit test (6) was used to assess how well the final model had fit the data. Deciles of risk were calculated (Table 7) based on the conventions established by Hosmer and Lemeshow (1989). The 6 result was not significant (p-0.31) indicating that, overall, the observed and expected frequencies of cowbirds were not significantly different from one another. The final cowbird predictor model is presented below in equation form. Cowbirds are predicted to be present within those habitats where the calculated cowbird probability (a) is greater than 0.50 and absent if x < 0.50. ms(x)/(1+es(x)) Where g(x)- [(-0.4009*HT) + (5.654*VCl) + (2.591*vc2) + (1.564*AD4) + (4.oos*ro1) + 4.182] 48 Table 7. Deciles of risk analysis for observed (OBS) and expected (EXP) frequencies for the presence (P) and absence (A) of cowbirds. Deciles of Risk Total Cowbird l 2 3 4 5 6 7 8 9 10 P OBS 4 0 2 0 2 5 4 4 8 10 39 EXP 1.7 1.6 1.3 1.4 3.7 3.8 5.9 3.7 7.7 11.2 41 A. OBS 45 10 3 4 6 3 5 1 1 2 79 EXP 47.3 8.4 3.7 2.6 4.3 4.9 3.1 1.3 1.3 0.8 76 Total. 49 10 5 4 8 7 9 5 9 12 118 Each decile refers to the probability (risk) of cowbirds being present (i.e., 1 - 0.1 or 10% probability) x2-13.803, P-0.31 with 8 d.f. DISCUSSION The prevalence of cowbirds within a forested region represents a landscape level problem (Robinson et al. 1993a, Thompson et al. 1993a). Many researchers have traditionally focused on identifying specific stand-level habitat components which influence cowbird distributions and then make management recommendations based on disjointed correlates. No attempt has been made to synthesize correlates into an applied management tool. The development of a cowbird predictor model is essential if effective long term cowbird management solutions are to be made. The cowbird predictor model developed from this investigation may allow resource managers to quickly assess which forested areas are currently at a high or low risk of attracting cowbirds. Projections of this model to current and proposed land management scenarios could potentially address the fate of neotropical migratory birds nesting in these forest stands. The cowbird predictor model was developed for use in northern Michigan hardwood forest stands. Its use in landscapes outside of the northern Great Lakes region would need to be evaluated prior to developing regional management recommendations. Our analysis of tree species abundance and host abundance, for example, indicated that these parameters were not significant cowbird predictors in northern Michigan hardwood forests; however, Robbins et a1. (1989) and Thompson et al. (1993b) noted their significance for forested sites located in Maryland and Missouri, respectively. In addition, Robinson (1990), noted that cowbirds were pervasive throughout the Shawnee National Forest of southern Illinois regardless of distance (> 400 m) from edges or 49 50 agricultural lands. He suggested that cowbirds in this region were so prevalent that the forest was simply saturated with cowbirds (Robinson et al. 1993a). This appeared not to be the case in northern Michigan forests, especially in the Hiawatha National Forest and Huron Mountain Club regions wherein relatively large tracts of forested land still occur and have little or no agricultural influences within 3 km of most study sites and no record of cowbirds. Results of the t-test, correlation, and logistic regression analysis indicated that five covariates (i.e., the percent vertical cover in the 0 - l m and l - 7 m strata, the occurrence of agriculture within the landscape, mean canopy height, and the total percent of non- agricultural openings within 0.5 km of a study site) were the most significant predictors of cowbird occurrence within northern Michigan hardwood forest stands. The most influential covariate was the occurrence of agriculture in the landscape. In the absence of agriculture, cowbird probability dropped noticeably across all covariates. Additionally, cowbirds were 3 - 3.5 times more likely to occur when agricultural lands were within 3.0 km of a study site, and the probability of finding cowbirds diminished appreciably when agricultural lands occurred outside of this 3.0 km buffer. The high relative cowbird abundance noted for study sites in Alcona County may best be accounted for by the high percentage of agriculture present in the landscape. Study sites within Delta County, however, indicated that agriculture was a more prominent feature of the landscape than in the PRCSF and yet the proportion of sites having cowbirds was relatively lower for Delta County. A closer examination of the data 51 revealed that each of the 4 study sites located in the Stonington peninsula of Delta County had cowbirds present. Most likely, this can be attributed to the prevalence of private landholdings and agricultural settings scattered throughout the Stonington area. The remaining 21 sites located to the north of the Stonington had (with l exception) no recorded cowbirds nor agricultural lands present within 3 km. Likewise, study sites in the Huron Mountain Club had no agricultural lands present within 10 km and no records of cowbirds. The relatively low proportion of cowbird occupied forest sites in the PRCSF may be attributed to the overall continuity of the forested area. The nearest agricultural lands were on the periphery of the PRCSF (z 5 km from the center of the PRCSF) and most sites (31 out of 43) were located more than 3 km from these areas. The presence of agriculture and urban settings within the landscape provides cowbirds with crucial foraging areas and may be their most limiting factor (Rothstein et a1. 1984, Robinson et al. 1993b, Thompson et al. 1993b). This suggests that managing for natural (non- agricultural and non-urban) ecosystem continuity in the landscape may further reduce the occurrence of cowbirds within those hardwood forest stands lying to the interior of these areas. As such, the removal of agricultural lands or other potential foraging resources (e.g., bird feeders, horse camps, camping areas) from forested landscapes, whenever possible, should be a top priority if effective cowbird management plans are to be implemented (Rothstein et al. 1984, Robinson 1993c). Outside the influences of agriculture, the percent vertical cover in the 0-1 m stratum was the next most important predictor of cowbirds. Indeed, relative cowbird abundance correlated significantly with this 52 covariate as did host abundance.' Further, cowbird probability was also found to increase as the percent mid-story cover (1 - 7 m) increased. It is well known that avian density tends to increase as the structural diversity within a forest stand increases (DeGraff et al. 1993, Thompson et al. 1993a). In forests, species segregation results from the use of these vertical layers (Chasko and Gates 1982). This may be attributed to the increased availability of nesting sites and foraging opportunities for a wider variety of birds occupying the same forest. Recently, researchers had noted that host abundance correlated significantly with cowbird abundance (Thompson et al. 1993b). This was not true of our investigation (p-0.58), however, host abundance did correlate significantly with the percent vertical cover in the 0 - 1 m stratum (p-0.0004) which, in turn, was also a significant predictor of cowbirds. This makes intuitive sense, if one considers that female cowbirds probably do not census each forest stand they visit. Instead, cowbirds may intuitively select conditions that support a relatively greater abundance of hosts, thereby increasing their parasitism opportunities. The correlation analysis of mean canopy height indicated a significant inverse relationship with relative cowbird abundance. Cowbird probability tended to increase with declining mean canopy height as well. Low canopy heights (5 19 m) in this study were indicative of young ( < 20 years) forest stands. By using the logistic regression model to extrapolate beyond the recorded minimum (13.4 m) canopy height observation, one could expect to have an ever increasing probability of attracting cowbirds to low (5 19 m) canopy forest sites. It is not 53 unreasonable to assume that these younger stands would have relatively higher percentages of ground cover (0 - l m) and mid-story cover (1 - 7 m), and higher sapling densities. Relatively higher abundances of neotropical migrants were observed in mature (> 40 years) northern hardwood forest stands wherein ground cover and mid-story cover were prevalent. Thus cowbirds may also be attracted to young, structurally diverse hardwood forest stands given the likelihood that more hosts and more parasitism opportunities will occur in these stands as well. The total percent of non-agricultural openings within 0.5 km of a study site (T01) proved to be a significant predictor of cowbirds as well. That is, cowbird probability tended to be the highest in those stands situated within a diverse and highly fragmented area. Ironically, greater avian densities and species diversity are associated with habitat heterogeneity as a result of forest fragmentation (DeCraff et al. 1993, Thompson et al. 1993a). This presents a paradox to land managers desiring to maintain populations of many neotropical migratory bird species and, at the same time, reduce the presence of cowbirds when both are positively influenced by structurally diverse and heterogeneous forested areas. Overall, average landscape heterogeneity surrounding a forest site appeared to be the greatest in Alcona County as noted by the high percentages of mixed openings and grasslands in Figures 3 and 4 respectively. This stands in stark contrast to the Huron Mountain Club region where stand disturbances had been minimized and cowbirds were 54 non-existent. This further suggests that the disturbance of surrounding forest stands may be an important factor in predicting where cowbirds will occur in the landscape once conducive land-use patterns are present (e.g., agricultural lands). Indeed, the chi-square analysis for cowbird occupied study sites indicated that the proximity of an edge was more significant than the distance to the nearest grassland opening or agricultural opening. This suggests that the proximity of primary and secondary roads, 5 20 year old clearcuts, and selective cut stands with < 501 canopy cover may play a significant role in determining whether or not cowbirds will occur in a stand once agriculture is present in the landscape. A distance of 300 m or more from an edge, for stands having > 702 canopy closure was important in reducing the probability of cowbirds; however, when this distance to a nearest edge was examined from a multivariate perspective (i.e., logistic regression), it was a much less significant factor. Instead, the total percent of non- agricultural openings within 0.5 km of a study site (T01), was a more significant predictor of cowbird occurrence than the distance to a nearest edge. This would indicate that the overall quantity of disturbed (edge) areas immediately surrounding a hardwood forest stand is more important than just its proximity. Indeed, the chi-square test of the 4-section-area forest data indicated that at least 802 of the area must occur in closed stands if cowbirds are to be limited. Thus, landscape level management solutions may be needed if diverse, heterogeneous areas are to provide edge-dependent and area-sensitive bird species with nesting habitats that are relatively free of cowbird parasitism pressures. The development of a landscape level management 55 plan for controlling cowbirds in northern Michigan will need to focus on reducing cowbird foraging bases (e.g., agricultural lands) in targeted areas and avoid managing other highly fragmented forested areas where agricultural lands are prevalent. Conclusion The occurrence of cowbirds in hardwood forest stands of northern Michigan is most limited by available foraging sites. Those forest stands which occurred within 3.0 km of an agricultural field had the highest probability of cowbird occurrence. In the presence of agricultural influences, habitat heterogeneity and intra-stand _ structural diversity become important factors in determining where cowbirds will occur. Once agriculture is present in the landscape, the maintenance of forest continuity will become essential for reducing cowbird occurrence in forested regions where heterogeneity and intra- stand structural diversity are known to provide needed habitat for many neotropical migratory bird species. Management Recommendations All the evidence gathered thus far indicates that landscape level management decisions will need to be developed first if the risks of cowbird occurrences are to be effectively reduced in the nesting habitats of area-sensitive neotropical migrants. Primarily, the removal of cowbird foraging areas (e.g., agricultural lands, human settlements, and artificial feeding sites) in a forested region is paramount. Unfortunately, many state and federal agencies do not possess the necessary financial resources for acquiring farmlands occurring within forested landscapes. A more reasonable alternative would simply involve 56 the identification and management of northern hardwood forest stands which occur more than 3 km from the nearest agricultural area or urban settlement. Silvicultural treatment of these hardwood forest stands may continue to occur provided cowbird foraging areas lie well outside the 3 km radius of the forest. Evidence for this comes from the observation that those stands situated in the Stonington peninsula of Delta County were managed no differently than those stands located in the northern regions of Delta County and yet cowbirds occurred in all 4 study sites of the Stonington peninsula versus just one out of the 21 sites located in the northern portions of Delta County where agricultural lands were virtually non-existent. The following is a potential projection of the cowbird predictor model to a management scenario for controlling cowbirds in managed forested areas; 1. Identify specific northern hardwood forest stands for neotropical migratory bird management which are no less than 3.0 km from the nearest agricultural or urban settlement (if land acquisition is not an option). If agricultural lands occur within 3.0 km of a potential management site, then identify hardwood forest stands within 4 section sized forest areas where more than 802 of the area occurs in closed canopies (> 701 canopy closure). 2. Within these management units, identify a core region where no roads (primary and secondary), openings (grassland or agricultural areas 2 0.4 ha), or edges (5 20 year-old clearcuts, upland and lowland brush, or areas with < 50% canopy 57 closure) occur within 0.5 km. 3. Move any potential cowbird foraging areas (e.g., bird feeders, camp grounds, garbage dumps, and horse camps) outside the 3.0 km buffer. 4. The removal of camp grounds and bird feeders may be unnecessary if these can be properly managed so as not to provide cowbirds with food resources. a. Implement public education programs (i.e., "Don't feed the cowbirds"). b. Institute low impact camping policies for sites located near management areas and ensure proper disposal of human refuse. c. Encourage the use of bird feeders which have restrictive openings (e.g., thistle feeders), and avoid using millet or corn. Discontinue the use of bird feeders if cowbirds continue to occur. 4. If agricultural lands occur within 3.0 km of a management area then locate large clearcuts and selective cut treatments that reduce canopy cover < 50% outside the 3.0 km buffer unless they comprise < 201 of the landscape in total openings. 5. Use the cowbird predictor model to evaluate the risk or probability that cowbirds will occur in a management area given a proposed forest management scenario. (Note: A calculated probability of 0.49 differs little from 0.51 and yet cowbirds would be declared absent from the 0.49 probability sites and present in the 0.51 sites. Therefore, one should set risk 58 factor goals to be well below the 0.50 probability level [i.e, as close to 0.0 as possible]). Land managers should take care and target significant acreages (> 80 ha) of those ecological landtypes (ELTP 230's - 240's) that are normally associated with northern hardwood forest stands, allow these areas to undergo natural successional changes, and manage these areas for area-sensitive neotropical migratory birds. This would not only ensure the long-term persistence of these northern hardwood forest stands, it would ensure the long-term persistence of the area-sensitive neotropical migratory bird species as well, given that these species are adapted to the natural disturbance regimes associated with these ELTP's. Refer to Haufler and Irwin (1993) for a further discussion of ELTP's and landscape planning. The cowbird predictor model can be a valuable tool in assessing the risk of cowbird occurrence within a management area. The risk (probability) of cowbird occurrence is a factor which is modified by the surrounding landscape. Large, forested regions having no agricultural lands within 10 km or more of'a forested area may require no significant changes in current management policies. However, if cowbird foraging areas become increasingly interspersed within the forest, significant alterations of current management policies may be necessary if neotropical migrants are to survive over the long-term. APPENDIX 59 Appendix A Table Ael. Habitat variables used in the logistic regression analysis. Variable Description SAPL Number of saplings (trees 3 - 8 cm dbh) per ha BA Tree basal area (n? per ha) HT Mean canopy height TSPP Number of tree species TDEN Number of trees per ha HOST Host species abundance RCBA Relative cowbird abundance HCl Percent horizontal cover (0 - 0.3 m) H02 Percent horizontal cover (0.3 - l m) HC3 Percent horizontal cover (1 - 2 m) H04 Percent horizontal cover (2 - 3 m) VCl Percent vertical cover (0 - 1 m) VC2 Percent vertical cover (1 - 7 m) VC3 Percent vertical cover (> 7 m) G1 Percent of area as grassland openings within 0.5 km radius of census point G2 Percent of area as grassland openings within 0.5 - 1.0 km radii of census point G3 Percent of area as grassland openings within 1.0 - 2.0 km radii of census point ‘ G4 Percent of area as grassland openings within 2.0 - 3.0 km radii of census point Al Percent of area as agricultural lands within 0.5 km radius of census point A2 Percent of area as agricultural lands within 0.5 - 1.0 km radii of census point A3 Percent of area as agricultural lands within 1.0 - 2.0 km radii of census point A4 Percent of area as agricultural lands within 2.0 - 3.0 km radii of census point M01 Percent of area as mixed openings within 0.5 km radius of census point M02 Percent of area as mixed openings within 0.5 - 1.0 km radii of census point M03 Percent of area as mixed openings within 1.0 - 2.0 km radii of census point M04 Percent of area as mixed openings within 2.0 ~ 3.0 km radii of census point DNO Distance (m) to nearest opening from census point 0812 Size of nearest opening (ha) DNE Distance (m) to nearest edge ESIZ Size of nearest edge area (ha) TOl Total grass and mixed openings within 0.5 km radius of census point IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII-IIIIIIIIIIIIIIIIIIII 60 Table Ar2. Common and scientific names of censused birds. On I. Scintific l-e Isotropical Migrants American redstart etc a c a Bay-breasted warbler penggica cgstgga Blackburnian warbler dro ca ca Black-throated blue warbler o a ca 1 c Black-throated green warbler Dggdroice vigegg Black and white warbler Qiotilta vgia Bdwiuk Baummmaamunmmm Cerulean warbler Ogdrgica cerulea Chestnut-sided warbler W Chipping sparrow is l a e a Eastern wood peewee W; Eastern phoebe o i oeb Field sparrow Wig; Golden-winged warbler Vgrmivorg chgggptgra Great crested flycatcher mm; Hermit thrush Cgthme ggttgtug Indigo bunting flaggeging cygga Least flycatcher mm Lark sparrow estes r c a Northern parula a is Orchard oriole W Ovenbird s a us Pine siskin mum Rose-breasted grosbeak e cus lu ov c us Red-eyed vireo Vireo olivgcgus Ruby-crowned kinglet W Scarlet tanager a v c Solitary vireo mm Swainson' s thrush W Wood thrush o i te i a Veery Cgthams {gageggens Hhite-throated sparrow Zonotrighia albicollis Yellow-bellied flycatcher Midonax glavivgntgis 61 Table A-2 (cont'd). Cm '- Sci-Itific I- Yellow rusped warbler W Yellow warbler on o is Hint-r wr-n W alert distance Migrnts Black-backed 3-toed woodpecker 2W Canada geese ta a s K1116“: W Cannon loon M Rina-billod cull Larss_§slszs£:nais Sandhill crane W Blue Jay oc ta c t a Brown etc-pox Winnie Gray «third MW Brown-headed cowbird oth a e American crow W Evans arc-huh 111W Northern Flicker Mm Cm srncklo Wiesel: Mourning dove a da a Red-breasted nuthatch mm “bite-breasted nuthatch t s s Broad-winged hawk tee a t Red-shouldered hawk W Rufous-sided towhee i e 0 a1 American robin d s ato u Red-winged black bird Wm Per-snot. Resid-Its Barred owl m Black-capped chickadee a s a u Ruffed grouse s e s 81>ch arous- were; Downy woodpecker co des s a Hairy woodpecker Picoidgs villgsus Pileated woodpecker c i tus North-m r-mn We; Table A-2 (cont'd). em I. Scintific l-e European sterling mmus vulgagis Tufted titmouse Barus bicolg; Wild turkey ris allo avo 63 Appendix B The following is a detailed description of the logistic regression procedure used to develop the cowbird predictor model based on the variable selection method proposed by Hosmer and Lemeshow (1989). The analysis of maximum likelihood estimates for the univariate logistic regression model is presented in Table B-1. Maximum likelihood estimators are analogous to the least squares estimates used in linear regression, that is, unknown parameters which maximize the probability of generating the observed data set are estimated (Hosmer and Lemeshow 1989). The log of these maximum likelihood estimates (log-likelihood) are simpler to use mathematically and can be used in statistical tests. The likelihood ratio test (G-statistic) incorporates this log-likelihood parameter in determining the significance of a logistic regression model with and withouta variable(s). The univariate G-test results for each variable's contribution to the model independent of the other covariates is noted in Table B-1. All covariates exhibited significance except for distance to nearest edge (DNE). However, prior chi-square testing indicated that this parameter was significant. As a result, DNE remained in the model for further analysis in an effort to ascertain potential interaction effects. The saturated logistic regression model consisted of all variables deemed significant from prior analysis (Table B-2). The concordance value noted near the bottom represents how well each observed case of cowbird occurrence matched the expected result calculated from the model. Those variables with a significant Wald chi-square statistic 64 Table B-1. Univariate logistic regression analysis to determine which covariates should be used in the final cowbird predictor model. Var.” 8 SE(B) 0 95% 01 . Log- 0 Likelihood Inter -0.706 0.196 -74.875 HT 0038 0.009 0.962 0.962, 0.963 -72.566 -4.62‘ BA -0.034 0.008 0.967 0.966, 0.967 -72.918 -3.91- DNE -0.003 0.001 0.997 0.997, 0.997 -74.082 -1.59 v01 -0.300 0.693 1.350 0.899, 2.031 -81.697 14.29 v02 -0.619 0.371 0.538 0.344, 0.845 -80.369 11.09 01 -o.067 1.763 0.935 0.743, 1.18 -81.791 13.8‘ (:2 -0.469 1.925 0.625 0.106, 3.67 -81.762 13.8' 03 -0.631 2.963 0.532 0.026, 3.13 -81.769 13.8‘ 04 -5.287 3.788 0.005 4.4E-20, 5.7El4 -80.767 11.8‘ Al 0.916 0.592 2.50 0.864, 7.23 -80.463 11.2. A2 0.916 0.483 2.50 1.05, 5.95 -79.799 9.85‘ A3 0.916 0.418 2.50 1.18, 5.30 -79.135 8.52‘ A4 0.693 0.369 2.00 1.41, 2.83 -79.923' 10.1'1 M01 1.649 1.543 5.20 0.035, 761 -81.205 12.7‘ M02 -1.686 1.497 0.185 0.001, 26.1 -81.144 12.5. M03 -2.975 1.732 0.051 2.1E-6, 1239 -80.268 10.8'I M04 -2.706 1.843 0.067 3.79E-6, 1622 -80.685 11.6“ B-estimated coefficients, SE(B)-standard error of coefficients, W-log- odds ratio, 95% CI- confidence interval for the log-odds ratio, G-log-likelihood ratio statistic ‘-x3 distributed with l d.f at a-0.10 bVariable descriptions are provided in Appendix A, Table A-1. 65 Table B-2. Results of logistic regression for the covariates used in the saturated cowbird predictor model. Variableb Parameter Standard Vald Prob > Estimate Error x3 x2 Intercept 5.486 3.623 2.293 0.1300 BA -0.0008 0.0682 0.0001 0.9907 HT. -0.4664 0.1521 9.408 0.0022 DNE -0.0059 0.0042 1.964 0.1611 VCl‘ 7.628 2.275 11.24 0.0008 VC2a 2.947 1.451 4.123 0.0423 01 3.174 5.126 0.3834 0.5358 G2‘ -26 . 20 11.33 5. 354 0.0207 G3‘ 52.19 20.81 6.292 0.0121 G4‘ -25.64 12.82 3.998 0.0455 A1 0.0182 1.857 0.0001 0.9922 A2 0.1461 2.400 0.0037 0.9514 A3 0.4592 2.331 0.0388 0.8438 A4 1.709 1.840 0.8632 0.3528 MOI‘ 11.16 4.832 5.334 0.0209 M02, -3.126 5.647 0.3063 0.5800 MO3‘ -23.72 11.94 3.949 0.0469 M04a 29.93 13.05 5.257 0.0219 Concordant -94.8X Tied-0.11 Discordant-5.l% 'Represents significant (p50.05) variables that should remain in the model hVariable descriptions are provided in Appendix A, Table A-1 66 remained in the model as did those variables deemed important by the chi-square test of independence. Results of the quantile analysis indicated that virtually all covariates had sequential increases or decreases in their respective estimated coefficients across quantiles thus demonstrating a linear or near linear response to cowbird probability (Table B-3). Quantile analysis could not be used for the percent agricultural opening (Al—A4) covariates given that all observations occurred in the fourth quantile. Instead, a bar graph was used to elucidate the relationship of cowbird probabilities for sites having agricultural lands present or absent within each of the 4 distance bands (Fig. 7 in the text). Results of this analysis indicated that cowbirds were 3 - 3.5 times more likely to occur at a site when agricultural lands were present in the landscape, regardless of its distance. Based on this fact, the occurrence of agricultural openings within a 3 km radius was modeled as dichotomous (present-l, absent-0). All other variables were modeled as continuous and included in the main effects model along with the new agriculture design variable (AD4) (Table B-4). 67 Table B-3. Results of quantile (Q) analysis for determining the appropriate covariate scale to be used in the logistic regression analysis. Var.‘ Q Mid N E 8 w 95% 01 BA 1 14.1 28 2 20.9 29 20.6 -1.86 6.41 0.003, 8.57 3 24.9 31 24.8 0.203 1.22 0.775, 1.94 4 35.8 30 31.4 0.615 1.85 0.451, 7.57 HT 1 16.3 27 17.2 2 20.4 30 20.1 -0.639 0.528 0.133, 2.09 3 22.3 31 22.2 -2.27 0.103 0.001, 13.8 4 27.0 30 25.3 -8.59 1.85E-4 1.13, 3.05 DNE 1 100 28 100 2 150 28 142 0.511 1.67 0.477, 5.83 3 250 32 209 0.435 1.54 0.541, 4.41 4 500 30 256 -0.024 0.976 0.912, 1.04 v01 1 0.013 28 0.008 2 0.074 31 0.069 3.74 42.1 1.9E-4, 9.486 3 0.190 29 0.177 4.74 114 1.6E-5, 8.388 4 0.580 30 0.454 6.00 403 1.3E-7, 1.3El2 v02 1 0.138 29 0.149 2 0.342 30 0.332 0.822 2.28 0.299, 17.3 3 0.510 29 0.490 0.613 1.85 0.474, 7.18 4 0.805 30 0.789 1.54 4.66 0.179, 121 01 1 0 32 0 2 0.013 24 0.014 2.00 7.39 0.045, 1207 3 0.044 30 0.036 1.38 3.97 0.155, 102 4 0.222 32 0.170 2.84 17.1 0.009, 3.1E4 02 1 0.005 27 0.004 2 0.019 31 0.018 3.03 20.7 4.9E-5, 8.786 3 0.049 29 0.048 4.44 84.8 1.2E-7, 5.9810 4 0.239 31 0.151 2.65 14.2 7.5E-5, 2.786 68 Table B-3 (cont'd). Var.‘ Q Mid N E 8 0 95% 01 c3 1 0.007 27 0.003 2 0.023 31 0.021 2.10 8.20 0.004, 1.785 3 0.040 30 0.039 2.36 10.6 0.003, 4.285 4 0.147 30 0.096 1.08 2.94 0.069, 124 G4 1 0.008 29 0.005 2 0 021 26 0.020 2.18 8.85 0.044, 1769 3 0.034 33 0.032 1.88 6.54 0.159, 269 4 0.134 29 0.086 * * * M01 1 0 45 0 2 0.013 9 0.015. -3.35 0.035 3.79, 3.287 3 0.069 34 0.060 0.652 1.92 0.508, 7.26 4 0.255 30 0.219 1.98 7.26 0.003, 1.685 M02 1 0.009 25 0.004 2 0.045 31 0.042 1.84 6.32 0.024, 1653 3 0.113 32 0.110 1.66 5.28 0.029, 975 4 0.308 30 0.205 0.420 1.52 0.329, 7.05 M03 1 0.008 29 0.006 2 0.049 30 0.052 31.4 4.2813 1 08, 1.61 3 0.108 29 0.107 30.0 1.1813 2.09, 5.46 4 0.214 30 0.178 26.7 3.9811 2 10, 7.38 M04 1 0.009 29 0.005 2 0.046 28 0.052 29.8 8.7812 7.81, 9.80 3 0.096 31 0.091 21.0 2.9813 9.16, 9.21 4 0.216 30 0.167 32.3 1.1811 3.33, 3.42 Mid! median of a quantile; N- number of observations in a quantile xi mean of observations in a quantile B—estimated coefficient for a quantile; W—log-odds ratio 951 CI- 95% confidence interval for the log-odds in each quantile (*) indicates missing values due to redundancy 'Variable descriptions are provided in Appendix A, Table A-1 69 Interaction terms were designed and evaluated within the main effects model (Table B-4). The design emphasis was placed on those covariates which exhibited non-significant Wald chi-square results in an effort to improve their contribution to the model. Many of the interaction terms displayed a significant log-likelihood result in the main effects model, however an examination of their respective Wald statistics indicated that these terms would not contribute significantly to the model and therefore none were included (Table B-5). The likelihood ratio test statistic (G) was calculated for those covariates displaying a non-significant Wald chi-square value in the main effects model (Table B-6). The removal of BA, DNE, and M02 had no significant effect on the model, indicating that these variables were‘ not good at predicting the occurrence of cowbirds nor were their interactions. The merging of grassland openings (G1) and mixed openings (M01) into percent total openings (T01) within a 0.5 km radius of a site, did contribute significantly to the model both in terms of the likelihood ratio test and the Wald chi—square test upon removal of the non-significant covariates (Table B-7). This indicates that grassland areas, recent clearcuts, and intensively selected cut areas may be equally attractive to cowbirds. 70 Table B-4. Main effects logistic regression model of the covariates used in developing the final cowbird predictor model. Variable‘ 8 SE Held x3 Prob> x3 Intercept 5.547 3.512 2.487 0.1148 BA 0.0022 0.0668 0.0011 0.9737 HT -0.4691 0.1509 9.664 0.0019 DNE -0.0063 0.0040 2.474 0.1158 VCl 7.803 2.216 12.40 0.0004 VC2 2.944 1.421 4.294 0.0383 AD4 2.235 0.8153 7.514 0.0061 G1 3.249 4.757 0.4665 0.4946 G2 -25.47 11.08 5.287 0.0215 G3 50.84 20.30 6.272 0.0123 G4 -26.40 11.50 5.270 0.0217 M01 11.21 4.778 5.501 0.0190 M02 -3.588 5.492 0.427 0.5136 M03 -24.57 11.10 4.898 0.0269 M04 31.40 12.10 6.728 0.0095 Concordant- 94.71 B—estimated coefficients SE- standard error of coefficients “Variable descriptions are provided in Appendix A, Table A-1 Tied— 0.1% Discordant- 5. 2: 71 Table B-5. Log-likelihood and likelihood ratio test (G) results for interaction terms used in developing the final cowbird predictor model. Interaction Log- G Interaction Log- G likelihood likelihood Main Effect -33.257 DNExGZ -36.988 7.46‘ BAxDNEh -33.018 0.478 DNExG3 -37.422 8.33‘ BAxHT -37.960 9.41‘ DNExG4 —35.618 4.72‘ BAxVC1 -33.842 1.17 DNExMOl -34.226 1.94 BAxVC2 -33.403 0.292 DNExMOZ -34.582 2.65 BAxGl -33.454 0.394 DNExMO3 -34.969 3.42 BAxG2 -33.606 0.698 DNExMO4 -37.752 8.99‘ BAxG3 -34.989 3.46 GleD4 -36.997 7.48‘ BAxG4 -35.230 3.95‘ G1xMOl -34.445 2.38 BAxAD4 -34.372 2.23 G2xAD4 -40.468 14.4' BAxMOl -33.558 0.602 G2xM02 -35.650 4.79‘ BAxMOZ -33.445 0.376 G3xAD4 -39.061 11.6‘ BAxMO3 -34.970 3.43 G3xMO3 -39.641 12.8‘ BAxMO4 -35.023 3.53 G4xAD4 -39.762 13.0‘ DNExHT -37.683 8.85‘I G4xMO4 -38.732 11.0‘ DNExVCl -37.812 9.11‘I MleAD4 -36.734 *7.0‘ DNExVCZ -37.112 7.71‘ MOZxAD4 -35.293 4.07‘ DNExAD4 -35.153 3.79 MO3xAD4 -33.476 0.438 DNExGl -34.553 2.59 MO4xAD4 -36.880 7.25' 'Significant interaction for x3 with l d.f. at a-0.05 bVariable descriptions are provided in Appendix A, Table A-1 72 Table B-6. Results of weld statistic for interaction terms included in the main effects model. Interaction Held X2 Prob > x3 BAxG4‘ 1.82 0.177 DNExG2 2.02 0.155 DNExG3 1.52 0.218 DNEXG4 2.17 0.141 DNExMO4 0.172 0.678 G2xM02 2.24 0.134 G3xM03 1.62 0.204 G4xMO4 0.068 0.795 I'Variable descriptions are provided in Appendix A, Table A-1 Table B-7. Log-likelihood ratio test (G) results for deletion of non- significant variables from the main effects model. Action Log-likelihood G Main Effects -33.200 Remove BA‘ -33.258 0.116 Remove DNE -34.711 3.02 Remove M02 -34.784 3.17 Merge Gl+MOl-T01 -35.477 4.551, “Variable descriptions are provided in Appendix A, Table A21 bRepresents significance for x2 with l d.f. at a-0.05 73 Table B-8. Main effects logistic regression model with non-significant variables removed with estimated coefficients (8), standard error of coefficients (SE), and Held chi-square results. Variable‘ 8 SE weld X2 Prob> x3 Intercept 4.456 3.022 2.175 0.1403 HT -0.4617 0.1417 10.61 0.0011 VCl 6.550 1.820 12.96 0.0003 VC2 3.145 1.318 5.692 0.0170 AD4 2.065 0.7537 7.507 0.0061 T01 8.557 3.181 7.235 0.0072 G2 -29.21 10.85 7.246 0.0071 G3 50.30 19.90 6.390 0.0115 G4 —21.93 10.41 4.437 0.0352 M03 -21.26 9.938 4.574 0.0325 M04 22.51 9.687 5.399 0.0201 Concordant- 93.61 Tied- 0.1% Discordant- 6.32 IIIIII-IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII-IIIIIIIIIIIIIII-I-Il 'Variable descriptions are provided in Appendix A, Table A-1 J; 74 All remaining covariates in the logistic regression model of Table B-8 had significant Wald chi-square results. Concordance diminished slightly from the saturated model but not appreciably. Regression diagnostics was performed to ascertain which observations were highly influential in terms of their deviance from the expected result. Visual inspection of iteration plots indicated that observations 21 and 47 showed consistent deviance from the expected for most covariates. Three diagnostic procedures were calculated for these observations. These included the change in the covariance matrix (A6), the change in the Pearson chi-square residual (AX3), and the change in the deviance (AD) as a result of deleting the respective observations from the data set (Table B-9). Both observations were expected to have cowbirds present and did not. Overall, the results of the three diagnostic procedures indicated that these observations fitted the model poorly (47 fitted better than 21). However the leverage value (hJ) was relatively low in either case indicating that these observations had little influence on the overall predictability of the model and their removal was unnecessary. A closer examination of the estimated coefficients in the logistic regression model of Table B-8 indicated that each of the land-use variables (G2, G3, G4, M03, and M04) had relatively large coefficients and were unstable (alternation of positive and negative signs). This is symptomatic of an over-fitted model (Hosmer and Lemeshow 1989) and, as such, these variables were removed from the final logistic regression model. 75 Table B-9. Regression diagnostics for observations 21 and 47. # HT v01 v02 AD4 801 1r A8 sz AD hJ 21 22.8 0.85 0.18 0 0.12 0.69 2.22 19.9 6.5 0.10 47 17.0 0.18 0.61 l 0.37 0.95 1.10 8.08 4.8 0.12 HT- mean canopy tree height VCl and VC2- percent vertical cover in the 0 - l m and 1 - 7 m strata respectively AD4-presence of agriculture within 3.0 km of a study site TOl- total grassland and mixed openings within 0.5 km of a study site «- probability of cowbirds occurring at a site Afl-change in covariance matrix upon removal of observation Axg-change in chi-square result upon removal of observation AD-change in deviance upon removal of observation 15-1everage value The deciles of risk analysis presented in Table 7 of the results section lent credence to the final logistic regression model as being a relatively effective predictor model for cowbirds in northern Michigan; however, the true test of a model's reliability comes through validation with an independent data set given that any statistical validation of a model which incorporates the original data set will tend to favor that model. Unfortunately, no independent cowbird data set existed. The use of jackknifing procedures on the current data set with just 118 observations and 4 continuous covariates would most likely generate an unstable model thus this procedure was not attempted (Hosmer and Lemeshow 1989). Further validation may be warranted as a means of improving the reliability and efficacy of the model. LITERATURE CITED LITERATURE CITED Airola, D.A. 1986. Brown-headed cowbird parasitism and habitat disturbance in the Sierra Nevada. J.Wild1. Manage. 50:571-575. Albert, D.A., S.R. Denton, and B.V. Barnes. 1986. Regional landscape ecosystems of Michigan. School of Natural Resources, Univ. of Michigan, Ann Arbor. 32pp. Ankney, C.D., and D.M. Scott. 1980. Changes in nutrient reserves and diet of breeding brown-headed cowbirds. Auk 97:684-696. , and . 1982. On the mating system of brown-headed cowbirds. Wilson Bull. 94:260-268. Askins, R.A., J.F. Lynch, and R. Greenberg. 1990. Population declines in migratory birds in eastern North America. Current Ornithol. 7:5-57. Bender, L.C. 1992. The Michigan elk herd: ecology of a heavily exploited population. Ph.D. Diss., Michigan State Univ., East Lansing. l99pp. Beyer, D.E. 1987. Population and habitat management of elk in Michigan. Ph.D. Diss., Michigan State Univ., East Lansing. l48pp. Blondel, J., C. Ferry, and B. Frochot. 1981. Point counts with unlimited distance. Studies in Avian Biology. Cooper Ornithol. Soc. 6:414-420. Braun, E.L. 1950. Deciduous forests of eastern north America. McGraw-Hill Book Co., New York. 596pp. Brittingham, M.C., and S.A. Temple. 1983. Have cowbirds caused forest songbirds to decline? Bioscience. 33:31-35. Burgham, M.C.J., and J. Picman. 1989. Effect of brown-headed cowbirds on the evolution of yellow warbler anti-parasite strategies. Anim. Behav. 38:298-308. Chasko, G.G., and J.E. Gates. 1982. Avian habitat suitability along a transmission-line corridor in an oak-hickory forest region. Wildl. Monogr. 82:1-41. Cox, G.W. 1990. Laboratory manual of general ecology. Sixth ed. Wm. C. Brown Publ., Dubuque, IA. 251pp. Darley, J.A. 1968. Territoriality and mating behavior of the male brown-headed cowbird. Condor 84:15-21. 76 77 Decapita, M.E. 1993. Brown-headed cowbird control on Kirtland's warbler nesting areas in Michigan, 1972-1993. Pages 28-29 in Proc. north American research workshop on the ecology and management of cowbirds. 88pp. In press. DeGraff, R.M., M. Yamasaki, and W.B. Leak. 1993. Management of New England northern hardwoods, spruce-fir, and eastern white pine for neotropical migratory birds. Pages 363-373 in D.M. Finch and P.W. Stangel eds. Status and management of neotropical migratory birds. 0.8. For. Serv. Gen. Tech. Rep. RM-229. 422pp. Faaborg, J. 1979. Qualitative patterns of avian extinction on neotropical land-bridge islands. J. Applied Ecol. 16:99-107. Fretwell, S.D. 1977. Is the dickcissel a threatened species? Am. Birds. 31:923-932. Friedmann, H., L.F. Riff, and S.I.Rothstein. 1977. A further contribution to knowledge of the host relations of the parasitic cowbirds. Smithsonian Contributions 2001. 235:1-73. Gates, J.E., and L.W. Gysel. 1978. Avian nest dispersion and fledgling success in field-forest ecotones. Ecology. 59:871-883. , and N.R. Giffen. 1991. Neotropical migrant birds and edge effects at a forest-stream ecotone. Wilson Bull. 103:204-217. Gill, F.B. 1990. Ornithology. W.H. Freeman and Company. New York, N.Y. 660pp. Gysel, L.W., and L.J. Lyon. 1980. Habitat analysis and evaluation. Pages 305-327 in S.D. Schemnitz, ed. Wildlife management techniques manual. The Wildlife Society. Washington, D.C. 686pp. Harrell, F.E. and Lee, K.L. 1985. A comparison of the discrimination of discriminant analysis and logistic regression under multivariate normality. Biostatistics: Statistics in Biomedical, Public Health and Environmental Sciences, ed. P.K. Sen. New York and Amsterdam: North Holland for Elsevier Science Publications. Haufler, J.B. 1990. Static management of forest ecosystems. Pages 123-130 in J.M. Sweeney, ed. Management of dynamic ecosystems. North Cent. Sect., The Wildl. Soc., West Lafayette, Ind. Haufler, J.B. and L.L. Irwin. 1993. An ecological basis for forest planning for biodiversity and resource use. Proc. Internet. Union Game Biol. Halifax, Nova Scotia. l6pp. (In Press). Hosmer, D.W., and S. Lemeshow. 1989. Applied logistic regression. John Wiley and Sons, Inc. 307pp. 78 Kelly, T., and M.E. DeCapita. 1982. Cowbird control and its effect on Kirtland's warbler reproductive success. Wilson Bull. 94:363-365. Lantz, G. 1976. Compartment prescription and environmental analysis report. U.S. For. Serv., 2410-1 (2/73), Harrisville, Mich. 32pp. Mayfield, H. 1965. The brown-headed cowbird, with old and new hosts. Living Bird 4:13-28. 1977. Brood parasitism reducing interactions between Kirtland's warblers and brown-headed cowbirds. Pages 85-91 in S.A. Temple, ed. Endangered birds: management techniques for preserving threatened species. Univ. Wisconsin., Madison. Michigan Weather Service. 1974. Climate of Michigan by stations. Mich. Dep. Agric. cooperating with NOAA - National Weather Serv. U.S. Dep. Commerce. East Lansing. Moran, R.J. 1973. The rocky mountain elk in Michigan. Research and Development Rep. No. 267. Michigan Dep. Nat. Resour. Lansing. 93pp. Mosby, H.S. 1980. Reconnaissance mapping and map use. Pages 277- 290 in S.D. Schemnitz, ed. Wildlife management techniques manual. The Wildlife Society. Washington, D.C. 686pp Norman, R.F., and R.J. Robertson. 1975. Nest searching behavior in the brown-headed cowbird. Auk 92:610-611. Ott, L. 1988. An introduction to statistical methods and data analysis. PWS-Kent Publ. Co., Boston, MA. 835pp. Payne, R.B. 1965. Clutch size and numbers of eggs laid by brown-headed cowbirds. Condor 67:44-60. . 1977. The ecology of brood parasitism in birds. Ann. Rev. Ecol. Syst. 8:1-28. Reese, R.F., and J.T. Ratti. 1988. Edge effect: a concept under scrutiny. Trans. N. Am. Wildl. and Nat. Res. Conf. 53:127-136. Robbins, 0.8. 1981. Bird activity levels related to weather. Studies in Avian Biology. Cooper Ornithol. Soc. 6:301-310. , D.K. Dawson, and B.A. Dowell. 1989. Habitat area requirements of breeding forest birds of the middle Atlantic states. Wildl. Monogr. 103:1-34. 79 Robertson, R.J., and R.F. Norman. 1976. The function and evolution of aggressive host behavior towards the brown-headed cowbird (Mglgghggg gteg). Can. J. 2001. 55:508-518. Robinson, S.R. 1990. Effects of forest fragmentation on nesting songbirds. Ill. Nat. Hist. Rep. 296:1-2. , J.A. Grzybowski, S.I. Rothstein, M.C. Brittingham, L.J. Petit, and F.R. Thompson. 1993a. Management implications of cowbird parasitism on neotropical migrant songbirds. Pages 93-102 in D.M. Finch and P.W. Stangel, eds. Status and management of neotropical migratory birds. U.S. For. Serv. Gen. Tech. Rep. RM-229. 422pp. , J.P. Hoover, and R. Jack. 1993b. Patterns of variation in cowbird parasitism levels in fragmented Illinois landscapes. Pages l7-18 in Proc. north American research workshop on the ecology and management of cowbirds. 88pp. In Press. 1993c. Plight of forest birds in northwestern Illinois. Illinois Natural History Survey Reports. 322:1-3. Rothstein, S.I., J.Verner, and E. Stevens. 1984. Radio-tracking confirms a unique diurnal pattern of spatial occurrence in the parasitic brown-headed cowbird. Ecology. 65:77-88. Scott, D.M., and C.D. Ankney. 1980. Fecundity of the brown-headed cowbird in southern Ontario. Auk. 97:677-683. Simpson, T.B., F.E. Stuart, and B.V. Barnes. 1990. Landscape ecosystems and cover types of the reserve area and adjacent lands of the huron mountain club. Occasional Papers of the Huron Mountain Wildlife Foundation. No. 4. 128pp. 800 Line Railroad Company. 1964. Upper Michigan forest resources. Forest Products Development Dept. Minneapolis, Minn. 30pp. Southern, W.E. 1958. Nesting of the red-eyed vireo in the Douglas lake region, Michigan. Jack-Pine Warbler. 36:185-199. Temple, S.A. 1977. The concept of managing endangered birds. Pages 3-6 in S.A. Temple, ed. Endangered birds: management techniques for preserving threatened species. Univ. Wisconsin., Madison. Terborgh, J. 1989. Where have all the birds gone? Princeton Univ. Press, Princeton. 207pp. Thompson, F.R., W.D. Dijak, T.G. Kulowiec, and D.A. Hamilton. 1992. Breeding bird populations in Missouri Ozark forests with and without clearcutting. J. Wildl. Manage. 56:23-29. 80 , J.R. Probst, and M.G. Raphael. 1993a. Silvicultural options for neotropical migratory birds. Pages 353-362 in D.M. Finch and P.W. Stangel, eds. Status and Management of neotropical migratory birds. U.S. For. Serv. Gen. Tech. Rep. RM-229. 422pp. , S.R. Robinson, and J. Faaborg. 1993b. Biogeographic, landscape, and local constraints on cowbirds: the importance of scale to managing brood parasitism. Pages 15-16 in Proc. north American research workshop on the ecology and management of cowbirds. 88pp. In Press. U.S. Department of Commerce. 1979. Climate atlas of the United States. National Oceanic Atmos. Adm. 80pp. Walkinshaw, L.H. 1972. Kirtland's warbler-endangered. Am. Birds 26:3- 9. Westover, A.J. 1971. The use of a hemlock-hardwood winter yard by white-tailed deer in northern Michigan. Occasional Papers of the Huron Mountain Wildlife Foundation. No. 4. 59pp. Whitcomb, R.F. 1977. Island biogeography and habitat islands of eastern forest. Am. Birds. 31:3-23. , J.F. Lynch, M.K. Klimkiewicz, C.S. Robbins, B.L. Whitcomb, and D. Bystrak. 1981. Effects of forest fragmentation on avifauna of the eastern deciduous forest. Pages 125-205 in R.L. Burgess and D.M. Sharpe, eds., Forest island dynamics in man-dominated landscapes. Springer-Verlag, New York, N.Y. Wilcove, D.S., and R.F. Whitcomb. 1983. Gone with the trees. Nat. Hist. 9:82-91. 1985. Nest predation in forest tracts and the decline of migratory songbirds. Ecology 66:1211-1214. , C.H. McLellan, and A.P. Dobson. 1986. Habitat fragmentation in the temperate zone. Pages 237-256 in M.E. Soule, ed. Conservation biology: the science of scarcity and diversity. Sinauer Associates. Sunderland, MA. Wolf, L. 1987. Host-parasite interactions of brown-headed cowbirds and dark-eyed jun¢os in Virginia. Wilson Bull. 99:338-350. Woodward, P.W., and J.C. Woodward. 1979. Survival of fledgling brown- headed cowbirds. Bird Banding. 50:66-68. "‘IIIIIIIIIIIIIIII“