Received: 10 January 2025 | Revised: 16 July 2025 | Accepted: 24 August 2025 DOI: 10.1002/wsb.70000 S P E C I A L S E C T I O N Is wild turkey habitat selection spatially consistent? A three ‐ decade meta ‐ analysis in Mississippi Ryo Ogawa 1 | Adam B. Butler 2 | Guiming Wang 1 | Scott A. Rush 1 | Darren A. Miller 1 | K. David Godwin 1 | Mark D. McConnell 1 | M. Kyle Marable 1 | Stanley R. Priest 1 | Brad D. Holder 1 | John E. Stys 1 | James E. Inglis 1 | Benjamin C. Jones 1 | Stephen J. Dinsmore 1 | L. Wes Burger 1 1 Department of Wildlife, Fisheries, and Aquaculture, Mississippi State University, Box 9690, Mississippi State, MS 39762, USA 2 Mississippi Department of Wildlife, Fisheries, and Parks, 1505 Eastover Drive, Jackson, MS 39211, USA Correspondence Adam B. Butler, Mississippi Department of Wildlife, Fisheries, and Parks, 1505 Eastover Drive, Jackson, MS 39211, USA Email: Adam.Butler@wfp.ms.gov Present addresses Darren A. Miller, National Council for Air and Stream Improvement, Box 9681, Mississippi State, MS 39762, USA. K. David Godwin, Mississippi Forestry Association, 6311 Ridgewood Rd. Suite W405, Jackson, MS 39211, USA. M. Kyle Marable, Alabama Cooperative Extension System, Auburn University, 1702 Noble Street, Suite 108, Anniston, AL 36201, USA. Stanley R. Priest, and Brad D. Holder, Mississippi Department of Wildlife, Fisheries, and Parks, 1505 Eastover Drive, Jackson, MS 39211, USA. Abstract Widely distributed species may alter habitat selection based on availability of resources or may respond in dissimilar ways to broadly classified landcovers that exhibit regionally vary- ing ecological distinctions. Failing to account for selection discrepancies may impair management decisions. Eastern wild turkeys ( Meleagris gallopavo silvestris ; hereafter, wild turkey) need heterogenous environments. Although numer- ous studies have investigated wild turkey habitat selection, little work has quantified whether the strength of selection differs spatially. To better understand habitat selection in wild turkeys across dissimilar landscapes, we examined telemetry data gathered across 7 study sites in Mississippi, USA, over a nearly 3 ‐ decade span. After first determining the appropriate scale at which to examine our data, we found wild turkeys exhibited a functional response to total forest cover. Across our study sites, strength of selection for forest cover was inversely related to availability within individual turkey home ranges. Our results further suggest wild turkeys exhibited regional differences in selection of hardwood Wildlife Society Bulletin 2025;49(S1):e70000. wileyonlinelibrary.com/journal/wsb | 1 of 17 https://doi.org/10.1002/wsb.70000 This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2025 The Author(s). Wildlife Society Bulletin published by Wiley Periodicals LLC on behalf of The Wildlife Society. James E. Inglis, Mirim Enterprises, 770 Pleasant View Street, Upper Sandusky, OH 43351, USA. Benjamin C. Jones, Ruffed Grouse Society, 100 Hightower Boulevard, Suite 101, Pittsburgh, PA 15205, USA. Stephen J. Dinsmore, Department of Natural Resource Ecology and Management, Iowa State University, Ames, IA 50014, USA. forests. Implications from our findings provide important considerations for managers and pose questions about the intricacies of wild turkey habitat selection that future research should examine more thoroughly. K E Y W O R D S eastern wild turkey, forest cover, functional response, habitat selection, hardwood forests, home range, landcover analysis, Meleagris gallopavo silvestris , meta ‐ analysis, Mississippi Understanding discrepancies in resource selection is important for explaining variability of wildlife population dynamics in heterogeneous environments (McLoughlin et al. 2010, Matthiopoulos et al. 2015, Sánchez ‐ Clavijo et al. 2016). Johnson (1980) proposed orders of hierarchical habitat selection including 1) the geographic range of a species (order I); 2) home range placement on a landscape (order II); and 3) fine ‐ scale resource or habitat selection within a home range (order III). For home ranges, resource availability likely influences the way in which animals distribute themselves across landscapes. Individual animals may be distributed spatially in proportion to resource availability to maximize energy intake (Fretwell and Lucas 1970). Likewise, if resource availability varies among ecologically similar cover types across regions, animals may exhibit region ‐ specific patterns of selection. When individual animals alter the strength of habitat selection depending upon distributions in resource availability, the process is defined as a functional response (Figure 1; Mysterud and Ims 1998, Bell et al. 2009, Leclerc et al. 2016). Consider, for instance, an animal with a specified feeding resource. If the resource is sufficiently available, the animal may not show a well ‐ defined selection for it relative to cover types used for other life history needs. Conversely, if the availability of the specified feeding resource is below a necessary threshold, selection for it will be positive and significant. Consideration of functional responses is necessary because habitat selection analyses at the population level may mask subtleties in individual animal behavior that can be significant for management (Mysterud and Ims 1998). Awareness of functional responses, should they F I G U R E 1 Theoretical representation of function responses in animal resource selection. In the example, dashed lines represent the boundary of home range, gray areas are food resource patches, and cross points are animal locations derived from telemetry. (A) Represents selection in a more limited availability of food patches, while (B) represents selection in more extant availability of food patches. 2 of 17 | OGAWA ET AL 23285540, 2025, S1, Downloaded from https://wildlife.onlinelibrary.wiley.com/doi/10.1002/wsb.70000 by University Of Florida, Wiley Online Library on [19/12/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License exist, is therefore crucial for providing expectations on how animals might respond to changes in their environment. Eastern wild turkey ( Meleagris gallopavo silvestris ; hereafter, wild turkeys) is one of the most common game birds in the United States. Wild turkeys usually do not fly long distances but are highly mobile (Pelham and Dickson 1992). Wild turkeys are found in a variety of landscapes, but can show strong selection during certain life stages (Porter 1992). Nesting and brooding typically occurs within early successional vegetation that offer female wild turkeys and their young concealment beneath low growing plant communites (Wood et al. 2019, Bakner et al. 2022). Nontheless, wild turkeys are synonymously associated with forest cover in the eastern United States, as total forest and hardwood forest cover have both been shown to be critically important components of wild turkey habitat (Miller and Conner 2007, Davis et al. 2018, Wang 2018). Previous studies of wild turkeys in the eastern United States have documented a specific selection for hardwood forests during portions of the annual cycle (Hurst and Dickson 1992, Miller et al. 1999, Marable et al. 2023). Hardwood forests contain important food resources, such as hard and soft mast, insect ‐ rich understories, and provide vegetation structure to facilitate other life history needs (Dickson 1990, Ross and Wunz 1990, Hurst 1992, Hollifield and Dimmick 1995). Heterogeneous forest covers, whether hardwood or otherwise, are predictive of turkey abundance (Davis et al. 2018, Farrell et al. 2019, Pollentier et al. 2021), as extensive forest diversity is necessary to address the changing needs of wild turkeys throughout various seasons (Miller and Conner 2007). Although ample support exists to suggest wild turkeys select a diversity of forest types and thrive in heterogeous landscapes (Farrell et al. 2019), it is not clear whether findings on forest diversity are the result of the collective selection decisions of individual animals (i.e., functional responses) or due to ecological variation across studies. For managers, the distinciton can be important. If individual wild turkeys demonstrate a func- tional response to forest cover, or specific types of forest like hardwood forests, it would indicate these features have within ‐ home ‐ range thresholds, below which the presence of that cover type becomes increasingly important. Conversely, if wild turkey selection exhibits regional differences in forests that are otherwise broadly classified alike, this may indicate certain ecological characteristics, which may vary regionally, likely play an unrecognized role in the needs of wild turkeys. To better understand the distinction, we investigated how wild turkeys respond to forest cover across varying landscapes. First, we hypothesized that individual wild turkeys may show functional responses to forest cover or certain forest types. Secondarily, we sought to determine whether regional ‐ specific patterns of selection, independent of individual animal variation, might exist. Our goal was to allow for more informed management decisions within these contexts, based on a deeper understanding of how wild turkeys respond to varying environmental conditions. STUDY AREA We used a meta ‐ analytical approach to retrospectivley examine radio ‐ telemetry data sets on wild turkeys collected across 7 study sites in Mississippi, USA, during the mid ‐ 1980s until early 2000s. The 7 study sites were Caston Creek Wildlife Management Area (CCWMA), located in Franklin and Amite counties; commercial forests in Kemper County (Kemper); Leaf River Wildlife Management Area (LRWMA), located in Perry County; Malmaison Wildlife Management Area (MWMA), located in Grenada and Leflore counties; private landholdings surrounded by agriculture to the north (QN) and south (QS) of the town of Marks, Mississippi, in Quitman County; and Tallahala Wildlife Management Area (TWMA), within Bienville National Forest, located in Jasper, Scott, and Smith counties (Table 1; Figure 2; further descriptions of our study sites are available in Supplemental Information). Our 7 study sites represented locations in all 5 wild turkey management regions designated by the Mississippi Department of Wildlife, Fisheries and Parks (Butler and Godwin 2017; Figure 2). WILD TURKEY HABITAT SELECTION | 3 of 17 23285540, 2025, S1, Downloaded from https://wildlife.onlinelibrary.wiley.com/doi/10.1002/wsb.70000 by University Of Florida, Wiley Online Library on [19/12/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License METHO DS Wild turkey capture and telemetry Caston Creek Wildlife Management Area We captured wild turkeys with cannon nets across CCWMA between middle January and early March 1999 and 2000 (Jones et al. 2005). Each captured wild turkey was equipped with a 90 ‐ g very ‐ high ‐ frequency (VHF) trans- mitter with a mortality sensor (Advanced Telemetry Systems, Isanti, MN, USA) affixed to the bird using a backpack harness. Captured wild turkeys were transported in National Wild Turkey Federation (NWTF) transport boxes and released. Triangulation was conducted for wild turkey relocations. Average telemetry error was 8.33 degrees (SD = 6.03; n = 225; Jones et al. 2005). Captured wild turkeys were relocated through reproductive seasons from March 1999 to July 2000. We relocated radiomarked adult female wild turkeys with broods 3 times per day during the brooding season (Jones et al. 2005). Kemper County We captured female wild turkeys between the third week of January and the second week of March and between late June and mid ‐ August during 1986 to 1992 (Miller and Conner 2007). We attached a 108 ‐ g VHF radio transmitter to each turkey with a backpack harness. We estimated locations via telemetry triangulation using 2 fixed ‐ telemetry stations with azimuths less than 12 minutes apart and differing by between 60° and 120° (Cochran and Lord 1963). We located all wild turkeys 3 times per day and 3 days per week during March to June, and twice a day and 2 days a week during the rest of the year (Weinstein 1994, Miller and Conner 2007). T A B L E 1 Study site descriptions for wild turkey studies at Caston Creek Wildlife Management Area (CCWMA), Kemper County site, Leaf River Wildlife Management Area (LRWMA), Malmaison Wildlife Management Area (MWMA), Quitman County North (QN) and Quitman County South (QS), and Tallahala Wildlife Management Area (TWMA), Mississippi, USA. Site Long Lat Ecoregion Total forest (%) Hardwood forest (%) Year TWMA 89°16 ′ W 32°12 ′ N Mostly forested area. Slope of 0 – 16%. 81.5 32.5 1991 – 1992 Kemper 88°31 ′ W 32°47 ′ N Mostly contiguous forested area. Flat with many streams. 84.1 22.7 1991 – 1992 QN 90°17 ′ W 34°19 ′ N Dominated by more agriculture fields than other sites. 16.5 15.7 2009 – 2010 QS 90°21 ′ W 34°10 ′ N Dominated by more agriculture field than other sites. 16.5 15.7 2009 – 2010 MWMA 90°00 ′ W 33°43 ′ N Within alluvial floodplain of the Yalobusha River. Eastern portion includes primarily loess hills. 48.6 45.2 2004 CCWMA 90°54 ′ W 31°24 ′ N Within lower thin loess. Southern Mississippi valley silty uplands soil resource area. 75.3 14.3 1999 – 2000 LRWMA 88°55 ′ W 30°58 ′ N Moderately rolling topography.Slopes from 0 to 16%. 84.3 26.6 1999 – 2000 4 of 17 | OGAWA ET AL 23285540, 2025, S1, Downloaded from https://wildlife.onlinelibrary.wiley.com/doi/10.1002/wsb.70000 by University Of Florida, Wiley Online Library on [19/12/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Leaf River Wildlife Management Area We captured wild turkeys with rocket and cannon nets in the core area of LRWMA between 1 February and 10 March, 1999 and 2000 (Inglis 2001). We equipped each turkey with a 100 ‐ g backpack style transmitter with a F I G U R E 2 Geographic locations of 7 study sites and boundaries of 5 wild turkey management regions of the Mississippi Department of Wildlife, Fisheries, and Parks, during 1984 – 2010. Study sites are Caston Creek Wildlife Management Area (CCWMA), Kemper County site, Leaf River Wildlife Management Area (LRWMA), Malmaison Wildlife Management Area (MWMA), Quitman County sites (north and south), and Tallahala Wildlife Management Area (TWMA), Mississippi, USA. All triangle points represent telemetry locations of wild turkeys. WILD TURKEY HABITAT SELECTION | 5 of 17 23285540, 2025, S1, Downloaded from https://wildlife.onlinelibrary.wiley.com/doi/10.1002/wsb.70000 by University Of Florida, Wiley Online Library on [19/12/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 2 ‐ hour mortality switch (Advanced Telemetry Systems, Isanti, MN, USA). We monitored turkeys from 1 March 1999 to 28 February 2000 (Inglis 2001). Average telemetry system error was 4.93 degrees (SD = 3.33; n = 3,313; Inglis 2001). Malmaison Wildlife Management Area We captured wild turkeys with rocket and cannon nets from 21 January to 13 March in 2004 (Holder 2006). We equipped each turkey with 90 ‐ g backpack ‐ style transmitters using nylon coated rubber tubing (Advanced Telemetry Systems, Isanti, MN, USA; Delahunt et al. 2010). We located each turkey at least 2 days a week from 15 March to 13 August (Holder 2006) via triangulation. Average telemetry error was 8 degrees (SD = 2.5; n = 45; Holder 2006). Tallahala Wildlife Management Area We captured female wild turkeys by cannon nets or with alpha ‐ chloralose from 7 January to 4 March and from 1 July to 25 August during 1984 to 1995 (Bailey et al. 1980, Miller et al. 1999). We fit each turkey with a 108 ‐ g VHF radio transmitter with a motion sensor via a backpack harness (Wildlife Materials, Carbondale, IL, USA; Delahunt et al. 2010). We estimated locations via telemetry triangulation using 2 fixed ‐ telemetry stations with azimuths less than 12 minutes apart and differing by between 60° and 120° (Cochran and Lord 1963). Average telemetry error was 7.2 degrees (SD = 6.3; n = 43; Palmer 1990). We located turkeys at least once per day from 14 March to 1 June each year and 3 or more times per week during the remainder of study periods. Brood ‐ rearing females were located 6 times per day and 3 times per week (Miller 1997). Quitman County – north and south sites We captured wild turkeys using cannon or rocket nets throughout Mississippi from January to March in 2009 and 2010 and relocated them to the Quitman County sites (Marable 2012). Mean distance between the Quitman County sites and trapping sites was approximately 207 km (SD = 133 km). We transported wild turkeys to the study site by a truck with NWTF wild turkey transport boxes (35 × 56 × 65 cm; International Paper, Memphis, TN, USA). We fit each turkey with a 71.2 ‐ g VHF radio transmitter (Model A1540, Advanced Telemetry Systems, Isanti, MN, USA) with a backpack harness. Radio ‐ tagged wild turkeys were located at least 2 days per week from February 2009 to August 2011 (Marable 2012, McKinney 2013). Mean telemetry error was 8.3 degrees (SD = 7.1; n = 40; Marable 2012). Land cover and landscape data We processed land cover data for the 7 study sites from the 30 ‐ m resolution National Land Cover Database (NLCD, https://www.mrlc.gov/) using package raster in Program R (Homer et al. 2015, Hijmans 2016, R Development Core Team 2016). We used NLCD 1992 for Kemper (study year: 1991 – 1992) and TWMA (1991 – 1992), NLCD 2001 for CCWMA (1999 – 2000) and LRWMA (1999 – 2000), and NLCD 2011 for Quitman (2009 – 2010). We used NLCD 2001 and 2006 to interpolate the 2004 land cover data of MWMA. Our response variables included hectares of total forest (i.e., combination of deciduous forest, evergreen forest, mixed forest, and woody wetland) and hardwood forest (combination of deciduous forest and woody wetland). We used CircAn in the program Biomapper to generate proportion maps (ranging from 0 to 1) of each forest type within a circular buffer around each focal 30 × 30 ‐ m cell or pixel (Hirzel et al. 2002). First, we booleanized the 6 of 17 | OGAWA ET AL 23285540, 2025, S1, Downloaded from https://wildlife.onlinelibrary.wiley.com/doi/10.1002/wsb.70000 by University Of Florida, Wiley Online Library on [19/12/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License raster images of total or hardwood forest, assigning 1 to a cell of total or hardwood forest and 0 to a cell of all other cover types. Then, we calculated the proportion of forest types for each grid cell. For MWMA, we calculated proportion of land cover types in the NLCD 2001 and 2006, respectively. Then, we averaged the 2 proportions of each land cover type to represent proportions during 2004. We used 3 different buffer sizes to calculate forest proportions: average wild turkey daily movement distance (390 m), radius of the mean wild turkey seasonal home range area (1,140 m), and radius of the mean wild turkey annual home range area (2,100 m; Phalen 1986, Lambert et al. 1990, Godwin 1991, Miller and Conner 2005, Marable 2012). Hereafter, we refer to the 3 buffer sizes as daily, seasonal, and annual movement scales. Statistical analysis We conducted fine ‐ scale habitat selection analyses (order III; Johnson 1980) for the period from March to August, because the period represented the only time available in our historic data in which radio ‐ tagged wild turkeys were tracked consistently across all 7 study sites. We only used adult female wild turkeys in our study because no male turkeys were tracked in Kemper, MWMA, CCWMA, and LRWMA during the study period. To quantify habitat use, we used telemetry locations, and to quantify habitat availability (i.e., the habitat conditions accessible to each individual), we generated random pseudo ‐ absence locations within each bird's home range. When individuals were located multiple times per day, we randomly selected a single location per day to reduce temporal autocorrelation. We required ≥ 30 daily locations per individual to reliably estimate third ‐ order habitat selection (Seaman et al. 1999). We first estimated 95% minimum convex polygons (MCPs) between March and August for each individual wild turkey using the package adehabitatHR in Program R (Calenge 2006). We then generated n random pseudo ‐ absence locations equal to the number of observed locations ( n = the number of telemetry locations) within individual wild turkey's MCP to quantify available resources. We computed the proportions of the total forest or hardwood forest within the individual MCPs as available forest proportion. In a preliminary analysis, we classified telemetry locations into the biological seasons of wild turkeys following the seasonal divisions of Chamberlain and Leopold (2000). Because habitat selection was not related to seasons (pre ‐ nesting: β = − 0.01, SE = 0.19, z = − 0.06, P = 0.96), we combined data over seasons to increase sample size. Habitat and resource selection by animals may vary across spatial scales (Van Moorter et al. 2013, McGarigal et al. 2016) and measurement of this includes 2 components: extent and grain size (Hobbs 2003). Spatial extent is defined by hierarchical levels of habitat selection (Johnson 1980), while grain is defined as the minimum spatial unit of landscape characteristics animals perceive (Wiens 1989). Misidentifying or neglecting grain size may cause biases in interpreting data on animal ‐ habitat relationships (Laforge et al. 2015, McGarigal et al. 2016). Therefore, to determine the optimal spatial scales of order ‐ III habitat selection (for total or hardwood forest) by wild turkeys, we fit single ‐ variable generalized linear models (GLMs) with a logit link function with combined data over individual wild turkeys following Laforge et al. (2015) to reduce computational costs. For total and hardwood forests, we fit 3 GLMs for each forest proportion at 3 different buffer sizes: daily, seasonal, and annual movement ranges. We conducted model selection with Akaike Information Criteria corrected for small sample size (AIC c ) using package MuMIn in Program R (Burnham and Anderson 2002, Barto ń 2016). The selected model had the lowest AIC c . Models with Δ AIC c < 2 relative to the top ‐ ranked model were considered competing models. The optimal spatial scale resulted in the lowest AIC c value among the GLMs at 3 different buffer sizes for a forest type. We used the best supported spatial scale for the subsequent GLMMs and meta ‐ analysis. We assessed functional responses of habitat selection by wild turkeys using GLMMs with a logit link function. The GLMMs included proportion of total or hardwood forest as a fixed effect, random intercepts for wild turkey identity nested within study site and year, and a random slope of the proportion of forest. Due to the low number of study sites ( n = 7), we also fit GLMMs with study site only as a fixed effect to determine necessity of study site as a random effect by comparing resulting AIC c value with GLMMs including study site as a random effect. We report WILD TURKEY HABITAT SELECTION | 7 of 17 23285540, 2025, S1, Downloaded from https://wildlife.onlinelibrary.wiley.com/doi/10.1002/wsb.70000 by University Of Florida, Wiley Online Library on [19/12/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License variance components ( σ 2 ) for random intercepts and random slopes to quantify among ‐ individual variation in habitat selection. We conducted model selection using a backward selection method (Burnham and Anderson 2002, Zar 2010). We developed all GLMMs using package lme4 in Program R (Bates et al. 2015). We extracted individual wild turkey selection coefficients (i.e., fixed slope combined with individual's random slope) of total and hardwood forests, respectively, from the top ‐ ranked GLMM. To test for functional responses, we regressed individual ‐ specific selection coefficients against within ‐ home ‐ range forest proportions of individual wild turkeys using linear models. We tested the null hypothesis concerning regression slopes of linear models at α = 0.05. We concluded functional responses of habitat selection were supported in wild turkeys if selection coefficients of individual wild turkeys were related inversely to the within ‐ home ‐ range proportions of total forest or hardwood forests. Differences in data collection methods, sample sizes, observers, geographic locations, and times likely resulted in different uncertainties in parameter estimation and heterogeneities in data quality among studies. To account for estimation uncertainties and data heterogeneities among the 7 study sites, we fit meta ‐ analysis models as an alternative method to examine the functional responses of habitat selection by wild turkeys (Schwarzer et al. 2015). We first developed resource selection functions using the GLMs of the logit link function for each wild turkey. We used a bootstrapping method to reduce a potential estimation bias caused by spatial autocorrelation among telemetry locations and standardize unequal sample sizes (i.e., number of telemetry locations) among wild turkeys. We bootstrapped 20 locations from the original telemetry locations and generated 20 random pseudo ‐ absence or background locations within MCPs for each wild turkey. We repeated the process 1,000 times. We fitted a GLM to each of 1,000 repetitions. We calculated the mean and variance of 1,000 estimated selection coefficients for each wild turkey. We conducted meta ‐ regressions of relationships among individual wild turkey's habitat selection coefficients and within ‐ home ‐ range forest proportion with study site as covariates. We used restricted maximum likelihood estimators with Knapp and Hartung adjustment for the unbiased standard errors of coefficient estimates (Knapp and Hartung 2003). To minimize multicollinearity, we did not include forest proportion and study site simulta- neously in the same models. The 2 forest proportions were significantly correlated with study site (total forest: F 6, 147 = 321.40, P ≤ 0.01; hardwood forest: F 6, 147 = 97.88, P ≤ 0.01). We selected the top ‐ ranked meta ‐ regression model using AIC c . Models with Δ AIC c < 2 were considered competing. In the preliminary analyses, we included square and cube terms of forest proportions as covariates, but neither were supported by AIC c . Therefore, we did not include nonlinear terms in our subsequent analyses. We also measured heterogeneity of selection coefficients with study sites. Cochran's Q statistic was used for evaluating significance of the heterogeneity (Schwarzer et al. 2015). All meta ‐ analysis models were developed using the package metafor in Program R (Viechtbauer 2010). R E S U L T S We included 16, 20, 33, 14, 16, 16, 15, and 18 wild turkeys and 4,672, 2,492, 2,068, 1,342, 3,972, and 3,510 telemetry locations in the analyses of order ‐ III habitat selection for CCWMA, Kemper, LRWMA, MWMA, TWMA, QN, and QS, respectively. In our single ‐ variable analyses, the GLM of daily movement scale had the lowest AIC c value for total and hardwood forests. Therefore, daily movement scale was supported as the optimal spatial scale of order ‐ III habitat selection for both total and hardwood forest (Table 2). The top ‐ ranked GLMMs of total forest and hardwood forest included random slopes of total and hardwood forest proportions, respectively, and random intercept (Table 3). Variance estimates ( σ 2 ) in the random slope for hardwood forest proportion was σ 2 = 0.28, indicating substantial individual ‐ level variation in the strength of selection for hardwood forest. In contrast, the variance in the random slope for total forest proportion was σ 2 < 0.01, suggesting little individual ‐ level variation in the selection strength for forest cover. The random intercept variance was comparatively small, indicating limited variation in baseline selection among individuals relative to 8 of 17 | OGAWA ET AL 23285540, 2025, S1, Downloaded from https://wildlife.onlinelibrary.wiley.com/doi/10.1002/wsb.70000 by University Of Florida, Wiley Online Library on [19/12/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License variation in the functional response. Inclusion of the random slopes in the top ‐ ranked model suggested a possibility of functional responses of order ‐ III habitat selection to availability of forests. However, in linear models regressing individual selection coefficients against within ‐ home ‐ range forest proportions, functional responses of order ‐ III habitat selection were not supported for either forest type (total forest: β = 0.01; SE = 0.01, t 130 = 0.18, P = 0.86; hardwood forest: β = − 0.15; SE = 0.10, t 130 = − 1.48, P = 0.14). In the meta ‐ analysis of the selection coefficients of bootstrapped GLMs, selection coefficients of total forest proportion were related inversely to within ‐ home ‐ range total forest proportion ( β = − 5.55, SE = 0.84, t 130 = − 6.59, P ≤ 0.01; Table 4, Figure 3). However, the selection coefficients of hardwood forest proportion were not related to within ‐ home ‐ range hardwood proportions ( β = − 1.46, SE = 0.84, t 130 = − 1.74, P = 0.08; Table 4). Study site was included in the top ‐ ranked model of hardwood coefficients (Table 4). In the heterogeneity test among study sites, total and hardwood selection coefficients were significantly heterogeneous among study sites, possibly suggesting that other factors might also influence the selection of total and hardwood forest by wild turkeys (total forest: Q 6 = 44.86, P ≤ 0.01; hardwood forest: Q 6 = 35.76, P ≤ 0.01; Figure 4). D I S C U S S I O N Our work confirmed the daily movement scale as most appropriate for understanding immediate habitat selection decisions by wild turkeys, insofar as those relate to forest cover types. We acknowledge our inference is limited to third ‐ order selection and may not be applicable to home range placement (second ‐ order selection). Nonetheless, in our meta ‐ analysis we found that wild turkeys exhibited a negative functional response to total forest cover. Across all study sites, as forest cover became increasingly available to them, wild turkey selection for it weakened. We failed to find a similar functional relationship specifically for hardwood forests. Rather, the strength of selection for hardwood forests was region ‐ specific and varied by study site. Although our study relied on retrospectively assembled location data, our approach illuminated relationships that were not obvious in the individual studies themselves. Awareness of functional responses is vital when evaluating habitat selection, because failure to do so may mask patterns resulting from heterogeneous resource availability (Mysterud and Ims 1998). Functional responses are a behavioral mechanism that signifies individual ‐ level trade ‐ offs between multiple habitats under changing propor- tional availability, each representing potential fulfilment of a particular life history need (Van Moorter et al. 2013). T A B L E 2 Single ‐ variable generalized linear models of order ‐ III habitat selection of forests by wild turkeys in Caston Creek Wildlife Management Area (1999 – 2000), Kemper County site (1991 – 1992), Leaf River Wildlife Management Area (1999 – 2000), Malmaison Wildlife Management Area (2004), Quitman County sites (2009 – 2010), and Tallahala Wildlife Management Area (1991 – 1992), Mississippi, USA. Total forest a Hardwood forest b Movement scale AIC c c Δ AIC d Coefficient e AIC c c Δ AIC d Coefficient e Annual 26767.67 16.73 − 0.017 26767.79 27.87 − 0.007 Season 26767.66 16.73 0.017 26767.20 27.28 0.059 Daily 26750.93 0.00 0.197 26739.93 0.00 0.302 a Total forest includes deciduous, evergreen, and mixed forest and woody wetland. b Hardwood forest includes deciduous forest and woody wetland. c AICc is Akaike Information Criteria corrected for small sample size. d Δ AIC c is the difference in AIC c from the top ‐ ranked model. e Coefficient is the selection coefficient of each land cover type. WILD TURKEY HABITAT SELECTION | 9 of 17 23285540, 2025, S1, Downloaded from https://wildlife.onlinelibrary.wiley.com/doi/10.1002/wsb.70000 by University Of Florida, Wiley Online Library on [19/12/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License As one example, wild turkeys might exhibit functional responses in selection of forest cover due to trade ‐ offs between foraging and thermal regulation (Rakowski et al. 2019, Nelson et al. 2022). Wild turkeys at our 2 Quitman sites, where 88% of the landscape was covered by open agricultural fields, may have selected forests more intensely during the hot summer months of our study for the shade and thermal refuge they provided (Rakowski et al. 2019). Whereas on TWMA, where the landscape was nearly entirely forested, forest selection may have been less intense because opportunities provided by other cover types (e.g., foraging for insects in open fields) became more important relative to the resources afforded from ubiquitous forest cover. Other studies have suggested functional responses in wild turkeys. Niedzielski and Bowman (2014) indicated a possibility of functional responses to several land cover types (e.g., deciduous forest, conifer forest, and mixed forest), and demonstrated temporal changes in selection that likely corresponded to variations in resource avail- ability. Gonnerman et al. (2023) demonstrated wild turkeys may modulate strength of habitat selection as a function of winter weather. Other avian species, such as little owls ( Athene noctua ) and greater sage ‐ grouse ( Centrocercus urophasianus ), have also been shown to modify the strength of habitat selection as a function of resource availa- bility, weather, and seasons (Sunde et al. 2014, Sandford et al. 2017). Our findings add further evidence suggesting T A B L E 3 Generalized linear mixed models of order ‐ III habitat selection of forests by wild turkeys in Caston Creek Wildlife Management Area (1999 – 2000), Kemper County site (1991 – 1992), Leaf River Wildlife Management Area (1999 – 2000), Malmaison Wildlife Management Area (2004), Quitman County sites (2009 – 2010), and Tallahala Wildlife Management Area (1991 – 1992), Mississippi, USA. Model a K b logLik c AIC c d Δ AIC c e w i f Total forest g fop + (fop | site/id) 8 − 13309.0 26633.94 0.00 0.731 fop + (fop | site/id) + (1 | yr) 9 − 13309.0 26635.95 2.00 0.269 fop + site + (fop | id) 11 − 13344.4 26710.83 76.89 0.000 fop + (1 | site/id) 4 − 13359.2 26726.47 92.53 0.000 fop + (fop | id) 5 − 13373.5 26756.94 122.99 0.000 1 + (1 | site/id) 3 − 13381.9 26769.80 135.86 0.000 Hardwood forest h hwp + (hwp | site/id) 8 − 13304.8 26625.51 0.00 0.731 hwp + (hwp | site/id) + (1 | yr) 9 − 13304.8 26627.51 2.00 0.269 hwp + (hwp | id) 5 − 13331.4 26672.81 47.3 0.000 hwp + site + (hwp | id) 11 − 13356.0 26733.95 108.44 0.000 hwp + (1 | site/id) 4 − 13366.4 26740.81 115.30 0.000 1 + (1 | site/id) 3 − 13381.9 26769.80 144.29 0.000 a Model terms are forest proportion (fop), hardwood proportion (hwp), study site (site), wild turkey identity (id), and year (yr). Proportions of forests were computed at daily movement scales. b K is number of parameter. c logLik is Log ‐ likelihood value. d AIC c is Akaike Information Criteria corrected for small sample size. e Δ AIC c is the difference in AIC c from the top ‐ ranked model. f w i is model weight. g Total forest includes deciduous, evergreen, and mixed forest and woody wetland. h Hardwood forest includes deciduous forest and woody wetland. 10 of 17 | OGAWA ET AL 23285540, 2025, S1, Downloaded from https://wildlife.onlinelibrary.wiley.com/doi/10.1002/wsb.70000 by University Of Florida, Wiley Online Library on [19/12/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License individual wild turkeys exhibit a degree of niche plasticity by adjusting space use to accommodate the variations in forest availability they face (Bolnick et al. 2003). On our sites, wild turkeys showed weaker selection to forest cover where it was most prevalent (on sites such as TWMA and Kemper), while most intensively selecting for it on sites such as Quitman where it was most lacking. While not implicitly tested, our results suggest that in the southeastern United States, wild turkeys require approximately 70 – 80% forest cover to meet their needs and beyond this threshold increasing forest coverage may become constraining. Wild turkeys did not show functional responses at the individual level to hardwood forest availability but instead exhibited site ‐ specific variation. While our study sites spanned a wide continuum of total forest cover, ranging from 12% of QS to 95% of TWMA, the proportion of hardwood forests was generally lower and relatively similar across sites. Wild turkey abundance in Mississippi has been shown to peak when the proportional availability of hardwood forests constitutes slightly under one ‐ third of landscapes equivalent in area to the mean annual dispersal distance of the subspecies, or roughly 40% of landscapes equivalent in area to the mean annual home range for wild turkeys (Davis et al. 2018). Few individual turkeys within our study had home ranges comprised of such level of hardwood forests; only a single site (MWMA) had hardwood forest cover >40%. If access to the ideal within ‐ home ‐ range proportion of hardwoods was almost universally lacking, individual variation in strength of selection for this cover type (i.e., a functional response) should not have been expected to be manifested in our analysis. In essence, the relative scarcity and narrow range of hardwood forests on our study areas likely constrained expression of a functional response in our results because hardwood availability did not exceed the minimal threshold necessary for most wild turkeys in our study. Conversely, we found selection for hardwood forests most likely varied by study site. While considerable research from across the southeastern portion of their range consistently demonstrates wild turkeys exhibit strong T A B L E 4 Meta ‐ regressions of order ‐ III habitat selection by wild turkeys in Caston Creek Wildlife Management Area (1999 – 2000), Kemper County site (1991 – 1992), Leaf River Wildlife Management Area (1999 – 2000), Malmaison Wildlife Management Area (2004), Quitman County sites (2009 – 2010), and Tallahala Wildlife Management Area (1991 – 1992), Mississippi, USA. Model a K b logLik c AIC c d Δ AIC c e w i f Total forest g Proportion 3 − 469.71 945.58 0.00 1.00 Site 8 − 472.46 961.92 16.34 0.00 Null 2 − 487.53 979.13 33.55 0.00 Hard