Drivers of support for citizen science across state wildlife management agencies in the United States

Citizen science (CS) is gaining recognition as a valuable approach to meet data needs for environmental projects while fostering collaboration between scientists and members of the public. Despite increasing implementation of CS by natural resource entities, organizations’ motivations for engaging in CS remain poorly understood. We examined the utility of the theory of planned behavior (TPB) and social exchange theory (SET) in identifying factors influencing support of CS by scientific organizations. To test predictions of the TPB and SET theories, we surveyed (quantitative, web based) state wildlife agency staff in the United States on their perceptions of organizational engagement in CS. We divided questions that measured TPB items into individual and organizational components to address the influence of personal‐ and organization‐level decision‐making on staff perceptions and attitudes. We used structural equation modeling to identify key constructs that influence staff support of CS in state wildlife agencies. The survey yielded 627 responses across 44 states. Both TPB and SET constructs accurately predicted staff support of CS; however, measures from SET (e.g., public engagement benefits and costs of CS to scientific credibility) were most influential (i.e., TPB constructs had less impact). Our findings indicate that organizational support for CS is primarily influenced by assessment of trade‐offs among perceived costs and benefits. Indicators of support for CS were further elucidated by including measures from the TPB model. Based on our results, we suggest that natural resource entities give careful consideration to CS project design, develop thorough communication and data management plans, and practice iterative evaluation of CS project productivity.

organizaciones científicas a la CC. Sondeamos (cuantitativamente y en línea) al personal de las agencias estatales de los Estados Unidos sobre sus percepciones de la participación organizacional en la CC para evaluar las predicciones de ambas teorías. Dividimos las preguntas que medían los objetos de la TCP en componentes individuales y organizacionales para abordar la influencia de la toma de decisiones a nivel personal y organizacional sobre las actitudes y percepciones del personal. Usamos modelos de ecuación estructural para identificar los constructos clave que influyen sobre el apoyo a la CC por parte del personal de las agencias estatales de fauna. El sondeo aportó 627 respuestas de 44 estados. Los constructos de ambas teorías pronosticaron correctamente el apoyo a la CC por parte del personal; sin embargo, las medidas de la TIS (por ejemplo, los beneficios de la participación del público y los costos de la CC en la credibilidad científica) tuvieron la mayor influencia (los constructos de la TCP tuvieron un menor impacto). Nuestros resultados indican que el apoyo organizacional hacia la CC está influenciado principalmente por la valoración de las compensaciones entre los costos y beneficios percibidos. Los indicadores del apoyo a la CC se esclarecieron más con la inclusión de las medidas del modelo de la TCP. Con base en nuestros resultados, sugerimos que las entidades de recursos naturales consideren cuidadosamente el diseño de proyectos de CC, desarrollen una comunicación minuciosa y planes de manejo de datos y practiquen la evaluación iterativa de la productividad de los proyectos de CC.

INTRODUCTION
Citizen science (CS) (or community science or public participation in scientific research) is the practice of engaging members of the public in scientific projects through the collection, analysis, or both of data and information (Silvertown, 2009). The practice has gained popularity over the past decade, particularly in ecology and conservation science (Kullenberg & Kasperowski, 2016). Proponents of CS argue there are several benefits to volunteers (participants in CS) and implementers (developers and administrators of CS) (e.g., Freitag, 2016;McKinley et al., 2017). For instance, involvement in CS by members of the public may increase volunteers' scientific literacy (Bonney et al., 2016), improve volunteers' knowledge of scientific processes and current conservation challenges (Gouveia et al., 2004;Jordan et al., 2011), and promote environmental stewardship and conservation behaviors . Implementers of CS also benefit because volunteers can collect high volumes of data over large spatial and temporal scales-an approach that is often impossible for 1 or a few professional scientists to achieve in a cost-effective manner (Edgar et al., 2017). In addition, CS can provide positive opportunities for scientists and organizations to engage with more diverse publics (Allf et al., 2022). Properly designed CS projects can positively affect scientific research by engaging a diverse pool of participants with unique skills, perspectives, and knowledge, including those with local and traditional knowledge (Danielsen et al., 2007). Vital to the benefits of CS being realized, projects must meet the needs of volunteers and implementers, and these actors must be willing to engage in all necessary activities. As such, a growing body of literature focuses on factors that motivate members of the public to participate in CS, arguing that these individuals are essential to the success of any volunteer-based initiative (e.g., Larson et al., 2020;Martin & Greig, 2019). However, there is substantially less research on factors that drive organizations to implement CS. Research on scientists' perceptions of CS suggests that drivers for CS engagement include a desire to promote scientific research (Golumbic et al., 2017), but skepticism about the quality of citizen-generated data limits the utility of CS for achieving scientific outcomes Clare et al., 2019). Although current research provides some insight into influential factors for CS administration, most studies fail to consider the impact of a scientist's employment agency. This perspective is important to consider because in many cases it is the agency, rather than individual scientists, that decides whether to implement CS. Therefore, more research is needed on the factors that influence scientists and agency managers to engage in CS.
We sought to understand how scientific staff employed at governmental natural resource agencies perceive CS from an individual and organizational standpoint. We applied the theory of planned behavior (TPB) (Ajzen, 1991) and social exchange theory (SET) (Blau, 1964;Emerson, 1962;Homans, 1961) to examine predictors of staff support of agency-implemented CS initiatives. Previous research in the CS field shows that the TPB has utility in predicting staff and volunteer support for the practice (Cigarini et al., 2021;Martin & Greig, 2019). Although SET has not been directly used to examine support in a CS context, the theory's underlying premise of assessing trade-offs between costs and benefits has been applied to evaluate drivers and barriers to CS success (Asingizwe et al., 2020;Nov et al., 2014). We tested predictions of the TPB and SET theories with perceptions of staff employed at state wildlife agencies (SWAs) in the United States for 2 reasons. First, the structure of these entities is hierarchical: a top-down approach to decision-making is often present (Wildlife Management Institute, 1987), and this structure mirrors that of other governmental natural resource agencies nationally and globally. Thus, decisions regarding CS administration may not always be at the discretion of all agency employees. This offers a unique opportunity to explore the influence of agency-level decision-making on staff support of CS. Second, SWAs are composed of several environmental departments that can administer multiple CS programs simultaneously. These programs range from local efforts targeting a single species (e.g., Gopher Tortoise Sightings [Burr et al., 2014]) to statewide initiatives that focus on multiple species and conservation questions (e.g., Snapshot Wisconsin ). Schulz et al. (2021) highlight the importance of understanding staff attitudes and perceptions of agency-implemented, volunteer-based initiatives by demonstrating how these attributes can affect program support and influence program outcomes. Perceptions and attitudes of staff can also be evaluated to determine drivers and barriers to CS success (Cigarini et al., 2021), which may help agencies develop programs that maximize benefits to volunteers and implementers and better achieve desired objectives. For entities that utilize CS across multiple departments and plan to implement additional CS projects in the future, staff reluctance to embrace CS could undermine agency-led CS initiatives. Vice versa, if staff are supportive of the practice but feel stifled by agency bureaucracy or hesitancy to use CS, it could lead to frustration, potential burnout, and staff retention problems. Thus, an overarching understanding of staff support of current agency-operated CS projects, potential new CS projects, and for the practice as a whole (i.e., through recommending CS to others) is vital. This type of analysis is also of interest because much of the current literature has examined support for CS based on only one focal study. Assessments of factors that affect staff support of CS across multiple domains are therefore critical to increasing the practice's utility as a scientific, educational, and collaborative tool. Based on our study's findings, we sought to develop a set of best practices for CS implementation that can be used as a guide by natural resource agency decision makers.

Theory of planned behavior
The TPB contends that an individual's intention to perform a particular behavior can predict their engagement in that behavior (Ajzen, 1991). This theory is an extension of the work by Fishbein and Ajzen (1977) on the theory of reasoned action, which modeled behavioral intention as a result of attitudes toward the behavior and perceptions of subjective norms (i.e., impact of influential people in an individual's life) related to the behavior. The TPB provides an additional predictor of behavioral intention called perceived behavioral control, which is the perceived ability of an individual to perform the intended behavior (Madden et al., 1992). Prior research that applied the TPB to CS shows that most, if not all, antecedents of the model (attitudes, subjective norms, perceived behavioral control) are significant predictors of engagement in particular behaviors. However, these results are from studies that almost exclusively focused on public engagement in CS (e.g., Groulx et al., 2019;Martin & Greig, 2019). Cigarini et al. (2021) drew on the TPB to determine strengths and barriers that existed for public libraries in Barcelona, Spain, to effectively administer CS to library patrons. Their findings indicated that items from all antecedents of the TPB model influenced support for the CS initiative, including librarians' attitudes toward learning outcomes, norms surrounding participant inclusivity, and perceived ability to implement CS given adequate training. This study is one of the only studies to our knowledge in which a TPB model was used to examine motivations for CS engagement and support at the organizational level.
We expanded the application of the TPB to CS by separating the 3 antecedents of the model into 2 levels: personal (i.e., Hypothesized (H) study model of perceptions of citizen science from wildlife professionals employed at state wildlife agencies in the United States (+, positive influence; -, negative influence). Personal refers to individual perceptions and organizational refers to state wildlife agency perceptions individual) and organizational (i.e., agency-wide). This modification provides a novel framework for examining the influence of individual-and group-level perceptions on staff support of CS and whether differences exist between the 2 ( Figure 1). Likely, scientists' experiences with CS in a professional setting and the experiences of colleagues, supervisors, and other staff can influence their attitudes and subjective norms. Positive attitudes are linked to a general enthusiasm and willingness to engage in CS (Golumbic et al., 2017); however, concerns about resource acquisition and citizen-generated data quality can act as barriers Clare et al., 2019). Therefore, we examined the following hypotheses: Hypothesis 1a: There is a direct positive relationship between positive personal attitudes and support for agency engagement in CS. Hypothesis 1b: There is a direct positive relationship between positive organizational attitudes and support for agency engagement in CS.
Similarly, we postulated that personal subjective norms (i.e., the influence of other scientific professionals) have a positive relationship with support for CS based on the collaborative nature of most natural resource agency jobs. Further, supporting initiatives that are perceived to be approved by an organization and its employees should be positively linked to engaging in those behaviors (Fugas et al., 2011). As such, we theorized that organizational subjective norms, or the degree to which an entity is perceived to approve or disapprove of scientific professionals engaging in CS, will have a positive relationship with support ( Figure 1). Therefore, we tested the following hypotheses in our model:

Hypothesis 2a:
There is a direct positive relationship between personal subjective norms and support for agency engagement in CS. Hypothesis 2b: There is a direct positive relationship between organizational subjective norms and support for agency engagement in CS.
Recent studies show that perceived behavioral control is a significant predictor of an individual's intention to engage in CS (Martin & Greig, 2019;Martin et al., 2016), but little research has examined the relationship between perceived behavioral control and organizational administration of CS. Cigarini et al. (2021) showed that factors related to workload, organizational support, and training influenced librarians' perceived ability to engage in cocreated CS projects. Similar factors, including job description constraints, managerial or political pressure, and access to resources within governmental natural resource entities, may affect an individual's autonomy to engage in CS within their job responsibilities. These elements can also be present at the organizational level. Thus, the perceived capacity to make decisions about CS administration within an individual's employment agency may influence their degree of perceived behavioral control. Therefore, we tested the following hypotheses in our model:

Hypothesis 3a:
There is a direct positive relationship between personal perceived behavioral control and support for agency engagement in CS. Hypothesis 3b: There is a direct positive relationship between organizational perceived behavioral control and support for agency engagement in CS.

Social exchange theory
The SET seeks to explain individual behavior through the lens of resource exchange. Its roots are in the early social psychology works of Homans (1961), Emerson (1962), and Blau (1964). The key underlying premise of SET involves an exchange between rewards and costs, where an individual's choice to engage in a behavior is a product of weighing the perceived costs and benefits associated with the behavior. As such, the principle of individual behavior is to maximize benefits and minimize costs. The SET has not been directly utilized to study motivations for CS participation; however, it has been applied in conjunction with other theoretical frameworks, such as self-determination theory (see Nov et al., 2014). The SET is a valuable theoretical framework for governmental natural resource agencies because these entities often operate under resource constraints (Jewell et al., 2020) and must make policy decisions that function under such limitations. Therefore, decisions regarding CS administration, and staff support of those decisions, are likely to be affected by how individuals and agencies make trade-offs among perceived benefits (e.g., increased data quantity) and costs (e.g., decreased data quality).
We applied the SET model by assessing the impact of perceived benefits and costs of CS on staff support ( Figure 1). Different classes of perceived benefits and costs (e.g., financial, emotional, professional) can have stronger, weaker, or insignificant influence on support (Coyne et al., 2015), suggesting that assessing a range of these predictors can help disentangle specific drivers and barriers to CS engagement. We expected that perceived benefits of CS would positively influence support, whereas perceived costs would negatively influence support ( Figure 1). We also expected that certain classes of benefits and costs would have greater influence on staff support than others. Therefore, we tested the following hypotheses in our model:

Survey distribution
To test our model, we distributed an electronic survey to staff employed in wildlife divisions or departments in all 50 SWAs in the United States (Appendix S1). Our target population was wildlife professionals employed in state agencies and included biological technicians, field biologists, regional biologists, supervisors (e.g., assistant chief), species specialists, species group specialists (e.g., nongame), and researchers. We chose a broad sample of employees to survey because we were interested in understanding perceptions of CS from staff at all levels with varying degrees of involvement in CS and control over decisions to implement CS. We identified wildlife division staff with the help of our personal and professional connections and with a search of employee directories on agency websites. Employees with whom the authors had personal or professional connections were contacted via email and informed about the objectives of the study and the population of interest. We also inquired about a contact's willingness to distribute the survey link to relevant staff in their agency, provide the contact information for an employee who could distribute the survey link, or provide a contact list. For agencies that did not respond to our initial request or for whom we did not have connections, we conducted an internet search of employee directories and limited our search to individuals who fit our sampling criteria. The University of Georgia's Institutional Review Board deemed this study exempt from human subjects approval because we collected no sensitive information from respondents. Survey questions are in Appendix S1. State agency contacts received an email announcing the forthcoming survey link in June 2021 (Dillman et al., 2014). The survey was distributed to contacts electronically via Qualtrics in July 2021 (Qualtrics, 2020). Most contacts were asked to personally distribute the electronic survey to relevant staff in their agency, and the email containing the survey link described our target population in detail to ensure that only individuals in the target population received the survey. In contrast, for agencies that provided a contact list of wildlife staff, we sent the survey link directly to individuals on the list. Following a modified Dillman et al. (2014) method to increase response rate, we sent a reminder email to complete the survey 2 weeks after the first email and a final reminder email 1 week later. We estimated that approximately 1670 wildlife professionals from 44 SWAs received the survey. However, the exact sample size could not be determined because some contacts distributed the survey through email listservs, which may have contained email addresses for individuals that were no longer employed in the division or contained additional employees outside the target population. The number of state agencies (6 of 44 [14%]) that reported this issue was minimal. We could not assess the representativeness of the sample because we depended on most state agency contacts to distribute the survey to wildlife division staff and information on employee demographics is not publicly available.

Construct measurement
We asked questions that measured constructs of the TPB model (attitudes, subjective norms, perceived behavioral control) and the SET model (perceptions of benefits and costs) within the context of SWA engagement in CS. Questions that pertained to the TPB model were asked twice to address the constructs at 2 different levels: personal (individual) and organizational (SWA) perspectives. We completed a literature review to understand how other researchers applied the TPB in a CS context to predict behavioral intention (Groulx et al., 2019;Martin & Greig, 2019;Martin et al., 2016) and adapted our survey questions that measured TPB constructs based on these studies. We also conducted exploratory factor analyses (EFAs) and confirmatory factor analyses (CFAs) on these constructs to assess their reliability and validity before including them in our model (Table 1).
We conducted a separate literature review of studies that documented benefits and costs of CS to scientists and scientific organizations to develop questions that measured items from the SET model. Benefits of CS included positive opportunities for scientists and organizations to engage with the public and promote conservation behaviors Danielsen et al., 2005), increased data collection efforts and scientific relevancy (Edgar et al., 2017), and reduced costs (monetary and time) for conducting scientific studies (Carr, 2004). Documented costs of CS included difficulties working with public participants on scientific projects (Golumbic et al., 2017), decreased data quality and scientific credibility (Lukyanenko et al., 2016), and increased workload for personnel (Burgess et al., 2017). We constructed statements for the benefits and costs of CS based on the aforementioned studies within the specific context of SWA engagement in CS (Table 1).
Our dependent variable, support for agency engagement in CS, was composed of 3 items: agreement that an SWA should continue to engage in CS (given that the SWA had at least 1 CS program in operation), agreement that an SWA should develop new CS initiatives, and likelihood that an individual would recommend that other SWAs engage in CS (Table 1). All questions were measured on a 5-point Likert-type agreement scale (from 1 = strongly disagree to 5 = strongly agree), with the exception of intention to recommend that other SWAs engage in CS, which was measured on a 5-point Likert-type likelihood scale (from 1 = very unlikely to 5 = very likely) (Likert, 1932). We also collected information on respondents' demographic characteristics, job characteristics, current agency-based CS projects, and previous CS projects.

Data analyses
In total, 740 individuals clicked on the survey link. We removed 113 responses based on data quality issues, including if the survey was completed in <50% of the median time (633 s), <50% of the survey was completed, or a respondent was outside our target population (n = 2). This resulted in a usable sample of 627 respondents who completed at least 60% of the survey and provided answers to the TPB, SET, and support for CS questions. Thus, the estimated overall response rate was 37.5% (627 of 1670), which is significantly higher than most electronic surveys (Hardigan et al., 2012). Survey completion rate was 85% (627 completed responses of 740 opened surveys). We received ≥1 completed response from 44 SWAs.
Prior to testing the theoretical models, we performed a series of EFAs and CFAs to identify the common factors that explain the order and structure among measured variables (Watkins, 2018), investigate causal relations among latent and observed variables (Hancock & Mueller, 2001), and maximize internal consistency and convergent and discriminant validity of the items (Raubenheimer, 2004). We conducted EFAs in the SPSS Statistics software package (IBM Corp, 2020) and CFAs in the AMOS software package (Arbuckle, 2014). For the EFAs, we removed all items loading below 0.50 or that detracted from Cronbach's alpha scores. We determined the number of suitable factors based on eigenvalues >1 (Hair et al., 2010).
We assessed the construct validity of the model with CFA. Construct validity assesses whether one is measuring what one intends to measure and is made up of 4 subdimensions of validity: face validity, convergent validity, discriminant validity, and nomological validity (Hair et al., 2010). Face validity pertains to how the wording of the items matches the definition of the construct that the items are trying to measure. We enhanced face validity by using the CS literature to word the items that measured benefits and costs and by using previously published scales to measure the TPB constructs. Convergent validity pertains to how well the items of a construct represent the construct through the amount of common variance they share (Boley & McGehee, 2014). We assessed this metric based on standard regression coefficients, average variance extracted (AVE), and construct reliability. Respective values of >0.05, >50%, and >0.70 indicated convergent validity. Discriminant validity assesses a construct's uniqueness from other constructs in the model. This metric was assessed by comparing each construct's squared correlations with other constructs (shared variance) and its own AVE (own variance). Nomological validity assesses how well constructs relate to other constructs that theory suggests they should. This metric was assessed through structural equation modeling (SEM).
Our 8 hypotheses were tested using a series of 4 SEMs in SPSS' AMOS software. Hypotheses were tested using both the statistical significance of the relationship at the 0.05 level and whether the relationship was positive or negative as hypothesized. We used comparative fit index (CFI) and root mean square error of approximation (RMSEA) as indices of goodness of fit of our models and for the CFA (Byrne, 1998). Model fit was considered acceptable with CFI values >0.90 and RMSEA values <0.08. Model 1 tested the influence of personal TPB constructs (attitudes, subjective norms, perceived behavioral control) on support for CS. Model 2 tested the influence of organizational TPB constructs on support. Model 3 measured the influence of perceived benefits and costs of CS on support using the SET framework. Our last model (model 4) included a combination of the TPB and SET items to see which of the previous significant predictors best explained support for CS.

Descriptive statistics
Respondents were primarily male (64.0%) with an average age of 44 years (SD 10.2, range 24-77). Most respondents (95.5%) had a bachelor's degree or higher. On average, respondents worked for a particular SWA for 14 years (SD 9.4, range 0-52). Job position varied substantially among respondents: 30.1% regional biologists, 7.5% regional supervisors, 29.0% statewide biologists, 8.8% mid-level managers, 4.0% upper-level managers, 1.9% assistant chiefs, 1.6% chiefs, and 17.1% other (e.g., technician, researcher). Before being asked questions pertaining to perceptions of CS, respondents were asked if the agency they work for had at least 1 CS program currently in operation. Most respondents (90.7%) thought that their agency had at least 1 CS program currently in operation, 3.0% did not think their agency had an operating CS program, and 6.2% did not know. Overall, support for CS among individuals employed in SWAs was high. A majority of respondents (78.5%) who thought their agency had at least 1 CS program in operation (n = 556) agreed (55.2%) or strongly agreed (23.3%) that their agency should continue to engage in CS. When asked if respondents would want their agency to develop new CS initiatives, results were mixed (n = 612): 30.8% neither agreed nor disagreed, 48.2% agreed, and 15.5% strongly agreed. Respondents were asked how likely they were to recommend that other SWAs engage in CS (n = 612): 42.1% were somewhat likely to recommend CS and 25.8% were very likely. Respondents primarily perceived CS as beneficial for providing positive opportunities for SWAs to interact with the public (mean [SD] = 4.26 [0.592]) but considered CS data to be messy compared to data collected by agency personnel (mean [SD] = 3.44 [0.858]) (Table 1).

Exploratory factor analysis
The EFAs of attitudes, subjective norms, and perceived behavioral control items produced single-factor solutions; each item loaded onto a single factor. The EFAs of the benefit and cost items each produced a 2-factor solution, and we removed any items that loaded onto more than 1 factor. This analysis identified 2 unique benefits (public engagement benefits, internal agency benefits) and costs (implementation costs, scientific credibility costs) of agency engagement in CS (Table 1). Public engagement benefits were characterized by items describing positive outcomes of CS for agency-public engagement. Internal agency benefits were characterized by items describing tangible gains from CS for an agency. Implementation costs were characterized by items describing outcomes of CS that could lead to increased workload for agency personnel. Scientific credibility costs were characterized by items describing possible negative impacts of CS to the scientific credibility of an agency or project. Support for CS was indexed based on aggregating 3 items measuring support for continuing CS projects, developing new CS projects, and recommending CS to other agencies.

Confirmatory factor analysis
All items of each scale demonstrated convergent validity (standard regression coefficients >0.7, AVEs >50%, and construct reliability scores >0.70), except for the implementation costs scale, which had an AVE <50%, indicating that the scale left more variance unexplained than explained (Table 1). The test of discriminant validity failed for the implementation costs scale, suggesting there were no significant differences between this scale and the scientific credibility costs scale (Table 2). Therefore, we removed the implementation costs scale from further analysis. All other scales demonstrated discriminant validity (Table 2), and the support for CS scale performed especially well (factor loadings all >0.7, AVE of 69%, and a construct reliability of 0.91). The CFI value for this analysis was >0.90 and the RMSEA value was <0.05, indicating good model fit.

Structural equation modeling
According to model 1, only personal attitudes toward CS had a significant relationship with support for CS (β = 0.731, p = 0.001); personal subjective norms and personal perceived behavioral control were not significant predictors (Table 3). In model 2, an SWA's attitudes (β = 0.251; p = 0.001) and perceived behavioral control (β = 0.304; p = 0.001) were significant predictors of support for CS, but there was no support for the influence of organizational subjective norms (Table 3).
For model 3 (SET model), public engagement benefits of CS were significantly positively correlated with support for CS (β = 0.296; p = 0.001), and scientific credibility costs were significantly negatively correlated with support (β = −0.689; p = 0.001) ( Table 3). Internal agency benefits were not significantly related to support for CS in this model.
Model 4 indicated that personal attitudes toward CS (β = 0.140; p = 0.012), organizational perceived behavioral control to implement CS (β = 0.113; p = 0.001), public engagement benefits of CS (β = 0.266; p = 0.001), and scientific credibility costs of CS (β = −0.532; p = 0.001) were all significant predictors of support for CS, and this model explained 81% of the variance (Table 3). An organization's attitudes toward CS were not a significant predictor in this model (Table 3). The CFI values for all 4 models were >0.90, and RMSEA values were <0.07.

DISCUSSION
We sought to evaluate staff support for engagement in CS by governmental natural resource agencies by using the TPB and SET as the underlying theoretical frameworks. Our approach revealed a variety of drivers and barriers to CS support and demonstrated the efficacy of the TPB and SET for assessing influential variables.

Support for CS
The SWA staff were generally supportive of agency engagement in CS. Most respondents (78.5%) agreed or strongly agreed that their agency should continue to engage in CS, and 63.7% agreed or strongly agreed that their agency should implement new CS projects. Similarly, 67.9% of respondents were somewhat likely or very likely to recommend that other SWAs engage in CS. This overall positive perception may be attributed to increasing adoption of CS projects by SWAs that has resulted in positive outcomes for individuals, communities, and natural resource entities (e.g., suburban human-coyote interactions, New York Department of Environmental Conservation [Weckel et al., 2010]; Snapshot Wisconsin, Wisconsin Department of Natural Resources ). Our final sample of individuals consisted of a diverse group of SWA employees with various job positions (e.g., regional biologist, upper-level manager). As such, natural resource entities should encourage positive perceptions of CS among employees of all job classifications to foster continued individual and organizational support for the practice.

Predicting support with the TPB
We expanded the body of literature on the TPB and its application to CS by applying the theory at the personal (individual) and organizational (agency-wide) levels. By addressing TPB

Best practice Constructs addressed Description
Communicate who is making decisions regarding citizen science administration.

Organizational perceived behavioral control a
Decision makers should specify if individuals outside of the agency are requiring citizen science implementation or if the decision is at the discretion of agency employees.
Define objectives in terms of contribution to organizational decision-making.
Personal attitudes b Consider how the objectives of a citizen science project can contribute to broader organizational goals and missions.
Include measurable public engagement objectives into all citizen science projects.
Public engagement benefits c Ensure that the objective(s) clearly states the type of public engagement that is desired. Develop methods for measuring public engagement objectives, such as using pre-and post-surveys of participants. See Lasky et al. (2021) for an example of how a state wildlife agency-led citizen science initiative incorporated outreach and engagement objectives.
Have an interdisciplinary team of collaborators on every citizen science project, including a communication and data expert. If unattainable, collaborate with individuals who have communication and data management skills.
Public engagement benefits c Communication experts can ensure that information and benefits of citizen science are properly communicated to project participants and to the broader scientific community. Data experts can offer support to mitigate potential data reliability issues that may arise during the data collection and analysis phase of citizen science projects.

Scientific credibility costs d
Develop communication and data management plans.
Public engagement benefits c Communication plans should include an outreach element that specifies pathways for communicating with volunteers, indicates how volunteers can communicate with project leaders, contain methods for soliciting volunteer feedback, and include plans for incorporating feedback into future project iterations. Data management plans should include a thorough understanding of what data will be collected (including metadata), how the data will be collected, used, and stored, and what constitutes "quality data" (Balázs et al., 2021). Consult existing literature on how to ensure data quality in citizen science. Publicly share summaries of communication and data management plans to promote transparency and support mutual accountability (Ganzevoort et al., 2017).

Scientific credibility costs d
Develop simple data collection protocols and training materials that match the skills of volunteers.
Scientific credibility costs d Reduce the complexity of protocols but do not oversimplify instructions. Project managers should test protocols themselves and with peers and volunteers to understand how protocols are being interpreted and to identify areas where clarification is needed (Rambonnet et al., 2019).
Conduct a check of citizen-generated data early on and throughout the duration of the project.
Scientific credibility costs d Conduct quality control checks, such as flagging outliers and identifying ambiguous contributions, early in the data collection phase to ensure that data are meeting quality standards outlined in the data management plan.
Conduct an internal evaluation of citizen science project productivity.
Personal attitudes b Practice iterative evaluation of citizen science projects on multiple dimensions (e.g., volunteer recruitment and retainment, decision support, team collaboration). See Wiggins et al. (2018) for guidance on how to evaluate citizen science projects with respect to science productivity.  Lukyanenko et al. (2016) for reviews on data quality in citizen science. a Perceived ability of a state wildlife agency to make decisions regarding citizen science implementation. b Individual attitudes toward state wildlife agency engagement in citizen science. c Items describing positive outcomes of citizen science for agency-public engagement. d Items describing possible negative impacts of citizen science to the scientific credibility of an agency or project.
antecedents across 2 domains, our results revealed 2 major findings. First, having positive attitudes toward CS positively influenced support. Other researchers have found scientists' positive attitudes are linked to a desire to use CS as an avenue to promote scientific research (Golumbic et al., 2017). This suggests that natural resource entities can increase positive attitudes about CS by framing projects in ways that emphasize advancing science. Project goals can also be defined in terms of their contribution to broader organizational missions, which may increase relevance for staff and further improve support for CS. Second, organizational perceived behavioral control, or the degree to which respondents perceived their agency as having the autonomy to make decisions regarding CS administration, had a significant positive relationship with support in our models. In SWAs, organizational perceived behavioral control may be limited because directors and politically appointed board members are often the only ones who have the authority to make decisions about certain management actions, including CS administration (Wildlife Management Institute, 1987). Decision makers can potentially bolster organizational perceived behavioral control by specifying where decisions regarding CS implementation originate. Integrating this simple flow of information with an understanding of logistical and financial needs for a project  will allow staff to understand the context of the decision and ensure that CS is being implemented effectively to reach its intended goals. Our personal TPB model (model 1) and organizational TPB model (model 2) explained 55% and 19% of the variance in staff support of CS, respectively (Table 3). Although organizational attitudes toward CS had a significant positive relationship with support in model 2, this factor was not significant when it was combined with SET constructs (model 4; Table 3). The interpretation of this result, combined with the low coefficient of determination for the organizational TPB model (R 2 = 0.19) and the weak relationship of organizational perceived behavioral control with support in model 4 (R = 0.113), is not immediately clear. These findings may be indicative of a weak influence of agency perceptions of CS on staff support or possibly a result of respondents' limited understanding about agency culture and decision-making processes. Nevertheless, significant predictors of staff support emerged from our personal-and organization-focused TPB models. Previous research on CS in organizations shows that organizational structure, history (i.e., past experiences within the organization), and receptivity to change can influence how individual employees and organizations as a collective approach implementation of CS (Anderson et al., 2020). Such factors can result in divergent perspectives of CS from individuals and organizations, which highlights the importance of evaluating micro-and meso-level perceptions. By partitioning the TPB antecedents into 2 levels, we demonstrated how the hierarchical nature of SWAs impacts staff perceptions of CS at the individual and organizational level and documented a need for future researchers to consider these multifaceted relationships when examining entities of similar structure.

Predicting support with SET
Our study is the first to apply the SET model as one theoretical framework for predicting staff support of agency engagement in CS. Constructs from the SET model, specifically, public engagement benefits and scientific credibility costs of CS, were the strongest predictors of staff support in our SEM analysis; 79% of the variance in support for CS was explained solely by the SET model (model 3; Table 3). When these significant SET constructs were analyzed with constructs from the TPB, the resulting model produced the highest coefficient of determination (R 2 = 0.81), and SET constructs remained the strongest predictors. In fact, negative perceptions about the scientific credibility of CS, which included difficulties utilizing CS data in wildlife management and defending CS outcomes, were more powerful than all 3 of our models' positive predictors (personal attitudes, organizational perceived behavioral control, public engagement benefits of CS).
Concerns about the quality of citizen-generated data and the acceptability of CS outcomes by the broader scientific community is well documented (Clare et al., 2019;Riesch & Potter, 2014). Factors such as careful consideration to project design, adequate volunteer training, and quality control measures have been suggested as potential remedies (Dickinson et al., 2010;Lasky et al., 2021). However, some research indicates that alleviating concerns about CS data quality requires a more holistic approach (Vann-Sander et al., 2016). Numerous interacting variables can influence the ability of CS participants to generate scientifically robust data, many of which are under the discretion of CS project teams. For instance, fostering positive relationships and establishing trust with volunteers and supporting them through ongoing engagement and training are frequently cited as essential elements to achieving scientific outcomes from CS efforts (Shirk et al., 2012). Because respondents in our study also perceived CS as beneficial for public engagement, including viewing the practice as a pathway for improving trust between SWAs and the public, synergistic volunteer and data management opportunities may exist for CS managers . This illustrates the importance of having a diverse project team with relevant skills to improve the capacity for adequate project management.

Management recommendations
To maximize the capabilities and positive impacts of CS within natural resource entities, project managers must systematically consider the needs of volunteers and implementers and formally integrate these considerations into project designs. As CS gains popularity within the scientific community, guidelines exist to aid in the development of CS programs (Rambonnet et al., 2019;Tweddle et al., 2012). Although these guidelines can be easily adapted to fit most CS projects, governmental natural resource entities often administer several projects for multiple species that differ in terms of spatial and temporal scale, data requirements, and administrative and participant effort. Thus, to devise guidelines relevant to these entities, we drew on our model results and the findings from existing CS literature to develop a set of best practices for natural resource agencies looking to increase the utility of CS for meeting agency goals (Table 4). We integrated our model results into each best practice to highlight specific ways agencies can bolster positive drivers of CS support (personal attitudes, organizational perceived behavioral control, public engagement benefits of CS) and mitigate perceived barriers (scientific credibility costs). These recommendations will provide natural resource entities with the foundation to develop and modify CS projects that contribute to conservation and management objectives and operate under resource constraints.

Limitations and future research
Although our study contributes to the limited research on motivations for organizational administration of CS, it has limitations. First, it lacked random sampling because there was not an updated database of SWA employee job titles or contact information at the national level. We suggest that future researchers coordinate with SWAs or other governmental natural resource entities to obtain a census of employees from which a random sample can be drawn. This will allow for more rigorous sampling and assessment of relevant metrics, such as nonresponse bias. Second, we evaluated perceptions of CS as an entire practice. Evaluating support for CS by project-specific factors, such as species, species groups, objectives (e.g., mapping presence or absence of species on a landscape), or other factors may further uncover variation in drivers and barriers to CS engagement.
Third, we surveyed a broad sample of individuals employed in SWAs, from field staff to upper-level managerial staff. It is possible that nonsupervisory staff may have substantially different perceptions of CS (and understandings of agency views) than supervisory staff. However, because many individuals in SWAs share supervisory roles, we could not differentiate supervisory from nonsupervisory staff. We suggest that future researchers identify managerial and nonmanagerial staff, segment the sample into these 2 groups, and run separate analyses on each group to evaluate if the predictors for CS support differ by job classification. In addition, our study applied microlevel theories (TPB and SET) to understand staff perceptions of CS. Researchers may benefit from applying mesolevel theories to better understand individual and organizational perspectives and the potential relationship between job classification and CS support. Finally, we constructed our own measures for the cost-benefit assessment of CS using SET as the underlying theoretical model. In our scale, items that comprised implementation costs could not be distinguished from those comprising scientific credibility costs, leading to the removal of the former dimension from our analysis. This result is likely a product of our scale items not being distinct enough from one another. Because items from the SET model were the most influential in determining staff support of CS in our models, future research should work to improve the reliability and validity of the implementation costs scale.
Despite our study's limitations, we uncovered specific factors that influenced staff support of CS. We highlighted the importance of understanding how staff perceive benefits and costs of CS and documented how organizations can practically apply our model results through a realistic set of best practices that can function under resource constraints. Our study provides the foundation for helping scientific organizations modify and develop CS programs that best achieve desired objectives and that maximize benefits to volunteers and implementers. Natural resource entities can use our findings to bolster internal agency support for CS, which may positively affect CS outcomes (Schulz et al., 2021) and increase the utility of CS for achieving agency conservation goals.