Scientists and conservationists increasingly rely on contributions by volunteers recruited from the wider public to work over large and ecologically meaningful spatial scales. Optimizing working in partnership with unpaid, volunteer citizen scientists and conservationists requires an understanding of the determinants of volunteer retention rates and how they are affected by management practices. To this effect, we present the novel use of the mark-recapture framework widely used in wildlife demography in order to quantify volunteer retention probabilities. We illustrate the versatility and power of the approach using a project that removed invasive American mink from 10,000 km2 in Scotland in partnership with volunteer citizen conservationists recruited from local communities. Wide scale adoption of the mark-recapture framework to analyze volunteer management will give novel insights into how volunteers interact with the conservation projects they are involved in and provide evidence for their optimal management.
Large scale, effective, and sustainable ecological monitoring and conservation projects are central to the conservation of existing biodiversity (Schmeller 2008). However, the current requirement for ecological evidence and conservation action informed by evidence is limited by the economic resources available to provide it. Using unpaid, volunteer citizen scientists and conservationists (volunteers hereafter) could be the only practical and cost-effective way to collect data and implement conservation action at functional scales (Dickinson et al. 2010). Such volunteers are characterized by working autonomously in order to perform tasks such as floral or faunal surveys, collecting presence/absence data, or implementing management actions, without financial reward (Bryce et al. 2011; Kremen et al. 2011; Hochachka et al. 2012).
Research into volunteer-based conservation and monitoring projects is still in its infancy, as such researchers from the biological and social sciences are currently adopting fundamentally different approaches. Within the social sciences, conservation psychology approaches predominate that have been principally concerned with the factors influencing an individual's decision to volunteer and the emotional benefits they receive (Miles et al. 1998; Bruyere & Rappe 2007; Asah and Blahna 2012; Hobbs & White 2012). In contrast, biological scientists have largely been interested in determining how accurately and reliably volunteers can implement project methodologies and how these can best be optimized (Kadoya et al. 2009; Devictor et al. 2010; Dickinson et al. 2010). Here, we argue that the disparate approaches adopted are missing crucial questions that are relevant to both disciplines, such as determining how the way volunteer-based projects are managed leads to variation in how effectively they retain their volunteer workforce (see Figure 1). Understanding the factors influencing volunteer retention rates will result in increased understanding of volunteer behavior, volunteer management, and project sustainability, ultimately providing a factual basis for utilizing limited financial and human resources optimally.
Understanding the determinants of volunteer retention is crucial to effective volunteer management. High volunteer retention rates are indicative of succeeding projects when volunteers feel happy, valued, and learn and develop in their task, project managers do not need to invest additional resources finding and recruiting new volunteers freeing up time and money to achieve project objectives (Bell et al. 2008). Management literature suggests that low retention rates will indicate a poor management structure, with volunteers feeling that their contribution to the project is either not effective, or valued, or that they are not progressing as individuals while performing their role (Stockard & Lehman 2004). This at best results in the need for ongoing investment in recruiting and training replacement volunteers. At worst, low retention rates indicate a failing and unsustainable project. Furthermore, increasing the number of volunteers can increase the scale or intensity of monitoring, the impact and effectiveness of conservation projects, and the precision and reliability of data from monitoring projects (Schmeller et al. 2009). Consequently, if volunteer retention probability is high, there will likely be greater statistical power to detect relevant changes in target species abundances, phonologies or population trends, and greater probability of a conservation project objective achievement.
The factors driving volunteer retention in conservation and monitoring projects are complex involving a variety of personal, social, and environmental elements. Studies across the breadth of social science are successfully using classical survey techniques such as structured interviews, participant observation, and focus groups to characterize such factors, including (but not limited to): motives (e.g., Measham & Barnett 2008), project management (e.g., Stockard & Lehman 2004), task specific factors (e.g., Millette & Gagné 2008), sense of responsibility (e.g., Jamison 2003), environmental values, (e.g., Campbell & Smith 2006), benefits to the individual (e.g., Wilson & Musick 1999), benefits to the surrounding communities (e.g., Bell 2003), and social-political factors (e.g., Vandzinskaite et al. 2010).
Such approaches rarely shed light on if and how differences in volunteer management practices, underlying volunteer demography or individual motives and values actually influence the overall project volunteer retention rates (but see Asah and Blahna 2012; Wolcott et al. 2008). Consequently, it is difficult to make a reliable assessment of current project health and long-term project sustainability using such techniques, especially for large projects.
The capture-mark-recapture framework
Here, we introduce novel use of the capture-recapture (CR) framework (e.g., Lebreton et al. 1992), a survival analysis tool developed and in widespread use by population ecologists, in order to quantify volunteer retention probabilities. Using this longitudinal approach, volunteer involvement within a project is simplified by scoring participation from recruitment to termination of their involvement over discrete time steps (e.g., 6 month blocks). For each time step, an individual is assigned a “1” if they are known to be active or a “0” if they are not active or if their status is unknown. These binary data are then modeled to quantify variation in retention probabilities according to covariates denoting individual time-invariant covariates (e.g., gender) or time-varying covariates (e.g., frequency of project staff-volunteer contact in previous time step). Individual heterogeneity in the volunteer detection process (e.g., some volunteers being more detectable than others) can cause bias in the estimation of volunteer retention probabilities if not accounted for (Gimenez & Choquet 2010; Cubaynes et al. 2010). Such heterogeneity might also arise from spatial differences in the detection process, some regions being more active than others due to dynamic local volunteer network managers. This can be incorporated into spatially explicit CR models to produce maps of retention probabilities or detectability (Péron et al. 2011).
Multistate versions of CR (Lebreton et al. 2009) make it possible to model the probability of a volunteer making a transition from one category of involvement to another one (e.g., untrained to trained). Recent advances in models even make it possible to incorporate uncertainty in volunteer involvement categories if training records or census data are incomplete (Gimenez et al. 2012). Survival analysis has been successfully applied in other disciplines, including: employee retention analysis (Somers 1996) and analyzing customer defection behaviors in banking (Brockett et al. 2008). However, in such cases, individuals are contractually obliged work, or there is perfect knowledge regarding the activity of each individual. Such methods are not applicable to volunteers as census data are often incomplete, a situation that often arises with self-reporting volunteers who may fail to file reports on their activities due to dropping out or temporary inactivity.
Owing to the flexibility and diversity of the CR toolkit, we do not present a complete guide to its use in examining volunteer retention. Instead, we outline several ways in which using the CR framework can address key hypotheses on the health and sustainability of volunteer-based projects (see Table 1). Then, we illustrate the interdisciplinary scope of this framework using a case study where volunteer citizen conservationists monitor and remove an invasive species by individually operating American mink (Neovison vison) monitoring and capture rafts in their neighborhood. The CR analysis presented demonstrates the utility of exploring variation in volunteer retention probabilities through time, space, and between groups of volunteers with different vocational backgrounds, and provides novel insights into how volunteer retention is influenced by project management. We believe that the technique will appeal to both biological scientists and project managers who have not previously considered how retention and the factors that influence it are affecting their projects, and to social scientists looking for a robust longitudinal framework to compliment more traditional analyses.
Table 1. Highlights hypotheses relevant to project managers, social scientists, and biological scientists who can be addressed using the CR framework and suggest types of model that can be employed to address them. The interdisciplinary hypotheses in particular will benefit from comparisons across multiple projects as responses are likely to be high heterogeneous depending on the methods employed and the geographical location of the project. CJS = Cormack-Jolly-Seber model, KF = known fate model, MS = multistrata model
Volunteer retention probability (S) or detectability (p) are influenced by:
Describing temporal trends
2) Demography (Age/Sex/Social Status)
Optimizing recruitment strategy
3) Frequency of contact with project staff
Optimizing project staff investment
4) Level of training
Optimizing development opportunities
5) Volunteering duration
Understanding determinants of retention
6) Tasks performed
Understanding determinants of retention
7) Task completion frequency
Understanding determinants of retention
8) Volunteer motives
Understanding determinants of retention
9) Volunteer satisfaction
Understanding determinants of retention
Interdisciplinary hypotheses: volunteers S or p are correlated with:
1) Data reliability
Volunteers with the highest S and/or p may also submit the most reliable data
2) Volunteer effectiveness
Volunteers with the highest S and/or p may be more effective task implementers
3) Project success
Successful projects may be or may not be projects with high S and/or p rates
4) Cost effectiveness
Cost effective projects may have higher S or p rates than wasteful projects
5) Management structure employed
The type of management adopted may influence between project S and p rates
The CR framework
The CR framework was developed in the 1930s in order to estimate the abundance of individuals within a given population. Put simply, individuals are marked when first captured and then released. After a known time step, individuals are recaptured with a given probability depending on whether they have survived and their “detectability” (the probability of being detected if they are still alive). Using maximum likelihood probability theory, capture histories of marked individuals can be used to estimate parameters of interest, including: survival and transition to different life history state probabilities for each time interval, and recapture probabilities at each time step.
The simple yet, to the best of our knowledge, novel application of the CR framework presented here depends on two key assumptions. First is to consider the retention probability of volunteers (the probability that a given individual will remain in a volunteer-based project over a given time step) as analogous to survival probability. Second is to assume the probability of an individual's current activity status being known—through a volunteer census scheme or the self-reporting of their current activity—as analogous to detectability. When treated as such, data regarding volunteer involvement within a project share the same structure as data on CR studies of wild populations. Thus, it becomes possible to use the well-established CR toolkit to address classical population demography questions using volunteer data and gain novel insights into the volunteer retention process.
Why use CR to study volunteers?
CR is widely used in ecological and evolutionary demography studies where data often consist of repeated encounters of individually marked animals where knowledge of the exact fate of each individual is unknown. Encounters of individuals are summarized as capture histories that are the product of two Bernouilli processes, survival, and detectability. In animal studies, uncertainty about the fate of an individual is usually due to low recapture rates of individuals owing limited trapping effort, trap shyness, large home ranges, migration out of the study area, seasonal activity patterns, or death without recovery. There is also uncertainty surrounding the activity of volunteers, although the underlying causes are different. Where volunteers work independently they idiosyncratically self-report (e.g., by submitting ringing and sighting reports in online data bases) can be autonomous and thus rarely contact project staff, and can silently drop out of projects. As in animal survival studies where it is hard to separate actual death from failure to capture an individual which is alive, it can be hard to determine whether a volunteer is active but not self-reporting, or no longer participating. Failure to take into account of these processes will produce negatively biased estimates of volunteer retention probabilities.
What can be modeled?
CR analysis can be used to model how the survival or retention probabilities of populations (S) from time (t) to the next occasion (t+1): this parameter is denoted as St. In ecology understanding how survival probabilities vary with time can give crucial insight into the fundamental processes that influence population dynamics (e.g., Siriwardena et al. 1999). When used to analyze volunteer data, estimating retention parameters allows project managers to compare current and previous project health, and determine if management practices are improving volunteer retention with time.
Additionally, ecological and evolutionary studies also use CR to explore how group-level covariates influence survival or recapture probabilities by estimating a separate survival parameter for each level of the covariate. Examples of group-level covariates that influence population dynamics are; sex (Ssex: reflecting gender related differences in survival probabilities), season (Sseason), and cohort effects (Scohort: individuals who share a common temporal experience such as year of birth). Existing social science literature has hinted that similar patterns may be observed in volunteer retention profiles (such as Straining: learning new methods or Smotives: volunteers with the same motivations) (Ryan et al. 2001; Asah and Blahna 2012); however, until now the framework required to quantitatively test such hypotheses on large data sets while accommodating for detectability of less than one was lacking.
CR is also used in ecology to determine how variation in environmental, physiological, and pathological conditions experienced by individuals and populations influence their survival probabilities, including rainfall, body condition, and parasite burden (Lebreton et al. 1992). Such factors are often coded as individual covariates—which are hypothesized to influence the parameters being estimated. CR allows the comparison of the relative importance of different covariates and if interactions occur between them (see Grosbois et al. 2008). Hypotheses related to volunteer retention analysis can also be addressed using covariates, for example: volunteer age, staff-volunteer contact, and successful task completion frequency have each been speculated to influence volunteer retention (Bell et al. 2008). Fitting such variables as covariates will allow quantitative assessment of their importance in determining volunteer retention rates.
Data collection for CR analysis is longitudinal by definition and thus differs markedly from previous work analyzing volunteer retention or frequency of involvement. Whereas Wolcott et al. (2008) and Asah and Blahna (2012) used surveys to determine anticipated future involvement in a project and volunteering frequency at one time point, CR requires volunteer activity data to be collected at defined, though not necessarily fixed, intervals. Systematic bias in such volunteer activity data can be reduced in the same way as biological survey data using rigorous and standardized sampling protocols (e.g., Dillman et al. 2009). One way to perform this would be for project staff to contact all volunteers at defined time intervals (e.g., every 6 months) to determine their current activity status. The optimum sampling frequency will be determined by the underlying retention rate as it needs to be long enough to allow individuals to leave the project but not so long that they are never censused.
It is possible that existing ecological and conservation projects that employ volunteers collect no data on volunteer activity whatsoever as it is not the focal subject of interest, or that directly contacting project volunteers is deemed costly or time-consuming. In these situations, it may be possible to use a proxy for volunteer activity to determine current status. For example, if volunteers are required to submit data records regularly, then activity can be broken into discrete time steps where individuals are defined as active if a record has been submitted within that period. Regardless of the census technique used, not all volunteers will reply (in the case of direct census) or submit reports (in the case of an activity proxy) and as such the detectability of volunteers will always be less than one. Failure to accommodate this using the CR framework will result in flawed estimates of volunteer retention and workforce size (Gimenez et al. 2008).
The type and frequency with which additional covariate information is collected should be motivated by the hypotheses at hand (see Table 1). For example, information that will not change through the duration of the study (e.g., volunteer gender) will only have to be collected at one time point, whereas covariates that do change (e.g., volunteer satisfaction or task completion frequency) will have to be collected at frequent intervals. A fully interdisciplinary approach is encouraged here as data of this nature are already routinely collected by social scientists and may already exist for a given project.
There are a huge number of different models available to users of CR, each of which designed to address specific hypotheses or accommodate underlying structure of the data available. As such, it is impossible to provide a generalized “recipe” of steps to follow in order to analyze any volunteer data at hand. Instead, to give an appreciation of the types of questions that can be addressed using this suite of techniques, we present a nonexhaustive list of available models, introduce data formatting and suggest a series of timely hypotheses that can be addressed using the CR framework.
Known fate (KF)
KF models are used in ecology to model survival of individuals monitored with perfect knowledge such as those that carry radio or satellite transmitters. They are applicable to volunteer retention analyses when there is full knowledge of the fate of each volunteer, such as when active volunteers are frequently censused by phone such as in Bryce et al. (2011). KF analyses model a single binary variable and are thus akin to logistic regression (see Figure 2). It is simple to accommodate staggered entry of individuals through time if recruitment is ongoing (see the case study below) and censoring of a small number of individuals with unknown fate (if an individual was not contactable at the time of census) using the CR framework.
CJS models are widely used in animal ecology as the probability of recatching or resighting a marked individual after a given time step is usually less than 1. They jointly model apparent survival (S) and detectability (p), the probability that an individual is alive and recorded. True survival cannot be estimated for open populations as it is confounded by migration out of the study area. Consequently, the parameter estimated is termed as “apparent survival.” As the probability of recapture is less than 1, CJS capture histories can include 0s embedded within strings of encounters (see Figure 2). In volunteer retention analyses, CJS models are applicable when there is incomplete knowledge of volunteer status, especially where it is difficult to distinguish between a volunteer silently dropping out, being temporarily inactive, or active but not self-reporting. While detectability is generally considered a nuisance parameter in population biology studies, it is a key parameter of interest in volunteer-based research projects. Here, an individual's detectability reflects their reliability and willingness to self report, which could be a key determinant of data quality and reliability.
MS models are used in ecological studies to determine state-specific survival and between-state transition probabilities, such as dispersal between islands or disease infection statuses (e.g., susceptible à infected à recovered) (Spendelow et al. 1995; Conn & Cooch 2009). As with CJS models, they accommodate imperfect detection rates and thus state-specific parameters (retention, transition, and detectability) can be included where appropriate (see Figure 2). MS models have huge potential to examine the probability of volunteers progressing through discrete levels of involvement within a project (e.g., levels of training or responsibility), and the influence of such transitions on their retention probability. Such information will allow project managers to optimize the availability of volunteer training and development opportunities and lend novel insight into how volunteers interact with the methods they are implementing.
Covariates in evolutionary and ecological sciences are variables that are believed to influence an individual or groups probability of survival or recapture. For example, Soay Sheep prewinter body mass was used as a covariate to demonstrate its influence on surviving the winter period (Bonner et al. 2010). Covariates can take several different forms: 1) covariates common to all individuals (e.g., winter temperature), 2) group-level covariates shared by subpopulations of individuals (e.g., brood size), and 3) individual covariates where each individual has its own unique value (e.g., body mass). Furthermore, such variables can be categorical or continuous, constant, or time varying. Covariates in particular offer considerable scope to integrate hypotheses relevant to both social and biological sciences. The effect size and direction of factors that have been suggested to influence volunteer retention can now be quantitatively assessed. For example, if a volunteer successfully completing a task is hypothesized to increase their retention probability, task completion frequency fitted as an individual covariate is predicted to have a positive effect. Multiple covariates can be compared to determine the most influential factors on volunteer retention through comparison of effects sizes, and possible interactions between covariates can be explored (Grosbois et al. 2008). For example, the effect of successful task completion may depend on their level of involvement within a project. Ultimately, if carefully selected, each covariate reflects a hypothesis to be tested. If they are informative, they will improve the fit of the model to the data and be retained during model selection.
Adoption of the CR framework to analyze volunteer retention and detectability profiles opens up research avenues of interest to both social and biological scientists studying conservation volunteers. We have collated a table of timely hypotheses relevant to both disciplines in order to motivate future research directions and included information on how they may be coded in the CR framework (see Table 1). Aside from simply increasing understanding of how volunteers engage with the projects they are involved in, CR analysis can help project managers optimize the recruitment, management, and training opportunities available to their volunteers.
Volunteer case study
The CR framework outlined above was applied to a conservation project that used volunteers to remove an invasive nonnative mammal species from a 10,000 km2 of Scotland. Volunteers, defined as individuals who implemented project methodologies without financial reward were censused by telephone every 6 months (June and December) between 2006 and 2010 to determine their activity status. Volunteers fell into one of five vocational groups determined by their profession: 56 game keepers, 41 salmon fisheries staff, 52 conservation professionals, 17 land managers, and 120 local residents (see Web Figure 1). No further demographic information (e.g., age, sex, or income bracket) was collected. The data represent a time-expanded version of the data analyzed by Bryce et al. (2011). Retention probabilities were determined using staggered entry known-fate models (Pollock et al. 1989). Individuals with unknown fates for a given time step, due to failure respond to project officer contact, were censored. This removed the individual from the retention probability estimate for that time step, preventing imperfect detection bias. The analysis was performed using the RMark package within the R statistical modeling environment that interfaces with Program MARK software (White & Burnham 1999). Ten a-priori hypotheses were selected based on the available literature and project staff expertise (see Table 2).
Table 2. This table states the a priori hypotheses of factors thought to be influencing volunteer retention probabilities in our case study and the explanations behind them. The table is divided into group-level covariates and individual-level covariates in order to make it clear how they would be coded in the CR framework. The type column denotes whether a covariate is constant (fixed) or time varying (varies with each 6-month time step), and the β-direction column denotes the hypothesized direction of the covariates influence; positive β-estimates means that as the covariate increases, so does the probability of retention; negative β-estimates means that as the covariate increases retention probability decreases. References: 1Bryce et al. (2011, 2Bussell & Forbes (2003), 3Jamison (2003), 4Millette & Gagné (2008), and 5Wolcott et al. (2008)
Group level covariates
Retention probabilities are unique to each individual and show no structure
Volunteers from different vocational backgrounds have different retention probabilities owing to different motivations for participating1
Retention probabilities vary depending on the when the volunteer was recruited due to factors such as changing management structure through time
Volunteers with the same staff member will have more similar retention probabilities than others as staff-volunteer relationship is important factor in volunteer retention
Level of involvement
Volunteers with more training and greater autonomy have a higher probability of being retained2
Method of recruitment
Volunteers who approach the project autonomously have higher survival probabilities than those who were asked as volunteer motivation has been implicated in influencing retention rate
Individual level covariates
Active volunteers within 5 km
High total surrounding volunteer density will increase retention probability as they reflect the amount social interaction and support4
Target organism detections
Detecting a mink will increase retention probability as it represents successful task completion
Target organism captures
Capturing mink will increase retention probability as it benefits the local environment5
Number of rafts monitored
The more rafts a volunteer monitors, the higher their retention probability as the number of rafts they choose indicates their motivation level
The CR analysis indicated that, of the five categorical covariates tested, a given individuals retention probability was principally dependent on the vocational group to which they belonged and which 6-month period in which they were recruited (Scohort+vocation). No evidence was found for the hypotheses that the method in which volunteers were recruited, level of involvement, or the staff member in charge influenced volunteer retention probability (see Web Table1).
Figure 3(A) shows that fisheries staff (6-month retention probability estimate (R) = 0.95, 95% confidence interval (CI) = 0.89-0.98) had significantly higher retention probabilities than game keepers (R = 0.83, CI = 0.76-0.88) at the 95% level. The differences observed were robust across cohorts (data not shown). Conservation professionals, land managers, and residents all displayed similar variable retention probability estimates (R ≈ 0.90). The variability in retention estimates of the remaining groups hints that other factors could be influencing their retention probability behavior such as; task training (Jamison 2003) and socioeconomic status (Ryan et al. 2001; Hobbs & White 2012).
Figure 3(B) shows the cohort retention estimates, regardless of vocation. Individuals from cohort 1 have very high retention probabilities (R = 0.98, CI = 0.92-0.99), cohorts 2 to 7 show variable but similar estimates (R ≈ 0.88), and cohorts 8 (R = 0.71, CI = 0.50-0.85) and 9 (E = 0.76, CI = 0.55 -0.88) show the lowest retention probability estimates with wide confidence intervals. A reduction in the number of full time project staff from three to one, and the relative inexperience of the newly appointed project staff concerned may have been responsible for the low retention probability of the last two cohorts.
Applying covariates using CR analysis demonstrated that as the frequency in which an individual detected the target organism in the previous 6 months increased, their retention probability in the following 6 months also increased as the β-parameter estimate for this covariate is large and positive (see Table 3). In contrast, no influence of surrounding number of active volunteers, number of target organism captures, or number of rafts monitored was found.
Table 3. Hypothesized covariate analysis. This table shows model selection for single additive combinations of the base mode (cohort + vocation) and each covariate. TV = time varying covariate, C = constant covariate, ΔAICc = change in AICc from the base model (null hypothesis), β- estimate = estimated coefficient of the covariate, UCI = upper 95% confidence interval, LCI = lower 95% confidence interval. All models had 12 parameters, except the base model that had 11
Target organism detections
Target organism captures
Number of rafts monitored
Active volunteers within 5 km
CR represents an obvious, yet underutilized longitudinal framework by which retention rates of volunteers within ecological and conservation projects can be analyzed. To date, we are aware of no other freely available analysis framework with a comparable wealth of resources available (software, literature, and online forums) that will better lend itself to the quantitative analysis of volunteer retention in conservation and monitoring projects. Owing to its widespread use, many of the issues that a researcher studying retention will face, such as nondetection or fate uncertainty, have already been accounted for.
It is important to stress that we do not envisage the CR framework be used in order to replace more traditional lines of enquiry, such as focus groups and questionnaires (Evely et al. 2008), but rather as an invaluable longitudinal analysis framework that should be integrated with them. The additional covariate data (e.g., volunteer motives) required to make full use of this framework necessitates an interdisciplinary approach whereby social and biological scientists share data collection and analysis techniques.
Volunteer retention and project effectiveness
In the case study presented here, the link between high volunteer retention rates and increased project effectiveness are clear—a large resilient monitoring network is robust to detecting reinvasion of the focal invasive species. However, how volunteer retention rates relate to conservation project effectiveness generally is unknown. To our knowledge, no comparable longitudinal retention estimates of conservation or monitoring project volunteers have been published within the primary scientific literature. The scarcity of volunteer retention data precludes the drawing of general conclusions about the relative benefits and costs of high or low retention probabilities.
Under some scenarios, it may be possible that high volunteer retention rates are not beneficial to a conservation project. If a project has a vital yet demanding methodology, high drop-out rates may be unavoidable or if selective drop out of volunteers leaves a more resilient and qualified workforce, it may be beneficial. If the methodologies that volunteers are employing are ineffective, or even counter-productive, then increased retention rates will not improve the projects chances of success. With wide-scale adoption of CR, it will be possible to explore the factors determining if high retention promotes or hinders conservation project success, and determine when introducing measures to increase volunteer retention are cost-effective. As retention is just one factor influencing project success, having a high retention rate will not by default mean that a project is effective (see Devictor et al. 2010 for further factors). As such, a holistic approach, integrating, economic, social, and biological perspectives will be required in order to fully assess a projects long-term outlook.
The case study presented illustrates the utility of the CR technique to examine long-term trends in volunteer retention probability and to address relevant hypotheses regarding that factors which influence it. We found that three key factors explained variation in retention rates, a volunteers vocation, the cohort into which they were recruited and the frequency with which they detect the target organism. That volunteers are more likely to be retained if they have caught a mink in the previous 6 months raises the possibility of negative feedback between the biological aims of the project and the aims of its volunteers. If they are successful at trapping and removing mink, mink detections will decrease resulting in lower volunteer retention rates. We recommend coupling volunteer retention data with biological data within other conservation projects in order to further explore if such feedbacks are common characteristics of volunteer projects.
The CR analysis presented here has simply scratched the surface of the novel insights that this longitudinal framework can bring to volunteer behavior and management processes studies. Future work can examine how other common phenomena in wildlife research, such as trap dependence and transience (Pradel et al. 2005), could also be meaningful characteristics in volunteer retention analysis. For example, trap dependence would result in volunteers being more detectable after a given time step if they were detected in the previous time step (as opposed to individuals not detected then detected). Such nonindependence of an individual's detection probability could be highly correlated with the reliability of the data in which they generate or the effectiveness of the methods they implement. As more volunteer retention estimates become available, it will become possible to correlate project-level retention estimates with data reliability, cost efficiency, and project effectiveness, and ultimately providing novel insight into project sustainability. For the CR framework to reach its full potential, researchers must start to perform the following: (1) collect longitudinal activity data, or if it already exists share such data between disciplines, (2) define hypothetical groupings (e.g., vocational or demographic) that are broad enough to give sufficient statistical power and specific enough in order to confer detectable common traits, and (3) identify and measure sociologically relevant individual covariates that will influence retention probabilities.
This methodology, while being a useful tool for project managers, will hopefully be the first step toward a robust and integrated framework for the analysis of volunteer recruitment, retention, and management structure across projects and disciplines. The next logical step is to look across the spectrum of volunteer projects to determine any common factors associated with high volunteer retention and successful project implementation. This process could ultimately lead to science-based codes of best practice for volunteer management, which optimize retention rates and enhance both the economic viability and long-term sustainability of volunteer-based monitoring and conservation projects.
We thank three anonymous reviews and Olivier Gimenez for invaluable comments on this manuscript. This article would not have been possible without the hard work and dedication of all the Scottish Mink Initiative (formerly Cairngorms Water Vole conservation Project) staff and volunteers, with special thanks to Sarah Atkinson for her generous help and support. We thank to René Van der Wal for inspiration and invaluable insight into the social sciences. We would also like to thank NERC for CB's Studentship, Scottish Natural Heritage, the Tubney Charitable Trust, Mammal Trust UK, NERC (NE/E006434/1), CNPA, Dee Spey Isla Deveron salmon fisheries, and rivers Trust for funding. XL was in receipt of a fellowship from the Leverhulme Trust.