Who Engages in Citizen Science Projects on Urban Wildlife Ecology? Findings on Participant Groups from Cluster Analyses

Participants in citizen science

CS projects would certainly be impossible without volunteers. They are at the core of the success of CS and, therefore, an indispensable part of CS projects. Several studies have sought to understand the demographic background of CS participants (e.g., Füchslin et al. 2019; Haklay 2018; James 2020; Pateman et al. 2021; Strasser et al. 2023; Waugh et al. 2023); however, there are still significant gaps and limitations within the current literature: One report documented demographic data in only 10% of papers (NASEM 2018), and a meta-analysis found only 7.5% of papers that report demographics (Waugh et al. 2023). Other studies suggest that CS participants’ demographic profile is unknown (Moczek et al. 2021; Strasser et al. 2023). In cases when demographics have been reported, most data came from the United States (NASEM 2018) and less is known from other countries (for exceptions see Dekramanjian et al. 2023; James 2020; Mac Domhnaill et al. 2020; see also Paleco et al. 2021). Therefore, we suggest using a person-centered approach (as opposed to variable-centered approaches; Füchslin 2019; Losi 2023; Schäfer et al. 2018) based on few demographic variables (Moczek et al. 2021; Strasser et al. 2023).

There have been several studies on participants’ demographics so far, but they presented mixed results: CS participants were mostly middle-aged or older (Crall et al. 2013; Curtis 2018; Domroese and Johnson 2017; Füchslin et al. 2019; Haklay 2018; Jönsson et al. 2024; Mac Domhnaill et al. 2020; Pateman et al. 2021; Waugh et al. 2023) or belonged to different age groups (Strasser et al. 2023), and were mostly male (Curtis 2018; Füchslin et al. 2019; Haklay 2018; Jönsson et al. 2024; Pateman et al. 2021; Strasser et al. 2023) or mostly female (Crall et al. 2013; Domroese and Johnson 2017) depending on the projects’ topic. CS participants tended to have a high level of education (Curtis 2018; Füchslin et al. 2019; Haklay 2018; Mac Domhnaill et al. 2020; Strasser et al. 2023; Waugh et al. 2023) or a low to medium level of education (Haklay 2018), and had a predominantly white ethnic background (Pateman et al. 2021; Waugh et al. 2023). Most CS participants also had a background in science (Strasser et al. 2023) and were mostly employed (Crall et al. 2013; Mac Domhnaill et al. 2020; Pateman et al. 2021).

In terms of CS topics, biodiversity projects were the most prominent projects, followed by health projects and physical science projects (Waugh et al. 2023). White and retired participants were more likely to engage in biodiversity projects than in health projects (Waugh et al. 2023), and female and educated participants were more likely to engage in online projects than in hands-on projects (Waugh et al. 2023). Such previous research has only relied on descriptive demographics of participants (e.g., Moczek et al. 2021), and better documentation of who is participating is needed (NASEM 2018), especially in countries other than the United States and for online projects (e.g., NASEM 2018; Waugh et al. 2023). It is also unclear whether there are specific participant groups and subgroups or clusters in the pool of CS participants. A person-centered approach could help demarcate these clusters. Knowledge about such clusters is highly important because it would provide a more comprehensive picture of the usual CS participants, and it could help synthesize the scattered results on CS participants’ demographics so far.

Clustering in citizen science

The process of clustering is a common scientific, person-centered method that will be further relevant in the future as it shifts the focus from populations to subgroups in populations (Metag and Schäfer 2018; Scheufele 2018). For instance, Füchslin (2019) provides an overview of segmentation (i.e., cluster) analyses from the field of science communication. We argue that such cluster analyses are also relevant for the CS context regarding participants’ demographics. So far, there has been some CS research that used clustering to categorize different CS projects (e.g., Kirschke et al. 2022; Pocock et al. 2017), literature on CS (e.g., De Filippo et al. 2020), engagement groups based on online engagement behavior (e.g., Aristeidou et al. 2017; Bruckermann et al. 2022; Ponciano and Brasileiro 2014), contributions to CS projects (e.g., Rosenblatt et al. 2022), or groups of participants with different motivations (e.g., West et al. 2021). Little is known, however, about whether there are also particular groups of participants based on their demographic data.

In the CS context, previous research has examined which groups of participants would potentially participate in CS projects by means of cluster analysis (Füchslin et al. 2019). These groups were determined based on demographic data and scientific measures (e.g., interest in science) in a representative science barometer study in Switzerland (Füchslin et al. 2019). This research found, for example, a Fully Employed Parents group with mostly middle-aged male participants who were all fully employed and had a medium to high level of education, a Senior Sciencephiles group with mostly middle-aged male participants who were mostly fully employed and highly educated, and a Free-Timers group with mostly middle-aged female participants who worked part-time or were already retired and had a high level of education.

Other research has examined the demographics of participants (besides other measures) of nine different CS projects in Australia, but did not use cluster analysis (James 2020). This research found, for instance, an Environmental Stewards group with mostly older female participants with a high educational background, a Science Enthusiasts group with mostly middle-aged, highly educated participants of both genders, and a Young People group with young participants. Extending this previous research (Füchslin et al. 2019; James 2020), we argue that it is also important to identify participant groups in CS projects by means of cluster analysis based on their demographics in order to characterize those participants who actually participate in CS (NASEM 2018; Waugh et al. 2023).

Such participant groups may also differ from another, for example, in how they contribute to an online platform of a CS project (Strasser et al. 2023). These differences may be quite informative for CS practitioners and researchers alike because they may describe different contribution patterns, which may influence the course of CS projects. So far, CS research has tracked online contributions and categorized participants into engagement profiles based on their activities (e.g., Aristeidou et al. 2017; Bruckermann et al. 2022; Ponciano and Brasileiro 2014). Yet, this research focused on the entire group of CS participants and did not differentiate among different participant groups.

Although there has been research on participants’ demographic background (e.g., Füchslin et al. 2019; NASEM 2018; Strasser et al. 2023; Waugh et al. 2023), person-centered approaches are lacking so far. Identifying relevant participant groups, which may have been overlooked so far, can help CS practitioners determine how to attract citizens’ attention and recruit potentially new participants for certain types of CS projects (e.g., by specifically framing advertisements and public relations campaigns; Schumann et al. 2024). Knowing more about participant groups could help align project goals to participants’ demographic background and establish a better fit between them. By increasing awareness, it could also indirectly support tailoring CS projects qualitatively to participant groups’ potential interests, and support managing CS projects in ways that retain participant groups in ongoing CS projects (Fischer et al. 2021). Clustering approaches can also help CS researchers identify more diverse groups of participants (Dawson 2018; Paleco et al. 2021; Pateman et al. 2021). Ultimately, knowing more about participants’ demographics enables CS researchers and practitioners to learn about the most valuable part of CS, that is, its participating citizens.

The current research

The research presented here aimed to (1) identify different groups of participants based on their demographics in two ecological online CS projects in Germany and (2) investigate how these participant groups differ in their engagement in the projects, which, to the best of our knowledge, has not been done before in the CS context. Although we acknowledge the increasing diversity of CS projects (e.g., Cox et al. 2015), we focused on ecological CS projects as they are the most prominent and best-established ones (Waugh et al. 2023). The number of ecological projects grew exponentially, with many active participants in the past decade (Strasser et al. 2023), which is why it is relevant to know more about the ecological project participants. To answer our research questions, we first performed cluster analyses in an exploratory manner (Jain et al. 1999). Second, we analyzed differences in engagement among the clusters based on participants’ contributions to the projects.

Ecological citizen science projects

We conducted two urban wildlife ecology CS projects in a German metropolitan city between September 2018 and December 2020. We planned, conducted, analyzed, and evaluated these projects together in an overarching joint research project. One CS project investigated urban terrestrial mammals, which we called Wildlife Researchers; the other one investigated urban bat species, which we called Bat Researchers (see below for details). Apart from outdoor and offline data collection, in both CS projects, participants performed all other activities on an internet platform we had specifically developed for these projects. In general, participants could obtain information about urban ecology, analyze their own data and the data from all participants, and discuss with other participants and scientists in a forum.

Procedure and participants

Besides the ecological research tasks, participants filled out a questionnaire at two measurement points, one at the start of the CS projects (T1) and one at the end (T2). The questionnaires assessed several individual learning outcomes (for results see Bruckermann et al. 2021a, 2021b, 2023; Greving et al. 2022, 2023; see also Phillips et al. 2018, 2019) and were identical except for the demographic data, which we assessed only at T1. In the samples, we included those citizens who filled out the first questionnaire at T1 and who were 18 years or older. Thus, we analyzed the demographic data of N₁ = 847 participants (M_age = 52.41, SD = 12.14, range = 18 – 83; 383 male, 461 female, 3 non-binary) in the Wildlife Researchers project and N₂ = 223 participants (M_age = 42.24, SD = 12.25, range = 18 – 78; 93 male, 129 female, 1 non-binary) in the Bat Researchers project. Notably, participants could participate in only one of the projects to ensure that there was no overlap between the samples.

Measures

We assessed participants’ age, gender, education, and employment with one item each. We measured age in numbers and gender with a single-choice item with three answer options (i.e., male, female, non-binary). Educational background was based on the International Standard Classification of Education (ISCED; OECD 2015) and a single-choice item with nine answer options (i.e., no qualification, secondary education certificate [ISCED 2], general secondary education certificate [ISCED 2], college entrance qualification [ISCED 3], university entrance qualification [ISCED 3], vocational training degree [ISCED 4], technical college degree [ISCED 4], college/university degree [ISCED 6 or 7], PhD/habilitation [ISCED 8]). Finally, we measured status of employment with a single-choice item with six answer options (i.e., non-employed, partial retirement, vocational training, marginally/occasionally employed, part-time employed, full-time employed). We did not assess ethnic background as we did not expect much ethnic diversity in our samples.

We assessed participants’ engagement in the projects by the number of contributions to the platform, which is a meaningful metric for the success of an online project (Strasser et al., 2023). We used logfiles from the platform for this purpose. Before analyzing them, we classified participants into three engagement categories, that is, non-active, visiting, and engaging participants, which we derived from previous research (Aristeidou et al. 2017; Bruckermann et al. 2022) and which were defined as follows: (1) Non-active participants did not visit and contribute to the platform at all. (2) Visiting participants actively contributed to the platform (e.g., uploaded pictures, identified species, or discussed in the forum) only once or twice, but could have scrolled through its content more often. (3) Engaging participants actively contributed to the platform at least three times or more often and could have also visited the platform more often. Subsequently, we used the number of participants in each category to analyze engagement frequency.

Data analysis

In both projects, we analyzed the demographic data using cluster analyses. Before performing the analyses, we z-standardized age. We used Euclidean distance as a distance measure and k-means as clustering procedure. To determine the number of clusters, we used the Silhouette score (Rousseeuw 1987), which varies between –1 and 1, where –1 indicates the worst and 1 the best clustering quality. Ideally, Silhouette scores should be around 0.5 or higher (Kaufman and Rousseeuw 1990). All steps of the cluster analyses were performed with R (version 4.4.1; R Core Team 2024; for the R script used, see Supplemental File 1).

Furthermore, we retrieved the logfiles for each participant with Matomo (v3.9.1). To test for differences between the clusters, we performed Chi-square tests and used α < .05 as criterion. Notably, we performed two separate tests for each project, one for the data collection phase and one for the data analysis phase, as previous analyses of the Wildlife Researchers’ tracking data had already indicated that engagement differed between these phases (Bruckermann et al. 2022). To perform the Chi-square tests, we used IBM SPSS Statistics (Version 29; IBM Corporation 2023).

Clusters of the projects

For both projects, we obtained similar Silhouette scores for two, three, and four clusters (see Figure 1). We chose the four-cluster solutions for the following reasons: (1) We had large sample sizes, and the chance that one of the clusters contained less than 5% of the samples, which would hinder optimal clustering (Anhalt-Depies et al. 2023), was still low with four clusters. (2) We wanted to avoid a purely dichotomous classification of participants. With only two clusters, which are oftentimes quite huge, especially with large sample sizes, important subgroups could be merged within these larger clusters and would remain undiscovered. (3) The cluster centers of two clusters are often not representative of the entire clusters (Kaufman and Rousseeuw 1990), and four clusters may represent the data structure much better. (4) Several smaller groups may be especially informative because they may provide a deeper understanding of the CS participants than few larger groups. Such an understanding may further inform practical decisions in the CS context.

Figure 1

Engagement frequency of the clusters

The Chi-square tests revealed no differences for the data collection phase of the Wildlife Researchers: χ²(6) = 3.15, p = .790. Participants were predominantly engaging on the platform with relatively few non-active or visiting participants (see Figure 2a; for absolute numbers see Supplemental File 3). For the data analysis phase, the clusters differed: χ²(6) = 14.77, p = .022. In the Academic Employees and Free-Timers clusters, there were more visiting than non-active or engaging participants, which had similar numbers. The All-Round Workers cluster had roughly equal numbers of non-active, visiting, and engaging participants. In the Academic Free-Timers cluster, there were more engaging than non-active or visiting participants (see Figure 2c).

Figure 2

For the Bat Researchers, there were no differences for the data collection phase either: χ²(6) = 11.72, p = .069. Participants were predominantly engaging on the platform with relatively few non-active or visiting participants. Still, the All-Round Workers and Free-Timers clusters had a bit more non-active participants than the other two clusters (see Figure 2b). For the data analysis phase, the clusters differed again: χ²(6) = 15.57, p = .016. The Academic Employees cluster had similar numbers of visiting and engaging participants and fewer non-active participants. In the All-Round Workers and Free-Timers clusters, there were more non-active than vising or engaging participants. In the Academic Free-Timers cluster, there were more visiting than non-active or engaging participants (see Figure 2d).

Implications

Knowledge about participant groups of ecological CS projects can help CS practitioners accelerate the development, design, and quality of future ecological CS projects in two ways. First, project goals can be aligned much better with the characteristics of the discovered participant groups. Second, the cluster analysis approach can also be helpful for developing recruitment strategies that can be tailored to the potential groups, for example, by using personas that had not been addressed before. The personas approach describes potential participant groups based on a typical individual’s characteristics (Chang et al. 2008; Cooper 1996; see also Schumann et al. 2024). If project coordinators discover underrepresented groups from clustering actual project participants, personas can help them better understand how to address these groups. This proposition is in line with the CS literature that has identified the need to reach a wider range of participants (Bonney et al. 2016; Haklay 2018; Lewenstein 2016). Knowledge about cluster analysis can also help CS researchers look differently at their data and help advance the typical composition of mostly middle-aged, male, and highly educated participants.

Strengths and limitations

We report on data from two ecological CS projects with robust sample sizes. To the best of our knowledge, cluster analysis has rarely been applied to CS participants’ demographics as most studies referred to demographics only descriptively (e.g., Füchslin et al. 2019; Strasser et al. 2023; Waugh et al. 2023). We also used common methods and metrics for the analysis. Nevertheless, different clustering methods, metrics, and decisions could yield different results, and the Silhouette score was not always above the recommended cutoff value. This, along with their exploratory nature, means that the current findings should be interpreted with caution. More research is needed to replicate and refine our results.

We could also have used more demographic categories. We opted for a smaller set to be able to make clear statements regarding the participant groups. Still, future research could investigate additional categories and include, for example, participants’ ethnic background. As the variety of CS projects is rapidly growing (Cox et al. 2015; Smith et al. 2023), our findings represent case studies and may not be applicable to other CS projects. It could also be the case that the type of our CS projects partly influenced which participant groups we found. Such influences should be kept in mind when setting up future CS projects and studies. The samples were also self-selected and not representative. Therefore, more studies are necessary to conduct broader research incorporating several case studies (e.g., James 2020) to gain a more solid picture of the demographic background of participant groups in CS.

Conclusion

This research is one of the first to describe participant groups in ecological CS projects based on their demographics by means of cluster analyses. In contrast to the scattered results on CS participants’ demographics so far, this person-centered approach provides an alternative description of CS participants. Our findings revealed two participant groups that can be easily overlooked (i.e., All-Round Workers, Free-Timers) besides more typical groups (i.e., Academic Employees, Academic Free-Timers). Notably, it is not one demographic characteristic alone, such as age, but the combination of demographics within these groups that highlight potentially relevant groups. Moreover, the participant groups also engaged differently on the online platform, especially during the data analysis phase. Academic Free-Timers were particularly engaging in or visiting the projects, whereas All-Round Workers and Free-Timers were mostly inactive. These insights can provide valuable information for CS researchers and practitioners alike for attracting, engaging, and retaining participants in future ecological CS projects.