Open data practices have become increasingly important as highly collaborative projects are needed to resolve important issues in science, such as biodiversity loss, climate change, and infectious diseases. Data-centered science, with a focus on open data, is increasingly seen as fundamental to solving these massive, interdisciplinary challenges and to increase the return on investment of research (Data sharing and the future of science, 2018). This requires a cultural shift in scientific practices that cannot take place without the concerted effort of all researchers, their institutions, and funding sources. In the past decade, the impact of sharing data on research quality and scientific progress has been studied extensively (Milham et al., 2018; Perez-Riverol et al., 2019; Zhang et al., 2021). Many international initiatives have been instrumental in illuminating sound data practices, workflow reproducibility (Baker, 2016; Fidler et al., 2017; Open Science Collaboration, 2015), and developing tools and guidelines to enable data sharing, such as the Enabling FAIR Data project (COPDESS, 2018; Wilkinson et al., 2016), GO FAIR (COPDESS, 2018; David et al., 2020), and DataONE (Michener et al., 2012). There has also been increasing interest in probing the attitudes of scientists toward data sharing and reuse to determine the social and institutional barriers that prevent it. Lack of recognition and rewards for publishing research data has been found to be a key barrier to sharing data (David et al., 2020; Fecher et al., 2015), and there is a positive relationship between the behavior and attitudes of researchers on data reuse (Curty et al., 2017; Ludäscher, 2016). Scientists who reported that reusing others’ data increased their own efficiency and saved them time, were more likely to reuse data shared by others. The availability of institutional resources supporting data sharing and reuse (e.g., education, technology, expert help) and more generally available resources such as repositories and metadata creation tools have been found to greatly increase rates of data reuse (Kim and Yoon, 2017).

Data management organizations and institutions must make assumptions about which of these factors can be used to influence researchers’ open data practices. DataONE (the Data Observation Network for Earth) was created in 2009 with the idea that attitudes and behaviors could be changed by (a) providing easy-to-use data-sharing infrastructure and training, and (b) helping influential researchers within the natural sciences adopt open data practices (Michener et al., 2012). Its initial focus was on the biological and environmental sciences. DataONE was funded by a large federal grant to create cyberinfrastructure that would address barriers preventing more open, global, and reproducible research (Michener et al., 2012). Part of the charter of DataONE was to determine what these barriers are by surveying researchers in different fields and developing personas to help address their needs.

The Findable, Accessible, Interoperable, Reusable (FAIR) initiative uses a multi-faceted approach to assess data sharing and guide training at research institutions (COPDESS, 2018; Wilkinson et al., 2016). Large institutions like the Office of Science and Technology Policy (OSTP) and the National Institutes for Health (NIH) in the United States are now mandating open data behaviors, using the FAIR principles as a guideline for compliance (Holdren, 2013; NIH, 2023). Such mandates highlight the importance of knowing whether these initiatives have had an impact and which factors have been influential on the adoption of open data behaviors.

Surveying researcher attitudes towards data sharing and reuse are one method of determining the effect of these research initiatives on the scientific culture. Three large surveys were designed and implemented for DataONE by a research group at the University of Tennessee and the results were published in 2011, 2015, and 2020 (Tenopir et al., 2011, 2015, 2018, 2020). These surveys have been highly influential (2011 survey has 754 citations, 2015 has 258 citations, and 2020 has 65 citations according to Scopus (date accessed: 2023/04/27)) and continue to serve as a benchmark for surveys of researchers’ data sharing attitudes and behaviors. The multidisciplinary team of researchers in the DataONE Usability and Assessment Working Group provided input to the surveys. The first two surveys were open from October 2009 to July 2010 and October 2013 to March 2014, respectively (Tenopir et al., 2011, 2015). The third survey was administered in two instances. This survey was first sent to members of the American Geophysical Union and was open from March 2017 to March 2018 (Tenopir et al., 2018). It was then sent to a global audience from December 2017 to May 2018 and the two instances were amalgamated in a final analysis (Tenopir et al., 2020). In each survey, scientists were asked about their general views on data sharing and to report the issues that prevented or encouraged them to share data. The first study found that lack of access to and preservation of data were significantly hindering research progress, but researchers would be willing to share data if they were to receive proper citations or could place some conditions on access. The second survey, as well as work by Curty et al. (2017), revealed an increase in the willingness to share data and also a perceived risk of sharing data. The last study demonstrated continuing favorable attitudes toward open data; however, it was found that sharing was impeded by a lack of resources or a lack of perceived benefits of data sharing (Tenopir et al., 2020). The degree of cross-comparison between surveys was mainly qualitative. We are working to connect them quantitatively.

To determine the importance of individual, social, and organizational factors on data sharing, as well as how attitudes to data sharing and reuse have changed over time, this paper describes an integrated analysis of the three surveys from Tenopir et al. (2011, 2015, 2018, 2020). We use two theoretical frameworks, the Theory of Planned Behavior (Ajzen, 1991) and the Technology Acceptance Model (Davis, 1989) to determine which of these three factors each question within the survey relates to. The impact that world region, research domain, work sector, and funding source have on data-sharing attitudes is also analyzed and compared within this framework.


To examine the changing behaviors of the survey respondents, we employed several conceptual frameworks that have been used to explain underlying influences on human attitudes and behaviors. The theory of reasoned action (TRA) links the perception of social norms to behavioral intention (Ajzen and Fishbein, 1980). The TRA has been used to design and analyze several surveys on data sharing and reuse behaviors (Kim and Zhang, 2015). An extension of TRA, the theory of planned behavior (TPB) (Ajzen, 1991), incorporates an individual’s perception that a goal is achievable and distinguishes between career benefits and risks. TPB has been used in several multi-level frameworks to describe data sharing and reuse attitudes and behaviors after survey data revealed the importance of ease-of-use and time commitments (Kim and Stanton, 2016; Kim and Yoon, 2017). The technology acceptance model (TAM) is another extension of the TRA that was popularized in the fields of information systems and communication (Davis, 1989). The TAM adds the variable of perceived usefulness, assuming that individuals are more likely to be open to behaviors that are beneficial to them. Individual and social factors are often contextualized with facilitating conditions (e.g, institutional support, training repositories, technical services, mandates) as these have a strong impact on research behavior in many research fields (Kim and Zhang, 2015; Talukder, 2012; Venkatesh et al., 2003; Yoon and Kim, 2017). It is clear from the literature on data sharing and reuse that individual factors related to reputation and plausibility of success, social norms, and organizational commitment need to be considered (Kim and Zhang, 2015; Talukder, 2012; Venkatesh et al., 2003; Yoon and Kim, 2017).

It was clear to us, based on the work of Tenopir et al. (2011, 2015, 2018, 2020) that the three main influences on attitudes toward data sharing and reuse were individual motivations, social influences, and organizational support. There were several psychological and social theories that incorporated some of these influences, but capturing the complexities of all three required combining elements from multiple theories. As such, we have developed a theoretical framework which is a combination of TPB and TAM. The key elements we hypothesize will influence attitudes to data sharing and re-use are:

  1. 1.

    Social influences (TAM), including the social norms around data sharing of a particular field, trust in colleagues’ data and/or academic integrity, and incentives to share and reuse data.

  2. 2.

    Organizational influences (TAM), including the availability of training, tools and repositories and funding mandates and guidelines provided by local or federal institutions.

  3. 3.

    Individual influences (TAM and TPB), including effort associated with data sharing (TAM and TPB) and career benefits and risks (TPB).

We hypothesize that attitudes to data sharing and re-use will be positively related to:

H1: Increased perception that the research climate is trustworthy and open will be associated with improved data sharing and reuse attitudes.

H2: Perceived career benefits and reduced risk will be positively associated with improved data sharing and reuse attitudes.

H3: Increased sharing mandates, tools and support will be associated with improved data sharing and reuse attitudes. (Fig. 1).

Fig. 1: Theoretical Framework.
figure 1

Framework for researchers’ attitudes toward open data behavior showing the key elements of hypothesized influence, and hypotheses (H = hypothesis).


Data cleaning

To detect attitudinal changes across time, questions that were comparable between the three Tenopir et al. surveys were selected and combined into a single longitudinal dataset. For the sake of simplicity in this paper, each survey is labeled by publication year. In several cases, there were slight changes in the wording of the question and/or choice offered. There were also questions removed and new ones added between surveys. For instance, as a result of comments from survey participants, skipped answers in previous surveys, and discussions with members of specific fields on their particular use case, some demographic questions were removed and new ones added between surveys (e.g., career stage, age, gender). We only included data for questions that were sufficiently consistent across the three surveys.

The data cleaning process for this project required making sure that the answer types and ranges of options were made consistent between surveys. When yes or no answers were changed to yes, no, or not sure, we changed not sure to blank responses (NA). In the cases where the Likert scales had both neither agree nor disagree and not sure as answers for some questions but not others, we combined them into just neither agree nor disagree. We also cleaned each dataset for non-response. Specifically, for each survey, respondents who did not answer more than 5 questions were excluded to prevent ‘no answer’ from dominating. Note that a higher proportion of people in the 2015 survey skipped questions in the survey (Table 1).

Table 1 Summary of survey responses according to respondent demographics.

The cleaned dataset comprised 42 questions and responses from 3197 individuals (1214 in 2011; 539 in 2015; and 1444 in 2020; Table 1, raw data available on Zenodo (Olendorf et al., 2022)). Four of the 42 questions concerned the demographics of the respondents: region of the world, scientific domain, work sector, and funding agency. For these questions, answer types and scales were adjusted to ensure comparability across surveys. Countries were aggregated into six regions: Africa and the Middle East, Asia and Southeast Asia, Australia and New Zealand, Europe and Russia, Latin America, and the USA and Canada. The countries were clustered in the same way as they were in the surveys by Tenopir et al. (2011, 2015, 2018, 2020). These divisions were based on the International Union for Conservation of Nature (IUCN) regions (IUCN, 2022) (Table S1). This clustering allowed us to make quantitative comparisons. A similar clustering was made for the other categories, for example, the Physical Sciences included respondents in physics, chemistry, and engineering, and the Natural Sciences included those in biology, medicine, and zoology (Table S2). Work sectors were divided into Academic, Corporate, Government, Non-Profit, and Other. Funding agencies were divided into five categories: Corporation, Federal & national government, Private foundation, State, regional or local government, and Other. A corporation refers to for-profit companies, while a private foundation refers to a privately funded non-profit (e.g., the Bill & Melinda Gates Foundation).

The majority of the variables concern attitudes or beliefs about data sharing. The attitudinal questions required either a binary (yes or no) response or a selection on a Likert scale (Tables S3 and S4).

The variations in the portions of responses from each survey demographic for the 2011 and 2015 surveys were 2.9 ± 2.5 (mean ± standard deviation), 7.4 ± 9.1 for the 2015 and 2020 surveys, and 7.9 ± 8.6 for the 2011 and 2020 surveys.

Data analysis

To determine the factors that explain the most variance in the survey responses we performed a multiple factor analysis (MFA) with the Factominer (Husson et al., 2020) and Factoextra (Kassambara and Mundt, 2020) packages in R. MFA reduces the dimensionality of a large set of variables to maximize the explanatory value with the fewest variables possible. The input for the MFA is the list of survey questions as columns and survey respondents as rows. The MFA outputs orthogonal dimensions in the order of how much statistical variance within the survey responses each dimension explains. Analysis of variance (ANOVA) using the R stats package was then performed to assess the effect of demographics on each factor, followed by Tukey’s honest significant difference (TukeyHSD) test to determine the significance of the differences between the answers of the various demographic groups. All survey data and code for these analyses are available on Zenodo (Olendorf et al., 2022).

Degrees of freedom, the sum of squares, mean squares, F-values, and p-values for all ANOVAs performed in this experiment are provided in Table S14. Bartlett’s test for homogeneity of variance and residual tests were computed for each survey and each demographic variable (Table S15). Non-homogeneous variance could result from differences in the samples or the sampling methods across the three surveys (Tenopir et al., 2011, 2015, 2018, 2020). Sources of variation due to region and funding agency prevented analysis of these variables. One contributing factor was undoubtedly the fact that all the survey question responses were categorical variables with limited response ranges (yes/no or 1–5 Likert), which led to non-homogeneous variances. The standardized residual plots generally showed zero correlation with either dimensions one or two for each of the surveys and for all combined. The normal-QQ plots (Figs. S2S33) generally showed linearity within the central four quantiles of responses, which suggests that these data plausibilities derive from the same distributions.


Defining the dimensions

The multiple-factor analysis of the cleaned survey data from the three DataONE surveys in 2011, 2015, and 2020 (Tenopir et al., 2011, 2015, 2018, 2020) showed significant loadings on just two dimensions. These explained 4.0% and 2.9% of the variance, respectively, and were the only ones greater than the average percentage of explained variance of 2.7% (Fig. S1). The total mean and standard errors across all three surveys for all demographics for each dimension are shown in Table S5. The significant Tukey’s HSD means and p-values for all demographics in the three surveys combined are shown in Tables S6S9 and for each survey individually in Tables S10S13. In the following material, Tukey’s means and ANOVA significance levels are presented.

The questions clustered within Dimension 1 all related to openness to sharing data and the risks/benefits of sharing data (Table 2). The questions illustrated in Table 2, although significant, are not representative of the bulk of the responses (Table S3). The answers to these questions negatively correlate with dimension 1 when more openness to sharing is expressed and positively correlate when concern over risks of data sharing is expressed which led us to interpret dimension 1 as willingness to share.

Table 2 Top 5 contributing questions for dimension 1 of the MFA.

The top five questions clustered with Dimension 2 related to processes, databases, data availability, and risk (Table 3). The questions illustrated in Table 3, although significant, are not representative of the bulk of the responses (Table S4). The answers negatively correlate with dimension 2, indicating more agreement, when the questions relate to satisfaction with processes, databases, and data availability, and positively correlate with dimension 2, indicating more disagreement, when the questions relate to risks and conditions on data. Hence, we interpreted dimension 2 as satisfaction with resources.

Table 3 Top 5 contributing questions for dimension 2 of the MFA.

As illustrated in Tables 2 and 3, we used the data sharing and reuse framework (Fig. 1) to categorize all the questions included in the MFA (Table 4). Individual benefits and risks are the strongest elements for each of the first two dimensions, but social and organizational elements play a more important role in researchers’ satisfaction with resources. Organizational influences have a much greater impact on satisfaction with resources.

Table 4 Contributions to the first two dimensions of the MFA derived by assigning each question to an element from the framework in Fig. 1.

Impact of demographics on willingness to share and satisfaction with resources

The effect of world region on willingness to share scientific data varied with the region and with the survey. Overall, respondents from Australia & New Zealand (0.288), USA & Canada (0.238), and Europe & Russia (−0.037) were more willing to share than those from Asia & Southeast Asia (−1.020) and Africa & Middle East (−0.939). Those from USA & Canada were more willing to share than respondents from Latin America (−0.468) (Table 5). Respondents coming from the USA and Canada showed a significant increase in willingness to share between 2011 and 2015 (−0.446 to 0.635, p < 0.001) and across all three surveys (−0.446 to 0.908, p < 0.001) as did those in Europe & Russia (2011–2020: −0.644 to 0.264, p < 0.001) (Table S10 and Fig. 2A). No other significant differences in willingness to share by region were observed between surveys.

Table 5 Significant pairwise differences between willingness to share for world regions computed with ANOVA and TukeyHSD.
Fig. 2: Region MFA.
figure 2

Willingness to share (A) and satisfaction with resources (B) by region (Erway and Rinehart, 2016; European Union, 2013; Evans, 2010a). The world is divided into six regions (Africa & Middle East, Asia & Southeast Asia, Australia & New Zealand, Europe & Russia, Latin America, USA & Canada) based on previous work.

Satisfaction with resources was significantly greater for respondents from Africa & Middle East (0.770), Asia & Southeast Asia (0.587), and Latin America (0.510) than those from USA & Canada (−0.163) (Table 6 and Fig. 2b). The same suite of countries were more satisfied with resources than Europe and Russia (−0.086). There were no significant differences observed between regions for individual surveys or between survey events.

Tables 6 Significant pairwise differences between satisfaction with resources for world regions computed with ANOVA and TukeyHSD.

Across regions and surveys, mean willingness to share was inversely correlated with satisfaction with resources (R2 = 0.8, p-value = 0.017, F-value = 15.67). There were also significant correlations for the 2011 (R2 = 0.82, p-value = 0.012, F-value = 18.68) and 2020 (R2 = 0.84, p-value = 0.011, F-value = 20.44) papers, but the ANOVA results for satisfaction with resources for these individual papers were not significant, so these results are not included.

For domains of study, physical science (0.291) and information scientists (0.660) were more willing to share than natural scientists (−0.128) (Table 7). Social scientists (−0.592) were far less willing to share than researchers in the physical sciences, natural sciences, and information sciences. Willingness to share increased substantially across time in the physical sciences (e.g., physics, chemistry, engineering) (2011–2015: −0.566 to 0.462, p < 0.01; 2011–2020: −0.566 to 0.585, p < 0.001) and natural sciences (e.g., biology, medicine, zoology) (2011–2015: −0.485 to 0.179, p < 0.01; 2011–2020: /0.485 to 0.216, p < 0.001; Fig. 3A). Satisfaction with resources, however, only increased significantly for the social scientists (2011–2020: −0.259 to 1.026, p < 0.05; Fig. 3B) (Table S11). There was no correlation across domains between willingness to share and satisfaction with resources.

Table 7 Significant pairwise differences between willingness to share for domains of study computed with ANOVA and TukeyHSD.
Fig. 3: Domain MFA.
figure 3

Willingness to share (A) and satisfaction with resources (B) by domain. The domains are divided into six categories (Natural Science, Physical Science, Information Science, Social Science, Other).

For work sector, people working for the government (0.756) were much more likely to share their data than people working in the academic (−0.046) or commercial (−0.264) sectors (Table 8). The effect of work sector on respondents’ willingness to share showed a significant increase over time in the government (2011–2015: −0.197 to 1.020, p < 0.01; 2011–2020: −0.197 to 1.249, p < 0.001) and academic sectors (2011–2015: −0.604 to 0.161, p < 0.001; 2011–2020: −0.604 to 0.155, p < 0.001) (Fig. 4A and Table S12). In 2015, respondents from the government sector (1.020) became more willing to share data than those in academia (0.155, p < 0.05) and the difference between them was greater by 2020 (1.249, 0.155, p < 0.001) (Table S12). Satisfaction with resources was substantially larger in the commercial (0.645) sector than the government (−0.015) and academic (−0.046) sectors (Table 9). There was no trend observable over time (Fig. 4B). There was no correlation across work sectors between willingness to share and satisfaction with resources.

Table 8 Significant pairwise differences between willingness to share according to work sector computed with ANOVA and TukeyHSD.
Fig. 4: Work sector MFA.
figure 4

Willingness to share (A) and satisfaction with resources (B) by work sector. The work sectors are divided into five categories (Academic, Commercial, Government, Non-Profit, Other).

Table 9 Significant pairwise differences between satisfaction with resources according to work sector computed with ANOVA and TukeyHSD.

The effect of funding agency was the last variable analyzed (Table 10 and Fig. 5). Researchers with federal & national funding (0.257) were far more willing to share data than researchers with state, regional & local government (−0.676), corporate (−0.555), or private funding (−0.312). There was a substantial increase in willingness to share for researchers receiving funding from federal & national governments between 2011 and 2015 (−0.323 to 0.605, p < 0.001) and 2011–2020 (−0.323 to 0.596, p < 0.001) (Fig. 5A and Table S13). Researchers with funding from corporations, however, showed a marginal, non-significant increase in their willingness to share between 2011 and 2020 (−1.305 to 0.318, p = 0.055). Satisfaction with resources showed no change overall or between years (Fig. 5B). There was no correlation across funding agencies between willingness to share and satisfaction with resources.

Table 10 Significant pairwise differences between willingness to share according to sources of funding computed with ANOVA and TukeyHSD.
Fig. 5: Funding agency MFA.
figure 5

Willingness to share (A) and satisfaction with resources (B) by funding agency. The funding agencies are divided into five categories (Corporation, federal & national government, Private foundation, State, regional & local government, Other).


Our data indicate that researchers’ analysis of the costs, benefits, and effort associated with sharing and reusing data has the greatest impact on their willingness to share and satisfaction with resources (Table 4). The top-scoring responses associated with willingness to share relate to a general sense that sharing data is a good idea in general (Table 2). Organizational and social influences have a significant impact on satisfaction with resources, while individual elements relate to satisfaction with current practices or knowledge about the existence of data-sharing tools (Tables 2 and 3). This indicates that individual researchers are happy with their current data management practices, but tend not to be satisfied with the tools and data provided by other researchers (Table 3).

The hypotheses derived from our framework (Fig. 1) are supported by our analysis of dimensions 1 and 2. The correlations between the survey responses to individual questions and willingness to share (dimension 1) (Table 2) indicate that researchers who are less concerned with the risks of sharing and reusing data tend to see it as beneficial to their careers and the research community in general (H2). The strong impact of social influences on willingness to share (Tables 2 and 4) relates to the concept that a trustworthy and open research community is one that ensures credit is given to the researchers that deserve it (H1). The correlations between question responses in Table 3 and the satisfaction with resources dimension confirm hypothesis H3, that providing training and tools to researchers improves their data sharing and reuse attitudes. It is also clear that researchers who are satisfied with their data resources tend to be less concerned with the risks of sharing and reusing data based on the correlations in Table 3.

There was a substantial increase in researchers’ willingness to share data in most parts of the world between 2011 and 2020 (Fig. 2). The USA & Canada, Australia & New Zealand, and Europe & Russia showed similar patterns, consistent with research on existing collaboration networks between member countries (Leydesdorff et al., 2013). In contrast, respondents from China and many Middle Eastern countries were not as willing to share their research data, which is possibly the result of a complex combination of historical, cultural, and governmental factors, although this situation is changing (Barrios et al., 2019; He, 2009) (Fig. 2 and Table S10).

The results by research domain (Fig. 3 and Table S11) are consistent with existing research on levels of collaboration across different fields (Barrios et al., 2019; Ding, 2011; Iglič et al., 2017; Kyvik and Reymert, 2017), which showed that researchers in the social sciences are more likely to have single-authored papers and are less likely to share their data than are researchers in the physical and natural sciences (Iglič et al., 2017; Kyvik and Reymert, 2017). This may result from the differences in data type and restrictions associated with human subjects data (Fecher et al., 2015), lower levels of interdisciplinary research in the humanities and social sciences (Bishop et al., 2014; Uddin et al., 2021), or relate to the more fundamental issue of funding. An increasing overlap between commercial and federal funding is increasingly making research more profit rather than theory-focused. This trend favors science, technology, engineering, and mathematics research as these produce highly cited research and bring in a great deal of grant money (Münch, 2016). Social science was the only domain in our study that increased satisfaction with resources (2011–2020: −0.259 to 1.026, p < 0.05). This change may have occurred due to the small number of respondents to the 2020 survey; however, it does align with other findings that researchers who are less likely to share their data, such as social scientists, feel satisfied with the resources they need for finding and sharing data (Tenopir et al., 2011, 2015, 2018, 2020).

Work sector has some notable impacts on willingness to share. Our observations of low willingness to share within the commercial sector in contrast to a high willingness to share in the government sector are consistent with results from the literature. Such results indicate that commercial involvement makes researchers more cautious about sharing data, while government involvement motivates open data practices through mandates and strong recommendations (Fig. 1C iii) (Evans, 2010b, 2010a). The sharp increase in the willingness to share data in the U.S. government sector (Fig. 4 and Table S12) may have been influenced by the Open Data Policy outlined by the U.S. Obama administration in 2013 (Holdren, 2013; Office of the Press Secretary, 2013). These data policies during the Obama administration resulted in a 400% increase in the number of datasets available on between 2012 and 2016 (Chief Information Officer’s Council, 2016). This is supported by the fact that the government sector of USA-Canada was the only sector to see a significant increase in willingness to share between 2011 and 2015 (−0.001 to 1.542, p < 0.05). The European Union has also devoted a great deal of funding and effort to research data openness with the creation of the Public Sector Information (PSI) Group in 2002 (European Union, 2013). It is likely that the academic trends in willingness to share will follow those within the government sector because academic research depends heavily on funding from the federal government, and thus governmental policies will eventually apply.

Indeed, consistent with this idea, researchers receiving federal and national government funding (Fig. 5A and Table S13) were the most willing to share research data. Our observed increase in willingness to share research data involving commercial and federal funding may result from the large proportion of individuals in the commercial sector (31%) reported receiving the majority of their funding from a government source. These answers do not address the possibility of partial funding from the government, which means that this survey likely underestimates the proportion of those in the commercial sector that must abide by open data policies to receive government funding. It is difficult to interpret all the factors that drive data sharing within science, but generally speaking, the trends reported in this work indicate an increasing willingness to share research data without a substantial change in the level of satisfaction with resources for sharing research. This finding aligns with previously reported trends (Fane et al., 2019).

There were some substantial sampling differences between each of the three surveys, which may have affected the results (Table 1). The proportional response among groups for 2011 and 2015 was quite similar where the most substantial change in willingness to share and satisfaction with resources occurred. The 2020 survey had the most substantial differences in sampling proportions between the demographic categories. Despite this, willingness to share and satisfaction with resources did not change significantly from the results in 2015. This indicates that sampling differences are unlikely to be the cause of the observed differences between the 2011 and 2015 surveys.

The insights offered by looking at regional differences over time show that there is an overall negative relationship between willingness to share and satisfaction with resources depending on the region of the world (Fig. 6). Despite a high willingness to share data in Europe & Russia, USA & Canada, and Australia & New Zealand, well-documented (European Union, 2013; Fane et al., 2019; Holdren, 2013; Mason et al., 2020) advances in the establishment of mandates, training and the availability of resources were not reflected by a correspondingly high satisfaction with resources. This contrasts with other regions, where willingness to share data may be low, but researchers are satisfied with the resources available. These results might appear to contradict H3, but really they indicate that data-sharing infrastructure, incentives, and social norms must be in place before data-sharing training and tools will have a meaningful impact. Mandates, tools, and training do improve data-sharing attitudes within organizations and regions (Fig. 2 and Table 3).

Fig. 6: Region MFA Correlation.
figure 6

Mean willingness to share (Dimension 1) versus mean satisfaction with resources (Dimension 2) by world region across all three surveys (All). The regression line is shown along with the standard errors of each point in each dimension.

On the whole, scientists in the Western world (United States, Canada, Australia, New Zealand, Latin America, Europe) who work in academia, government, or corporations and who focus on the natural and physical sciences show an increased willingness to share research. There are many factors that could explain these trends. For example, there has been a substantial increase in international collaboration within Western countries over the past decade, but this scientific network has largely excluded African, Middle Eastern, and Asian countries (Leydesdorff et al. 2013). This outcome may be due to funding efforts that focus on collaboration across borders, for example, the work the Centers of Excellence for Europe (Bloch et al. 2016) and the International Panel on Climate Change. There have also been efforts to encourage collaboration and openness by federal agencies in the United States of America such as the Department of Energy with its Environmental Research Centers (Boardman and Ponomariov, 2011). Funding for collaborations between industry, government, and academia (Boardman and Gray, 2010; Gray et al., 2001; Lin and Bozeman, 2006) have also been prioritized with the growth of multidisciplinary, multipurpose university research centers. These collaborations have resulted in exponential growth in the funding and output of scientific research for the past decade (Clark and Llorens, 2012). This focus may partially explain why satisfaction with resources did not change for most of these categories throughout the three surveys. The increase in the need for collaboration and sharing has outpaced the investment in technologies and resources for sharing data data (Bijsterbosch et al. 2016; Erway and Rinehart, 2016; NASEM, 2017). A survey of European research institutions showed that 56% devoted no funds to research data infrastructure and most respondents either did not know anything about said funding or reported it as insufficient (Bijsterbosch et al. 2016). Data management is typically viewed as an indirect cost of research, rather than fundamental to the research process (Erway and Rinehart, 2016).

Finally, we want to mention that this work is a case study in data reuse, in particular its strengths and challenges. One of the primary objectives of open science is to ensure the reproducibility of results and hence reuse of data (Peng, 2011). Methodological consistency is a critical component of any type of data integration and sharing, and our experiences demonstrate this point. The results described from the combined data presented in this work demonstrate the importance and impact of good data stewardship. We, therefore, recommend that the final, combined data set we prepared for this work should be utilized for any future longitudinal research that involves these data. It is always challenging to balance a desire for long-term consistency and an ability to capture new trends. We believe that more robust findings could have been drawn between surveys had there been greater design consistency across surveys and their deployment. Despite these challenges, the trends we extracted from these data clearly indicate improvement in willingness to share data over time. We hope that future surveys on scientific attitudes toward data sharing and reuse will benefit from the lessons on the consistency that we have learned in this work.


We have shown that a trustworthy and open research climate and a perception of individual benefit increase data-sharing behaviors. The fact that willingness to share has increased since the 2010s in government and academic circles suggests that mandates and societal norms work. This is emphasized by the increase in willingness to share by researchers that receive federal or national funding. Such funding sources are dominated by research funding agencies such as the National Science Foundation and the Australian Research Council, all of whom have increasingly set an expectation of open data and open science for their grantees in alignment with recommendations from the International Science Council and the World Data System.

The difference in domain uptake of data-sharing attitudes suggests, as others have found (Specht and Crowston, 2022), that research that uses instruments or shared infrastructures such as astronomical telescopes, satellites or large marine vessels leads to an expectation and perhaps requirement that the research emanating from use of such facilities will be shared. Here the importance of social or community norms is paramount. For social scientists and other similar domain specialties work is more individual, resulting in them being late adopters of the open data concept.

We have confirmed that researchers in domains that have an existing high reliance on instruments or shared resources, researchers reliant on funding from national or federal sources, work sectors where internal practices are mandated, and regions where communication networks are highly effective have all improved in their willingness to share data. It is only when you move to action that you begin to test the tools and procedures necessary to engage in the practice. We encourage, based on this observation, increased collaboration with countries with fewer research resources to improve their access to data-sharing tools and expose them to open data norms that could foster a substantial increase in willingness to share data. Also, we encourage increased investment in data-sharing resources throughout the world as it will translate a willingness to share into actual sharing of data.

This work is limited by the fact that these data management surveys were not designed to be longitudinal or with our theoretical framework in mind. The questions were designed to capture a wide array of behaviors and causes rather than identify specific ones. As a result, dimensions 1 and 2 did not contribute as strongly to the MFA as they would with a survey designed for this purpose. However, the results from this analysis are still significant and show interesting trends in the attitudes of researchers. In the future, we recommend such surveys should have a consistent theoretical framework to enable more precise diagnoses and the questions should be standardized over time to ensure the results are comparable.