Forwards is a task force that was set up by the R Foundation in 2015 to address the under-representation of women that has since widened its scope to encompass other under-represented groups. The task force is organised as a core team comprising leaders from a number of sub-teams that focus on particular aspects:
- Community
-
General outreach to help people from under-represented groups get into R and develop as useRs. Members have represented the task force at events such as AlterConf, Trans*Code International Transgender Day of Visibility Hackathon, National Federation of the Blind Convention, and Society for Advancement of Chicanos/Hispanics and Native Americans in Science. The team promotes outreach schemes such as NASA Datanauts and works alongside others in the R community seeking to widen participation, such as R-Ladies or other sub-teams, for example co-ordinating the diversity scholarship scheme for useR! 2017.
- Conferences
-
With a particular focus on R Foundation conferences, this team liaises with the organizers/program committee on policies and inclusion initiatives. For example this team initiated the conference buddy scheme for useR! 2017 and collaborated with R-Ladies to host a special session for newcomers.
- On-ramps
-
Creating paths for useRs to develop their skills and make contributions to the R/Bioconductor package ecosystem. Activities have included speaking at the useR! 2017 newcomer session and R-Ladies meetings on collaborative coding.
- Social Media
-
Managing the Twitter account and the recently started Facebook group to support people from under-represented groups. Maintaining the website and co-ordinating the blog.
- Surveys
-
Running and analysing community surveys; publishing corresponding reports and data. For example, this team ran a survey at useR! 2016 to find out about the demographics, programming experience and community involvement of useR! attendees.
- Teaching
-
Working on methodology, materials and workshops designed for under-represented groups. In particular, this team have developed materials for two workshops: one aimed at high school girls, on creating a data analysis web application and another aimed at women, on package development. So far these workshops have been run in Australia and New Zealand, with repeats planned for Europe and North America.
This new column provides an outlet for news about the work of the task force as well as more detailed reports. This inaugural column presents an article on the results of the useR! 2016 survey.
1 A First Survey on the Diversity of the R Community
The R environment for statistical computing is an open-source project for data analysis. It provides an opportunity for users worldwide to benefit from an extensive number of software tools for a wide spectrum of data analysis and also allows users to participate in the project. This participation may take many forms. Beginners may attend tutorials on data analysis with R and contribute through their feedback while advanced users may write their own code and even publish it on one of the public repositories. As a whole, we will refer to the participants of the R project as the “R community” – but be assured of our awareness that this community is very hard to grasp.
Several studies have already investigated important aspects of the R community. A central place in the development of R is taken by the R Core Team. It is responsible for the development of the basic R software and the maintenance of infrastructure which is necessary for its distribution. Several interesting aspects of the organization and work of the R Core Team have been summarized by Fox (2009), who conducted a series of interviews with its members. Mair, E. Hofmann, K. Gruber, R. Hatzinger, A. Zeileis, and K. Hornik (2015) investigated the motivation and values of the authors of R software packages published on the CRAN, Bioconductor and R-Forge repositories by the means of a questionnaire. Like the R Core Team, package authors are a well-defined subgroup of the R community that is quite easily accessible because their names are made public. However, very little is known about the remaining part of the R community, which consists of regular R users as well as developers who have not published R packages on public repositories so far. They are usually anonymous and can not be directly addressed.
Therefore, in a first attempt to start gathering information on the R community, we turned to a subsample of the community that was more clearly defined, namely the attendees of the useR! conference 2016, held in Stanford, CA. The advantage of the useR! conference as a platform for the survey – besides feasibility – was that it is attended by both R users and R developers, which is an important prerequisite for getting a broader insight than the previous studies on R Core members and CRAN package developers.
Yet we are fully aware that the useR! attendees are not a representative sample from the R community as a whole, because conference attendance itself depends on a variety of resources. This is particularly true for an expensive location like Stanford. Therefore, in this study we don’t claim that our results can be generalized to other parts of the R community – nevertheless they tell us something about a quite relevant part of this community.
Speaking in terms of study design, we can only consider the useR! attendees – not the R community as a whole – as the population, from which again only a sample answered the questionnaire. We did, however, achieve a rather high return rate of 455/904 and also checked the representativeness of this sample as compared to the population of useR! attendees with respect to demographic information provided at or derived from registration1. Like most studies, we found that women were slightly more willing to participate in the survey compared to men, however this imbalance only becomes relevant when men and women differ in their responses, in which case we present results for men and women separately.
All attendees of the useR! conference 2016 received an invitation to participate in the survey. The questionnaire was presented online and contained 26 questions, which concerned demographic information, experience with and opinions about R, and involvement in the R community. Our results will be presented in two sections: In the first section, we report results on the demographic data. In the second section, we summarize the responses concerning the participants’ usage of R and their involvement in the R community.
The main goal of the survey was to obtain information from R users and developers that may help in setting up more inclusive infrastructure for the usage, development and teaching of R. The results may also point out prerequisites for successfully developing R expertise and hopefully initiate a dialogue with under-represented subgroups of the R community in order to help them formulate their needs.
While the initial focus of the survey, as well as the task force itself, was on addressing the under-representation of women, we did also investigate the representation of minority groups identified by race and sexuality. We will show that these minorities are severely under-represented. However, for a detailed statistical analysis the group sizes were so small that conclusions would have been questionable and identifiability of individual respondents would have become an issue.
Given that this survey was the first attempt towards addressing a broader part of the R community, both the structure of the questionnaire and its analysis are of a rather explorative nature. This, together with the above-mentioned fact that the respondents are a representative sample from the attendees, but not necessarily from the R community as a whole, has led us to formulate any conclusions carefully and emphasize effect sizes rather than significance in our statistical analysis.
Based on the work of Cohen (1988), we calculated Cohen’s Omega as an effect size measure for describing associations between categorical variables. Cohen also provided guidelines for interpreting these measures: For Cohen’s Omega, values around 0.1 correspond to a small effect, values around 0.3 correspond to a medium effect, and values around 0.5 correspond to a large effect. For all results presented here, a Cohen’s Omega of 0.13 or higher also corresponded to a p-value below 0.05.
Demographic information
The sample consists of 455 respondents. We first investigated the basic characteristics of this sample, which include the age of the respondents, the number of women and men respondents, and the educational level of the respondents, as well as variables related to under-represented groups. Unless noted otherwise, all reported percentages refer to the relative size compared to the entire sample of 455 respondents.
Most respondents identified as men or women: an extremely small number of attendees had a gender outside the binary, but are not included in this analysis to avoid risking identifying them without their consent. Of the remaining respondents, 169 (37.22%) identified as women. 27 respondents (5.93%) reported to identify as members of the LGBT community, whereas 11 respondents (2.42%) did not reply to this question. When asked about their ethnicity and country of origin, a wide array of responses was given. We used a free text field in order to avoid deficiencies in standard race classification systems, which may not represent the respondents’ identity. The open answer format does, however, make summarizing the results more difficult, so for this analysis we did compare the free answers to a standard classification system. There were 409 responses on the subject of racial identity. This number was used as the reference for the following percentages: 302 respondents (73.83%) identified as White/Caucasian. Only 7 attendees (1.71%) identified as Black, Native American (including Pacific Islanders) or Middle Eastern.
To put this in context, more respondents identified primarily as Jewish (8 responses, corresponding to 1.96%), than as Black, Middle Eastern or Native American combined. Of the other attendees, 67 (16.38%) fell within what the standard American Federal criteria would class as Asian, 12 (2.93%) as Hispanic/Latinx, and the remainder as mixed-race.
Given these results, there appears to be a severe under-representation of non-white attendees compared to the general population. These results might further indicate an under-representation of members of the LGBT community, although this question is more difficult to address. In the adult population of the United States, around 3.5% of the adult population consider themselves as part of the LBGT community, as was shown in the report of Gates (2011). However, these figures might constitute a severe underestimation, as was argued by Coffman, L. C. Coffman, and K. M. M. Ericson (1950). We are not able to further investigate these minorities and their possible reasons for not attending the conference for two reasons: First, the small sample sizes of the under-represented groups would make it difficult to generalize any conclusions. Still we do not find it justified to merge ethnic minorities into larger groups just for the sake of the analysis. Second, if a group consists of very few attendees, reporting more detailed findings could compromise the anonymity of individual respondents. Therefore, the remaining analysis will focus only on gender differences.
We do, however, want to point out that the under-representation of non-white attendees we found in the useR! survey was more severe than we would have expected and led to a broadening of the focus of the task force to represent not only the concerns of women but also of other under-represented groups, be they identified by race, gender, sexuality, class, or disability.
With respect to gender differences, we found that women and men respondents differed with regard to their age (approximated from their birth year, see Figure 1), but not with regard to their educational level (with response options “max. secondary school”, “undergraduate degree”, “Masters degree”, “Doctorate” and “Professional degree”).
The median birth year of men respondents was earlier than that of women respondents, with men respondents on average being in their early 40s and women respondents being in their mid-30s.
Therefore, when comparing women and men respondents in the remainder of the analysis, care should be taken with the interpretation of gender differences in bivariate analyses, because they might be confounded with age differences.
Gender differences were indeed found in the reported employment status of the respondents (see Figure 2): A higher rate of men respondents tended to be employed in industry or to be permanently employed in academia. On the other hand, a higher rate of women respondents tended to be students. Part of these differences might be due to the age differences reported earlier.
Note that here and in all following figures comparing the answers of men and women respondents, the graphs display conditional relative frequencies of answering in a certain category given the gender. Accordingly, the percentages within one gender add up to 100% (missing values are not considered).
116 respondents (25.49%) reported to be caregivers for children or adult dependents on a regular basis, with men attendees being more likely to be caregivers than women attendees. 28.97% of the men respondents reported to be caregivers, while only 21.83% of the women respondents did. Even though this difference was relatively small (Omega = 0.08), we decided to report it since it demonstrates the self-selection effect inherent in this study: Our sample contained only those members of the R community who were able to attend the useR! conference in Stanford (and were also willing to answer the questionnaire).
At first glance, the fact that men respondents were more likely than women to be caregivers may sound like the R community might have overcome traditional role models already. But the more probable interpretation, that is also supported by the free text answers, is that traditional role models do still make it easier for men to leave children at home with their partner (if applicable) than for women to do the same, resulting in a self-selection effect in which women with children are less likely to be able to attend – and as a result women who do attend are less likely to be caregivers.
This again reflects the general problem of this survey: We cannot draw any conclusion about reasons why women or other under-represented groups did not attend the conference, since we do not have any information on them.
Opinion on R
After having reported demographical information about the useR! sample, we now report our findings on the opinions reported by the respondents towards R and working with R. Since the demographic information reported in the last section suggested that women respondents came from a different professional background than men respondents, we were particularly interested in further gender differences in their opinion towards R.
A first question was whether the respondents would recommend R to friends or colleagues as a programming language to learn. This was agreed to by 418 (91.87%) respondents, while 21 (4.62%) respondents disagreed and 19 (4.18%) respondents did not answer. Asked about their number one argument for or against learning R by selecting one argument from a list, numerous responses were given. We summarized these reponses discarding all answers that were given less than 10 times – this left only arguments for using R. Figure 3 summarizes these most frequently given answers.
These data seem to suggest that the respondents would recommend learning R because of it being a good tool for statistical analyses, followed by its use as a tool for reproducible research. Men and women respondents did not differ in their willingness to recommend to learn R, or in their arguments for or against R (note, however, that for the arguments the cell frequencies may have been too small for a valid analysis).
Further questions of the survey asked the respondents to indicate to what extent they agreed with certain statements about R. Figure 4 summarizes the responses to the statements that
Writing R is fun.
Writing R is considered cool or interesting by my peers.
Writing R is difficult.
Writing R is a monotonous task.
Percentages are in reference to the number of all given answers to the respective question.
We did not find any gender differences in these questions. Attendees of the useR! conference – unsurprisingly – regard R mostly as something fun and interesting and not very monotonous or difficult.
It seems interesting to note that a large part (160 respondents, 35.16%) of the sample reported to use R not only in a professional or educational setting, but also in their free time. When compared with men respondents, women respondents tend to use R less often in their free time, and more often in an educational or professional setting (see Figure 5). Percentages again add up to 100% for each gender respectively, while missing values are not considered.
In a chi square test, we found a medium effect size (Cohen’s Omega = 0.25) for this gender difference.
Given that the higher share of R usage in their free time might also correspond to longer exposure times for men, which might again affect subjective or objective expertise as well as self-confidence for actively participating in activities like package writing, it might be worth investigating the factors behind this different usage behavior – be they motivational or due to structural differences like inequal distributions of household or childcare duties – in future research.
Experience with R
A significant part of the survey concerned the respondents’ experience of working with R. Generally, the respondents tended to be rather experienced R users. 369 respondents (81.10%) reported that they had already worked with R for 2 years or longer, with 338 respondents (74.29%) stating that they had already had programming experience before working with R.
Since women respondents tended to be younger, it could be expected that they would also have shorter experience in working with R. As can be seen in Figure 6, our analysis shows that this is indeed the case (Omega = 0.19). Again, percentages add up to 100 per gender.
There were also gender differences when the respondents were asked about their previous programming experience before using R (Omega = 0.18). While 82.25% of the men who answered this question reported to have previous programming experience, the corresponding percentage among women was 66.06%.
A related question concerned whether the respondents use only existing functions of R or whether they also write and publish their own functions. A majority of the respondents (389, 85.49%) reported to have written R functions for their own use. A smaller part of the sample (253, 55.60% of the respondents who answered) reported to have written their own R package or have contributed to an R package. 155 respondents (33.07% of the respondents who answered) reported to have published their own R packages on CRAN. These results are further illustrated in Figure 7. Percentages in this plot are in reference to the respective gender again. They do not add up to 100% for each group though because multiple answers per person were possible. Notably, men were more predominant when it comes to R package development.
So far, our results indicate that women respondents tended to be younger and have used R for a shorter amount of time than men respondents. Furthermore, we found that women respondents have contributed to R packages less often. From these bivariate analyses, however, we cannot assess whether the gender difference in package development is confounded with the usage length and programming experience, or whether there are gender differences beyond these effects, that may again be confounded with the younger age of the women participants. Therefore, in the next section we conduct a multiple logistic regression model to assess the partial effects of gender and the experience variables on package writing.
Modeling gender differences in contribution to R packages
As was outlined in the previous sections, women respondents were less likely to contribute to R packages, but also tended to have less programming experience and a shorter length of R usage. The observed gender differences in the contribution to R packages could be confounded with these variables.
In order to assess the partial effects of these and further variables, we employed a logistic regression model. It should be clearly stated that this was a fully exploratory analysis, as no clear hypotheses about the factors affecting R package contributions in the general R community were available to guide the design of the survey or this statistical anlysis. As will be outlined in the following, we explored the association between package contributions and those survey variables that seemed plausible.
In the logistic regression models, we predicted whether someone had contributed to an R package in any form (i.e., the categories “contribute to packages”, “have written package” and “written and released package” from Figure 7 were combined to form response category 1). Different models were compared, with contribution to R packages as outcome variable and gender, length of R usage and previous programming experience as a first set of candidate predictors. Length of R usage was coded as an ordered factor variable that consists of 5 response categories that correspond to less than 1 year, 1-2 years, 2-5 years, 5-10 years, and 10 or more years. Previous programming experience was coded as a dichotomous variable indicating whether someone had previous programming experience before using R or not.
The results were inconclusive as to whether gender differences remain after accounting for the differences in length of R usage. With respect to AIC, the model with length of R usage and gender as predictors was the best model, whereas the BIC and the Likelihood ratio test preferred the model with only length of R usage as a predictor. The variable previous programming experience had no additional effect and was excluded by all criteria. From this first logistic regression analysis, it looked like a large part – but maybe not all – of the gender differences in the contribution to R packages were caused by differences in the length of R usage (with women showing shorter usage lengths, as displayed in Figure 6).
To explore potential effects of additional survey variables, we also included employment status, purpose of R usage and community in the analysis. Employment status was coded as a factor variable with eight different categories like in Figure 2. Purpose of R usage is the factor variable from Figure 5, and community is a binary variable indicating whether someone stated to feel as part of the R community or not (with descriptive statistics for this variable presented in the next section).
The logistic regression model with gender, length of R usage, employment status and community had the lowest AIC and was also the best model according to the Likelihood ratio test. The BIC again preferred the more sparse model without gender. Purpose of R usage did not improve model fit for any of the criteria.
With respect to the interpretation of the effects of the additional variables included in this model, for employment status we find that people working permanently and non-permanently in academia contribute most to R packages, whereas those working in government/non-profit and industry are slightly less likely to contribute to R packages. Feeling as part of the R community goes along with contributing more to R packages. Of course, the direction of this association may also be the other way round, since people who have already contributed to R packages are more likely to feel as part of the R community.
Again, the analysis does not give a clear answer to the question whether any gender differences in package contributions remain beyond the differences already captured by the other predictor variables. Any remaining differences might depend on a variety of other individual and structural factors, such as differences in motivation or self-confidence, in access to information, or in networking and peer support for contributing to R packages. After this first exploratory study, it would be very interesting to further question women R users that are on the verge of becoming R developers what might be keeping them back – as well as asking men R users that did cross over and become developers what helped them take that step.
The respondents as part of the R community
A final set of questions in the survey concerned the role of the respondents as part of the R community. Asked whether they considered themselves to be part of the R community, 361 (79.34%) respondents agreed, whereas 69 (15.16%) respondents disagreed and 28 (6.15%) respondents did not answer. Men and women respondents did feel as part of the R community to a comparable extent.
The respondents further reported to use a variety of resources to support their work with R. The respective question in the survey provided a list of possible resources, and also allowed the respondents to enter additional resources that were not listed as free text. Among the listed resources, StackOverflow and the R mailing list were the most frequently used resources. We found no gender differences here. Reporting only given answers that were chosen more than 4 times for brevity, the results are displayed in Figure 8.
Further asked about their preferred medium for R community news, the respondents most frequently chose the website (24.18%) and the mailing lists (21.10%). Other selected options were blog (18.02%), Twitter (15.16%) and Facebook (4.62%). Given these preferences, communication across a range of media is necessary to be confident in reaching a large proportion of R users and developers. A sizeable proportion (7.91%) did not select any option, so does not have a preference or is not interested in R community news. Nonetheless, as conference participants, they may receive some news in person.
We further investigated whether men and women respondents differed with regard to their preference of individual resources. Therefore, we tested for every single response category of the previous question whether the relative frequency with which it was selected differed between men and women respondents. When correcting for multiple testing, none of them was statistically significant.
Among the 455 respondents of the survey, 163 (35.82%) respondents reported that they attend meetings of a R user group, whereas 268 (58.90%) responded that they did not, and 27 (5.93%) respondents did not reply. Among the 163 respondents attending R user group meetings, 134 (82.21%) respondents indicated that they attended a general user group, whereas 16 (9.82%) respondents reported that they visited a user group within a university. Other types of user groups were less often named.
The respondents who did not visit R user group meetings were further asked about their reasons for this decision. Again, the respondents could answer this question by either selecting statements from a list or by entering new statements. The arguments that were chosen most frequently were that the respondents were too busy or that no active user group was available.
Finally, we investigated whether men and women respondents differed with respect to their attendance of R user group meetings, and how these meetings could be made more attractive for new attendees. In a first step, we found gender differences with regard to the reported attendance of R user group meetings. Men respondents reported to attend R user group meetings more often than women respondents with a small to medium effect (Omega = 0.21).
In a second step, we investigated gender differences to the question what measure would make the respondent more likely to attend user group meetings. For the individual response options, the following differences were found (only those answers are listed that were chosen by at least 10 respondents, with n indicating the total number of respondents that chose the respective option):
New R user group near me (n = 122): no gender differences (Omega = 0.02)
New R user group near me aimed at my demographic (n = 31): More positive responses by women respondents with a small effect (Omega = 0.19)
Free local introductory R workshops (n = 61): no gender differences (Omega = 0.02)
Paid local advanced R workshops (n = 61): no gender differences (Omega = 0.08)
R workshop at conference in my domain (n = 73): no gender differences (Omega = 0.08)
R workshop aimed at my demographic (n = 20): More positive responses by women respondents with a small effect (Omega = 0.12)
Mentoring (e.g. first CRAN submission/useR! abstract submission/GitHub contribution) (n = 87): no gender differences (Omega = 0.09)
Online forum to discuss R-related issues (n = 62): More positive responses by women respondents with a small effect (Omega = 0.14)
Online support group for my demographic (n = 18): More positive responses by women respondents with a small effect (Omega = 0.15)
These answers indicate that most people who do not attend user group meetings do not have a user group near them or (as was suggested by free text answers) do not know that there are user groups. Other measures that could help to make people attend user group meetings would be R workshops at conferences in their domain, an online forum or local R workshops of different kinds. When it comes to gender differences, women might be more attracted by user groups explicitly aiming at women, but might also hint in the direction of women being less willing or less able or both to spend their free time with R, as we have seen in an earlier question.
Discussion
Our results draw a complex picture of the sample of attendees who agreed to respond to our survey. When looking at the sample as a whole, the respondents to this survey tended to have programming experience prior to working with R, and usually used R for 2 or more years. Most respondents further wrote their own R code, either to create functions for their own work or to contribute to R packages. Moreover, the average respondent had a positive opinion towards working with R. These results are not very surprising, given that this sample was collected among the attendants of a useR! conference. However, as no comparable results on the community of R users have been published so far, these findings are nevertheless valuable. Future studies could use the results reported here as a benchmark for the evaluation of long-term developments in the R community.
In accordance with the initial question of the taks force, we did find that women were under-represented at the useR! conference 2016 – but even more strikingly, that non-white users were even more severely under-represented.
When looking at gender differences in more detail, results from three areas of the questionnaire stand out, that may however be confounded as discussed above. First, women respondents tended to be younger and have shorter experience in R usage than men respondents. Second, women used R less in their free time and contributed less to R packages. Third, women agreed less to feeling part of the R community.
The results of the exploratory logistic regression analysis suggest that the survey questions captured some factors associated with gender differences in contributions to R packages, such as the length of R usage, employment in academia and a feeling of belonging to the R community, that were positively associated with contributing to R packages. Yet the results were not conclusive as to whether there may be further gender differences, for example due to personal or structural factors, that may discourage women from writing R packages and would be important to investigate.
An additional hierarchical cluster analysis of the same data found three different groups of people: The first group (around 38% of the sample) are experienced R users, who tend to be men, from academia and people with doctorate. They use R not only in a professional setting, but also for recreational purposes. The second group (around 57% of the sample) are intermediately experienced users who use R for less than 2 years and mainly apply existing packages. They tend to be female, and are either undergraduates or have a master degree. They are using R mainly in professional or educational settings. The last group (around 3% of the sample) are the least experienced users who are using R during their free time. Details on this analysis as well as further multivariate analyses of the data can be found at UseR! 2016 R community: a multivariate analysis and UseR! 2016 participants: a multivariate analysis.
An anonymised form of the survey data, which minimises disclosure risk by excluding some demographic variables and recoding others, is available on CRAN (forwards). This data set includes aggregated versions of all demographic variables used in the logistic regression analysis and the majority of demographic variables used to aid interpretation in the multivariate analysis. Apart from the suppression of a small number of data values and a few free text variables that contained sensitive/identifying information, the responses to the programming and community questions are provided as recorded.
As stated already in the beginning of this text, due to the limitations of the study design, the results from the conference sample might not generalize to other subgroups of the R community, in particular not to those individuals who could not attend the conference due to factors associated with their being part of an under-represented group. Further studies are necessary to try to obtain a better picture of the R community as a whole.
Moreover, several topics which could be of further interest for the support of R users had not yet been included in the survey in order to keep it as short as possible. Further potentially interesting questions include, besides the ones already mentioned above, for example, what programming languages R users had already used before they started working with R, whether being in contact with other users that contribute to R packages makes it more likely to contribute to R packages oneself, etc.
Still the answers from this questionnaire give some indication to measures that could be taken to advance the participation of women, for example that user groups and other means of exchange explicitly targeting women might make them more accessible. Even though this would be methodologically challenging both due to the unspecific nature of the R community and the confounding of any interventions with different developments over time, it would be important to evaluate the development of under-represented groups over time.
The fact that women useR! attendees are on average younger and have lower R usage lengths might stimulate the hope that, once they reach the age and experience of their men counterparts, any gender differences might disappear automatically. However, this may not be the case, since multivariate analyses (see for instance Josse and H. Turner (2016)) tend to suggest that after adjusting for the age, women are still less involved in the R community. We still expect that if the R community and the opportunities to contribute are not equally attractive for women, they will not have the motivation to develop the skills to become contributors.
Although our initial framing looked specifically at gender as an axis of exclusion, the results show that race, not gender, is the area where there is the greatest disparity, and it is vital that inclusivity efforts factor this in. While the survey results give some ideas for improving inclusion in general, specific efforts should be made to reach out to people from under-represented races, for example through diversity scholarships, invited conference contributions, or invitations to serve in community roles. Further information on who is under-represented, and for what reason, would support such efforts.
CRAN packages used
CRAN Task Views implied by cited packages
Note
This article is converted from a Legacy LaTeX article using the texor package. The pdf version is the official version. To report a problem with the html, refer to CONTRIBUTE on the R Journal homepage.