The focal article by Murphy (Reference Murphy2021) succinctly articulates the impressive advancements and improvements that have been made in statistical methods over the past few decades, with the application of these complex tools appearing in many top-tier industrial-organizational (I-O) psychology journals. No doubt, these methods are incredibly powerful and important for answering the complicated questions that are posed by researchers. However, Murphy makes the argument that, in favor of these more complicated methods, researchers have neglected the importance of the “simpler” descriptive statistics (i.e., “Table 1”) and their influence on the interpretation and application of findings to applied practice. We concur, and in the following pages, we focus on the role that statistical graduate training has in highlighting the importance of descriptive statistics for applied data science practitioners and data visualization.
Drawing from our experience designing and instructing an online applied master’s program in I-O psychology, we first highlight the types of jobs that I-O practitioners are engaging in today and the statistical skills needed for such jobs (spoiler: mostly descriptive statistics). Then, we focus on the rapidly growing science of data visualization and the statistical skills necessary for effective data visualization (another spoiler: mostly descriptive statistics). In doing so, we hope to reinforce Murphy’s (Reference Murphy2021) suggestion that scholars, especially I-O graduate students, should not neglect the importance of Table 1 even as the more advanced methodologies are taught and mastered.
Where we are: The current state of applied I-O statistical training
More than 100 I-O psychology graduate programs in the United States offer a master’s, doctorate, or both. I-O graduates are employed across four employment sectors (academia, consulting, industry, and government), with a majority going applied (Zelin et al., Reference Zelin, Lider and Doverspike2015). In recent years, there has also been growth in applied master’s programs, sometimes called Masters of Professional Studies (MPS). Employers continue to recognize the benefits of leveraging and incorporating I-O psychology knowledge and competencies to a variety of jobs. Consequently, graduate programs are preparing students for a wide range of jobs including traditional I-O positions (e.g., consulting, applied research) and those in human resources (HR) and management.
As graduate I-O programs produce students who are increasingly entering the applied workforce, it is important to evaluate the specific competencies that employers desire. In developing the curriculum plan for our own MPS I-O program, we reviewed knowledge, skills, abilities, and other characteristics (KSAOs) from numerous job advertisements that were recruiting master’s-level I-O practitioners. The most common KSAOs included “apply expertise in people research, quantitative analysis, data science, and data visualization to provide insights on talent management and leadership development initiatives,” and “track record in interpreting data creatively and delivering impactful insights (e.g., going beyond the “what” of a research inquiry into the “so what” and “now what” and “what haven’t we thought of yet?” In alignment with these job ads, the Society for Industrial and Organizational Psychology’s (SIOP) Professional Practice Committee conducted a career study and found that oral communication is a top competency for I-O psychologists across employment sectors (Zelin et al., Reference Zelin, Lider and Doverspike2015). This evidently growing interest in the communication of data and results leads to the question, does our statistical graduate training adequately prepare students for the applied workforce?
The Guidelines for Education & Training (SIOP, 2016) recommends covering both descriptive and inferential statistical methods, spanning both parametric and nonparametric statistical methods and covering both quantitative and qualitative research methodology. However, a cursory review of several syllabi from various I-O graduate programs reveals that there is much more emphasis on advanced statistical courses focused on, for example, structural equation modeling, multilevel analysis, and longitudinal analysis. This anecdotally reflects Murphy’s (Reference Murphy2021) findings that the focus of statistical analyses in I-O psychology, both in research and in education, tends to be toward these “advanced” methods. In contrast, relatively little time is spent on understanding, interpreting, and using descriptive statistics, reflected both in research (Murphy) and in education (our review of syllabi).
Importantly, this latter skill is a critical KSAO that is necessary to prepare the growing number of students entering applied practice. We surveyed about 100 master’s and PhD I-O graduates working in applied practice, asking them what software and analytical techniques they used. The most popular software was Tableau for data visualization (81 used it regularly in their jobs), followed by Excel (79), and then R(48). The most popular analyses used were correlation (62 answered “5” on a 5-point scale from not at all to a lot), data visualization (55), regression (49), and t tests (41). Remarkably, the majority of practitioners only used advanced techniques (e.g., structural equation modeling, longitudinal, multivariate) not at all or a little in their daily jobs. Clearly, the industry is calling for data visualization as a critical skill, potentially even more important than advanced analysis methods. In the following section, we dive deeper into what we argue is missing in I-O graduate training: using descriptive statistics for data visualization.
Where we need to go: The importance of data visualization
The “academic–practitioner” divide is well known, with dozens of articles, commentaries, and scholarly discussions in recent years bemoaning the lack of translation of the “science of I-O psychology” to actual workplace settings and calling for movements to change this (e.g., Aguinis et al., Reference Aguinis, Ramani, Campbell, Bernal-Turnes, Drewry and Edgerton2017; Anderson et al., Reference Anderson, Herriot and Hodgkinson2001; Latham, Reference Latham2019). The most recent articulation of SIOP’s strategic goal acknowledges this, emphasizing the importance of “translat[ing] scientific knowledge to promote individual and organizational health and effectiveness” (SIOP, 2021). Put simply, the complex and advanced statistical models that are used in most top I-O journals are valuable, but unless the results are effectively summarized and communicated succinctly to an applied (and largely nonstatistical) audience, such research is likely to have little influence. For SIOP to grow as an organization and our field to build influence in the workplace, effective and concise communication of data-driven results must be a preeminent skill for all I-O psychologists. In most situations with an applied audience, communication of results does not take the form of blocks of dense text or tables but rather of high-quality visual graphs and charts, that is, data visualization.
Data visualization is built almost entirely from descriptive statistics. Granted, this is an oversimplification, as there are certainly ways to incorporate inferential statistical models into visuals. But even then, these are often not much more than regression lines or error bars to depict standard errors (see Healy, Reference Healy2018). It would be tough to visualize a latent state–trait model with multiple indicators and autoregression across four measurement occasions in a simple, succinct, and “beautiful” way. Instead, the best visuals, in terms of effectively communicating and persuading the audience, focus on not much more than means (e.g., bar charts), variances (e.g., scatterplots), and groupings (e.g., social networks). Just as Murphy (Reference Murphy2021) demonstrated in the focal article, descriptive statistics can and often should be used to communicate the results of more advanced statistical methods. In a modern world driven by highly engaging, interactive, and easy-to-understand charts and graphs, this becomes even more important.
Data visualization also exemplifies the rule of “less is more” (Berinato, Reference Berinato2016; Knaflic, Reference Knaflic2015). Such a sentiment often goes in direct contrast to the approach taken in most advanced statistical methods. Take for example the classic problem of endogeneity in most cross-sectional and panel studies. The solution? Researchers must add the appropriate control variables, and they must use a variety of statistical methods to correct for and control endogeneity bias (e.g., Antonakis et al., Reference Antonakis, Bastardoz and Rönkkö2021; Zaefarian et al., Reference Zaefarian, Kadile, Henneberg and Leischnig2017). To be clear, this is not to say that endogeneity is not important; it is, and these recommended corrections are highly valuable. However, we bring this up as an example of how data visualization tries to keep things simple, but the advanced methods found in research studies require the use of more complicated techniques to get the right answer. This tension is not often recognized or valued, which is why we argue for more attention to be paid to when and why we should try to keep things simple.
To clarify, we are not saying that data visualization is easy; it is incredibly difficult to do well. For example, Sawicki (Reference Sawicki2020) articulates a 10-point “grammar of data visualization” that should generally be followed when creating data visuals. Moreover, bad visualizations can be found everywhere. At best, they are not effective in communicating what they are trying to communicate. At worst, they are misleading and can perpetuate misinformation among the public; the recent trend of misleading data visuals related to COVID-19 is a prime example (Leybzon, Reference Leybzon2020). In some ways, poorly conducted data visualization (and by extension, poorly used descriptive statistics) can have even more widespread negative effects due to public accessibility and audience size, than a poorly conducted structural equation modeling. Therefore, if we as I-O psychologists hope to pursue SIOP’s goals of bridging the academic–practitioner gap, much more focus needs to be put on the importance of descriptive statistics and data visualization.
Conclusion
We conclude by offering a few primary action-oriented recommendations for readers to consider. First, educators should make space for descriptive statistics and data visualization in their graduate statistics courses. As a start, we should review syllabi and incorporate or expand these topics to provide adequate training in these areas. Second, students must be open to learning both descriptive statistics and advanced statistics, with the goal of understanding when and why each should be used. The ability to conduct advanced and complex statistics is incredibly important to answer complicated research questions, but unless students are able to subsequently communicate their results and their data to a nonstatistical audience, there will be limited ability for the research to influence actual business practice. Moreover, our brief survey of I-O graduates suggests that data visualization skills may be even more valuable for getting a job than are advanced analytical methods. Finally, as a field, we must try and inform the public on how to interpret data visualizations. Unfortunately, misleading data visualizations (e.g., base rate fallacy, omitted variables, scaling) are rampant in popular culture. As scientists, we should be striving to improve public understanding and awareness of how to use data visualization and when it can go wrong. One way this can be accomplished is to continue to broaden inclusivity in our field and cater to students from all types of educational and career backgrounds (see Zhou & Ahmad, Reference Zhou and Ahmad2020) so that they can leverage I-O knowledge in their respective workplaces. In short, Murphy (Reference Murphy2021) is correct to point out the underappreciated nature of descriptive statistics and their importance to scientific research. Hopefully, we have built on his argument to illustrate one direct and powerful benefit of good descriptive statistics: the communication of data in the form of data visualizations.
You have
Access
The focal article by Murphy (Reference Murphy2021) succinctly articulates the impressive advancements and improvements that have been made in statistical methods over the past few decades, with the application of these complex tools appearing in many top-tier industrial-organizational (I-O) psychology journals. No doubt, these methods are incredibly powerful and important for answering the complicated questions that are posed by researchers. However, Murphy makes the argument that, in favor of these more complicated methods, researchers have neglected the importance of the “simpler” descriptive statistics (i.e., “Table 1”) and their influence on the interpretation and application of findings to applied practice. We concur, and in the following pages, we focus on the role that statistical graduate training has in highlighting the importance of descriptive statistics for applied data science practitioners and data visualization.
Drawing from our experience designing and instructing an online applied master’s program in I-O psychology, we first highlight the types of jobs that I-O practitioners are engaging in today and the statistical skills needed for such jobs (spoiler: mostly descriptive statistics). Then, we focus on the rapidly growing science of data visualization and the statistical skills necessary for effective data visualization (another spoiler: mostly descriptive statistics). In doing so, we hope to reinforce Murphy’s (Reference Murphy2021) suggestion that scholars, especially I-O graduate students, should not neglect the importance of Table 1 even as the more advanced methodologies are taught and mastered.
Where we are: The current state of applied I-O statistical training
More than 100 I-O psychology graduate programs in the United States offer a master’s, doctorate, or both. I-O graduates are employed across four employment sectors (academia, consulting, industry, and government), with a majority going applied (Zelin et al., Reference Zelin, Lider and Doverspike2015). In recent years, there has also been growth in applied master’s programs, sometimes called Masters of Professional Studies (MPS). Employers continue to recognize the benefits of leveraging and incorporating I-O psychology knowledge and competencies to a variety of jobs. Consequently, graduate programs are preparing students for a wide range of jobs including traditional I-O positions (e.g., consulting, applied research) and those in human resources (HR) and management.
As graduate I-O programs produce students who are increasingly entering the applied workforce, it is important to evaluate the specific competencies that employers desire. In developing the curriculum plan for our own MPS I-O program, we reviewed knowledge, skills, abilities, and other characteristics (KSAOs) from numerous job advertisements that were recruiting master’s-level I-O practitioners. The most common KSAOs included “apply expertise in people research, quantitative analysis, data science, and data visualization to provide insights on talent management and leadership development initiatives,” and “track record in interpreting data creatively and delivering impactful insights (e.g., going beyond the “what” of a research inquiry into the “so what” and “now what” and “what haven’t we thought of yet?” In alignment with these job ads, the Society for Industrial and Organizational Psychology’s (SIOP) Professional Practice Committee conducted a career study and found that oral communication is a top competency for I-O psychologists across employment sectors (Zelin et al., Reference Zelin, Lider and Doverspike2015). This evidently growing interest in the communication of data and results leads to the question, does our statistical graduate training adequately prepare students for the applied workforce?
The Guidelines for Education & Training (SIOP, 2016) recommends covering both descriptive and inferential statistical methods, spanning both parametric and nonparametric statistical methods and covering both quantitative and qualitative research methodology. However, a cursory review of several syllabi from various I-O graduate programs reveals that there is much more emphasis on advanced statistical courses focused on, for example, structural equation modeling, multilevel analysis, and longitudinal analysis. This anecdotally reflects Murphy’s (Reference Murphy2021) findings that the focus of statistical analyses in I-O psychology, both in research and in education, tends to be toward these “advanced” methods. In contrast, relatively little time is spent on understanding, interpreting, and using descriptive statistics, reflected both in research (Murphy) and in education (our review of syllabi).
Importantly, this latter skill is a critical KSAO that is necessary to prepare the growing number of students entering applied practice. We surveyed about 100 master’s and PhD I-O graduates working in applied practice, asking them what software and analytical techniques they used. The most popular software was Tableau for data visualization (81 used it regularly in their jobs), followed by Excel (79), and then R(48). The most popular analyses used were correlation (62 answered “5” on a 5-point scale from not at all to a lot), data visualization (55), regression (49), and t tests (41). Remarkably, the majority of practitioners only used advanced techniques (e.g., structural equation modeling, longitudinal, multivariate) not at all or a little in their daily jobs. Clearly, the industry is calling for data visualization as a critical skill, potentially even more important than advanced analysis methods. In the following section, we dive deeper into what we argue is missing in I-O graduate training: using descriptive statistics for data visualization.
Where we need to go: The importance of data visualization
The “academic–practitioner” divide is well known, with dozens of articles, commentaries, and scholarly discussions in recent years bemoaning the lack of translation of the “science of I-O psychology” to actual workplace settings and calling for movements to change this (e.g., Aguinis et al., Reference Aguinis, Ramani, Campbell, Bernal-Turnes, Drewry and Edgerton2017; Anderson et al., Reference Anderson, Herriot and Hodgkinson2001; Latham, Reference Latham2019). The most recent articulation of SIOP’s strategic goal acknowledges this, emphasizing the importance of “translat[ing] scientific knowledge to promote individual and organizational health and effectiveness” (SIOP, 2021). Put simply, the complex and advanced statistical models that are used in most top I-O journals are valuable, but unless the results are effectively summarized and communicated succinctly to an applied (and largely nonstatistical) audience, such research is likely to have little influence. For SIOP to grow as an organization and our field to build influence in the workplace, effective and concise communication of data-driven results must be a preeminent skill for all I-O psychologists. In most situations with an applied audience, communication of results does not take the form of blocks of dense text or tables but rather of high-quality visual graphs and charts, that is, data visualization.
Data visualization is built almost entirely from descriptive statistics. Granted, this is an oversimplification, as there are certainly ways to incorporate inferential statistical models into visuals. But even then, these are often not much more than regression lines or error bars to depict standard errors (see Healy, Reference Healy2018). It would be tough to visualize a latent state–trait model with multiple indicators and autoregression across four measurement occasions in a simple, succinct, and “beautiful” way. Instead, the best visuals, in terms of effectively communicating and persuading the audience, focus on not much more than means (e.g., bar charts), variances (e.g., scatterplots), and groupings (e.g., social networks). Just as Murphy (Reference Murphy2021) demonstrated in the focal article, descriptive statistics can and often should be used to communicate the results of more advanced statistical methods. In a modern world driven by highly engaging, interactive, and easy-to-understand charts and graphs, this becomes even more important.
Data visualization also exemplifies the rule of “less is more” (Berinato, Reference Berinato2016; Knaflic, Reference Knaflic2015). Such a sentiment often goes in direct contrast to the approach taken in most advanced statistical methods. Take for example the classic problem of endogeneity in most cross-sectional and panel studies. The solution? Researchers must add the appropriate control variables, and they must use a variety of statistical methods to correct for and control endogeneity bias (e.g., Antonakis et al., Reference Antonakis, Bastardoz and Rönkkö2021; Zaefarian et al., Reference Zaefarian, Kadile, Henneberg and Leischnig2017). To be clear, this is not to say that endogeneity is not important; it is, and these recommended corrections are highly valuable. However, we bring this up as an example of how data visualization tries to keep things simple, but the advanced methods found in research studies require the use of more complicated techniques to get the right answer. This tension is not often recognized or valued, which is why we argue for more attention to be paid to when and why we should try to keep things simple.
To clarify, we are not saying that data visualization is easy; it is incredibly difficult to do well. For example, Sawicki (Reference Sawicki2020) articulates a 10-point “grammar of data visualization” that should generally be followed when creating data visuals. Moreover, bad visualizations can be found everywhere. At best, they are not effective in communicating what they are trying to communicate. At worst, they are misleading and can perpetuate misinformation among the public; the recent trend of misleading data visuals related to COVID-19 is a prime example (Leybzon, Reference Leybzon2020). In some ways, poorly conducted data visualization (and by extension, poorly used descriptive statistics) can have even more widespread negative effects due to public accessibility and audience size, than a poorly conducted structural equation modeling. Therefore, if we as I-O psychologists hope to pursue SIOP’s goals of bridging the academic–practitioner gap, much more focus needs to be put on the importance of descriptive statistics and data visualization.
Conclusion
We conclude by offering a few primary action-oriented recommendations for readers to consider. First, educators should make space for descriptive statistics and data visualization in their graduate statistics courses. As a start, we should review syllabi and incorporate or expand these topics to provide adequate training in these areas. Second, students must be open to learning both descriptive statistics and advanced statistics, with the goal of understanding when and why each should be used. The ability to conduct advanced and complex statistics is incredibly important to answer complicated research questions, but unless students are able to subsequently communicate their results and their data to a nonstatistical audience, there will be limited ability for the research to influence actual business practice. Moreover, our brief survey of I-O graduates suggests that data visualization skills may be even more valuable for getting a job than are advanced analytical methods. Finally, as a field, we must try and inform the public on how to interpret data visualizations. Unfortunately, misleading data visualizations (e.g., base rate fallacy, omitted variables, scaling) are rampant in popular culture. As scientists, we should be striving to improve public understanding and awareness of how to use data visualization and when it can go wrong. One way this can be accomplished is to continue to broaden inclusivity in our field and cater to students from all types of educational and career backgrounds (see Zhou & Ahmad, Reference Zhou and Ahmad2020) so that they can leverage I-O knowledge in their respective workplaces. In short, Murphy (Reference Murphy2021) is correct to point out the underappreciated nature of descriptive statistics and their importance to scientific research. Hopefully, we have built on his argument to illustrate one direct and powerful benefit of good descriptive statistics: the communication of data in the form of data visualizations.