Project-based learning is a model that organizes learning around projects, which can be defined as “complex tasks, based on challenging questions or problems, that involve students in design, problem-solving, decision making, or investigative activities; give students the opportunity to work relatively autonomously over extended periods of time; and culminate in realistic products or presentations”  [Thomas 2000]. Project-based learning builds on the constructivist and constructionist ideas that learning is something that is actively built by learners rather than transmitted linearly from teacher to student - what [Freire 2000] called the “banking model” of education. In project-based learning, projects help students encounter the central problems and questions of a discipline, involve students in a constructive investigation, and are student-driven and realistic [Thomas 2000].

What does a project-based learning approach look like for data analysis and information visualization? First, project-based learning is inquiry driven so the first step of project-based pedagogy with data is helping students to understand what kinds of questions can be answered (or at least partially answered) with data and whether the data one needs has been collected in the first place. Determining which institutions may have collected such data (and why) involves creative political thinking as well as savvy information retrieval skills. Once learners understand what kinds of questions may be answered with data, a project-based learning approach grounded in Freire's model of popular education guides them to ask questions that have meaning for themselves and their communities. While many tutorials teach with datasets such as car performance metrics or server logs, these are data for which most new learners do not have a rich context of lived experience. If learners can bring their own context and lived experience to the table, this will enable them to ask better questions, draw a richer picture of the data's limitations and perhaps even challenge the data collection practices. Rather than focusing the majority of attention on the “right” and “wrong” visual tactics to create polished data visualizations, taking a project-based learning approach focuses more attention on the exploring and analysis stages of the process. This involves helping learners ask better questions about the data, spot missing or bad data, and understand the limitations of what they may and may not be able to conclude from any given data set. Finally, a project-based learning approach to information visualization scaffolds learners through the whole data analysis and visualization pipeline rather than using different datasets and subjects at each stage. In this way, the process becomes iterative, rather than idealized and procedural, and the learners may return to prior stages as they encounter hurdles or identify better questions to ask.

Hands-on learning emphasizes the experiential nature of learning over the transmission (or “banking”) model of education where an instructor talks to a room of students who sit quietly and listen. Doing and making with the active participation of one's hands and body is central to this approach, articulated well by Progressive Era educators [Dewey 1987] [Montessori 2013]. For Dewey, learning was not about the passive reception of knowledge which exists a priori out in the world but about “active learning” that engages the learner's hands and five senses in the construction of knowledge. Building off of Vygotsky's “thinking out loud” approach described above, Papert shifts the focus from verbalizing thoughts to creating “objects to think with”  [Papert 1980]. In particular, Papert showed how learning about computational thinking - the seemingly abstract digital processes of personal computers and robots - could be grounded in making concrete representations and tangible objects. Externalizing the learning process through embodied creation aids in learners' constructing their own knowledge about the world [Ackermann 2004].

Creating situations for hands-on learning may seem to be a challenge for information design, whose outputs one most often sees in two dimensions on screens.  Increasingly, however, educators and artists are experimenting with ways to “visceralize”, “physicalize” [Huron et al. 2014] [Willett and Huron 2016] [Perovich] [DataPhys wiki 2016] [Bhargava 2016] and otherwise make data tangible, embodied and felt. While creative information displays such as these may not be the end goal of a particular pedagogical process, showing these examples can expand learners' thinking about what constitutes information design and what possible outputs of a data analysis process may look like. There are also a number of ways that hands-on learning can be incorporated into the process (rather than the product) of data analysis. Warm-up exercises, introduced at the beginning of the learning process, might include making data sculptures from a simple data set or doing a group critique of a large printed infographic. Hands-on learning at this stage of the process helps to build confidence for non-technical learners and create a low barrier to entry for technical subject matter, essential for later learning [Zimmerman 2000]. In later parts of the data analysis process, collaborative sketching and low-fidelity paper prototyping can be extremely important methods to focus learners on the story they want to tell with data rather than on the mechanics of operating a particular software program or writing working code [O’Hear 2016]. The sculptures, sketches, and paper prototypes constitute “objects to think with” that help learners simultaneously see and reflect on their data analysis process, as well as test it out with others, in order to take the next step.

Constructivist approaches to education believe that learning is grounded in experience [Dewey 1938] and human experience, according to psychologists such as Vygotsky, is rooted in a social context with sustained relationships. This points towards incorporating peer learning into educational situations so that learners may work together on meaningful problem-solving activities, negotiating their learning through language. Peer learning can be defined as the acquisition of knowledge through active helping and supporting among status equals or matched companions [Topping 2005]. Because this type of learning requires peers to verbalize concepts, a form of simultaneously teaching oneself and another, this helps embody and crystallize thought into language [Vygotsky 1980]. Peer learning connects back to both project-based learning and hands-on learning because the basic premise is that people learn better through collaborative, meaningful problem-solving activities than through listening to an expert or working through exercises alone.

Working with data involves piecing together a patchwork quilt of tools.  The sheer number of tools to choose from has made it impossible to introduce even the most important of them all to students. The co-authors of this paper have, in separate work, logged more than 500 free or freemium tools designed for non-technical users to collect and analyze data or create data visualizations. More critically, there isn't well-established criteria for helping students understand how and when to piece together this quilt.  To address this challenge we look to peer learning approaches.

In our undergraduate and graduate courses at MIT and Emerson College, respectively, we have students write reviews of tools for each other as a way to distribute expertise in the classroom.  We created a website called NetStories[3] with reviews of online tools for working with data, where the reviews are crowdsourced as assignments to the participants in our classes.  In parallel, we maintain an informal list of any tool for working with data that may or may not have yet been reviewed. In class, we have learners pick the top 3 tools they are interested to learn, and then assign each person one tool to learn and write a review about. Often these are tools that we, as the instructors, have never used. The audience for each review is other learners, including other teachers.  We have each student present the tool they reviewed for 5 minutes in class, allowing for broader awareness across the class. For the duration of the course, students are then considered the class “expert” in that tool and the instructor refers other students to them if they need to use it in their final projects.

Another example involves having students teach tools to each other to establish criteria for assessing when and how to use them in appropriate ways.  As an example, Bhargava uses this technique to introduce his undergraduate and graduate students to useful tools for mapping.  He assigned half the class a tutorial that introduced them to CARTO, and assigned the other half a tutorial for mapping in Tableau Public.[4]  Each group was assigned the task of analyzing the same dataset geographically using the tool specified.  In the subsequent class the groups were paired across tools; each pair comprised of one person that learned CARTO and one that learned Tableau.  They were then given 15 minutes each to introduce the other to what they made, how they had done it, and why that tool might be useful.  The activity concluded with reflections about the comparative advantages, affordances, and limitations of each tool.  

Whereas the Just-in-Time experts examples, above, were for undergraduate and graduate college students, we typically run the WordCounter Sketch a Story activity[5] in more ad-hoc, one-time workshops for digital humanists, journalists, nonprofit and government workers. The activity introduces learners to basic concepts of quantitative text analysis. Learners often approach text as qualitative data, unaware of the potential for quantitative analysis of text that computational approaches provide. Counting words is a fundamental strategy of most computational text processing, including more complex techniques such as machine learning and topic modeling.

To introduce the tool and the activity we discuss motivations for quantitative text analysis and try to give domain-specific examples. In the humanities, for example, scholars are looking to computational methods for alternate kinds of insights into large bodies of text, which is often referred to as “distant reading”  [Moretti 2013]. In journalism, news organizations are increasingly under pressure to make sense of large document dumps like the Panama Papers or the WikiLeaks diplomatic cables. This is text at a scale which would be impossible for a single person to read and must be treated quantitatively for the sheer reason of scale. And in government, cities are increasingly gathering citizen ideas and comments in qualitative form and need a way to start to draw out patterns. For the WordCounter activity, we introduce the idea of analyzing music lyrics quantitatively and show several inspirational examples of prior work that analyzes music lyrics, including the Rap Research Lab and “Spotimap”.

Once we have explained the motivations, we demo the tool for just five minutes. WordCounter is a simple web application in which users can paste a block of text or upload a plain text file. The back-end server uses open source libraries to count the words, bigrams (two-word phrases) and trigrams (three-word phrases) in the uploaded text. After submitting their corpus, the user sees tables listing the most frequently used words, bigrams, and trigrams along with an option to download a CSV file (Figure 1). WordCounter can operate on sample data we have included, text the learner pastes in, a file the learner uploads, or text from a website URL the learner pastes in.  It includes two advanced options - toggles for case sensitivity and the removal of common words (“stopwords”).

Working in groups of three, learners have fifteen minutes to decide which lyrics to run through WordCounter in order to find a “story” to tell with the results. The sample data for the tool includes lyrical corpora for a variety of popular musical artists in English, Spanish and Portuguese (including Beyoncé, Elvis Presley, Maná, Legião Urbana, etc.). Each team uses crayons and large pads of paper to create a sketch of a visual presentation of their story to share with peers for feedback. The visual outputs resemble those seen in [Wolf 2015]. In our courses, the class then spends 10-15 minutes discussing the stories and using them as a jumping off point for further discussion of text mining and text analysis concepts.

While the activity is short, the focus on finding a “story” is a prompt for learners to use the tool, interpret and filter the results using prior knowledge of musical artists, and translate their organizing concept into visual language. The activity constitutes a simple model for a project-based learning approach to data visualization in a low-stakes environment with fun subject matter. Peer groups of three learners work together to negotiate each stage of the process, including selecting relevant data, suggesting focus ideas, and developing visual representations with color, hand-drawn icons and pictures. Translating the “story” into visual language is a key part of the hands-on learning aspect of this activity. Learners often start to make key visual design decisions, such as using color keys to represent data about individual artists, using familiar icons to represent concepts such as “love”, and using size and shape to denote differential quantities. How novices use visual mappings and narratives in these types of data-driven sketches is an active area of study in the information visualization field [Walny, Huron, and Carpendale 2015]. In the post-activity shareback, facilitators work to highlight these emerging design decisions and point out how they lead to more intentional and systematic strategies for visual information design. Simple, short activities can embody all three principles of project-based learning, hands-on learning and peer learning.

Our second example is related to constructing strong narratives in explanatory visualizations.  Our hands-on activity guides students through telling their data-driven story in storybook form, to flesh out the narrative structure and the key “plot points”.[6]

While introducing the idea of telling stories with visualizations, educators often break them into two buckets - exploratory or explanatory [Iliinksy and Steele 2011].  Exploratory visualizations provide some interface, visual or interactive, to manipulate the data being shown.  They allow the user to explore the data and find a story of their own.  Explanatory visualizations use data to tell a strong and clear story through the use of visual mapping and graphical symbols. They rely on narrative structure to convey the story of the data to the reader; built through time, physical space, or both.

Explanatory visualizations that strive to tell a story need strong narrative structure.  We find students often critique this aspect of examples we discuss.  These critiques open lively discussions, but don't provide a space for students to practice the art of stitching together a narrative.  To allow this, we offer the storybook activity as a playground where students can explore their story's narrative, without worry about the details of the visual mappings or symbolical glossary they will build on.

The activity works best once a group of learners has started to narrow in on the main story they hope to tell. We often use this in classroom settings after a group of students has picked a dataset they will be working with and analyzed it to find a particular narrative that they find engaging. Learners are asked to bring their current story in the form of a template that says “the data say _______, we want to tell that story because _______”.  We then give each group a large piece of paper and a pair of scissors.  We then lead them through a classic technique of folding that paper, with one small cut, into a small book with 3 two-page spreads, a front cover, and a back cover. Once the paper is in book form, each group is instructed to write “once upon a time…” on the cover and “the end” on the back.  The rest of the pages are available for them to sketch out their story in a form similar to a children's storybook, using large graphics and story text.  We offer crayons or thick markers as the implements to write with, to both suggest the playful approach and keep the visuals they choose to include at low fidelity.

This activity lets participants focus in on the backbone of their story to make sure it is strong and clear.  It doesn't argue against nuance and detailed story-telling, but rather provides the skeleton from which to hang the details from.  If their story can't be reduced to this simple form, then their narrative needs further iteration.  As with the WordCounter sketching exercise, this activity gets learners off the screen and back into a space they might be more familiar with - namely crayons and paper. The “story-time” we host at the end, where each group reads their story to the rest of the class, offers a context for the kind of social construction of knowledge that Vygotsky discusses.  The storybook artifact itself is a classic “thing to think with” from a Constructionist point of view; an embodiment of the learning process that can be passed around, discussed, and iterated upon until a final polished data visualization is created.  In addition, it brings a simple element of fun into learning settings in a way that makes sense.  This isn't fun for fun's sake, but rather fun that is had while doing the precise activity that is being learned.

To truly rethink approaches to teaching data visualization, one must take a step back and reconsider what “visualization” implies, who it includes, and who it excludes.  The term itself has become loaded - evoking mental pictures of strong, well designed graphics that depict some data-driven image of truth. This mythology of the all-knowing “data visualization”, occasionally propagated by designers themselves [McCandless 2010], can be broken down by renaming it as an information presentation.  This recasting is both non-intimidating and welcoming.  Most of our audiences are intimidated by data, but feel comfortable with information.  Few work on visualization, but most give presentations.  We choose our words carefully in workshop and classroom settings to be more inclusive.

Similarly, the materiality and media used to create data visualizations can be questioned to move beyond technological fetishism.  Technological expertise need not be a barrier to creating data visualizations, if one calls them simple visual presentations of information.  This opens a door to using paper printouts of data tables and pie charts with a community group to collaboratively design a mural, for example. Working together with community groups, Bhargava has painted 10 such murals around the world [Bhargava 2016].  Similarly, one can look to classic low-tech media from novel sources such as kindergarten to offer a data “sculpture” activity that allows learners to sketch their data-driven stories with pipe-cleaners, googly eyes, and other such craft materials.  These types of creative invitations align more closely with the pedagogy we described earlier. They specifically follow a constructionist approach to “building out loud”, and borrow Freire's approach to starting with materials that are familiar to and owned by the learner. While data literacy does not end with pompons and googly eyes, these can function as a productive starting point to get more people in the door (and to make the door a little larger).