DS 100: Data Speaks Louder than Words

Learning Outcome #1: Students will identify and apply major concepts used in the social sciences to explain individual and collective human behavior including, for example, the workings of social groups, institutions, networks, and the role of the individual in them.

We will employ hands-on analysis of real-world datasets, including curated economic data, data scraped from digital collections, social networks, and more. In this context, the course will expose you to social and legal issues surrounding data analysis, including issues of privacy and data ownership, and will highlight the many ways in which data could be used (or misused).

In this course we will be looking at data from multiple vantage points. For example, by looking at data characterizing COVID-19 infections, hospitalizations, deaths, vaccinations, we will be able to differentiate between phenomena (e.g., correlations) identified at the macro scale (federal and state) versus those identified at the micro scale (cities and communities) and draw conclusions or make statements supported with evidence from data (e.g., impact of socioeconomic background).

We will encourage you to apply what you learn on societally-relevant case studies of your choice (e.g., case studies similar to those presented in https://www.callingbullshit.org/) by applying the tools and techniques covered in class to analyze data sets in order to support or debunk hypotheses.

Learning Outcome #1: Students will be able to craft and deliver responsible, considered, and well-structured arguments using media and modes of expression appropriate to the situation.

We will use real data to understand relationships and patterns while also introducing critical concepts and skills in computer programming and statistical inference. In order to build your arguments, you will use multimodal data analysis and visualization in ways that are appropriate to the task at hand. This will include:

• Generation and interpretation of scatter plots, histograms, bar charts, and box plots
• Making predictions using simple regression
• Characterizing data quality and communicating associated uncertainties
• Establishing confidence in reproducible predictions
• Reaching defensible conclusions about real-world questions

These skills will be taught and evaluated in both problem sets and laboratory exercises, as well as in exams (e.g., asking you to critique statements made in light of a specific visualization, or asking you to select from a set of suggested visualizations the one that would either support or deconstruct an argument).

Learning Outcome #2: Students will be able to demonstrate an understanding of the capabilities of various communication technologies and be able to use these technologies ethically and effectively.

As part of DS-100, we will introduce you to multiple forms of data visualization and presentation, including histograms, scatterplots, word clouds, heat maps, infographics, etc. Each one of these forms of communication can be particularly effective (or even misleading) in certain settings. For example, the choice of different scales (e.g., absolute vs relative change) on an axis could over or under-emphasize particular conclusions from the data.

Given the multitude of sources from which the data is collected, you will be exposed to proper ways of handling the data. For example, to preserve the privacy of individuals or communities in a large data set, and be introduced to the use of randomization techniques (blurring the data). As another example, to deal with the scale of data it may be necessary to only consider/analyze a subset of all observations. In that context, we will introduce you to various ways in which selection bias may influence conclusions you may be able to reach with implications on reproducibility.

Learning Outcome #3: Students will be able to demonstrate an understanding of the fundamentals of visual communication, such as principles governing design, time-based and interactive media, and the audio-visual representation of qualitative and quantitative data.

We will teach you how to use Python to organize and manipulate data in tables, and to visualize data effectively. Furthermore, you will be able to use computation to help your data tell a story through fundamental principles and methods of data visualization. The data used throughout this course will include longitudinal data (time series over long-time scales), geospatial data (data overlaid on apps), or both. These modalities will offer you different ways to interact with the data. For example, with time series data, you will be able to develop animations to show how phenomena or inferences may evolve over time. As another example, with geospatial data sets, you will be able to develop animations or heat maps that may project different messages/narratives based on the level of aggregation (e.g., achieved by zooming in and out).

In all of the above learning outcomes, we note that some of your work products will be in the form of multimedia reports, in which data visualization is coupled with narratives or video clips. For example, in a report on deforestation due to climate change, you may add audio or video clips to demonstrate change over time. You may also include your own narration to supplement and/or add texture to the graphs, heatmaps, etc.

We will teach you critical concepts and skills in computer programming and statistical inference, in conjunction with hands-on analysis of real-world datasets, including economic data, document collections, geographical data, and social networks. In discussion sections (worth 10% of your grade), you will work in small teams, working under the supervision of the teaching fellow to frame a question or test a hypothesis using a set of potential data sources. The key phases of that process are the exploration and identification of relevant data sets, the formulation and reformulation of the questions based on the identified data, the development of a set of data processing/analytics steps leading to an answer, and the interpretation and/or validation of the answer. To a large extent, going through these phases mirrors the six steps of the data science research process shown below.

This course emphasizes learning through doing: we will work on large real-world data sets through interactive assignments to apply the skills you learn. Throughout, the underlying thread is that data science is a way of thinking, not just an assortment of methods. We will focus on honing your interpretation and communication skills, which are essential skills for data scientists. Moreover, we will explore the proper way to complete the research process, eliminating bias (including your own).

As such, you will be trained in the processes underlying robust inference from real-world data from a variety of domains, run experiments and test your hypotheses, know the correct statistical tools to use depending on the task, quantify and understand uncertainty in data, and understand and utilize computation and simulation in data science. You will learn to articulate the benefits and limits of computing technology for analyzing data and answering questions.

Learning Outcome #1: Students will be able to search for, select, and use a range of publicly available and discipline-specific information sources ethically and strategically to address research questions.

In addition to publicly-available datasets, you will also be encouraged to proactively identify data sets available online which may be relevant to the question at hand – e.g., to refine findings or to compare findings across populations – possibly leading you to rephrase the question. In that respect, steps #2 and #4 of the data science process will be exemplified through cases covered in lectures, problems that you will have to work on as part of homework assignments, and student group activities pursued in discussion sections. A set of recommended sources can be found on the course website.

As part of your framing of questions in step #1 and your selection and exploration of data sets in steps #2 and #4, you will be introduced to and encouraged to think through the social issues surrounding data processing and analysis in steps #3 and #5, such as transparency, privacy, and inclusive design. As with the identification and exploration of data and information sources, the consideration of ethical dimensions will be covered in the use cases used in lectures and discussion sections. You will be expected to reflect on them and to apply them in write-ups of your reports in which you communicate your results in step #6.

Learning Outcome #2: Students will demonstrate understanding of the overall research process and its component parts, and be able to formulate good research questions or hypotheses, gather and analyze information, and critique, interpret, and communicate findings.

You will learn the underpinnings of the overall data science research process by repeatedly covering its key phases (problem framing; data identification and exploration; data processing and analysis; and interpretation and communication of the results). To that end, the course uses a spiral approach, in which you will iterate over these phases (steps #1 to #6), with each iteration being more involved by virtue of your use of new approaches or consideration of additional data sets. For example, while a first iteration may consider descriptive statistics, a second iteration may consider building confidence around key statistics, a third may involve correlative analysis or regression, and a fourth may involve hypothesis testing using new data sets.

While different homework assignments and projects will focus on one or more of the phases of the data science research process (shown in the figure above), the entire process will also be covered explicitly in one of the early lectures and/or discussion sections in the course, and will be reinforced throughout the semester through the iterative application of different (and increasingly sophisticated) methods. In doing so, you will develop an understanding of the entire process in addition to learning (and being assessed on) the specific research methods involved in each one of the phases. Finally, the course will end with a “tour de force” of going through the entire data science research process in the last “putting it all together” part of the course.

One of the themes emphasized throughout the course is the importance of reproducible conclusions. For example, you may be given a data set to support a hypothesis, and then could be given another data set to try and prove the same hypothesis.

Assessment of and feedback on Research and Information Literacy (RIL) learning outcomes will be provided to you as part of the regular evaluation of homework assignments and discussion section activities.

The following are examples that illustrate how you will be assessed on your choices of data sets on the one hand, and your choices of research methods on the other.

• In homework assignments or programming project assignments asking you to apply a particular research method in order to answer a question (e.g., using correlation to expose racial disparities in COVID infections), you will be given choices of data sets and will have to justify (and will be graded on) your choice of the data sets (or subsets) you will use to develop your arguments.
• In homework assignments or programming project assignments asking you to choose a method out of many to apply to a given data set (e.g., using clustering versus regression, or using selection followed by a join or vice versa), you will have to discuss and elaborate on the societal or ethical implications of your choice (with a distinct percentage of the grading rubric allocated to that aspect).