In this introductory tutorial, Clara Bowen explains why protecting privacy in publicly released datasets is essential, especially for sensitive domains like health care and government salary information. She outlines the four‑chapter structure of the course: (1) basic anonymization methods such as removing identifiers and rounding continuous values, (2) an introduction to differential privacy and the Laplace mechanism, (3) key properties of differential privacy including composition and post‑processing, and (4) practical differentially private data synthesis for public release. The tutorial demonstrates the workflow on two real datasets—a White House staff salary dataset from Socrata and a male fertility dataset from the UCI repository—showing how to preview data, identify private variables, and apply simple anonymization steps using the dplyr package (e.g., replacing names with numeric IDs and rounding salaries to the nearest thousand). The video sets the stage for later, more advanced privacy techniques.
Chapters
Hi, I'm Clara Bowen and I'm excited to teach you how to use R to apply basic anonymization techniques so you can publicly release data sets. You'll also learn about a powerful mathematical principle called differential privacy. With social media and data being generated in large volumes everywhere, data privacy has been a growing public concern. Entities such as the U.S. census Bureau are promoting and implementing better privacy techniques to address this concern. There are several different meanings of data privacy. As mentioned earlier, you will be focusing on ensuring the privacy of data sets released to the public. For example, health care data in the United States, taxpayer money is used to collect health care information. Therefore, as a taxpayer, you and the general public should have access to this data. Additionally, this type of data is extremely useful for medical researchers to discover potential contributors of cancer or improve personalized medicine. However, no one, not even those medical researchers, should know specifically who in the data set has cancer. Furthermore, the government should not deny access to health care data or or other data sets that could benefit society. In chapter one, you will learn basic anonymization techniques such as removing identifiers and generating synthetic data. In Chapter two, you'll learn the concept of differential privacy and an algorithm that satisfies differential privacy called the LAPLACE Mechanism. In Chapter three, we'll cover various properties of differential privacy, such as the combination rules and post processing. Finally, in Chapter four, you will learn how to release simple datasets publicly using differentially private data synthesis techniques. You'll practice applying the various data privacy methods on two data sets. The first is a White House staff salary data set from Socrata that contains information for individual staff members about their employment details such as salaries. The second is a male fertility data set from the UCI Machine Learning Repository that was collected in order to determine infertility contributors such as level of physical activity of a person and how smoking affects the health of sperm. Before you start applying any privacy methods, you should always preview the data. Let's take a quick look at our first dataset called White House. We see there are a couple of variables containing personal information that we should employee names and salary. Both are considered private information since most people would not want their names linked to a specific income. Under the salary column, we see that the salaries are reported to the closest dollar, so we might want to change the salaries to be less specific. Note that this course will make extensive use of the dplyr package to manipulate data. Let's get started. One basic method for anonymizing data is removing identifiers or, in the case of our data set, removing the names. We can accomplish this by replacing the names with numbers we saw from the TIBBLE output. There are a total of 469 observations, so we will replace the names with a sequence of numbers from 1 to 469. Another basic method is rounding continuous values based on the salaries we saw in the data set. We could round the salaries to the nearest 100 or thousand. Let's round to the nearest thousand by using the round function and setting digits to minus 3. With these methods in mind, let's try some examples.
