.svg)
Questions that start with who, what, when, or how deal with quantitative data. Questions that start with what if deal with prediction, that is, in the causal relationships between data.
Though humans infer causation every day ("The bus is late? Probably the weather"), rigorously and objectively inferring causation is complicated. That said, the benefits of doing so are significant enough to be worth the struggle.
Uber engineers, for example, use causal inference to fine-tune features and create better personalization. As a result, they can determine whether customers are dissatisfied with a feature or whether the user experience is simply unclear.
So, you want to learn about causal inference. We've gathered information about four online courses so that you can determine which is best for you – and what to expect from each.
Course: A Crash Course in Causality: Inferring Causal Effects from Observational Data
Best for: Beginners
Price: $0 - $49
Time to complete: 18 hours
Flexible schedule: Yes
Prerequisites: No
Accredited: No
The course description does not specify who would benefit most from it. Instead, it offers a broad education about causal effects and how learners can implement and interpret related statistical methods. Those who attend can apply the techniques they learn to example data provided.
The course is taught by Jason A. Roy, the Chair of Biostatistics and Epidemiology at the Rutgers School of Public Health and previously a professor of biostatistics at the University of Pennsylvania. His research focuses on causal inference, especially as it pertains to Bayesian methods, which makes his knowledge even more relevant to this course. His work also tends to use data from public health contexts, such as electronic health records.
Like most Coursera courses, this one is entirely online, enabling learners to start whenever they want and engage with it on their schedules. There are quizzes and assignments, but learners can shift deadlines to accommodate their schedules. The course, according to Coursera, takes about 18 hours to complete.
Unlike some of the other courses on this list, this course offers little in the way of networking opportunities. Learners who pay for the course can share their certificates on LinkedIn, which might allow other learners to find and connect with them. Still, Coursera does not appear to facilitate this. It will be tough for learners to find others taking the course at the same time as them, meaning it will be hard to share assignments and compare notes.
Coursera offers two pricing options:
Those who complete the course will get a Coursera certificate. However, Coursera courses are not accredited by a third party like the Continuing Professional Development (CPD) Standards Office.
Course: Introduction to Causal Inference
Best for: Beginners
Price: $0
Time to complete: 11 hours
Flexible schedule: Yes
Prerequisites: The basics of probability
Accredited: No
This course material is created for anyone "who is interested in learning the basics of causality." Neal provides one major prerequisite, however: that learners know the basics of probability before taking the course.
Additionally, he shares that statistics and machine learning topics will come up throughout the course, so familiarity with those topics will be helpful but unnecessary.
Neal is a Ph.D. student studying causal inference and machine learning at Mila, the Quebec AI institute. He completed a Master's degree at the same institute and a Bachelor's at the University of Pennsylvania.
Since posting this course in 2020, Neal has become a Senior Research Scientist at Dataiku, a company that provides software for designing, deploying, and managing AI applications. He is also an AI teacher at Astra Nova School, an experimental online school for children between the ages of 10 and 14.
His interests lie in fundamental and applied causal inference, and he appears most interested in applying his learnings to public policy.
The course last took place in 2020; considering how quickly causal inference techniques evolve, some of the material might need to be updated. The course is primarily conducted via YouTube videos. Learners can watch everything on their own time, including guest lectures by people like Alberto Abadie, Associate Director of the Institute for Data, Systems, and Society at MIT.
The free course takes 15 weeks to complete, with one video and a set of slides per week. Though learners do not receive a certification or diploma, they can expect to learn the basics of causal inference, causal discovery with and without experiments, causal representation learning, and more.
While the course was active, Neal offered a Slack workspace for learners to connect to other learners, but the workspace no longer appears active. Neal also offered office hours and 1:1 email correspondence, but it's unclear whether he still provides these resources in 2022.
All sessions are available to the public on YouTube, so the course is free.
Learners do not receive a certification.
Course: Machine Learning & Causal Inference: A Short Course
Best for: Economists, researchers, policy designers, and behavioral scientists
Price: $0
Time to complete: 9 hours
Flexible schedule: Yes
Prerequisites: No
Accredited: No
Stanford offers an online course called Machine Learning & Causal Inference: A Short Course aimed at anyone curious about how they can use machine learning to measure the impact of an intervention.
Stanford doesn't identify skills or experience that learners should come to the course with. However, they call out four groups of learners who will most likely be interested in the material and benefit from the course:
…an economist, researcher, or policy designer who uses evidence from randomized controlled trials or A/B tests to make decisions about what works, or a behavioral scientist or practitioner who wants to understand how increasingly large data can be used for good.
Two Stanford professors teach the course: Susan Athey and Jann Spiess. Athey is an economics professor at Stanford and served as Microsoft's consulting chief economist for six years. In this role, Athey worked as an economic adviser, recommending major deals and corporate strategy to Microsoft's senior leadership team and board of directors.
Spiess is an econometrician who also works at Stanford but appears to have less significant experience applying causal inference in professional contexts. His research focuses on microeconomics and statistical decision theory as well as data-driven decision-making informed, primarily, by causal inference.
The course consists of roughly nine hours of video material that's free and hosted on YouTube. A tutorial is also provided, which the instructors say learners can use as lecture notes and programming examples.
The tutorial includes example code and examples of the methods described in the course applied to real data. Because the tutorial is written in Markdown, learners can download, modify, and execute the chapters. However, the tutorial is an "ongoing project," meaning it is incomplete, and new chapters are still being developed and uploaded.
Networking opportunities for this course appear to be slim. There's no stated way to interact with other learners or even to interact with the instructors.
The course is free. However, learners with a computer that can run one of the available GUI tools for R — such as RStudio or Jupyter, to use the examples given in the course's tutorial — will benefit most.
Course completers do not receive a certificate or diploma, but the Stanford name carries weight in many circles. Learners who can say they completed a course from Stanford might be trusted more than learners who complete a course from a provider with less brand recognition.
Course: Applied Causal Inference
Best for: Experienced data scientists and ML engineers
Price: Starts at $700
Time to complete: 12 hours over 3 weeks
Flexible schedule: Yes
Prerequisites: Yes
Accredited: Yes
At Sphere, we offer a course called Applied Causal Inference, designed for data scientists and ML engineers interested in using causal inference to shape business decisions. Potential learners should be proficient with Python or R, be familiar with probability and statistics at a college level (at least), and have a general understanding of supervised machine learning.
Expert practitioner Rob Donnelly leads the Applied Causal Inference cohort at Sphere. After earning his Ph.D. in economics from Stanford, Donnelly worked on economic models of competition and network growth as a research scientist at Facebook. Before then, he was a senior economist at Instacart, working on advertising, pricing, and marketplace growth.
During six two-hour live sessions (2 sessions a week for 3 weeks), learners can ask questions and learn from the instructor and others in the cohort. In addition, in the Sphere platform, they can access optional pre-reads and supplementary materials, as well as messages from the instructor answering questions between sessions.
By the end of the course, learners will be able to:
The community of professionals that learners are exposed to is an important component of the learning experience at Sphere. Interact with learners in two optional networking sessions, meet them in breakout rooms during live sessions, and connect via private messages in the Sphere platform.
Professionals who have taken Sphere courses have grown their personal and professional networks and even found their next job opportunity.
Sphere's causal inference course starts at $700 per seat, which we encourage learners to expense to your learning and development department. Employers who help cover the cost of this course will benefit from employees who use data to inform their decisions and predict their outcomes. They will be instrumental in contexts where A/B testing is impossible, such as when randomized experiments are – for ethical or legal reasons – not practical.
Because Sphere is accredited by the Continuing Professional Development Standards Office, you will receive a certificate to add to your resume, LinkedIn profile, and future cover letters.
Numerous course options are available for those interested in learning about causal inference. Curious learners new to the topic can start with those, understand whether learning about causal inference will be useful, and then try a paid course.
Additionally, learners will want to consider specialization. Though all the courses will likely teach similar fundamentals, learners interested in immediate applicability might be better served by the Sphere course.
.jpg)
.svg)
