The Professional Certificate in Data Science is HarvardX's flagship data series, nine courses numbered PH125.1x to PH125.9x and taught by Rafael Irizarry, professor of biostatistics at the Harvard T.H. Chan School of Public Health. The PH prefix tells you where it comes from. This is a public health statistician's view of data science, where the point of the exercise is drawing valid conclusions from messy data, and the machine learning course arrives eighth, after you have earned it. The sequence runs R Basics, Visualization, Probability, Inference and Modeling, Productivity Tools, Wrangling, Linear Regression, Machine Learning (now listed as Building Machine Learning Models on Harvard's catalogue) and a Capstone.
Most courses are listed at 8 weeks and 1 to 4 hours a week. The capstone is listed at 2 weeks but 15 to 20 hours a week, and Harvard is upfront that you get much less instructor guidance in it. Along the way you build the tidyverse workflow, ggplot2 and dplyr, learn enough Unix, git and GitHub to work reproducibly in RStudio, and work through case studies that are a cut above the usual iris dataset: trends in world health and economics, US crime rates, the 2007-08 financial crisis, election forecasting, building a baseball team on a budget, and a movie recommendation system. The machine learning course covers the standard algorithms, principal component analysis, regularisation and cross-validation, and ends with the recommender. What makes it good is the order and the teacher.
Irizarry does not let you fit a model until you understand why a sample mean has a distribution, what a confidence interval does and does not say, and how Bayes' theorem changes a prediction. The election forecasting module alone, which walks through poll aggregation and hierarchical models, teaches more honest statistical thinking than most paid data science bootcamps manage in twelve weeks. He also publishes the whole thing as a free open textbook, Introduction to Data Science, which I would read alongside the videos. What makes it dated is everything else. The courses were recorded years ago and the certificate pages signed by Irizarry stretch back to 2019 with the same content.
There is nothing on neural networks beyond a mention, nothing on language models, nothing on Python. R is the right language for what Irizarry is teaching, and tidyverse R is a pleasure for analysis, but the job market for data science and machine learning has moved decisively to Python, and a learner who finishes all nine courses will still need to learn pandas and scikit-learn before most interviews. Harvard's own newer data courses, Introduction to Data Science with Python and Machine Learning and AI with Python, are the tacit admission. Then there is the price. Each course is free to audit, and that includes the videos and most of the exercises.
Verified certificates cost $149 for Machine Learning and the Capstone and up to $219 for R Basics, and the full series is listed at $1,481 on pll.harvard.edu. A Business Insider piece from 2020 quoted $491 for the same programme. Tripling the price of unchanged content is an edX decision rather than a Harvard one, and the recurring 20 percent promotion codes suggest they know it, but it moves this from an easy yes to a careful no. The certificate is a HarvardX credential issued through edX, not a Harvard Extension School transcript, and no hiring manager I know treats it as more than evidence of persistence. The capstone deserves a specific note.
You produce a data analysis report on your own, with far less scaffolding than the earlier courses, and the reports are peer graded. That is fine as a learning exercise and poor as assessment, because the quality of the feedback depends entirely on which other learners draw your submission. It is also the point at which the 58 listed weeks start to feel long; a motivated person with some programming background can do this series in three or four months of evenings, and the suggested 17-month pace is for people fitting it around a full life. My honest recommendation is to audit the entire series and do every exercise, especially Probability, Inference and Modeling, and Linear Regression, because those three courses are the ones most self-taught data people skip and later regret. Buy a certificate only if an employer will pay or a programme asks for it.
Then move to Python. If you want the same statistical rigour with a credential that carries real weight, MIT's MicroMasters in Statistics and Data Science is harder, pricier and worth far more, and I have reviewed it too.