Note: I graduated in June 2024. The program is still in its development so the information here may quickly become outdated.

Read time: Around 10m

TL;DR

• This MSc. focuses on mathematical and statistical modelling to tackle real-world problems using data.
• Courses cover a range of topics including mathematical modelling, Bayesian statistics, frequentist statistics, and machine learning. 
• The degree includes a work or research term. I did a work term at the Workers’ Compensation Board of Alberta.
• I recommend coming into the program with at least a minors worth of experience in math and in stats, and some programming experience. 
• Graduates of the MDP program can expect diverse career opportunities in data science, with potential earnings around CAD 90k.

As I write this, I am reflecting back on my newly completed master’s degree at the University of Alberta, a Master of Science in Mathematics and Statistics with a specialization in Modelling, Data, and Predictions. I chose it, of course, because it has the longest name. From here on, I’ll spare you that title and call it the MDP program. I’ve fielded many questions about this degree from curious family, friends, and incoming students, so I figured I’d collect my thoughts formally and publish this overview. While the math department has just rebuilt the program’s website, which should address many questions, I’ll add my opinions about courses, program content and who should take the MDP degree below. 

The MDP program can be considered a Data Science degree. However, given that data science is an extremely broad term that covers several entirely separate fields of study, saying ‘data science’ doesn’t provide much insight. Instead, I’d call it a modelling degree.

The MDP degree aims to answer one set of questions: “Given zero, small, or big data, what approaches can be taken to provide inferences about a problem?”. To answer this question the MDP program teaches the mathematical underpinnings of an array of toolsets, so the practitioner has enough background to decide on the relevant method and apply it to a given scenario. The degree includes a research or work term (I chose work) so you can practice these techniques in the field. I’d describe the MDP curriculum’s focus as mathematical and statistical modelling; the use of math and stats to represent, analyze, and make predictions about real-world systems (or occasionally imaginary systems – much to the bemusement of my friends, I built a statistical model for a Dungeons and Dragons game to predict my party’s likelihood of winning a boss fight). With its coursework shared between math and stats, I’d consider it a statistics degree enriched with mathematical modelling coursework and with some numerical methods thrown in for good measure.

“Zero Data?” I hear you ask. Techniques for zero or limited data are often left out of data science adjacent degrees, so I’ll start there. 

Often in fields such as economics or epidemiology, decisions must be made before data is available. Two techniques popular here are mathematical modelling and simulations.

Mathematical modelling typically begins with the known facts about the situation and, from those facts, parameterizing the rates of change to understand how tweaking underlying variables can influence a scenario. The MDP program brought in Dr. Betsy Varughese, a senior mathematical modeller for the province of Alberta, to teach the mathematical modelling course. Her practical applications made math modelling my favourite course in the degree. She also taught science communication, including how to break down a model and its findings into a minimalistic briefing note to present the essential outcomes to policy makers. This skill served me well as I wrote many executive summaries during my work term. 

Under Dr. Varughese we studied the SIR model for disease transmission as a case study, and then students chose scenarios relevant to their domains for a term paper. I paired up with a couple of friends and modelled homelessness and housing flows in Edmonton.

The other technique used hand-in-hand with modelling is simulation. The primary tool for this is the Monte Carlo simulation, which showed up in three of the classes I took. Essentially, a Monte Carlo simulation runs a scenario thousands (or more) times with random sampling to calculate the mean, the variance, or other statistical measures of the scenario. It’s often used in Bayesian statistics for generating posterior distributions and sees significant use in finance and economics as well. One of my favorite math facts is an alternative use of Monte Carlo simulations: by scaling an integral into a probability distribution, sampling random points, and calculating their average, Monte Carlo simulations become one of the best methods for integration in high-dimensional spaces!

Thinking through the construction of a mathematical model is an engaging process, and if I were a betting man, I’d wager that modelling will be the last part of this skillset to be automated – after, of course, creating a model and running it through a Monte Carlo simulation to calculate the odds of my bet.  

“Okay, sure, but I want to hire you as a data scientist because I have data.” Well, thanks for the offer!  The MDP program core courses also included Bayesian statistics, frequentist statistics, and machine learning. Bayesian and frequentist statistics are two different ways to parameterize unknown information, with different philosophical interpretations. Bayesian statistics focuses on incorporating evidence to update a belief about the world, whereas frequentist statistics deals with changing one’s mind away from some default action. (See this article by Cassie Kozyrkroz for some plain-English explanations of statistics terminology) 

Both Bayesian and Frequentist statistics have applications in different contexts, and the mathematical underpinnings and assumptions change depending on the technique. Frequentist statistics has more closed form solutions (aka easier math) and is the preferred set of methods to teach to undergrads at UAlberta. Having taken many frequentist statistics courses in my undergraduate degree, the graduate course STAT 537 – Statistical Methods for Applied Research II, was a straightforward extension of that. On the flipside, Bayesian modelling was new to me. This fascinating set of methods was taught by Professor Jay Newby in MATH 509. For new students going through the course catalogue, ignore the title and description of MATH 509. Under Professor Newby it’s actually the Bayesian statistics course; the math department just hasn’t gotten around to updating the course description and name.

These days you can’t mention statistics without someone bringing up machine learning, and the MDP program is no different. We covered big data techniques and machine learning from a statistics perspective in STAT 541 – Statistics for Learning. 

Machine learning, often known by its sexier name Artificial Intelligence, is kind of the be-all end-all of prediction, when you have the gobs of data required to train it. Most of what we covered was supervised learning: the set of techniques used to analyze relationships between data and an outcome predict outcomes in new, unseen data. STAT 541 covered the mathematical underpinnings, derivations, and proofs of many of these methods, from LASSO to random forests to neural networks, and it had some implementation assignments as well. Unsupervised machine learning are techniques used to identify patterns, structures, or relationships in data without a specified outcome variable, and we covered all the standard techniques from K-means clustering to PCA. We also covered some miscellaneous topics, including differential privacy which I found particularly neat.

Option Courses 

The MDP curriculum has all students take at least 3 optional courses, two from the math and stats department and one from any department as long as it fits the students’ academic and career goals. I ended up taking 5.

In my first semester, I studied MATH 535 – Numerical Methods taught by Dr. Bin Han. It was incredibly helpful mathematically; I learned the inner workings of many optimization algorithms and other solving procedures, which made some of my undergraduate math finally click. This course also took the most work of anything in my degree, and it showed in the class enrollment – we started with about a dozen participants and dropped down to three.

I took the core course STAT 513 – Statistical Computing co-listed in my undergrad so the department let me replace it with STAT 553 – Risk Theory. I also took MATH 508 – Computational Finance, my first exposure to that field. The simulation techniques used there are similar to those in economics, and it was interesting to apply them to a new situation.

Two of my options came from the economics department. ECON 593 – Prediction and Machine Learning for Economics was a fantastic chance to apply previous knowledge. The professor, Dr. Sebastian Fossati, ran an exceptionally engaging competitive assignment where we had to make predictions on a housing dataset. He plotted our scores on the board so we could see how everyone did, offering bonus marks to the top performers. I took a second course of his outside of my degree requirements, ECON 493 – Economic Forecasting, which covered time-series forecasting, a skill every econometrician needs.

Capstone term

The MDP program requires either a work or research term, which is used to write up the degree’s capstone report. Most students opt for a work term, and the university offers resume review and interview practice to support it. The first year the MDP program ran, everyone who wanted a work term found a position. Chatting with my classmates from the second cohort, it looks like only around 2/3rds of us were hired and the last third had to switch to a research term as a backup.

My internship was with the Workers’ Compensation Board of Alberta, where I ran an econometric analysis to determine the efficacy and ROI of an intervention program they ran a couple of years back. This program predicted who would end up permanently disabled* so they could assign a specialist team to get these workers back on their feet. Using difference in differences estimation, I found that the models predictions combined with the specialist team was extremely effective, reducing the number of people on disability at the 14-month mark from a projected 76% to 55%. It was great to be able to show those teams that their hard work paid off.

I also assisted in implementing their AI ethics and bias framework, and I audited two machine learning models. Through this process I assessed the models to ensure they were unbiased, had effective fall-backs for handling false negatives, followed industry best practices, and performed better than the pre-existing manual approaches. I suggested improvements where models may violate these best practices and where they had potential bias. It was fascinating stuff, and I had an excellent time there.

*Permanently disabled is defined as a decrease in lifetime earnings resulting in permanent wage-replacement payments.

What’s missing?

The MDP program is pretty comprehensive, but there’s one small hole in its curriculum: databases and SQL. While the basics of those concepts required for data science are a quick self-study, I think it’d be worth adding a course to go in-depth on data modelling, database structures, and similar skills.

Why I took the MDP program

At the end of my undergraduate degree, I knew a lot of theory, but I was missing critical skills on using that theory to understand the world around me and find answers to real world problems. I knew I wanted an applied master’s degree to round out my understanding, and my parents gracious support in letting me live at home rent-free encouraged me to seek out options in Edmonton. Conveniently, the University of Alberta is extremely well-regarded in machine learning and adjacent fields, including a good math and stats department. Given that the University of Alberta does not have an econometrics master’s degree**, the MDP program stood out as the optimal choice to further my interests. I was especially drawn to how general it was, rather than specializing in a particular field of math or stats. Doing a work internship instead of a thesis was also a natural fit, as I currently aim to end up in government, an NGO, or industry instead of academia.

This strong background in inference sets me up well for working with economic data and models, and combined with my next degree will leave me an excellent foundation for my career. I’m satisfied with the two years I spent taking the MDP program. 

**Sources in the econ department tell me they are working on an econometrics and data science master’s program, which they hope to start in the next few years.

Undergraduate Degree Fit

The MDP program is a math and stats master’s degree, and the people who found this degree the most straightforward were those with an undergraduate stats major. It was also very reasonable for those with an undergraduate degree in math who have taken substantial statistics coursework (if I do say so myself).

Students coming in with a computer science major found that the MDP program required much more learning on their own time than it did for me. However, talking to my peers I was told that they felt the program really helped round out their knowledge, so I encourage computer science majors interested in data science to consider the MDP program. 

The admissions committee for the MDP program is pretty responsive to math skills learned outside of a math degree if you can provide sufficient proof of knowledge. I know someone in the program with a business degree who then took all their calculus, linear algebra, and statistics courses remotely through Athabasca, and was accepted to and thrived within the MDP program. 

Minimal Prerequisites

Below are the courses that I’d put as the unofficial minimal prerequisites for the MDP degree courses. Click here to visit the University of Alberta course catalogue with the descriptions and topics list. I think it would be extremely hard to succeed without having taken the following courses or equivalents:

• 4 courses on Calculus and 2 courses on Linear Algebra
• STAT 252 – Intro to Applied Stats II
• STAT 265 and STAT 266 – Probability and Statistics I and II
• STAT 378 or ECON 497 – Regression. Lots of statistics and machine learning come down to clever takes on regression algorithms. Some students without a regression course struggled immensely. 
• A course on optimization
• A course on differential equations
• Some programming knowledge, ideally in R and Python.

Although MATLAB is also used, licenses are expensive. I’d recommend holding off until you have access to the university’s licences and learning MATLAB alongside your other courses.

Recommended Additional Prerequisites
Some of my other undergraduate courses really helped me out, including: 

• STAT 413 – Computing for Data Science <- random number generation, bootstrapping, and computational statistics.
• STAT 371 – Probability and Stochastic Processes <- A prerequisite for many graduate-level courses.
• Mathematical Modelling <- ECON 482 under Prof. Dave Chetan was taught as a modelling course when I took it, and I’m sure the math department’s undergrad modelling courses would be similarly useful.

I’ve also heard that STAT 441 – Statistical Methods for Learning and Data Mining, was very useful covering the background behind machine learning algorithms.

Finally, ensure you have the background for the elective courses you want. While grad school prerequisites are flexible (and you can often be exempted from them), the courses strongly build upon prior knowledge. Professors are usually happy to send syllabi and background requirements in advance, and I’d recommend emailing them to make sure you have what you need to succeed. I strongly recommend getting your options settled ahead of time, or at least within the add/drop deadline of the first two weeks. Withdrawing from courses is possible up to ~2/3 through the semester, but it typically causes timetabling problems. Feel free to email me if you want a fellow student’s advice on choosing options.

Degree alternatives at the University of Alberta 

For a master’s degree, the MDP program is relatively general and intended to be a terminal degree in math and statistics. Taking the MDP program doesn’t create the publication paper trail that is useful for further academia. For those interested in a Ph.D. I expect a thesis MSc. program would be more useful, and there is some funding available for thesis programs. 

If you are looking for the master’s to Ph.D. pipeline at the University of Alberta with similar subject matter available to the MDP program, I’d recommend an MSc. in Applied Mathematics, Statistics, Statistical Machine Learning, or Mathematical Finance, depending on the direction you want to take your studies. 

Finally, I’ll give an honourable mention to the MA in Digital Humanities. If you want to learn everything from information ethics to cultural representations of AI to games and software as an artistic and cultural medium, this is the degree for you. It’s a social science program studying the implications of digital technology, AI, and all that jazz on humans, cultures, and society. I know someone who took it and they highly recommend it.

The MDP program was an intellectually stimulating use of a couple of years, fulfilling my intellectual curiosity about applying math in the real world. It provided me with an excellent modelling background, improved my coding skills, and strengthened ability to work with data.

It also strengthened my application to the London School of Economics, where I will continue my academic journey studying for a MSc. in Local Economic Development next year, after which I plan on moving beyond academia and making an impact out in the world.

I appreciate all the support of my family, professors, classmates, and friends through my master’s degree, and I can’t wait to see where the next leg of my journey takes me.

Please feel free to post any questions you might have about the program in the comments section below. I’ll do my best to answer them or point you in the right direction.

Q1: What types of positions and salaries can graduates expect after completing the MDP program?

From publicly available data on LinkedIn, there are a few alumni data scientists, and I got a data scientist offer as well. There’s a couple of data-warehouse or data-management positions, data analysts, and some miscellaneous roles reflective of people’s undergrad fields combined with the statistics skills from the MDP program.

Entrepreneurship is also a viable option. One of my classmates, Colby Jamieson, is currently launching Auditable AI, a company that focuses on legal compliance, efficacy, and ethics analysis for machine learning models. If you need to prepare your companies AI for audit or want to explore explainable AI solutions I’d highly recommend him.

I was able to find salary information for two of the roles at around 90k. If anyone is willing to anonymously send me salary information so I can repost it here, please do so.