Data Analysis for Social Scientists

Writing a review for this course is hard. The content of the course is ambitious and the promise is considerable. I am grateful that the Professors and MIT have made this course available online. That being said, I find it hard to recommend this course.

As an overview, each week contains 2-3 lectures, mostly probability mixed with some stats, with 'finger exercises (FEs)' at the end of each lecture segment to test knowledge. At the end of each week there is a more in-depth set of questions covering all the material and some more practical aspects with R. Here is a quick summary of the different course aspects:

Lectures - Some students felt that the MOOC course was just a video of the in-person lectures, without the MOOC in mind, but I felt this was one of the benefits i.e. identical coverage

FE - good tests of comprehension and linked to the lectures well. Some of the wording was a little confusing, at times it felt as if the questions were setup to trick you

Homework - this and particularly the sections using R, were more enjoyable and felt a little better thought out, at least when the grader worked correctly

Staff Support - both the staff and community TAs were great, one student commented ' I cannot believe the response time of the staff (responses to comments are so quick!). Thanks!'

Discussion - the discussion on the boards was initially lively and the discussion on each individual item, e.g. part of a lecture, was useful both as an aid to understanding and also to identify any human error or typos in the lecture, of which there were some. Over time however, it seemed like students were either leaving the course or getting frustrated, so their used diminished

The course got off to a pretty good start over the first couple of weeks. The lectures are well structured and linked to the FEs. The introductory lecture by Professor Duflo was incredible and really whetted the appetite for the rest of the course. The use of R was introduced with a course specific set of walkthroughs (modules) in the R package SWIRL. Whilst I'd heard of SWIRL before, I hadn't actually used it. Personally I found it fine, although a number of people on the course seemed to feel SWIRL was a little dated and using something like an interactive Jupyter or R Markdown document, would have been better. There was an issue with the grader for some of the R code for the first homework, meaning 3 of the 20 questions were ungraded. Staff pinned an item on the discussion group as soon as this was identified, students were also emailed. Overall, an interesting introduction, if a little frustrating due to the grader issues.

The following week covering the Fundamentals of Probability was good, albeit I reviewed some of the lecture sections more than once and tried to dust down some of my probability from the dark recesses of memory. The lecture and FEs were well structured and flowed logically, the questions by students in the actual class and by those taking the online version were helpful. Some students felt that the content suffered from not using the same terminology as the MIT class 6.431x (Probability - The Science of Uncertainty and Data) and that the content of this course was more akin to a review of probability for those already familiar with it. Given this is supposed to be taken after that course, this may be a fair assessment. Some of the lecture parts, for instance the cumulative probability function, felt like they were overly complicated explanations of things which were intuitively quite straight forward, at one point the lecturer even noted 'intuitively they make complete sense', so it wasn't just me! There were multiple errors in the content, both in some of the lecture explanations and in the homework.

Next was an R lecture was on Gathering and Collecting data. I thought this was a good introductory lecture and gave enough of an understanding about how, for instance, to webscrape in r using rvest. However, I suspect those already familiar with web scraping may find this a little too introductory. Those coming more from a programming background may find it useful to cover some of the other aspects the lectures covered.
This included things like the process of data collection, piloting questionnaires, having a data management plan and seeking ethical approval for the intended research. However, these concepts were really just touched upon rather than covered in any great depth. They did give pause for consideration.

In week three more problems started to emerge. Firstly there was an error in a histogram on a FE question - four large bars which, when combined, appears to exceed 100%. The staff TA noted "Yes, you are right. This is embarrassing". The homework had multiple issues. In one example, the data used teenager fertility rates from 1960s from the online World Bank data catalog. On review it appeared the underlying data had changed, resulting in the grader being based on older and therefore different data resulting in erroneous marking. A data file for the course content was missing and was provided by a Community TA from their own google drive, from a previous iteration of the course. There were multiple calls on the forum for a github repository to be created for the course. Examples in R were often provided in the lectures, but the underling R code was not available to students. One of the TAs had requested the R code from the lecturers, however it didn't arrive. On some occasions, the code was provided by other students.

By week four the course started to really feel like a slog. The lectures seem to cover the material at too high a level, but at the same time required a lot of prior understanding of concepts. The first lecture was on Functions of Random Variables (RVs). It was demonstrated how RVs were calculated, but it felt a little too mathematical with little connection to the real world. At one point, the lecturer presents information on Probability Integral Transformation, then looks to the audience and notes "I'm seeing some puzzled looks", which after a period of silence, a student asks the question possibly everyone is wondering - "could you do some examples of when this would be used"? I certainly felt this needed more relation to the real world. A further couple of questions from the lecturer are met with silence, until one student hazards a rather uncertain sounding response. After further explanation the lecturer rhetorically asks 'what's the application' and answers the question themselves stating 'I happen to think it's pretty cool'. The discussions after the lectures were still there, but with many fewer students involved. There was much less engagement from students in the actual class. The TAs (staff and community) were still providing a good response and supporting students, however one could question - should such support be required if the lectures stand up on their own?

One student noted, directing their comment at staff "I am afraid that there are just too many questions that are unclear, confusing, badly worded, etc. in this course. It wastes time and creates uncertainty in my mind when I'm looking at all questions now. I keep on having to ask myself if there could be a problem with the question itself. This shouldn't be happening on an MIT course!"

It was at the end of week four I had to decide not to invest any more time in the course. How could this course be improved?

1 - Be clearer about the pre-requisites - single and multivariable calculus plus an introductory level of probability and statistics is required
2 - Consistency with terminology - both with common practices and MITs other courses
3 - Correct the errors - there were too many errors both in lectures and questions, this makes students doubt their own knowledge
4 - Provide R code - the course is supposed to cover both theory and R and providing the R code would make the course more tangible and applicable
5 - Provide more examples - the material was often too abstract for me, or the examples given were too simplistic and not related to the course objectives, namely providing skills for Social Scientist

I would like to come back and complete the course at a later date, currently it feels a bit too ambitious and requires some polishing before being ready. It may be necessary to split the course over two semesters or two parts, if more examples and code is provided, to better achieve the course objectives.

I did not enjoy this course at all. Here are the main reasons why:

1) Lecture videos. The lecturers themselves might be masters in their respective fields, but the lecture videos are not suitable for an online course. The videos are literally from the actual MIT course. The most annoying thing about the videos is that the lecturers sometimes make reference to something shown on the board or the screen, but the video just shows one thing at a time: either the view of the lecturer or what is being projected on the screen. It feels weird and I don't think they really thought how students who are online would feel. Also, because the videos are recordings of a live class, there are students who ask questions, but most of the time you can't really hear what they are saying (the transcripts often just say "inaudible"). This is quite frustrating, as questions are asked often.

2) Quality of the material. I think they didn't explain many concepts properly. The lecturers would introduce concepts in a way that they would assume that we already know what they are talking about. Notes that are shown on screen are also quite vague. I would like more detailed explanations and worked examples. Speaking of examples, the lecturers would do a few examples, but I think the choice of examples were sometimes too confusing. I would have liked to see more examples worked out properly. Also, R programming was only briefly covered in the lecture videos

3) Quizzes and homework. The quizzes are literally just questions based directly from whatever the lecturers mentioned. While this tests if you actually watched the video, I feel like they could have added more questions requiring us to work out/calculate more things. It felt like these questions were made just for the sake of it. Many of the answer explanations are not very well explained.

There are also a few other issues and problems I had with this course. I really hope they improve this course on the next iteration.

tl;dr - poorly put together MOOC trying to cram too many things into one course. Doesn't leave you with a lot of confidence that you can analyse data independently on big projects. Confusing approach without concrete examples and demos of how to run full tests in practice. I found it a frustrating experience.

This course tries to cover a lot of ground in very little time, so, the treatment is very superficial. In the end, what you're left with is a hodge-podge of techniques and methods with no real intuitive understanding of what they mean and without a solid understanding of how they are to be applied. It would be a much more useful course if it went into more depth and focused on a narrower range of topics. Many of the questions in the homework are vaguely worded (the answer keys have several issues as well) and/or there's a significant jump in difficulty compared to what's covered in the lectures. I often find myself struggling to extrapolate and apply what I've learned in class to homework problems and other courses/projects. Contrary to what the course advertises, it assumes some prior knowledge of probability theory and core ideas of statistics. Despite the impressive range of topics on paper, it feels like an unorganized and poorly thought out MOOC (I don't know about the on-campus version of the class), in that the goals for the class were not clearly laid out and executed. Think of this class as a 'survey' course of data analysis techniques relevant to the social sciences designed for people who already know some probability and statistics. That being said, I found the included R course (and supplemental resources) helpful as I hadn't used it earlier.

I am taking this course because it is required for my edx micromasters. It is a second MIT course in my specialization and a 5th MIT course on edx. All of the courses have been great. None of them were introductory or easy. It's a lot of work (although this course is somewhat slower in pace than a couple others - either that or I am just getting used to the process and hopefully also developing a good foundation). I was almost afraid to take this course because of the many below average reviews, but it's a good course. It may not be perfect, but it is very rewarding. Yes, some of the questions and answers are still argued about on the forum, but, even if (and that's a big IF) they are not perfect, the value of the course is not diminished in any way. I believe the course is listed as introductory, but it would be easier if one had appropriate background (multivariate calculus is or was listed as a prerequisite, and some familiarity with statistics, probability and/ or R would help). Just like in any other field, however, a first serious course is usually challenging, but later courses come a lot easier. MIT MOOCs are more demanding. Before I spent months on a slow-paced "perfect" MOOC (in Python, for example), but despite my expectations I came out with very limited knowledge and skills. MIT covers a lot in a short time, but they are good at what they do, and with some faith and a lot of work, I believe the results will impress you.

I've just completed the course. I'd high expectations and I was hoping to learn a lot more practical skills in R. However, the course tries to cram way too much content into shorter term and covers the topics at a very shallow level. The instructors are obviously highly qualified, however the quality of this course is below average. It's not meant to be taught as an online course. It's probably a lot better in person. The finger exercises and homework are full of questions that aren't designed to teach anything to the students rather test their ability to understand sentences written in twisted ways. I'm probably being too harsh on it. However, when I compare it with the Probability course from MIT, there is huge difference in quality of teaching and material. I believe the only thing I've learnt at the end of the course is the subject titles and I only have some basic understanding of the subject.

This is just my experience and it's possibly because I didn't have a lot of the background knowledge required for this course. Also it's possibly because I expected too much. I'm not taking the paid exam at the moment, rather wait and learn the content in depth in near future before attempting the exam in the next session. However, I haven't been able to find a book or online resource that covers the whole course. I've found a few YouTube channels and some online articles that might help achieve my goals. Looking forward to the Machine Learning course next.