As a statistician now working as a machine learning engineer, my response is “definitely not”. It’s all the frustrations of software development, but on top of that you are now frequently dealing with clients/colleagues whose requests are now not simply impractical, but usually defy the laws of mathematics and probability, and an ever present pressure to put out work that amounts to fraud. Analytics have a lot of value to provide many organizations, but it requires planning, foresight, and a willingness to sacrifice a little now for the sake of a payoff in insights later, which very few organizations have, in my experience. So it essentially becomes a buzzword and people throw worthless data at you to wave a magic wand over so you can tell them what they want to hear. Doing so would essentially require lying, so instead, we would perform the awful, worthless analysis, it usually didn’t provide much insight, and we would include a litany of disclaimers about why the little insight it did provide wasn’t al that trustworthy, which would just disappoint and infuriate the people we were working with. We would also provide detailed guidance on how to execute moving forward to make the process much more valuable the next time around, which without fail went in one ear and out the other.
So essentially we became figureheads. Our work rarely was used in any significant way or provided much value, but we were kept around because the company wanted to be able to tout its “data driven” culture.
It was so bad that at one company I worked for, they had the data science/analytics department start putting on a yearly intracompany conference on analytics that became a huge deal. One year they got Stephen Levy, the author of Freakonomics, to be the keynote speaker. At one point he shared a story about how he was consulting with a company on their marketing, and they found that they had accidentally not been running ads in a particular metro area, and were able to leverage this to act as a control to assess the materials effectiveness. But when asked to intentionally do something similar moving forward, the company balked. It was so close to home that my colleagues and I wondered if the head of our department had fed him the need to talk about it. And yet, not a single thing changed at the company during my time there.
I currently work in a role much closer to software engineering, and I have all of the same problems described by the person in the original post and that many are describing here. But I consider it a strict upgrade over my time working as a statistician.
Seconded. I work as a data analyst in medical research (bioinformatics postdoc). I am often introduced as a statistician, even though I'm not, because I can do a bit more than a t-test.
The situation in research is exactly as you describe -- we are figureheads who are put into place and highly pressured to confirm whatever hypothesis a PI wants for their latest grant or paper. They would never ask us to commit fraud, only perhaps to "double check" an analysis 10 times until it shows what they want to see.
If I were working for a company, this would at least be understandable, as companies don't even have a theoretical commitment to truth and scientific integrity, and there are no real consequences to a faulty analysis.
But it is immensely galling to see in research. Here we are, paid by the public to supposedly pursue truth and improve human health, and instead the job is to constantly be finding ways to avoid fraud and fabrication without pissing off the collaborator. The result is, as you say, useless analyses if the analyst is honest, and fabrications if they are not.
There is absolutely no doubt in my mind that this is one of the key reasons the ROI on science has declined drastically in the last few decades. It makes me laugh bitterly every time I see (increasingly frequently) political exhortations for plebeians to "trust the science".
Thank you for sharing this, these are fair comments. My experience has been more positive, but I see the truth in many of your points. For what it's worth, I have always been involved in controlled experimentation, and never anything related to ML or modeling. Our biggest issue is usually ensuring we are getting accurate counts of events, and when I see all the challenges in that, I wonder how anything more advanced ever gets done.
I suppose, as with anything, your mileage may vary. And in this case, your mileage will depend largely on your organization. In my experience, most organizations are not equipped or prepared to do what is necessary to make most analytics efforts worthwhile. But in the instances where this is not true, working as a statistician is very rewarding. If you have had a more positive experience than I, more power to you, and kudos to you and the people you work with/for.
So essentially we became figureheads. Our work rarely was used in any significant way or provided much value, but we were kept around because the company wanted to be able to tout its “data driven” culture.
It was so bad that at one company I worked for, they had the data science/analytics department start putting on a yearly intracompany conference on analytics that became a huge deal. One year they got Stephen Levy, the author of Freakonomics, to be the keynote speaker. At one point he shared a story about how he was consulting with a company on their marketing, and they found that they had accidentally not been running ads in a particular metro area, and were able to leverage this to act as a control to assess the materials effectiveness. But when asked to intentionally do something similar moving forward, the company balked. It was so close to home that my colleagues and I wondered if the head of our department had fed him the need to talk about it. And yet, not a single thing changed at the company during my time there.
I currently work in a role much closer to software engineering, and I have all of the same problems described by the person in the original post and that many are describing here. But I consider it a strict upgrade over my time working as a statistician.