Is Data Analysis Stressful?

Being a request queue. That's what I am most days. Somebody in marketing wants a pull. Somebody in merch wants a report. The VP of e-commerce wants to know why the conversion rate on mobile dropped last Tuesday. And each of those requests comes with the phrase "when you get a chance" which actually means "today." I counted once. In one week I had fourteen ad-hoc requests from nine different people, none of whom knew the others had also asked me for something.

The stressful part isn't the volume. I can write SQL fast. The stressful part is that none of those fourteen requests are in my official priorities. I have a project plan. Deepa, my manager, and I agreed on quarterly goals. I'm supposed to be building a customer lifetime value model. I have not touched it in three weeks because every day gets eaten by requests that feel urgent to the person asking and trivial to the work I should be doing. But if I say no to the VP of e-commerce, that's a career problem. So the LTV model waits, again, and Deepa knows and I know and we both pretend it's fine in our one-on-ones.

Watching people use my data to tell a story it doesn't support. This happened last quarter. I built a report showing that our premium tier had a 6% monthly growth rate. Good number. Real number. Our head of growth, Dante, put it in a board deck. But he showed it as a graph that started at 4% on the y-axis instead of 0%, which made it look like a rocket ship. And then in his narrative he wrote "premium tier is growing exponentially." It's not growing exponentially. It's growing at 6% monthly, which is good but linear.

I said something. I dm'd Dante and said, "Hey, the y-axis truncation makes the growth look more dramatic than it is, and 'exponentially' isn't technically accurate." He said, "I hear you, but the board doesn't care about the axis. They care about the trend." And he sent the deck. And I sat there knowing that a room full of investors was looking at a graph that I built and that had been manipulated to tell a more exciting story. My name isn't on the deck. But the data is mine. And it's being used wrong. And I can't do anything about it.

That's the stress. Not the SQL. Not the data cleaning. The moment when your analysis leaves your hands and becomes someone else's narrative.

That a mistake I make could ripple into a clinical decision before anyone catches it. I've been doing healthcare analytics for nine years and I still triple-check everything. Last month I was calculating readmission rates for our cardiology service line and I found a discrepancy between my numbers and the numbers our quality team had reported two quarters earlier. A 1.8 percentage point difference. Which, in readmissions, is significant. It could change whether CMS penalizes us.

I spent four hours finding the source of the discrepancy. It was a date filter. The quality team had used discharge date and I'd used admission date. Both defensible choices, depending on the methodology you're following. Neither was "wrong." But they produce different numbers and different numbers produce different decisions. I wrote a three-page methodology document reconciling the two approaches and sent it to my director, Dr. Pham, at 9 PM because I couldn't sleep without resolving it.

The stress isn't making the mistake. It's the knowledge that somewhere in a pipeline or a dashboard there might be another date filter, another edge case, another 1.8 points that I haven't found yet. And that somebody might make a staffing decision or a protocol change based on a number I built that has a tiny invisible flaw. That thought is always there. Always.

Imposter syndrome that I'm starting to think might just be regular syndrome. I came out of a bootcamp. General Assembly, the data analytics immersive. Twelve weeks. Before that I was a barista at a specialty coffee shop in Brooklyn for four years. And I'm now sitting in meetings with people who have master's degrees in statistics from Columbia and I'm trying to figure out if I should nod along when they say "heteroscedasticity" or admit that I had to Google it during the meeting. I Googled it. On my phone. Under the table.

The thing is, I can do the work. My SQL is solid. I built a churn prediction report last month that my manager, Eric, said was "one of the cleanest analyses I've seen from a junior." That felt great. For about four hours. Then I was in a data modeling session where our senior analyst, Tomoko, started talking about slowly changing dimensions and I had literally no idea what she meant. I went home and read about it for two hours. And now I know. But that's my life. Learn a thing, feel competent for a few hours, encounter a new thing I don't know, feel like a fraud, go home and learn it, repeat.

My therapist says this is normal for career changers. But the people around me don't seem to be learning on the job the way I am. They seem to just... know things. Tomoko doesn't go home and read about slowly changing dimensions. She learned that in grad school seven years ago. I'm running to keep up and I've been running for a year and the distance doesn't seem to be closing.

The ceiling. I can see it and it's right there. I've been at this company for four years. I'm good at what I do. I built the entire BI layer in Looker. I automated twelve reports that used to be done manually in Excel by three different people. I saved the operations team about twenty hours a week. My manager, Frank, tells me this in every performance review. "You're doing great work." And then my raise is 3% and my title is still "Data Analyst."

The next level at our company is "Senior Data Analyst," which pays about $12,000 more and requires, according to the job framework that HR published, "demonstrated ability to lead cross-functional data initiatives." I've done that. I led the Looker migration. I trained the finance team on self-serve analytics. Frank agrees I meet the criteria. But the promotion cycle is once a year and there are two senior slots and four people eligible, and last year they promoted the person who'd been waiting longest, not the person who'd done the best work. Which was me. I know that sounds arrogant. I don't care. It was me.

The stress isn't the money. $12,000 is nice but it's not life-changing. The stress is the feeling that this career has a very low ceiling unless I become a manager or learn machine learning and rebrand as a data scientist. I don't want to manage people. I want to analyze data. And the market has decided that analyzing data is worth $90,000 and managing people who analyze data is worth $140,000. So the thing I'm good at is worth less than supervising the thing I'm good at.

Being ignored. I built a model last year that predicted which small business loan customers were likely to default within 90 days. Not a machine learning model, I'm not a data scientist. A logistic regression based on four variables: days past due on first payment, business age, industry code, and loan-to-revenue ratio. Simple. It flagged 23 accounts. I sent it to the commercial lending team lead, a guy named Garrett. He said thanks. Nothing happened.

Three months later, eleven of those 23 accounts had defaulted. $2.1 million in losses. Garrett's team did a post-mortem. They brought me in. I showed them the model. I showed them the email I'd sent three months earlier. The room went quiet. Garrett said, "I thought it was just a list." Just a list. He'd treated a predictive model the same way he'd treat a FYI email. Because to him, analytics output is noise until it's a crisis.

That's the stress for me. Not building the model. Building the model is fun. The stress is knowing that the model works, knowing it could prevent a $2.1 million loss, and having it sit in someone's inbox because they don't understand what they're looking at and I don't have the organizational authority to make them act on it. I can find the signal. I cannot make anyone listen.