How I Learned to Think Like a Statistician

The three lessons life taught me about statistics that classes never did

When I first studied statistics in high school, it appeared to me not as a revelation, but as a jumble of bizarre formulas that defied explanation. I could not for the life of me make sense of it. The calculations were mechanical and joyless. Compared to the elegance of physics, these equations felt unappealing, artificial, and just plain wrong. There was no reasoning to follow, only rules to memorize. I decided with absolute finality: statistics was not for me.

Bias Has a Direction

Later, during my undergraduate years, I began working with a mathematician who had turned himself into a statistician. He was newly interested in artificial intelligence and had brought me on to help with programming. Since it was not statistics, I felt comfortable that I could contribute. That margin of comfort mattered because it lowered my guard. I wasn’t on the defensive, worrying about formulas I didn’t understand. I could simply watch how he worked, and in that space I began to absorb ideas about statistics without even realizing it.

The turning point came at a conference. I watched as he and his wife, who was also a statistician, questioned a student about his poster. They were polite but relentless. For every result he presented, they asked whether it might be biased. Then they pressed further: in what direction would the bias move the estimates? Would it push the numbers too high, or too low?

That line of questioning stopped me in my tracks. Until then, bias had seemed to me like a stain that ruined the work entirely. Either a result was biased or it was not. But what they demonstrated was something deeper: the recognition that bias has consequences, and those consequences can be reasoned through. To dismiss a study as biased is easy. The real test is to ask what the bias implies, whether it understates or overstates the truth.

Imagine, for instance, an analysis that concludes a city will be underwater in five years. Suppose you learn the study is biased. It might be tempting to dismiss it outright. But what if the bias points toward underestimation? That would mean the danger is even closer at hand, and the need for alarm even greater. Bias does not always weaken a conclusion. Sometimes it strengthens it.

This was my first lesson that statistics was not just about formulas, but about disciplined judgment. It was the beginning of a habit of thought that has stayed with me. Years later, during the Covid-19 pandemic, that same lesson shaped how I interpreted evidence and judged risk.

Truth as an Aesthetic Standard

My second lesson came from reading Edward Tufte. I was open to Tufte’s message because it did not read like statistics. It looked like design, even art, and that made the ideas all the easier to take in. His books on graphs and charts were not mere textbooks but works of art. The elegance of the visualizations, and the passion evident in every page of text, argued with unexpected force that to present data sloppily was not only a logical failure but also an ethical and aesthetic one. To mislead with a chart was to betray the truth. To clutter a figure was to tolerate ugliness.

It reminded me more than anything of how some people feel about grammar. For them, grammar is not just a set of rules but a reflection of character. A missing comma is not merely a technical mistake, it marks a tolerance for disorder, a willingness to live with imprecision. To such people, grammar is a moral, aesthetic, and professional concern all at once. It signals to others the standards by which a person orders their mind. Tufte made me see graphs in the same light. A poorly-made plot was not just technically flawed, it was an admission of lowered standards, a sign of what the analyst was willing to accept in their work.

Uncertainty Matters Most

The third lesson came when I worked as a data scientist in a cancer biology group. I did not think of myself as a statistician. I was comfortable in the role because it was just working with data, not statistics. I was plotting results, exploring patterns, interpreting them in the ways any intelligent person could. That was safe ground for me. I thought of myself as a programmer who supported the scientists, by making the data more visible to them, not as someone doing any serious thinking myself.

One day I was discussing with my supervisor our estimates of cancer mutations in patient data. I cared deeply about the cancer patients and I wanted to do a good job. I was sure we were handling the uncertainty badly. Some estimates are always more certain than others, and in my view our methods underestimated the true uncertainty. I lost that argument. My concerns did not change the project one bit. I was overruled.

But something in me had shifted. My frustration was not about coding or visualization, it was about the way numerical evidence was being judged. I realized that the unease arose out of a deeply statistical concern. I wanted the numbers to mean something scientifically, not just exist as sloppy guesstimates in a table. All my separate insights about the understanding of data had been converging to a single point. I was no longer just making data easier for others to see. I was starting to think systematically about how best to determine what the data meant. I was no longer just a programmer helping out the scientists. I had begun to think like a statistician.

Less than a year later, I quit to begin a statistics PhD at Harvard.

What I Learned

It was not a single revelation but a gradual unfolding. Statistics was never about the pile of formulas that I hated in high school. It was about asking sharper questions, about refusing to accept vagueness, about telling the truth even when it is uncomfortable.

Once I saw that, I saw that statistics had been part of my life all along, just waiting for me to recognize it.