Still learning to read
Published on LinkedIn as an article on 2026-01-18, and announced in a post on 2026-01-18. The text below is the article as published.
I’m still learning how to read. I see this most clearly when I’m working with other learners, whether in math, statistics, or another field like epidemiology. When a learner meets a text, I meet it with them and take special care to reflect on how that goes.
I’ve been learning how to read since I was 3 years old. That’s when, according to family lore, my elder sister taught me how to read for the first time. I’ve continued learning how to read ever since. Not least when I was in my very first semester as an undergraduate, taking a New Testament course taught by a former Jesuit priest at a large, secular, state university. I often say that’s where I really learned to read, because the historical-critical method of exegesis labors to tease apart the layers of a text as a form of scholarly interrogation.
Suppose you’re in high school math, in a unit on mean and variance and reading a word problem where you need to apply a concept that is new to you:
The first 5 persons enrolled in a nutrition study weigh in at 149, 144, 148, 136, and 139 pounds, respectively. Suppose the sixth enrollee weighs 122 pounds. How does the mean of the first 6 compare to the mean of the first 5? The variance?
Some students will try to calculate the means and variances right away before attempting to answer the questions. That’s not wrong, but it is suboptimal, because there is more to learn here than the answers to 2 questions.
As a math tutor, I prescribe 3 steps for engaging a word problem:
What do you see? What are the words on the page? Pause before trying to make sense of them. This is the most surprising step, and it often requires the greatest discipline. I’ve seen time and again a learner try to move directly from the words on the surface to the solution. Then I keep the song cue to myself (“Slow down, you move too fast …”).
What are you being asked to do? What is the nature of the problem? What inputs are presented? Outputs requested? This step typically calls for some humility. How do I even know if I understand what is being asked of me, especially where the word problem involves a new-to-me concept.
What tools might help you do what is asked? How might you move from inputs to output? In some ways, this is the easiest of the 3 steps in structured learning. Having figured out what the words say and what is being asked, the learner can bring in other clues from the lesson. If you’re in a unit on mean and variance, then you have a pretty big hint that mean and variance formulas could figure into the solution. That doesn’t make it obvious, but the context clues narrow the solution space.
In the nutrition and weight example, you can answer both questions without applying any formulas: the mean weight goes down (because the new value is lower than all previous values, hence lower than the previous mean), and the variance goes up (because the new value is outside the range of the previous values). The formulas allow you to verify the answer. Both methods of solving the problem are useful, and each method shows understanding in a different way.
Now suppose that you’re reading a public health report that uses unfamiliar concepts.
[B]efore release of the CDC Guideline, the odds of receiving an opioid prescription increased 6.4% per year in [the most rural] counties (95% multiplicity-adjusted Wald CI = 2.1–10.8) …. [A]fter release of the CDC Guideline, the odds of receiving an opioid prescription decreased significantly in all county groups. [García et al. 2019]
This is a real-world problem in a public health report, not a word problem in a math class. This report doesn’t give context clues for what to figure out, so the learner has an extra step to make explicit.
What do you see?
What do you need to understand about what you see?
What claims are you being asked to assess?
What tools might help you assess those claims?
In this example, a learner needs to grasp the notions of a confidence interval, perhaps the Wald method, 95% confidence, multiplicity adjustment, annual percent change (APC) in odds, and comparisons between APCs before and after a specified event. Further context shows that these estimates come from piecewise linear logistic regression models. That can be a lot to unpack. It’s relatively easy to be put off by the text. It takes a bit of patience, and perhaps courage, to slow down and tease apart the elements. Indeed, the reader doesn’t need to understand every element to develop a working sense of the claims. They do, however, need every element and more to make complete sense of the claims.
Learning how to read a statistical claim starts with seeing the words, follows by figuring out what concepts and what claims need to be understood, and proceeds with sensing out that comprehension. All of this happens before actually assessing the claims and drawing conclusions. Even in data analysis, reading can be hard work. Especially in data analysis, reading is fundamental.