Most people don’t misread statistics because they’re bad at math, they misread them because no one ever explained what a p-value or a confidence interval actually promises. In this article, I have simplified five ideas that let you read any health or research finding with real confidence.

Know What Kind of Data You’re Looking At

Every variable in a study is either a category or a number. Blood type, smoker status, and pain severity (e.g., mild, moderate, severe) are categories. Height, age, and blood pressure are numbers.

Numbers split further into two kinds: discrete (whole counts, like number of ER visits – you can’t have 2.5 visits) and continuous (any value in a range, like weight – 68.3kg is perfectly valid).

This distinction isn’t academic trivia. It decides which summary statistics and which charts are even valid. You can’t average “pain severity” the way you average height in centimeters, because the “distance” between mild and moderate pain isn’t a fixed, equal unit the way a centimeter is.

Before you trust any claim about a number in a study, ask: what kind of data produced it?

Descriptive Statistics: Mean, Median, and the Outlier trap

Mean, median, and mode are three different ways of answering “what’s typical?” The mean is the mathematical average. The median is the middle value when everything is sorted. The mode is whichever value shows up most often.

When data has no extreme outliers, all three roughly agree. The trouble starts when one or two extreme values show up.

Take a waiting room with patients aged 22, 24, 25, 26, and 27. The mean is 24.8, the median is 25 – basically the same story. Now swap that last patient for a 90-year-old: 22, 24, 25, 26, 90. The mean jumps to 37.4, even though no one in the room is actually “37-ish.” The median barely moves, staying at 25 – a far more honest description of a typical patient.

This is exactly why hospital bills, ER wait times, and income are usually reported as medians, not means. This is because, a small number of extreme cases can drag a mean far away from what’s actually typical. Whenever you see an average, ask whether a few outliers might be doing the dragging.

What a p-value actually tells you

A p-value answers one narrow question: if a treatment truly did nothing, how surprising would data like this be?

Think of it like a coin flip. Flip a coin 10 times and get 9 heads – a fair coin could produce that, it’s just surprising. A p-value measures exactly that kind of surprise, assuming the “null hypothesis” (nothing is really going on) is true.

A p-value below 0.05 is simply a conventional line for “pretty surprising.” It is not a magic truth detector. Three myths worth retiring for good:

• “p < 0.05 means there’s a 95% chance the drug works.” False – the p-value says nothing about the probability a hypothesis is true. It’s about how surprising the data would be if the drug did nothing.
• “p > 0.05 proves the drug has no effect.” False – absence of evidence isn’t evidence of absence. The study may simply be too small to detect a real effect.
• “Statistically significant means clinically important.” False – a large enough sample can make a trivially small, meaningless difference statistically significant.

Confidence Intervals: The Honest Range

A point estimate is your single best guess. A confidence interval is the honest range of guesses that are also plausible, given the data.

Picture it as a fishing net instead of a single hook. A 95% confidence interval means: if you repeated the study 100 times and cast a net like this each time, about 95 of those nets would contain the true value. It is not “a 95% chance the true value is in this specific range” – that 95% describes the method’s long-run reliability, not this one interval.

Two things to check every time you see a confidence interval:

• Width – a narrow interval means a precise, trustworthy estimate. A wide one means real uncertainty remains, even if the average sounds impressive.
• Does it cross the “no effect” line? – zero, for a difference, or one, for a ratio like relative risk. If it crosses that line, the result is not statistically significant, no matter how promising the point estimate looks.

Two trials can report the identical average effect and tell completely different stories once you look at their intervals – one precise and clearly real, the other too uncertain to rule out “no effect at all.”

Reading a Results Section Like a Pro, and saying it in plain language

Put the first four ideas together and you get a simple checklist for reading any results section:

1. Who was studied, and how many? A result from 40 people in one hospital generalizes very differently than one from 4,000 people across 20 sites.
2. What’s the effect size, not just the p-value? “Significant” doesn’t tell you if the effect is big enough to matter.
3. How wide is the confidence interval? Precision matters as much as the headline number.
4. Does statistical significance mean clinical significance here? Ask explicitly – even a real effect might be too small to change a decision.

The last step is the one people skip: saying it in plain language without losing the honest uncertainty. A results section might report “a mean reduction in systolic blood pressure of 8 mmHg (95% CI: 5–11 mmHg; p < 0.001).” Translated for a patient, that becomes: “On average, this treatment lowered blood pressure by about 8 points, and we’re fairly confident the true effect is somewhere between 5 and 11 points – not just a fluke.”

Same facts. Same honesty about uncertainty. Zero jargon. That’s the actual skill – not just understanding a finding, but being able to hand it to someone else intact.

Need Help With Your Own Data?

oores Analytics® offers affordable biostatistical consulting, data analytics, and hands-on R training – with a free initial consultation on every project. Get in touch or email info@ooresanalytics.com to talk through your research.