The Scientific Method as Scientists Actually Use It

The flowchart is a teaching aid Every science classroom has the poster: question, hypothesis, experiment, analysis, conclusion, arrow back to the top. It is a useful scaffold and a poor description of research. Real investigation is messier — hypotheses change mid-study, unexpected results redirect the question, and much of science is observational rather than experimental. Astronomy has produced enormous knowledge without ever assigning a star to a control group.

What survives the simplification is the part that matters: a claim is scientific when it could have come out otherwise, and when someone else can check.

Hypotheses must forbid something A hypothesis that is compatible with every possible outcome tells you nothing. "Plants respond to light" is not testable in a useful way. "Bean seedlings grown under blue light will grow taller than those under red light over fourteen days" forbids a specific result, and that is what makes it worth testing.

The strongest hypotheses make quantitative predictions. Predicting "taller" is fine; predicting "roughly 20 percent taller" gives you far more information when you are wrong.

Variables, in plain terms The independent variable is what you change on purpose. The dependent variable is what you measure. Controlled variables are everything else you deliberately hold constant so that they cannot explain your result.

The classic student error is changing two things at once. If the blue-light group also sat closer to a window, you have no way to attribute the difference. Every uncontrolled difference is an alternative explanation, and the entire value of an experiment lies in eliminating those.

What a control group is for A control group is the version of the experiment where the thing you care about is absent. It exists to answer "what would have happened anyway?" Seedlings grow over fourteen days regardless of light colour; without a baseline you cannot tell how much of the growth your treatment caused.

In human studies the control usually gets a placebo, because expectation alone changes outcomes measurably. In double-blind trials, neither the participant nor the person taking measurements knows the assignment — this exists because researchers unconsciously measure more generously in the group they hope will win.

Sample size and why one trial proves nothing Biological and physical systems vary. One seedling can be unusually vigorous. Repeating the measurement across many units lets you separate signal from noise, and it is why every result should come with a spread — a standard deviation, a range, error bars — not just an average.

A rough rule for classroom work: at least five units per condition, three repetitions of the whole experiment where time allows. Report the variation honestly. An experiment with wide overlap between groups has not shown a difference, and saying so is a legitimate result.

Correlation, causation, and the third variable Two things moving together can mean A causes B, B causes A, both are caused by C, or the pattern is coincidence. Ice cream sales and drowning deaths correlate strongly; the third variable is summer. The only reliable way to establish causation is intervention — you change one thing and hold the rest steady — which is precisely why controlled experiments carry more weight than observational data.

Writing up so someone can check you A method section should let a stranger repeat your work without asking you a question. Quantities, durations, equipment, exact procedure. If you cannot write that, you did not run a defined experiment.

Report results before interpreting them, and keep the two visually separate. Then state limitations plainly: what you could not control, where measurement was imprecise, what you would change. Acknowledging limits is not weakness in a lab report — it is the part that shows you understand what your data can and cannot support.