Hypotheses and Judging a Good Experiment

A hypothesis is a proposed, testable explanation for an observation. A prediction states what you expect to observe if that explanation is useful. An “if, then” statement can express a prediction, but the wording alone does not make an idea testable. Specify what you will measure and what outcome would challenge the explanation.

Start with the explanation you want to test, then design a comparison that could distinguish it from a reasonable alternative.

What Makes a Hypothesis Testable

A useful hypothesis leads to measurable predictions. For a simple experiment, specify the factor you change and the response you measure. “If plants get more sunlight, then they will grow taller” is testable: you can change the sunlight and measure the height. Compare that to “plants like sunlight,” which is vague and cannot be measured directly.

Notice the shape: if [independent variable changes], then [dependent variable responds]. The structure helps organize a prediction, but you still need measurable definitions, a feasible comparison, and a result that could challenge the proposed explanation.

Judging a Good Experiment

An experiment is trustworthy when it does a few things well. In a simple controlled experiment, it deliberately changes one factor at a time. Factorial experiments can vary several factors systematically, using enough comparisons to separate their effects. It keeps all other conditions the same. It uses a control group for comparison. And it tests enough subjects, and repeats the trial, so a single fluke cannot decide the result. When an experiment skips one of these, the test may ask you to spot the weakness.

So when you evaluate an experiment, run down the checklist: a comparison that separates the factors of interest, relevant conditions controlled, and enough independent observations. Identify the weakness that matters for the stated conclusion. A study may have several limitations; the most important one is the limitation that prevents the claimed inference.

Bigger Samples, Repeated Trials

A two-plant comparison gives very little information about normal variation. One plant may differ from the other for reasons unrelated to the treatment. Using many plants in each group, and running the test more than once, makes the outcome far more believable. Sample size should be judged alongside variability, study design, and the size of the effect being measured. Larger samples and independent repetitions reduce some kinds of uncertainty. They do not repair biased sampling, a faulty measurement, or an uncontrolled confounding variable.

What would a fair seedling comparison look like?

Suppose a class proposes that additional light increases bean-seedling growth. Use seedlings of the same variety and similar starting size. Randomly assign several plants to each light condition, hold soil and watering as consistent as practicable, and measure the change in height over the same interval. Record every plant rather than reporting only the tallest.

If all brighter-light plants also receive fertilizer, light and fertilizer are confounded. Extra growth could be due to either factor or their interaction. Adding more plants to that same flawed arrangement would improve precision without separating the causes. A redesigned comparison is needed.

Watch: A Short Video Lesson

Watch how an explanation becomes a measurable prediction. Pause before each example is completed and identify what would be changed and what would be measured.

A Routine for Hypothesis and Design Questions

1. Write predictions as “if [I change this], then [this will happen].”
2. Check that the prediction names something you can measure.
3. To judge an experiment, confirm: a comparison that separates the factors of interest, relevant conditions controlled, and enough independent observations.
4. If asked for the flaw, look for the missing item on that checklist.

Practice

1. Rewrite “exercise is good for the heart” as a measurable if-then prediction.
2. Name two features of a well-designed experiment.
3. Why are two subjects usually not enough?
4. An experiment compares only two groups: one has dim light and sandy soil; the other has bright light and clay soil. Why cannot it separate the effects of light and soil?
5. What does a control group let you do?
6. How does a prediction differ from a hypothesis?

Answers

1. Sample: “If a person exercises more each week, then their resting heart rate will decrease.”
2. Any two of: isolates the effect being studied, controls relevant conditions, uses an appropriate comparison, includes sufficient independent subjects, repeats the study.
3. One unusual subject could swing the result; more subjects reduce that risk.
4. Light and soil are confounded in this two-group comparison. Separate or factorial comparisons would be needed to estimate their individual effects.
5. Compare the treatment with an appropriate baseline, which may receive no treatment, a placebo, or an established treatment.
6. A hypothesis proposes an explanation; a prediction states a measurable result expected under that explanation.

Where This Fits in Your Science Prep

This lesson ties together variables and controls into the full picture of a trustworthy experiment. It also connects to judging data quality through sample size and repetition. See all topics on the Science Topics Hub.