Science: Experiments

This is Unit Twenty, part two, of ACT Preparation. Part one closed on the placebo; this part adds blinding and study types, then reads research summaries the way the science section presents them. The habit carries over: name what was changed, what was measured, and what was held the same, before reading any result. By the end you will be able to tell an observational study from a randomized experiment. To explain why repeated trials matter. To compare two experiments by what changed. To predict another trial and modify an experiment. And to decide whether results support a hypothesis.

Four entries on study design: blinding, study types and repetition.

Blinding.

ACT 1918 Blinding

In a blinded study, subjects do not know which treatment they receive; in a double-blind study, the researchers working with them do not know either, as when neither patients nor nurses know whose pill is real. Researchers who know can, without meaning to, judge subjects differently; double-blinding prevents that.

Observing, or experimenting.

ACT 1919 Observational Study Versus Experiment

An observational study records what happens without assigning treatments; an experiment assigns them. Surveying cyclists about helmets is observational; assigning two helmet designs at random is an experiment. Observational studies can show association, not cause, because confounding variables remain possible.

Classifying a study, step by step.

ACT 1920 Experiment or Observational Study?

Ask whether the researchers assigned the condition or found subjects already in it, then whether assignment was random. In invented study one, a survey of 500 adults finds coffee drinkers sleep less: observational, association only. In invented study two, 100 adults are randomly given coffee or decaf each evening: a randomized experiment, so a difference in sleep can be attributed to caffeine. Claiming an observational study proves a cause overstates it; claiming an experiment shows only association understates it.

Repetition.

ACT 1921 Repeated Trials and Sample Size

Repeating trials and using more subjects make results more reliable, because chance variation averages out: averaging three bounce measurements smooths one unlucky drop. Repetition reduces chance error; it cannot fix a biased design.

Seven entries on research summaries, ending with a full one.

Reading a research summary.

ACT 1922 Reading a Research Summary

A research summary describes one or more related experiments, each with its procedure and results, usually in a table or graph. Experiment one might vary temperature at fixed light, experiment two light at fixed temperature. Read what was done before the numbers; questions ask about design as well as results.

Comparing experiments.

ACT 1923 Comparing Experiments

When a passage gives several experiments, the difference between them is often one change in procedure. Experiment two repeating experiment one with salt water instead of fresh traces any change in results to the salt. Read each new description for the one thing that is new.

Two experiments together, step by step.

ACT 1924 Reading Two Experiments Together

State each experiment's variables and controls, find what changed between them, and compare results only where other conditions match. In an invented ramp study, a ball travels 1.2 meters on wood and 0.8 on carpet at twenty degrees, and 1.8 and 1.2 at thirty degrees: on each surface, the steeper ramp sent it farther. Comparing 1.2 on wood at twenty degrees with 1.2 on carpet at thirty mixes two changes and shows nothing.

Predicting another trial, step by step.

ACT 1925 Predicting Another Trial

Find the trend as the independent variable changes, place the new condition within or beyond the tested range, and predict from the trend. On wood, twenty degrees gave 1.2 meters and thirty gave 1.8, so twenty-five degrees should give about 1.5. A choice of 2.0 would exceed the thirty-degree result and break the trend. Check a prediction against both neighbors.

Modifying an experiment, step by step.

ACT 1926 Modifying an Experiment

State the new question as a factor and a result, make that factor the only independent variable, keeping every other condition, and keep a control. To test whether water matters in the bean study, keep every pot under white light and give pots fifty, one hundred and 150 milliliters a day. A design that also varies light color answers neither question. The best modification is the smallest change that isolates the new factor.

Does the result support the hypothesis, step by step.

ACT 1927 Does the Result Support the Hypothesis?

State what the hypothesis predicts for this experiment, compare the actual results, and check that the answer's reason matches the data. The invented hypothesis that heat, not light, wilts seedlings predicts that shaded seedlings in full light stay upright; they did, and heated seedlings in low light wilted, so the results support it. Choices saying "proves" overstate, and choices citing another trial's data misread the evidence.

The whole unit in one research summary.

ACT 1928 A Research Summary

Research summaries test experimental design as much as results. In an invented passage, students stirred ten grams of sugar into 200 milliliters of water at twenty, forty, sixty and eighty degrees, and it dissolved in ninety, sixty, forty and twenty-five seconds, each an average of three trials. In experiment one, temperature is independent, dissolving time dependent, and the sugar, water volume and stirring controlled. Experiment two repeated forty degrees without stirring, so its 180 seconds against sixty shows the stirring's effect. At fifty degrees with stirring, the time should fall between sixty and forty, about fifty. And the hypothesis that warmer water dissolves sugar faster is supported, because times fell steadily as temperature rose. Name the variables first, compare only matched conditions, and say supports, never proves.

Look at what you can now do. You can explain blinding and tell an observational study, which shows association, from a randomized experiment, which tests cause. You can explain why repeated trials make results reliable, and why they cannot fix a biased design. You can compare experiments by the one thing that changed, matching conditions first. You can predict another trial from the trend and modify an experiment with the smallest isolating change. And you can decide whether results support a hypothesis. Unit twenty-one judges explanations: models and competing viewpoints.