Collect Reliable Data and Reduce Error

A fair test design only matters if the data you collect is trustworthy. Reliable data is data that another scientist could collect again, with the same tools and the same method, and get a result very close to yours. To get there, you must plan how you will measure, watch out for the small mistakes that creep in, and repeat your trials so a single bad measurement does not fool you.

What "Reliable" Means

• Reliable means repeatable: do the same thing again, get the same answer (or very close).
• A single trial can be lucky or unlucky. Several trials reveal the real pattern.
• Reliable does not mean "what I want to see" — it means "what really happened, every time".

Common Sources of Error

Error is the gap between what you measured and the true value. Some error is random — a tiny wobble of the hand, a small breeze, a quick reaction time. Some error is systematic — a thermometer that always reads 2 degrees too high, or a ruler that starts at 1 cm instead of 0. Random error gets smaller when you repeat trials and average. Systematic error stays — you have to fix the source.

• Parallax error — reading a scale from above or below instead of straight on at eye level.
• Equipment drift — a tool that has slowly gone out of calibration since the last check.
• Biased sample — picking only the subjects you expect to give the answer you want.
• Single-trial trap — running the experiment once and trusting the result.
• Unit confusion — writing 12 instead of 12 cm (about 4.7 in), or mixing grams (1 g ≈ 0.035 oz) with kilograms (1 kg ≈ 2.2 lb).

How To Reduce Error

1. Use the same instrument throughout the experiment, and check it before you start.
2. Read scales at eye level so the line on the tool lines up with the mark on the scale.
3. Repeat each trial at least three times, then take the average to smooth out random error.
4. Pick subjects randomly so your sample reflects the real group, not just your friends.
5. Always write the unit next to every number — cm, g, sec — so nobody guesses later.

Why Averages Help

If you measure once and get 18 cm, you do not know whether that was lucky or accurate. If you measure five times and get 18, 17, 18, 19, 18 cm, the average of 18 cm tells you the real value is very close to 18. The variation between trials also tells you how much random error you have — large spread means be cautious, tight cluster means be confident.