Reliability, validity and accuracy are three different words

What each of the three actually means

Reliability is about repeating. An investigation is reliable when doing it again, the same way, gives you close to the same result. It is a property of the method, and the way you improve it is to repeat trials and average them.

Validity is about whether you measured the thing you set out to measure. An investigation is valid when the method tests the hypothesis, one variable changes, the rest are controlled, and there is something to compare against. It is a property of the design, and repeating an invalid experiment gives you the same wrong answer more precisely.

Accuracy is about closeness to the true value. It lives in the instrument and in how carefully it was read. A stopwatch started late is inaccurate no matter how many times you repeat the run and no matter how well the experiment was designed.

• Reliable - repeating it gives the same result. Fix by repeating.
• Valid - it tests what it claims to. Fix by controlling variables.
• Accurate - close to the true value. Fix with a better instrument.
• Precise - the readings agree with each other. Not the same as accurate.

Telling which one the question is about

The question almost always names one of them, and the mark is for a sentence about that one. If it says reliability, talk about repeats and consistency. If it says validity, talk about the variable you changed, the ones you held constant, and the control.

Where people lose the mark is by writing a paragraph that mentions all three and commits to none, on the theory that one of them must be right. A marker reading that sees somebody who cannot tell them apart, which is the thing being tested.

And when a question asks you to improve an investigation, improve the one it named. Adding more trials to an experiment with an uncontrolled variable is a longer experiment with the same fault, and saying so is often worth more than the improvement itself.

Accurate and precise are not synonyms either

Precision is about how close repeated measurements are to each other. Accuracy is about how close they are to the truth. They come apart in both directions, and the example is worth keeping.

A scale that reads two hundred grams too heavy every time is precise and not accurate: every reading agrees with every other reading, and all of them are wrong by the same amount. A scale that wobbles either side of the true value is accurate on average and not precise.

Which matters depends on what you are doing. A systematic error - the scale that is always heavy - does not average out however many times you repeat, and it is the one that repeating cannot fix. A random error does average out, which is exactly why repeating is the fix for reliability and not for accuracy.

Naming the variables properly

The independent variable is the one you change. The dependent variable is the one you measure. The controlled variables are the ones you keep the same, and they are the part of the answer people leave out.

Name them specifically. Temperature is not a controlled variable; the water being kept at twenty-five degrees is. A list of vague nouns reads as somebody who has memorised the categories without applying them to this experiment, and it is easy to tell the difference.

The control is not the same as a controlled variable, and mixing them up is a common and costly slip. A control is a run with no treatment applied, there to show what happens anyway. A controlled variable is a condition held constant across every run.

What a conclusion may claim

A conclusion answers the hypothesis in one sentence and says what the evidence was. It does not claim more than the data supports, and the marks are often in the limit rather than in the claim.

Correlation in your results is not causation, and saying so when it applies is worth a mark. So is naming what your investigation could not rule out: a single trial, one range of values, one set of conditions. An answer that states its own limits reads as somebody who understands the method, which is the whole subject.

And if the data did not support the hypothesis, say that. An investigation that disproves its hypothesis is a successful investigation, and rewriting the conclusion to agree with what you expected is the one thing this course exists to train out of people.

Where it usually goes wrong

Five habits, none of them about knowing less science than the person next to you.

• Using reliable and valid as if they were the same word.
• Improving reliability when the question asked about validity.
• Listing vague controlled variables instead of specific conditions.
• Confusing a control with a controlled variable.
• Writing a conclusion that claims more than one trial can support.

What to practise next

Take five investigations you have already written up and, for each one, write three separate sentences: one about its reliability, one about its validity, one about its accuracy. If any two of the three sentences could be swapped between investigations, they are not saying anything specific yet.

Then take one of them and write the improvement paragraph twice - once improving reliability and once improving validity. Doing it twice is what makes the difference between the two stop being something you have to think about.

Check yourself

Three questions on what is above. Have a go before you open them - reading an answer you have not tried to give is the version of this that does nothing.