Three ideas cause more confusion in IB Psychology than any others, and all three are misused in roughly the same way: they get treated as verdicts when they are actually just descriptions.
A significant result is not automatically an important one. A strong correlation is not automatically a cause. And p is not the probability that a finding is wrong.
Getting these right is worth real marks on Paper 3, and it is the fastest way to sound like someone who understands evidence rather than someone repeating a phrase.
What p actually means
Researchers set a significance level, conventionally p < .05.
That means: if the null hypothesis and the assumptions of the test were true, data this extreme or more extreme would occur less than five per cent of the time. If the calculated p value falls below the threshold, the result is called statistically significant and the null hypothesis is rejected.
That is the whole claim. It is narrower than most people assume.
What p does not mean
It does not mean the effect is large.
It does not mean the effect is important.
It does not mean the result is certainly true.
And it does not mean that p is the probability the finding is due to chance. That last one is the most common misreading in student answers, and it is worth unlearning deliberately.
Significance and size are separate questions
This is the distinction that earns marks.
A very large sample can produce a statistically significant result for an effect so small it has no practical meaning. A small sample can miss a real and important effect entirely.
So a strong answer asks two questions, not one. Would a difference this large be unusual if there were really no effect? And is it big enough to matter?
Here is what holding both looks like:
The difference between conditions reached significance at p < 0.05, so if there were really no effect, a difference at least this large would be relatively unusual. However the means differ by less than one word out of twenty, so the practical importance of the finding is limited, and the researchers' claim that notifications meaningfully damage learning goes beyond what this effect size supports.
That paragraph does not reject the study. It calibrates it, which is the more sophisticated move.
Effect size
Effect size is how you answer the second question, and two measures come up.
Cohen's d describes how big a difference between two groups is. Roughly, 0.2 is small, 0.5 medium, 0.8 and above large.
The correlation coefficient r describes the strength and direction of a relationship, running from −1 through 0 to +1. The sign tells you direction. The distance from zero tells you strength.
If a source reports a significant result and no effect size at all, that absence is itself worth a sentence, because you cannot judge whether the finding matters.
Reading r honestly
An r value tells you strength and direction. It does not tell you significance, and it does not tell you cause.
It also depends heavily on sample size in ways that are easy to miss. A correlation of 0.9 in a sample of five people is far less convincing than 0.4 in a sample of five hundred, because small samples produce unstable coefficients.
So do not read r as a measure of quality. Read it alongside N.
Correlation and causation, done properly
Everyone knows the phrase. Almost nobody uses it well.
"Correlation does not show causation" is true of every scatterplot ever printed, which is exactly why it earns almost nothing on its own. It says nothing about the study in front of you.
To make it a real point, do one of two things.
Name the reverse direction and say why it is plausible. If a study finds that exercise is associated with better mood, say why lower mood would reduce exercise, not just that it might.
Name a specific third variable and say why it is credible here. Not "there could be a third variable" but the actual one, and the reason it would produce this exact pattern in this exact sample.
That is the difference between one mark and three on Paper 3 question 1, and it is the same difference in Paper 2 Section B.
Type I and Type II errors
A Type I error is a false positive: rejecting the null hypothesis when it is true. Claiming an effect that is not there.
A Type II error is a false negative: failing to reject the null when it is false. Missing an effect that is really there.
The two trade off against each other. Lowering the significance threshold reduces Type I errors and increases Type II errors.
Small samples, insensitive measures and ceiling effects all increase the risk of Type II errors, which is a useful thing to say about a study that reports no significant difference. No difference found is not the same as no difference existing.
Confidence intervals
A confidence interval is an interval estimate expressing uncertainty around a result. Under repeated sampling, 95% of intervals built by a 95% procedure would contain the true value.
The practical use in an exam is simple. A wide interval means the estimate is imprecise, so a single reported number should not be treated as exact. If two groups' intervals overlap substantially, that is a caution about how clearly separated they are, though overlap is not itself a significance test.
Where each of these gets examined
Paper 3 question 1 often turns on a graph with no inferential test reported, or a scatterplot inviting a causal reading it cannot support.
Paper 3 question 2 asks you to analyse findings and state a conclusion linked to the claim, which almost always means checking whether the researchers' verb matches their design and whether their effect size supports their language.
Paper 2 Section B can hand you an unseen study reporting any of this. No Paper 2 question at either level asks you to calculate or formally interpret statistics, but reading accurately what a study reports is what makes an evaluation specific rather than generic.
The habit worth building
Whenever you meet a reported result, ask three things in order.
What exactly was found. How confident can we be that it is not noise. And is it big enough to matter?
Most students answer the first, some answer the second, and the ones who answer all three are the ones whose conclusions get called calibrated rather than descriptive.
If you want the full notation decoder and worked results tables, that is what the PsychExcel Paper 3 Complete Guide is built for.