Question 1 is worth 3 marks and asks you to explain one issue that limits what can be read from the data in a named source.
One issue. Not three.
That is the first thing to get right, because listing three limitations in a 3-mark answer usually scores lower than explaining one properly. The marks come from depth, not coverage.
What separates a 1-mark answer from a 3-mark answer
A 1-mark answer names a limitation. A 3-mark answer names it, explains the mechanism, and says what it does to what can be concluded.
Take the most common answer students give: correlation does not show causation.
On its own that is weak. It is true of every scatterplot ever printed, and it says nothing about the source in front of you.
Push it. Name the plausible reverse direction, or name a specific third variable, and say why that variable is credible here. That is the difference between one mark and three.
The same is true of every limitation below. The name is the cheap part.
Know what you are looking at
Each graph type is built to show one thing, which means each one hides specific other things. Knowing what a chart is for tells you immediately what it has concealed.
| Graph | Built to show | What it hides |
|---|---|---|
| Bar chart | A summary value across separate categories | Everything except that summary. Group size, spread, individual differences, anyone unusual |
| Bar chart with error bars | Summary values plus a layer of spread or uncertainty | Usually what the error bars actually represent |
| Histogram | The shape of a distribution across continuous intervals | Bin width choices can change the apparent shape |
| Line graph | Change across time or an ordered variable | Whether the same people were followed, and what happened between points |
| Scatterplot | The relationship between two measured variables | Direction of the relationship, and any third variable |
| Box plot | Median, quartiles, spread and outliers | The shape of the distribution inside the box |
| Distribution curve | The spread of scores | How much two groups actually overlap |
Every bar is a crowd of people squashed into a single number. That sentence, held in mind, generates most of what you need for question 1.
Error bars are the highest-value skill on this question
Most students do not know how to read them, which is exactly why they are worth learning.
An error bar is the line sticking out of the top of a bar or point. Before you say anything about it, you have to know which of three things it is.
Standard deviation shows the spread of scores in the sample. Standard error shows how precisely the mean has been estimated. A confidence interval is an interval estimate expressing uncertainty around the estimate.
They look identical on the page. They are different sizes and answer different questions. If the graph does not say which one it is, you cannot judge what the bars are telling you at all, and saying so is a complete 3-mark answer.
If the bars do overlap, that is a caution rather than a significance test. Overlapping bars mean the two groups share a range of plausible values, which weakens any claim that they are clearly separated.
A bank of limitations that transfer
You will not know what the booklet contains, but the problems repeat. These are the ones worth having ready.
No measure of spread. A mean with nothing around it tells you where the middle is and nothing about whether people clustered there.
No sample size given. A small sample gives a less precise estimate, and one unusual participant can shift the result substantially. Careful though: a large N does not fix a biased sample or a poor measure.
Group averages hide individual differences. A group average can describe nobody in the group. Some participants may show the opposite pattern, and the graph makes them invisible.
No inferential test reported. Without a test result or interval, you cannot judge whether a difference meets a statistical threshold. Whether it matters practically is a further question again, which no test answers.
Broad or vague categories. When the x-axis groups things into bands like one to three hours, someone at one hour and someone at three hours land in the same column.
Unclear operationalisation. If you do not know how engagement or wellbeing was measured, you cannot judge whether the numbers reflect the thing they claim to.
Cross-sectional snapshot. Data from a single point in time cannot tell you whether a pattern is stable, how it developed, or which thing came first.
Outliers. Extreme values inflate the range and pull both the mean and the line of best fit. A pattern may be driven by one or two unusual cases.
Distributions overlap substantially. Even if group averages differ, heavy overlap means many individuals in each group scored similarly. The difference between groups may be smaller than the differences within them.
No comparison group. Without one, you cannot tell whether the pattern is distinctive or completely typical.
Two that deserve their own explanation
The truncated y-axis. When a bar chart's axis does not start at zero, small differences fill the whole chart. Bar length is what a reader interprets as magnitude, so cutting the axis breaks that relationship. A y-axis starting at 70 can make a 1.7 percentage point gap look enormous.
Be careful applying this beyond bar charts. On a line graph or scatterplot, a non-zero range can be perfectly legitimate. A related and separate problem is unequal tick spacing, which distorts any graph type.
Skew. In a distribution with a long tail stretching one way, the mean gets dragged toward the tail, away from where most people actually are. If most patients attended a few sessions and a handful attended many, the mean overstates what a typical patient did, and the median would describe them better.
How to write the answer
Three sentences is usually enough.
Name the feature you can see in the source. Explain the mechanism, meaning what that feature does to the data. Then state what it means for what can be concluded.
Point at something actually visible in the graph. An answer that could have been written without looking at the source is a general observation about graphs, not an analysis of this one.
How to practise
Take any graph, from your textbook, a news article, anywhere. Give yourself sixty seconds to write one limitation with its mechanism and its consequence.
Do that ten times and the recognition becomes automatic, which is the whole point. On the day you have about ten minutes and you want to spend it writing, not searching.
If you want the full set of graph types with worked examples across the bands, that is what the PsychExcel Paper 3 Complete Guide is built for.