A graph can be mathematically simple and conceptually misleading. The tallest bar might represent the largest total, the largest value per cell, or the largest change relative to a separate baseline. Those are different claims. Before deciding what a figure means, identify the quantity it actually displays.
Data interpretation is part of the MCAT's published scientific reasoning scope. This article uses original constructed data to practise a four-question reading routine: What is measured? Compared with what? In what units? With what uncertainty? AAMC content outline
Read the axis as a sentence
Translate an axis label into plain language. “Fluorescence, arbitrary units per viable cell” means each reported value has already been divided by a cell count. “Percent of vehicle control” means the reference condition defines the scale. “Change from baseline” describes a difference, not the final value.
Check whether the scale is linear or logarithmic. On a linear axis, equal distances represent equal numerical increments. On a logarithmic axis, equal distances represent equal multiplicative changes. Never infer the scale from the spacing of the plotted points alone; read the labels.
Then identify the groups. A colour may represent treatment, time, genotype, or a combination. A correct calculation between the wrong groups remains the wrong comparison.
Worked example: the denominator reverses the ranking
Researchers measure secreted product from three fictional cell cultures:
| Culture | Total product units | Viable cells | Product units per cell |
|---|---|---|---|
| A | 900 | 300 | 3.0 |
| B | 1,200 | 600 | 2.0 |
| C | 800 | 200 | 4.0 |
If the question asks which culture has the greatest total product, B is correct. If it asks which has the greatest output per viable cell, C is correct. Culture A is intermediate in the per-cell comparison even though its total exceeds C's.
The useful habit is to write the requested ratio before inspecting the options. Here it is product divided by viable cells. It is not viable cells divided by product, and it is not total product alone.
This table does not establish why C has greater output per cell. The cultures could differ in other ways. Normalization changes the quantity being compared; it does not automatically remove every confound.
Distinguish percentages from percentage points
A second constructed figure reports that 20% of control cells and 30% of treated cells express a marker. The absolute difference is 10 percentage points. The relative increase is (30 − 20)/20 = 0.50, or 50%.
Both statements can describe the same data, but they answer different questions. A distractor saying “10% increase” is ambiguous unless it clearly means percentage points. In your own working, label the quantity explicitly.
Now imagine the fraction falls from 30% to 20%. That is a 10-percentage-point reduction and a relative reduction of one-third, approximately 33%. The relative denominator is the starting value, so reversing a change does not preserve the percentage.
Do not invent a significance test from the picture
Error bars may display standard deviation, standard error, or a confidence interval. These quantities answer different questions. Read the legend before interpreting their length. A figure that does not identify the bars leaves an important uncertainty unresolved.
Standard deviation describes the spread of observations. Standard error describes sampling uncertainty in an estimate such as a mean. For independent observations from a common population, the usual standard error of the mean is SD/√n; repeated readings of the same sample cannot simply be substituted for independent n. A confidence interval adds a model-based range around an estimate. Small standard errors can therefore coexist with substantial variation among individual observations.
Overlap is not a universal significance rule. The interpretation depends on the kind of interval, the design, and the statistical comparison. A mean difference also need not be biologically important merely because it is statistically detectable. Cumming, Fidler and Vaux: Error bars in experimental biology
For a typical passage question, use an explicitly supplied test result where relevant. If no test or sufficient uncertainty information is provided, describe the observed pattern without declaring it significant or proving there is no effect.
A worked question about incomplete evidence
A fictional study reports mean product per cell of 3.0 in controls and 3.4 after treatment. The graph has no error bars, sample size, or statistical result.
Which statement is justified?
A. Treatment has no effect because the means are close.
B. Treatment reliably increases output in the population.
C. The reported treated mean is higher, but the supplied information does not establish the uncertainty of that difference.
D. Every treated cell produces more than every control cell.
C is best. It distinguishes the observed means from an inferential conclusion. A turns missing evidence into evidence of absence. B adds reliability that has not been demonstrated. D converts a group average into a claim about every observation.
This is not a reason to ignore the figure. You can still calculate the reported difference and describe its direction. The limit applies to claims the figure cannot support.
Check what the sample size counts
If three cultures are each measured four times, there are twelve readings but not necessarily twelve independent biological samples. The relevant unit depends on how samples were prepared and assigned. A legend should help you understand that design.
During practice review, note whether you misread the units, comparison, normalization, or inference. “Graph mistake” is too broad to guide the next exercise. A specific record, such as “I treated percent of each group's baseline as a common absolute scale,” points to a repair.
Try a fresh calculation
Two fictional populations start with 40 and 80 product units respectively. Both increase by 20 units. Calculate the absolute and relative changes before reading the answer.
Both have an absolute increase of 20. The first increases by 50%, while the second increases by 25%. A graph normalized to each baseline could therefore show different relative changes even when the absolute increases match.
Your next action is to take one figure and write two sentences: one that accurately describes its data, and one tempting conclusion it does not establish. If you can explain the difference, you are reading beyond the bar heights.
Sources and further reading
- AAMC: What is on the MCAT Exam? Content Outline — Data interpretation and scientific reasoning are within the published exam scope.
- Cumming, Fidler and Vaux (2007): Error bars in experimental biology — Independent replication and distinctions among error-bar types.