Common mistakes · Biology · Chemistry · 9 min read
15 IA mistakes that cost marks in Biology and Chemistry, with before-and-after fixes
The fifteen failure patterns that pull science IAs down, shown as what students write next to what earns the marks, so you can find and fix them in your own draft.
On this page
Science IAs usually lose marks in a predictable set of places: a research question that is too broad, too few levels or repeats, raw data that is never processed, uncertainties that appear once and vanish, misread R2 values, a conclusion with no numbers, an evaluation built on "human error", and a word count blown on background. Each mistake below is shown as a before-and-after pair, grouped by the criterion it hurts most, so you can check your own draft against it.
This guide is about failure patterns. If you want to know what each criterion rewards and how marks are given, read the science IA criteria explained first.
Which mistakes start in the research design?
1. A research question too broad to answer
A broad question produces a vague method, and a vague method produces data that answers nothing in particular.
Loses marks"How does temperature affect enzymes?"
Earns marks"How does temperature (20–60 °C, 10 °C intervals) affect the initial rate of oxygen production by catalase in potato extract, measured with a gas syringe over the first 30 s?"
The fix names one enzyme, one source, one range and one way of measuring. See how to write an IA research question for more before-and-after examples.
2. Too few levels or too few repeats
Three levels cannot show the shape of a relationship, and one run per level gives no idea of the spread.
Loses marksThree concentrations (low, medium, high), one run each.
Earns marksSix concentrations, more closely spaced where the pilot showed the rate changing fastest, five repeats each, with the number of repeats justified by the spread seen in the pilot.
The IB sets no minimum. Five or more levels and three to five repeats is a sensible floor in the lab; fieldwork usually needs more samples. If time is short, fewer levels done well beats many levels done once.
3. A borrowed method with no reasons
A method copied from a textbook or video, with no explanation of why it suits this question, earns little in Research design, however carefully it was carried out.
Loses marks"The method was adapted from [website]. 1. Measure 10 cm3 of hydrogen peroxide…"
Earns marks"A gas syringe was used rather than counting bubbles because bubble size varied by a factor of two in the pilot. Extract was equilibrated in the water bath for 5 minutes because the pilot showed a 6 °C drop when cold extract was added to warm substrate. The procedure is based on [cited source], changed as follows…"
4. Safety, ethics or environment missing
Where these issues are relevant, considering them is part of Research design. Leaving them out loses marks; a paragraph of generic lab rules does not gain any.
Loses marks"Safety goggles and a lab coat were worn at all times."
Earns marks"The thiosulfate–acid reaction releases sulfur dioxide, a respiratory irritant, so volumes were kept to 50 cm3, the room was ventilated and each mixture was rinsed away immediately after timing. Waste was neutralised before disposal."
For human participants, write down consent, the right to withdraw and anonymity. For fieldwork, write down permission and how you avoided damaging the site.
What goes wrong in data analysis?
5. Raw data that is never processed
Averaging is not processing. The data has to become a quantity that answers the research question.
Loses marksA table of times to end point and their means, then a graph of mean time against temperature.
Earns marksTimes converted to rates (1/t), plotted against temperature, with an Arrhenius plot giving an activation energy, and one fully worked calculation shown.
Ask yourself: what single number or relationship answers my question? Then make sure your processing produces it.
6. Uncertainties that appear once and disappear
Writing ±0.05 cm3 in a table header and never using it again earns almost nothing in Data analysis.
Loses marksUncertainties in the raw-data headers; none in the processed data, graph or conclusion.
Earns marksUncertainties carried into the processed values, shown as error bars, used to estimate the uncertainty in the gradient, and quoted in the conclusion when the result is compared with a published value.
The rules for propagating them are in our guide to uncertainties in the IA.
7. The wrong kind of graph
A bar chart of a continuous independent variable hides the shape of the relationship and makes a trendline meaningless.
Loses marksA bar chart of mean rate at 20, 30, 40, 50 and 60 °C, default spreadsheet title, no error bars.
Earns marksA scatter graph with temperature on the x-axis, rate with units on the y-axis, error bars defined in the caption, and a best-fit curve chosen to match the expected shape.
Use a bar chart only for categories, such as different species or brands. More in graphs for the IA.
8. Misreading R2
R2 tells you how closely a fitted line follows the plotted points. It is not a measure of accuracy, it does not show cause and effect, and a high value on a straight line can hide a curve.
Loses marks"R2 = 0.95, so the results are 95% accurate and prove that rate is proportional to concentration."
Earns marks"A straight line fits the means closely (R2 = 0.95), but the residuals are positive at both ends and negative in the middle, so the relationship curves. Proportionality would also need the line to pass through the origin, and the intercept is 0.012 ± 0.004 s−1, which excludes zero."
9. Anomalies deleted or ignored
Silently dropping a result looks like data manipulation; ignoring an obvious outlier distorts your means.
Loses marksThe 40 °C mean includes a run of 112 s when the other four were 45–49 s, with no comment. Or: the run simply is not in the table.
Earns marksAll five runs in the raw table; the 112 s run flagged against a rule stated in the method (more than twice the median of the other runs), the mean shown with and without it, a likely cause (the tube was not pre-warmed, noted at the time), and the run repeated.
Why do conclusions lose marks?
10. A conclusion that never quotes the data
If your conclusion could have been written before the experiment, it is not consistent with your analysis.
Loses marks"As the concentration increased, the rate increased, which supports my hypothesis."
Earns marks"Rate increased from 0.012 ± 0.001 s−1 at 0.25% to 0.051 ± 0.004 s−1 at 1.5%, then rose by less than its uncertainty up to 3.0%, consistent with the active sites becoming saturated above about 1.5%."
11. No scientific context, or context nobody can trace
The Conclusion criterion asks for a comparison with accepted science, and the sources have to be traceable.
Loses marks"This agrees with collision theory." Or a literature value with no source.
Earns marksYour value set beside a published value with a full citation, the percentage difference compared with your percentage uncertainty, and the theory used to explain any gap.
Finding and citing sources properly is covered in sources for the IA.
12. Claiming more than the data show
"Proves", "always" and "causes" are rarely justified by one school investigation, and fieldwork correlations never show cause on their own.
Loses marks"This proves that soil moisture causes taller plants."
Earns marks"Plant height correlated with soil moisture (rs = 0.82, p < 0.01, n = 18) across the 20–55% range sampled. Because light also varied along the transect, the data cannot separate the two effects."
What makes an evaluation weak?
13. "Human error" and other generic weaknesses
Weaknesses that could be written about any experiment sit at the bottom of the Evaluation criterion.
Loses marks"There may have been human error. The equipment may not have been accurate. The sample size was small."
Earns marks"The end point was judged by eye. At 60 °C the runs lasted about 10 s, so a 0.5 s reaction-time error is about 5% of each reading, against about 1% at 20 °C. The high-temperature points, which set the slope, are therefore the least reliable, and this was the largest weakness."
Name the step, say what it did to these numbers, and rank it against the others.
14. Improvements that do not match the weaknesses
An improvement earns credit when it is realistic and fixes the weakness you identified.
Loses marks"Use more accurate equipment and do more repeats." Or: "Use a mass spectrometer."
Earns marks"Replace the visual end point with a light sensor logging every 0.1 s, which removes the reaction-time error at the short high-temperature runs. Pre-warm the extract in the same bath for 5 minutes to remove the temperature drop on mixing."
Which write-up mistakes cost marks?
15. Going over 3,000 words, usually on background
The report is capped at 3,000 words; tables, graphs, equations, calculations, citations, the bibliography and headers do not count. Treat the cap as hard, and ask your teacher how over-length work is handled. The usual cause is a textbook-style introduction that the criteria do not reward.
Loses marks1,100 words on the history and structure of enzymes, 250 words of evaluation, 3,400 words in total.
Earns marksAbout 450 words of background, every paragraph leading to the prediction; the saved words moved into the conclusion and a ranked evaluation; 2,850 words in total.
How do you check your own IA for these mistakes?
Read your draft once for each question below, rather than once for everything:
- Does the research question name the variables, the range and the system? (1)
- Can I point to a reason for every range, interval, repeat number and control? (2, 3, 4)
- What one processed quantity answers my question, and where is it calculated? (5)
- Where does each uncertainty go after the raw-data table? (6)
- Is every graph the right type, labelled, with error bars explained? (7, 8)
- Is every raw value in the table, and is every exclusion explained? (9)
- Does the conclusion quote my numbers with uncertainties and a cited comparison? (10, 11, 12)
- Could any weakness I listed be pasted into someone else's IA? If so, rewrite it. (13, 14)
- What is the word count, and where are the words going? (15)
Use these questions on your draft before your teacher sees it. Your teacher comments on one draft only, so the more of these you fix first, the more useful that feedback will be. If you are still at the idea stage, our Biology IA ideas and Chemistry IA ideas are written to avoid mistakes 1 and 2 from the start.
Questions students ask
Why did my IA get a low mark when my experiment worked?
Because the criteria reward reasoning about the experiment, not the experiment working. Half the marks are for the conclusion and evaluation. A report with clean data but no processed quantity, no uncertainties, no comparison with published values and only generic weaknesses can score low even if every run went to plan.
Can I still get a good mark if my results don't support my hypothesis?
Yes. Nothing in the criteria rewards being right. A conclusion that honestly reports an unexpected result, quantifies it and explains plausible reasons scores better than one that bends the data to fit. Changing or inventing data is malpractice.
Should I delete anomalous results?
Never silently. Keep every raw value in your table, state the rule you used to identify anomalies before you apply it, show the processed result with and without them, and repeat the run if you can. Then discuss the likely cause in the evaluation.
Is it a problem if I followed a method from a textbook or website?
Using a published method is fine and normal, as long as you cite it and explain why it suits your question and what you changed. Copying it without reasons leaves Research design thin, and not citing it is an academic integrity problem.
Try one: Biology and Chemistry IA
A student plots the mean rate at each of five substrate concentrations, adds a linear trendline, and reports R2 = 0.97. Each mean comes from three repeats. Which statement is justified by this R2 value alone?
- Changing the substrate concentration causes 97% of the change in rate.
- The hypothesis is proven for all substrate concentrations.
- A straight line describes the trend in the five means closely, but this says nothing about the spread of the repeats behind each mean.
- The rate measurements have an uncertainty of about 3%.
Show the answer and why
Answer: C
R2 measures how closely the fitted line follows the plotted points, here the five means. It says nothing about causation (A), cannot prove a hypothesis or extend beyond the tested range (B), and is not an uncertainty (D). Because it is calculated from means, it also hides the variation between repeats, which is why error bars and the spread of the raw data still need reporting.