IB Physics IA Guide: The Mistakes That Cost Marks

The IB Physics internal assessment is a single scientific investigation that you design, carry out and write up yourself, in a maximum of 3000 words. It is marked out of 24 against four criteria worth 6 marks each, it counts for 20% of your final Physics grade at both SL and HL, and your teacher's marks are moderated by the IB.

The four criteria.

Research design, Data analysis, Conclusion and Evaluation. Each is worth 6 marks, so no criterion is a side dish: an investigation with beautiful data and a lazy evaluation loses a quarter of the total.

Where the data comes from.

IB Physics allows three routes: a hands on experiment you run yourself, an investigation built on a published database, or one built on a simulation. All three can reach full marks. What changes is the methodology you have to describe, not the ceiling.

The IA is not the collaborative sciences project.

The collaborative sciences project is a separate activity of roughly 10 hours, shared with Biology and Chemistry students, and it contributes no marks to your Physics grade. Keep the two apart in your planning calendar.

What the word count does not include.

Tables of raw and processed data, graphs, equations, calculations, references and short captions sit outside the 3000 words. The limit therefore applies to your argument, and students who spend 900 words restating the theory of simple harmonic motion have simply given away the space they needed for Evaluation.

Time.

Around 10 hours of the course are allocated to the investigation. In practice most of that disappears into data collection, which is exactly why the planning mistakes in the next section are so expensive: a badly framed research question wastes the lab time you cannot get back. If you are still deciding between the two levels, IB Physics HL vs SL explains what changes in the taught course, since the IA itself does not change.

IB Physics internal assessments lose Research design marks because the research question is not measurable. A question like "How does temperature affect resistance?" names no range, no material, no method of control. The fix is mechanical: name the independent variable, the dependent variable, the numerical range and the physical system, in one sentence.

Mistake 1: a question with no range.

Compare "How does the length of a pendulum affect its period?" with "How does the period $T$ of a simple pendulum vary with string length $L$ over the range $0.20\,\mathrm{m} \le L \le 1.00\,\mathrm{m}$, for a small angular amplitude below $10^{\circ}$?" The second version already tells the examiner what you measured, over what interval, and under which physical assumption. It also silently justifies your method, because the small angle approximation $\sin\theta \approx \theta$ is what makes $T = 2\pi\sqrt{\frac{L}{g}}$ valid.

Mistake 2: no theory linking the variables.

Research design rewards a question grounded in physics, not curiosity alone. Write the relationship you expect and the equation behind it before you collect anything. If you cannot predict the shape of the graph, you cannot design a sensible range of values.

Mistake 3: control variables listed but never controlled.

Writing "I kept the mass constant" is not enough. State how: the same bob, measured at $(50.0 \pm 0.1)\,\mathrm{g}$ on the same balance, for every trial. Examiners look for the method, not the intention.

Mistake 4: safety, ethics and environmental impact skipped.

These belong in Research design and take three honest sentences. A pendulum investigation has almost no risk, and saying so explicitly, with the one real precaution (clamping the stand so it cannot topple), is worth more than a paragraph of copied laboratory boilerplate.

Mistake 5: a topic that cannot generate enough data.

If your apparatus gives you only three usable values of the independent variable, the investigation is dead before you start. Choose a variable you can step at least five times without the effect vanishing into the noise.

Choosing a relationship that is trivially linear.

If your prediction is a straight line through the origin and the data confirm it in one graph, there is little for you to do in Data analysis. Relationships with a power, a square root or an exponential decay give you a linearisation to perform, and linearisation is where analysis marks live. Good candidates include the pendulum, $T = 2\pi\sqrt{\frac{L}{g}}$, the discharge of a capacitor, $V = V_0 e^{-t/RC}$, and the inverse square fall off of intensity, $I \propto \frac{1}{d^2}$.

If you do not have a topic yet, the list of 300 IB Physics IA ideas by syllabus topic gives each one as a measurable research question, with its variables, the graph to plot and how many other sites already list it.

An IB Physics investigation needs at least five values of the independent variable, spread across a wide range, with three to five repeated trials at each value. Below that, the gradient of your graph is not defensible and you have no honest way to estimate random uncertainty, which caps Data analysis immediately.

Mistake 6: one trial per setting.

A single reading gives you no spread, so the uncertainty you quote can only be the instrument precision. Repeats let you use half the range of your readings as the random uncertainty, which is the standard IB Physics convention and takes one line to justify.

Mistake 7: uncertainties invented or omitted.

Every raw measurement carries an uncertainty from the instrument. A metre rule read to the nearest millimetre gives $\pm 0.5\,\mathrm{mm}$ at each end, so a length measured between two points carries $\pm 1\,\mathrm{mm}$. A digital timer displaying hundredths gives $\pm 0.01\,\mathrm{s}$ as a display resolution, but human reaction time dominates, so quoting $\pm 0.2\,\mathrm{s}$ for a hand timed start and stop is more honest and examiners reward the honesty.

Mistake 8: fighting reaction time instead of designing around it.

Time 20 oscillations rather than one. Dividing a total time by 20 divides the reaction uncertainty by 20 too, so a $\pm 0.2\,\mathrm{s}$ human uncertainty on the total becomes $\pm 0.01\,\mathrm{s}$ on the period. That single design decision is worth more than any apology in the evaluation.

Mistake 9: raw data tables without units or uncertainties in the header.

The IB convention is to put the quantity, unit and uncertainty in the column heading, for example $L \pm 0.001\ (\mathrm{m})$, and to keep every entry in the column to the same number of decimal places. Mixed decimal places in a single column signal that the data was tidied afterwards rather than recorded properly.

Mistake 10: processed data that hides the raw data.

Both tables belong in the report. Show at least one full worked calculation of a processed value, including the uncertainty propagation, so the examiner can follow your arithmetic rather than trust it.

A strong IB Physics Data analysis section linearises the relationship, plots the graph with error bars, extracts a gradient with an uncertainty from maximum and minimum lines, and turns that gradient into a physical quantity. Plotting a curve and describing it in words is the single most common way to stall at the middle marks.

Mistake 11: leaving the data as a curve.

For a pendulum, $T = 2\pi\sqrt{\frac{L}{g}}$ gives a curve of $T$ against $L$, which tells the examiner nothing quantitative. Square both sides: $$T^2 = \frac{4\pi^2}{g}L$$ Now a graph of $T^2$ against $L$ is a straight line through the origin with gradient $m = \frac{4\pi^2}{g}$.

Step by step, with real numbers.

Suppose you measure five lengths from $0.200\,\mathrm{m}$ to $1.000\,\mathrm{m}$, timing 20 oscillations five times at each length.

• Process each mean total time into a period: $T = \frac{t_{20}}{20}$, and propagate the uncertainty the same way, $\Delta T = \frac{\Delta t_{20}}{20}$.
• Convert to $T^2$ and propagate: for a squared quantity the fractional uncertainty doubles, so $\Delta(T^2) = 2T\,\Delta T$. These become your vertical error bars.
• Draw the best fit line, then the steepest and shallowest lines that still pass through every error bar. Suppose the best fit gradient is $4.05\ \mathrm{s^2\,m^{-1}}$, with a maximum of $4.15$ and a minimum of $3.95$.
• Extract the physics: $g = \frac{4\pi^2}{m} = \frac{39.48}{4.05} = 9.75\ \mathrm{m\,s^{-2}}$.
• Extract the uncertainty from the extreme gradients: $\frac{39.48}{3.95} = 10.0$ and $\frac{39.48}{4.15} = 9.51$, so half the range is about $0.24$.
• Quote the result properly: $g = (9.8 \pm 0.2)\ \mathrm{m\,s^{-2}}$, with the uncertainty to one significant figure and the value rounded to the same decimal place.

Mistake 12: too many significant figures.

A gradient read off a hand drawn graph does not justify $9.7531\ \mathrm{m\,s^{-2}}$. Your uncertainty decides your precision, always.

Mistake 13: a graph that cannot be read.

Label both axes with quantity and unit, use a scale that fills the page, and never force the line through the origin unless the physics requires it. An intercept that should be zero and is not is evidence of a systematic error, which is free material for your evaluation.

$R^2$ is not accuracy.

A coefficient of determination such as $R^2 = 0.998$ only describes the scatter of the points about the fitted line. It says nothing about a systematic error that shifts every point the same way, so never quote it as proof that the result is accurate.

Conclusion and Evaluation lose marks in IB Physics because students describe instead of judging. A conclusion earns marks by answering the research question with a number, comparing it against an accepted value or theory, and deciding whether the difference is explained by the uncertainty. An evaluation earns marks by naming specific weaknesses and their direction of effect.

Mistake 14: a conclusion that never compares.

Continuing the pendulum example, the result $g = (9.8 \pm 0.2)\ \mathrm{m\,s^{-2}}$ must be set against the accepted $9.81\ \mathrm{m\,s^{-2}}$. The percentage difference here is under one per cent, and the accepted value falls comfortably inside your uncertainty range, so the correct sentence is that the data supports $T^2 \propto L$ with no evidence of a significant systematic error. That is a judgement, not a description.

Mistake 15: confusing percentage difference with percentage uncertainty.

They are different quantities and examiners notice. If your percentage uncertainty is 2% and your percentage difference from the accepted value is 8%, something systematic is present and you must say so. If the difference is smaller than the uncertainty, you cannot claim to have detected a systematic error at all.

Confusing agreement with accuracy.

If your uncertainty is 40%, almost anything agrees with anything. Say so. A candid line noting that the uncertainty range is too wide to distinguish competing models is worth more than a triumphant claim of confirmation.

Claiming proportionality when the intercept is not zero.

A relationship is proportional only if the line passes through the origin within uncertainty. If your intercept is $-0.05 \pm 0.01$ s$^2$, that is a systematic effect, and it belongs in the conclusion as a finding, not hidden.

Extrapolating beyond the range investigated.

You measured lengths between 0.20 m and 1.00 m at amplitudes below $10^{\circ}$. You cannot conclude anything about a 3 m pendulum or a $45^{\circ}$ swing. The overreach is noticed immediately, and it undermines the parts of the conclusion that were sound.

Mistake 16: writing human error.

It means nothing. Replace it with the specific mechanism and which way it pushed the result.

Mistake 17: improvements that are really wishes.

"Use more accurate equipment" is not an improvement. "Use a photogate at the lowest point of the swing, removing the reaction time uncertainty entirely" is, because it names the instrument and the uncertainty it removes.

Use a sentence template that forces the detail.

Because the pendulum bob was released by hand, the first oscillation was slightly larger than the rest, which makes the measured period systematically too long by roughly 0.5%, pushing my value of $g$ below the accepted value. Releasing the bob with an electromagnet would remove the extra initial displacement. That single sentence contains a cause, a direction, a size and a fix.

Separate random from systematic.

Random effects show up as scatter and as the spread between repeats, and more repeats reduce them. Systematic effects shift every point the same way and show up as a non zero intercept or a wrong gradient. Repeating the experiment more times does nothing to a systematic effect, which is why more repeats is such a weak improvement to propose.

Rank the weaknesses.

The marker wants to see judgement, not a list of eight equally weighted apologies. If the timing uncertainty is 0.7% and the length measurement is 2%, the length is where your error budget actually lives, so discuss it first and give it the most space.

Mistake 18: no extension.

Evaluation asks what comes next. A good extension follows from your own data, for instance repeating the investigation at amplitudes above $10^{\circ}$ to find where the small angle approximation visibly fails.

Plan the IB Physics investigation backwards from the school deadline in four blocks: choose and refine the question, collect data, analyse, write. Around 10 hours of course time are allocated, and the students who suffer are those who leave analysis and writing to the same fortnight as their first mock papers.

Mistake 19: choosing the topic last.

Decide the research question a full month before data collection, because the first version is almost never measurable and the second draft usually needs a pilot run. A pilot of twenty minutes tells you whether your range gives a visible effect, and it costs far less than a wasted afternoon.

Mistake 20: wasting the one draft.

You are entitled to teacher feedback on a single draft, in writing, and after that your teacher cannot mark further versions for you. Send a complete draft, including graphs and evaluation, not an introduction. Feedback on half a report buys you half the value.

Mistake 21: writing the introduction first and running out of words.

Draft in criterion order but write the introduction last, once you know how much of the 3000 words the analysis and evaluation actually needed. Three hundred words of context is plenty for most investigations.

Mistake 22: treating the IA as separate from exam preparation.

Uncertainty propagation, graph linearisation and evaluating experimental design all appear in the written papers too, so the skills transfer directly. Our IB exam preparation strategy shows how to fold IA deadlines into a revision calendar, and the general IB internal assessment guide covers how the IA sits alongside the Extended Essay and the rest of the Core.

Mistake 23: citing badly or not at all.

Every accepted value, database and simulation needs a reference, and references do not count towards your word limit. Academic honesty problems in an IA are not a marking issue, they are a diploma issue.