IB Physics IA guide: how to plan, write and score well (2025 criteria)

What is the IB Physics IA?

The Physics internal assessment (IA) is one scientific investigation, written up in a report of at most 3,000 words. You choose a question, collect or select quantitative data, analyse it and answer the question. Your teacher marks the report and the IB moderates a sample. It counts for 20% of your final Physics grade, and the task and the criteria are the same at SL and HL.

The subject guide recommends about 10 hours of teaching time, but most students also spend time outside lessons on data and writing, so plan for it. Your question need not go beyond the syllabus or be new to science. It has to be yours, measurable, and big enough that every criterion has something to mark.

The rules that trip students up

• Your raw data are yours. You may not hand in the same data set as a classmate, even if you shared the apparatus.
• Groups are allowed, reports are not shared. Up to three students may work together, but each one must vary a different independent variable, measure a different dependent variable, or analyse different data. Every student writes an individual report.
• One draft. Your teacher may read one draft and give oral or written advice, but does not edit it. The next version you hand in is the final one.
• Reference everything. Missing or improper referencing counts as academic malpractice.
• No double counting. The same piece of work cannot count for both the IA and the Extended Essay.

What changed from the pre-2025 criteria

Older exemplars, websites and some siblings talk about Personal engagement and Communication. Those belong to the previous model, which had five criteria. The 2025 course has four equal criteria of 6 marks and one clear limit of 3,000 words. There are no separate marks for presentation or personal engagement.

The marks that once rewarded effort and polish now reward reasoning, and the rest of this guide is built around that.

How the Physics IA is marked: the four criteria

Each criterion is worth 6 marks and is marked in three bands: 1 to 2, 3 to 4 and 5 to 6. Your teacher gives whole numbers, choosing the band that fits best. A top mark needs a few things done clearly, not a flawless report.

The bands are separated by command terms. To describe is to say what you did. To explain is to say what you did and why. A long, detailed account of your set-up is still a description. Look for “because” in the right places.

Criterion A: Research design (6 marks)

What it rewards. A question set in a clear context, sensible choices about collecting data, and a method another person could repeat. Top work names the variables and the system, and gives a reason for each choice: range, interval, repeats, instrument, controls.

Where marks go. A topic instead of a question; four values over a narrow range; controls that are named but never measured; an angle, mass or length with no reason given; a page of textbook derivation that never reaches a prediction; a list of apparatus that does not match the range in the question.

Our tip. Next to every number in your method, write “because” and finish the sentence. If you cannot, the choice is not yet explained.

Criterion B: Data analysis (6 marks)

What it rewards. Data recorded, processed and presented so the reader can follow them (clear) and so the conventions are right (precise): units, decimal places, significant figures, labelled graphs. Top work carries uncertainties from the raw data to the final result and onto the graph as error bars, and quotes each gradient or intercept with its own uncertainty.

Where marks go. A stopwatch uncertainty that is really a reaction time; the uncertainty of a squared quantity left unscaled; graphs without error bars; maximum and minimum lines drawn without regard to the bars; a line forced through the origin; degrees fed to a function that expects radians; raw data to three significant figures and results to eight.

Our tip. Choose one processed number and trace it back to the raw data. If a step is missing, your reader cannot follow it either.

Criterion C: Conclusion (6 marks)

What it rewards. An answer to your question that fits your analysis, including its uncertainties, and accepted physics. Top work opens with a sentence containing both variables and a number, stays inside the measured range, and compares with a cited value whose source and conditions a reader can trace.

Where marks go. Retelling the method; a trend claimed where the error bars allow a flat line; “close to the accepted value” with no number, no uncertainty and no source; a value for different conditions; the word “proves”.

Our tip. Underline the variables in your question, then check that your first conclusion sentence contains both.

Criterion D: Evaluation (6 marks)

What it rewards. An honest look at your method. Top work names weaknesses that can be seen in this data (scatter, an offset, a curve, a wide error bar), weighs them against each other with numbers, and says what each improvement would change.

Where marks go. Friction, air resistance or “human error” named but never estimated; a long list with no ranking; “use more accurate equipment” with no instrument named; a weakness that your own data show no sign of.

Our tip. For each weakness, ask where a stranger could see it in your data. If the answer is nowhere, cut it or look harder.

Three habits separate the bands in all four criteria: reasons, not lists; numbers, not adjectives; and links between the criteria, so that a prediction in A is tested in C and a weakness in D shows in the graph in B.

How to choose a Physics IA topic

A good topic is less about how impressive it sounds and more about whether it can give clean, repeatable numbers, with enough depth for every criterion to be marked. Try these seven tests.

1. Clear variables. You can say what you will change and what you will measure, and your school equipment can measure both to a useful precision.
2. Range and repeats. You can vary the independent variable smoothly over a wide range, with at least five values in our experience, and you can afford repeats in the time you have.
3. A law to test against. A known relationship gives you a prediction for Criterion A, a comparison for C and assumptions to examine in D.
4. Feasible and safe. It fits your time, your equipment and your school’s safety rules. A pilot run in the first fortnight saves weeks.
5. Yours. Not the standard school practical with the same variables and ranges, which every classmate will pick too.
6. Room for uncertainty work. One measurement should dominate the error so you can analyse it. If every value is exact or nothing varies, there is little to say.
7. Honest scope. Small and clean beats broad and thin.

Start from the apparatus you can use reliably, then ask what you could vary over a wide range and what the theory predicts. Our full list of ideas is in 93 IB Physics IA ideas and EE topics, and the free 30 Physics IA topic guides each come with a set-up diagram, a sample graph and a plan. Treat ideas as starting points. The final question, plan and data must be your own.

How to write a Physics IA research question

The IB describes a question “with context” as one that refers to the dependent and independent variables (or two correlated variables), briefly describes the system, and has background theory of direct relevance. In practice that gives four parts, shown below on a simple pendulum. We use the pendulum because everyone knows the physics. It is also a common choice, so a real IA needs a less familiar variable or a twist of its own.

Weak, better, strong

These three examples are our own, written to show the difference. They are not topics to copy.

What changed each time? The strong versions give a range with units, name a way of measuring the dependent variable, fix the system, and point to a relationship (a rolling-ball model, resistivity, Newton’s law of cooling) to test in Criterion C. They are longer, and that is fine if every part earns its place.

The 5-point check

Before you commit, score your question on five points. It is the same check our Coach applies to a research question: focus, variables, feasibility, data, and fit with the criteria.

Variables, controls and how much data you need

The independent variable is the one you change, the dependent variable is the one you measure, and control variables are everything else that could change the result. A control only counts if you say how you hold it steady and how you would notice a drift. The table shows what a good plan writes down for the pendulum in Figure 4.

Controls need numbers too

“Small angle” is an assumption, and you can put a number on it. For a simple pendulum the period grows with amplitude.

At 10° the effect is smaller than the timing uncertainty in Figure 6, so the assumption holds. At 30° it is 3.5% in T², pushing g too low by about that much. That is the reasoning Criterion D rewards.

Why five values and three repeats

Five values with three repeats is our working minimum, not an IB rule. Five values over a useful range show the shape of the relationship, which three cannot. Three repeats show the scatter, so you can tell a trend from noise and justify your uncertainty.

On range: for a straight-line law, a factor of two or three, like 0.30 to 0.80 m here, is a fair start. For a power law, a factor of ten or more pins the exponent down far better. Use even steps, and add values where the curve bends.

The three runs differ by 0.09 to 0.13 s on times of 11 to 18 s, under 1.2%. That is smaller than the 0.2 s a hand-held stopwatch adds, a fact we use below. Explain each choice of range, step and repeat in Criterion A.

Measurement methods that give clean data

Every technique has a classic trap. These are the five we see most, each with a fix for your plan.

Timing: stopwatch, light gate or video

The trap. Timing one short event by hand and quoting the display, ±0.01 s, as the uncertainty. Your reaction time is the problem, typically about 0.2 s.

The fix. Time many repeats of a periodic event and divide, so that the same 0.2 s is shared across all of them, as in Figure 6. Better, remove the person: use a light gate, or a phone video at a known frame rate. At 30 frames per second the frames are 0.033 s apart, so state that as your resolution. At 240 it is 0.004 s.

Counting has a limit. A pendulum loses amplitude as it swings, so check that the amplitude barely changes between the first and last swing.

Length and angle: rules, callipers and protractors

The trap. Measuring to the wrong point. For a pendulum the length is to the centre of the bob, not its top. Parallax and zero errors lurk in rules, callipers and micrometers alike.

The fix. Define the quantity in the plan and measure it the same way every time, with your eye level with the scale. Record each zero error. Use a calliper or micrometer for a diameter that enters a formula squared. A protractor gives about ±1°, so it is often better to measure a height and a distance, calculate the angle and propagate the uncertainty.

Temperature: thermometers and probes

The trap. Reading a thermometer before it has caught up, or trusting one reading from unstirred water. Heat also leaves a cup from the start, so the rise you see is smaller than the ideal one.

The fix. Stir gently, wait for a steady reading and keep the probe off the walls. Use a lid and an insulated cup, and log temperature against time rather than reading one number. A temperature rise uses two readings: a thermometer good to ±0.5 °C gives ±1.0 °C on a rise of 8.4 °C, which is 12%. A probe good to ±0.1 °C gives 2.4%.

Electrical meters: current, voltage and resistance

The trap. Letting the wire heat as the current flows, so that resistance climbs and the graph curves for the wrong reason. Another is ignoring the resistance of leads and contacts, which adds a fixed amount to every reading of a small resistance.

The fix. Keep currents low and readings short, and let the wire cool between runs. Short the leads together and record the zero reading, or read the offset from the intercept of a graph. Take the uncertainty from the meter’s manual, where a digital meter usually quotes a percentage of the reading plus a few digits.

Sensors and data loggers

The trap. Trusting the screen. A logger is only as good as its calibration, sampling rate and noise, and a smooth curve can hide a drifting zero.

The fix. Check the sensor against something you trust, such as a ruler for a motion sensor or a known mass for a force sensor, and record the check. Choose a sampling rate that captures the shortest feature, record the stated resolution, and log the noise with nothing happening so you know the scatter before you start.

Uncertainties in the Physics IA, with a worked example

Uncertainty is the most common reason a Data analysis mark stops at 4. Every measurement is a range, not a point. You record it as ±, carry it through every calculation, draw it as an error bar and use it to interpret your result.

The Physics guide lists the skills: propagate uncertainty through addition, subtraction, multiplication, division and powers; give it as absolute, fractional or percentage; draw uncertainty bars; construct lines of maximum and minimum gradient by eye, considering all the bars; and find the uncertainty in a gradient or intercept.

Absolute, fractional and percentage uncertainty

They describe the same thing three ways. For 0.600 ± 0.002 m, the absolute uncertainty is 0.002 m, the fractional uncertainty is 0.002 ÷ 0.600 = 0.0033, and the percentage uncertainty is 0.33%. Only the absolute one has units.

Instrument uncertainties

These are typical school values. Prefer the figure in the maker’s manual, and if your repeats spread more than the instrument suggests, use the spread.

Propagation rules

A worked example: g from a pendulum

Here is a full calculation with invented numbers for one length. The string is 0.600 ± 0.002 m, pivot to centre of the bob. Three timings of ten swings, 15.52, 15.61 and 15.61 s, have a mean of 15.58 s. They differ by only 0.09 s, but hand timing adds about 0.20 s of reaction time, so we use ±0.20 s. From T = 2π√(L/g), g = 4π²L ÷ T².

1. Period. T = 15.58 ÷ 10 = 1.558 s. Dividing by an exact number divides the uncertainty too, so ±0.20 ÷ 10 = ±0.020 s. As a percentage, 0.20 ÷ 15.58 = 1.28%.
2. T². 1.558² = 2.427 s². Powers double the percentage uncertainty: 2 × 1.28 = 2.57%, so T² = 2.427 ± 0.062 s².
3. Length. 0.002 ÷ 0.600 = 0.33%.
4. Add the percentages (g is a product and a quotient of these): 2.57 + 0.33 = 2.90%. The factor 4π² is an exact number, so it adds nothing.
5. The value. g = 4π² × 0.600 ÷ 2.427 = 9.758 m s⁻².
6. Back to absolute. 9.758 × 0.0290 = 0.283, so ±0.28 m s⁻².

Result: g = 9.8 ± 0.3 m s⁻² (2.9%), with the result rounded to the place of its uncertainty. A single length gives one value of g. A graph over many lengths gives a better one, as we show below.

Figure 8 tells you which uncertainty matters. Had we timed one swing, the same 0.20 s would have produced 26%. It also shows what the sum leaves out: resolution and scatter are in the 2.9%, but a systematic effect such as measuring to the top of the bob is not, and belongs in the evaluation.

Uncertainty on a graph

The pendulum data give a straight line if you plot T² against L, because T² = (4π²/g) × L. Each point gets a vertical bar of ±2TδT. The horizontal bars are left out, because ±0.002 m is smaller than the dot.

To find the uncertainty in the gradient, draw the steepest and shallowest straight lines that still pass through all the bars, and take half the difference. A spreadsheet’s regression statistics are another route, but they measure scatter about the line and ignore your error bars, so they usually give a smaller figure. State which you used.

The extreme lines in Figure 9 have gradients of 4.21 and 3.84 s² m⁻¹. Half the difference is 0.18, so the gradient is 4.04 ± 0.18 s² m⁻¹, or 4.5%. Then g = 4π² ÷ gradient = 39.48 ÷ 4.04 = 9.77 m s⁻², with the same 4.5% uncertainty, which is ±0.44. Rounded properly:

g = 9.8 ± 0.4 m s⁻² (4.5%).

Now the intercept. The best fit crosses the axis at −0.002 s² and the extreme lines at −0.085 and +0.108 s², so it is 0.00 ± 0.10 s². That band includes zero, as it should for a length measured to the centre of the bob. A 2 cm offset would have given an intercept near 0.08 s². But ±0.10 s² is wide enough to hide an offset up to 2.5 cm, so the graph alone cannot rule out a small systematic error. Measure the bob and say how.

Graphs and linearisation: from raw data to a result

Table conventions

• Put the quantity and its unit in the column heading, not in every cell: L / m, t / s.
• Give each measured value to the decimal places the instrument allows, and give its uncertainty to the same decimal places as the value.
• Number every table and graph and refer to it in the text. Keep qualitative notes beside the numbers they explain.
• Round processed values to the significant figures your data justify, and the final result to the place of its uncertainty.

Processed data, with a worked example per type

Show one worked example for each type of calculation, then the processed table. For the pendulum, Figure 5 gives these values, with δ(T²) = 2T × 0.020.

Choosing and linearising graphs

Plot the graph that answers your question. Where theory gives a straight line after a transformation, plot that, so that the gradient and intercept carry meaning. A polynomial fitted only because it passes near the points is hard to defend.

Take the capacitor discharge in Figure 11. The voltage falls as V = V₀e^(−t/RC), a curve. Natural logs give ln V = ln V₀ − t/RC, a straight line with gradient −1/RC and intercept ln V₀. A voltmeter with 0.1 V divisions gives ±0.05 V, so the bars on ln V are ±0.05/V: 0.8% at 6 V but 8% at 0.63 V.

The best fit has gradient −0.0453 s⁻¹, and the extreme lines give −0.0465 and −0.0443 s⁻¹. So the gradient is −0.0453 ± 0.0011 s⁻¹ (2.4%) and RC = 22.1 ± 0.5 s. The intercept is 1.790, so V₀ = 5.99 V, matching a 6 V supply. The nominal RC is 22.0 kΩ × 1000 µF = 22 s, but an electrolytic capacitor can be ±20% from its label, and that sets the real limit on the comparison.

Power laws and log-log graphs

If y = kxⁿ, then ln y = n ln x + ln k, so a log-log graph is a straight line with gradient n. Use it when you do not know the exponent. For the pendulum, a log-log plot of T against L tests n = 0.5 without assuming it. Use the same kind of log on both axes. The unit is divided out before taking a log, which is why the axis above reads ln (V / V).

R² is not an uncertainty

For the pendulum fit, R² is 0.999, and it is tempting to write “very reliable” beside it. Do not. R² tells you how closely the points follow the line, not how far the gradient could be from the truth, and a curved set of points can still give R² above 0.99. Take the gradient uncertainty from the extreme lines or the regression statistics.

Three habits that protect marks

• The intercept needs a reason. Forcing a line through the origin throws away information. If theory says it should be zero, test that with the intercept and its uncertainty, as above.
• Radians and degrees. A spreadsheet works in radians, so sin(30) means 30 radians. Convert first.
• Outliers. Identify them, say why you keep or remove them, and show the effect. The guide does not ask for a statistical test.

How to write the conclusion

A conclusion earns its marks by being specific, honest and traceable.

Answer with numbers and uncertainty

Begin with one sentence that contains both variables and a result, with the key value and its uncertainty. For the pendulum: T² is proportional to L over 0.300 to 0.800 m, with gradient 4.04 ± 0.18 s² m⁻¹ and an intercept consistent with zero.

Then say whether the trend is bigger than your uncertainties. If the error bars would let a flat line through the data, you cannot claim a trend. Use the language of evidence (“supports”, “is consistent with”), stay within the measured range, and say whether your prediction from Criterion A held.

Give the accepted value, with its source and conditions

Quote the accepted value or law, the conditions it applies to, and a reference a reader could find: author or publisher, title, edition or database, and the page or table. A bare website name will not do. Conditions matter in physics. Free-fall acceleration is about 9.78 m s⁻² at the equator and 9.83 m s⁻² at the poles, so a value for another latitude is not a fair test of yours. The standard value, 9.81 m s⁻², sits between the two.

Percentage error against percentage uncertainty

These are different things, and the conclusion should read them together. The percentage error says how far you are from the accepted value. The percentage uncertainty says what your own measurements can justify. Figure 12 shows how to decide what to say.

Two cautions. A tiny percentage error is not proof of accuracy, since a systematic effect can hide inside the uncertainty. And when the error is far smaller than the uncertainty, as in our 0.4% against 4.6%, check whether the uncertainty is generous. The extreme-line method gives the widest gradient that the bars allow, and a regression would give a narrower one.

Explain the physics

Close the loop with the physics behind your data: why T² should grow with L, why a discharge is exponential. Keep the explanation tied to what you measured. If your data cannot show why something happens, say so rather than reaching for a mechanism you did not test.