Experimental programme › Internal Assessment

Internal Assessment

The IA is your own scientific investigation: you ask a question, collect data to answer it, and write a report. It is worth 20% of your final grade, at SL and HL. This page takes you through it in the order you will actually do the work, and shows where students win and lose marks.

1. The IA at a glance

2. Introduction slides

These are the slides from our IA introduction lessons: the criteria, choosing a topic, research questions, variables, data analysis, uncertainties, conclusion and evaluation, with examples to discuss.

Open the slides in Google Drive

The slides are only visible to school accounts. If you see a sign-in message or a blank box, sign in to your school Google account in this browser, or use the button above.

3. The four criteria in plain language

I will give you the official IB criteria; read them. The tables below translate each part (each strand) into what you actually need to put on the page. The criteria are marked with a best fit: you don't need every word of a descriptor to get the mark, and the top band doesn't mean perfect.

Research Design (6 marks)

How well you communicate your methodology: what you did, and why.

Data Analysis (6 marks)

How well you record, process and present the data in ways that answer your question.

Conclusion (6 marks)

How well you answer your research question, using your analysis and the accepted science.

Evaluation (6 marks)

How well you judge your own methodology and suggest improvements.

Not assessed under Evaluation: the strengths of your investigation, and extensions to it. Leave them out. One major weakness, properly explained with a proper improvement, can earn full marks; a long list of generic ones earns very few.

4. The stages and the one-draft rule

Your Student IA Workbook follows the same stages, with a task to complete at each one. Do the work in this order. Most weak IAs have the same cause: data collection left so late that there was no time to fix anything when the first attempt went wrong. Try to keep the gap between starting your experiment and writing your draft to about two weeks.

5. Choosing your investigation

You are not marked on having a clever idea; you are marked on what you do with it. A pendulum experiment done well beats a badly done attempt to measure the mass of an electron. What matters is sound analysis, appropriate theory and a sound method, not originality.

The five-question test

A topic works if you can answer yes to all five:

1. Is there one independent variable I can change and measure, and one dependent variable that responds to it? (Two independent variables is the most common design failure examiners report.)
2. Can I get at least five values of the independent variable, over a range wide enough that the dependent variable changes a lot, and repeat each one about five times?
3. Is there an equation, or a known relationship, that tells me what shape the graph should be?
4. Can my equipment detect the change I'm looking for? If the effect only shows in the third decimal place, an instrument that reads to two can't answer the question.
5. Is there something to compare my answer with: an accepted value, an accepted relationship, or at least a plausible range?

Also ask: do I have the materials and equipment? Can I collect plenty of reliable data in the time? Can I control the other variables? Can I understand the theory with what I know, or with a bit of research?

Topics that are too simple

Some investigations leave nothing to analyse: the resistance of a wire against its length, current against voltage for a fixed resistor, and anything you have already done as a class practical, unless you add something extra (for example, a cooling correction in a specific heat capacity experiment). A study that only gives qualitative results, shown in a bar chart, is well below the standard.

Popular topics, and making one your own

Some topics come up every year: viscosity and temperature, a pendulum's period, a bouncing ball, pencil graphite resistance, specific heat capacity of salt solutions, string tension and frequency, refractive index and temperature, Hooke's law, projectiles. These are fine, but examiners find many reports with the same pictures, references and theory, copied from the same online IA. If you choose one:

• Reference any previous IA, video or website you took the idea or method from. This is required, and it doesn't cost you marks.
• Change something real: a different material, a wider range, a better measuring technique, or a variable others ignored.
• Write all the theory yourself, in your own words.

If the physics is already known

Some questions have textbook answers: the frequency of a string against tension, projectile range against angle. Don't pretend you don't know the answer. Make your aim one of these instead:

• "to confirm that … within experimental uncertainty";
• "to determine a precise experimental value of … using …";
• "to test the limits of … by extending the range beyond …".

Then use the known physics to design better: it tells you the precision you need, how many points, and what shape the graph should be.

Some ideas to get you thinking

The best questions have the form "How does y depend on x?" For example, how does…

• the period of a pendulum depend on the angle it is released from (testing the small-angle approximation)?
• the rebound height of a ball depend on the air pressure inside it?
• the damping of an oscillating mass depend on the area of a card attached to it?
• the deflection of a ruler clamped at one end depend on the mass hung on its free end, or on its length?
• the time to empty a funnel of sand depend on the diameter of its opening?
• the frequency of the first harmonic in a tube closed at one end depend on the tube's diameter (the end correction)?
• the rate at which water cools depend on its temperature above the room?
• the current that melts a fuse wire depend on the wire's diameter?

Hobbies help too: sport (projectiles and bounce), music (strings and pipes), astronomy (database investigations, such as star brightness or white dwarf masses). Topics like dark energy or black holes are fascinating, but you can't collect data on them in a school lab.

6. Research question and context

This is Research Design strand i, but it quietly controls your Conclusion too: a vague question can't be answered, so the conclusion has nowhere to go.

What a research question contains

It doesn't have to be a question; an aim works equally well ("to determine $g$ using a simple pendulum"). It must name quantifiable variables:

Avoid open questions such as "What factors affect the efficiency of a wind turbine?" Questions starting "What is the relationship between…" are allowed but often lack focus.

What "context" means

Context is the specific physics your question sits inside, usually half a page to a page. Include:

• the equation or relationship you expect, explained in your own words, with what each symbol means;
• the assumptions the theory makes (small angle, negligible air resistance, laminar flow…) and whether they hold in your set-up;
• where the theory stops being valid;
• roughly what size of answer to expect, so you can tell later whether your result is sensible.

Context is not: the history of physics, a scientist's biography, a general essay on the topic, or why you became interested. Don't derive an equation that is in the data booklet; examiners specifically criticize pages of derivation. Personal engagement is no longer assessed.

The hypothesis

If your investigation tests a relationship between two variables, it needs a hypothesis. Write it after your research question:

• Explain what will happen to the dependent variable as you change the independent variable, and why, using the physics from your context.
• Mention any control variables that matter, and why.
• End with a clear statement: "If the independent variable changes in this way, then the dependent variable will change in this way."
• Be specific. Don't just say "increases" or "decreases": is it directly proportional, inversely proportional, proportional to a square root?

For example: "For a longer string, the restoring force on the bob is smaller for the same sideways displacement ($F \approx \dfrac{mg}{L}x$), so the bob accelerates back more slowly and each swing takes longer. Theory predicts that the period is proportional to the square root of the length, so if the length is increased, then $T^2$ will increase in direct proportion, and a graph of $T^2$ against $L$ will be a straight line through the origin with gradient $\dfrac{4\pi^2}{g}$." That last sentence tells you exactly what to plot. If you already know the equation, say so: you are confirming it, not discovering it.

Determining a constant? If your aim is to measure a value, such as $g$ or the specific heat capacity of a metal, you don't need a hypothesis. Instead, state the value you expect, with its source, so you can compare your result with it in your conclusion.

7. Designing the method

This is Research Design strands ii and iii. They are separate: you can describe a poor method beautifully, or run a good method and describe it too vaguely to repeat. Aim for both.

The decisions you must explain

The key word is explain: give the reason for each decision.

How much data? As a guideline, use five values of the independent variable with five trials of each. It is a guideline, not a rule: some experiments need more values or trials (a curve whose shape you need to see clearly, or results that vary a lot) and some need fewer (very consistent readings, or a slow method). Justify your choice either way.

Variables table

Make a table of every variable: its symbol and unit, how it is measured or controlled (with the instrument and its uncertainty), and why it matters. That last column is where the marks are. If you calculate the independent variable from another quantity, say how.

The method itself

• Say what you did, focusing on how you changed the independent variable, measured the dependent variable and kept the controls fixed.
• Write the method as a series of clear steps, naming specific apparatus (with sizes or models) and specific quantities, so someone else could repeat it.
• Leave out steps about how everything connects together (such as plugging the sensor into the computer), unless it is an important methodological consideration.
• Include a labelled diagram of the set-up. A clear sketch usually communicates better than a photograph.
• Avoid unnecessary or repeated detail.

8. Recording data and raw uncertainties

An uncertainty you didn't record in the lab can't be invented afterwards. See also Physics skills: recording uncertainties.

Table conventions

Every column heading has four things: the quantity, its symbol, its unit and its uncertainty, for example "Length L / cm, ±0.1". All values in a column have the same number of decimal places, matching the uncertainty: (87.4 ± 0.2) cm is right; (87.4 ± 0.05) cm and (87.4 ± 2) cm are both wrong. Put the independent variable in ascending order. If you paste a spreadsheet screenshot and can't fit the uncertainties in the headings, state them right next to it.

Timing ten swings and dividing by ten makes the uncertainty in one period ten times smaller. Measuring many and dividing is a good habit (for example, the thickness of 50 sheets of paper).

Where a raw uncertainty comes from

Check for zero errors before you start (for example, a micrometer that reads −0.15 mm when closed) and correct for them.

Outliers and observations

A point that doesn't fit is a possible outlier, not a mistake to delete quietly. To reject it, give a cause you can identify; "the graph looks better without it" is not a reason. Never reject more than one point. Better still, measure that point again.

Write down what you notice as well as what you measure: bubbles, a sphere drifting to the side of the tube, a spring twisting, condensation. These observations cost almost no words and are the raw material for a strong evaluation.

9. Processing and propagating uncertainties

Data Analysis marks are usually good, and lost to the same few mistakes. The full rules, with worked examples, are on the Physics skills page. In short:

Carry one or two extra digits through the working and round only at the end. Averaging five readings does not give you an extra significant figure.

Writing the final value

Give the uncertainty to one significant figure (two if it starts with a 1), with the value to the same decimal place: (12.2 ± 0.4) m and (14.23 ± 0.12) s are right; (12.2 ± 0.36) m and (14.23 ± 0.1) s are not. If an uncertainty is bigger than the value, say so and deal with it.

Show one of each calculation

Show one worked example of each different calculation: the equation, the substituted values and the result, with its uncertainty. Don't show forty of them; examiners criticize pages of repeated arithmetic. A sentence is enough for routine spreadsheet steps such as taking a mean.

10. Graphs that answer the question

Linearize: let the theory choose the axes

A graph of the raw variables is often a curve that tells you nothing quantitative. Rearrange the expected equation into the form $y = mx + c$, and the gradient becomes your answer. Linearizing isn't required, but the strongest reports almost always do it.

The log–log graph is worth knowing. If you don't know the power, it finds it for you, and you then have a justified fit instead of a shapeless curve. Dimensional analysis can also predict a power: for sand flowing from a funnel, it suggests the emptying time depends on the opening's diameter to the power $-\tfrac{5}{2}$.

Maximum and minimum gradients

Here the best-fit gradient is $4.04\ \text{s}^2\,\text{m}^{-1}$, the steepest line gives 4.44 and the shallowest 3.64. So the gradient is $\dfrac{4.44 + 3.64}{2} \pm \dfrac{4.44 - 3.64}{2} = (4.0 \pm 0.4)\ \text{s}^2\,\text{m}^{-1}$, a 10% uncertainty, giving $g = \dfrac{4\pi^2}{\text{gradient}} = (9.8 \pm 1.0)\ \text{m s}^{-2}$. Find the intercept's uncertainty in the same way. More detail: Physics skills: uncertainty in the gradient.

Checklist for every graph

• A descriptive title naming both quantities, numbered (Figure 1, Figure 2…) and referred to in the text.
• Independent variable on the x-axis, dependent on the y-axis.
• Both axes labelled with quantity, symbol and unit: $T^2$ / s².
• Uncertainty bars on the dependent variable (and on the independent variable when they matter). The same absolute bar on every point is usually clearest; say how you chose it. Never enlarge the bars just so the line touches them all.
• A best-fit line or smooth curve, never points joined dot to dot.
• Not forced through the origin: that would hide a systematic error.
• A fit justified by theory or dimensional analysis. A high-order polynomial that passes through every point proves nothing.
• Steepest and shallowest lines drawn using all the bars, not just the first and last points.
• Gradient and intercept quoted with their uncertainties and units, and only if they mean something physically.

11. The conclusion

The Conclusion has two strands, and many students only write the first. The second is often two or three sentences of work for a whole mark band.

Strand i: answer the question

These don't count as an answer: describing in words the numbers already on your graph; repeating your method; a purely qualitative statement ("as the tension increases, so does the frequency"); or talking about the statistics instead of the physics.

These do:

• a number with its uncertainty range: "the specific heat capacity of aluminium was found to be (932 ± 15) J kg⁻¹ K⁻¹";
• a relationship named correctly: linear, directly proportional, inversely proportional, exponential. A straight line with a negative gradient is not an inverse proportion;
• what the gradient and intercept mean physically, including a non-zero intercept where theory expects zero;
• how the uncertainties affect your confidence in the answer;
• a "no relationship" result, if that is what you found. Reporting it honestly, without forcing a fit, earns credit.

Strand ii: compare with the accepted science

Your result doesn't have to agree to reach the top band. What earns the marks is an honest, quantitative comparison and a sensible explanation of any difference. For example, a measured internal resistance of (0.59 ± 0.09) Ω against a manufacturer's 0.48 Ω for a new cell can still score highly, if the difference is explained (the cell's age, its temperature).

Sentence starters

• "The gradient of (… ± …) [unit] gives a value of … of (… ± …) [unit], since the theory predicts that the gradient equals …"
• "The accepted value of … is … (source). This lies within / just outside the experimental range of … to …, a difference of …%."
• "The non-zero intercept of … suggests a systematic shift of …, which is consistent with …"
• "No relationship was found between … and …: across the whole range, all values lay within … of each other, comparable to the uncertainty of …"

12. The evaluation

Examiners call this the hardest criterion. Students know the textbook list of random, systematic and human errors but don't connect it to what they actually did. Two or three real weaknesses, followed all the way through, beat a page of generic ones.

The four-step chain

Take every weakness through all four steps. Most students stop after step 1, which is why most students score in the middle band.

The word relative in the top band means the examiner wants your judgement: which weakness mattered most, and why?

Where to look for real weaknesses

• The approach: was this the right technique? What did choosing it cost you?
• The apparatus: how did your choice of instrument limit the data?
• The procedure: what was hard to judge? Where did you estimate?
• The assumptions: did small-angle, laminar-flow or no-air-resistance assumptions really hold across your whole range? An assumption that fails at one end of the range is a strong point.
• The range: why did your data start and stop where it did? Does the relationship still make sense outside it?
• The graph itself: is there a non-zero intercept where there shouldn't be? Does the scatter grow at one end?

Systematic shifts are the richest source of marks. A non-zero intercept means something. In one pendulum investigation the graph missed the origin; the student realized the length should be measured to the centre of the bob, not to the top of it, measured the bob, and found the missing length matched the intercept. That is one weakness, fully explained, with a real physical cause.

Vague versus explained improvements

Improvements must be possible in a school lab. Calibrating instruments and extending the investigation are not expected.

13. Writing it up

There is no required format, and evidence for a criterion can appear anywhere. But this order is easy to follow and easy to mark:

That adds up to about 2,200 words. Reports of 25 to 40 pages that use all 3,000 words are not better; words spent on background you didn't need are words you can't spend on the conclusion and evaluation.

Cut these

• A cover page and contents page.
• A personal engagement section, or opening lines like "Ever since I was a child…".
• A summary at the end: your conclusion already is one.
• Derivations of equations in the data booklet.
• Pages of repeated arithmetic that belongs in a spreadsheet.
• A list of strengths in your evaluation.
• Photographs of yourself in the lab.
• Bibliography padding: list only what you actually used. Examiners check.

Style

• Write equations with an equation editor, not as "a = F/m". Use × for multiplication and $5.3 \times 10^2$, not 5.3E2.
• Variables in italics, units upright, with a space between units: m s⁻¹ (ms⁻¹ means "per millisecond").
• Number every table and figure, and refer to each one in the text.
• Use precise language: the ball leaves the foot with a high speed, not a large force. If you write "at high temperature", say high compared with what.
• Don't use terms you can't explain, and don't repeat yourself.

14. Group work, referencing and AI

Working in a group (optional)

• Groups can be no larger than three, and you can collect data together.
• Each of you needs your own research question: a different independent variable, a different dependent variable, or a different part of a shared data set. You must not submit the same raw data as another student.
• Every word of the report, including the method, must be written by you alone. Put the candidate codes of your group members on page one.

Referencing

Reference anything you took an idea, method, image, equation, expected result or value from, including previous IAs you found online. Any consistent style is fine (in-text, footnotes or numbered), plus a bibliography. For websites include the URL and the date you accessed it. Missing or improper referencing is treated as academic misconduct. You don't need to reference course-level physics such as $F = ma$ or Ohm's law.

Artificial intelligence

15. Two moderated exemplars

Both are real investigations by former students, with real moderated marks, and neither is perfect. Read my comments alongside each one (they open in Google Drive with your school account). Seeing why each one lost marks is more useful than reading a model answer.

Exemplar A: temperature and the viscosity of honey (19/24)

A steel sphere dropped through honey at five temperatures from 25 °C to 65 °C, filmed at 240 frames per second; viscosity from Stokes' law; $\ln\eta$ plotted against $\dfrac{1}{T}$ to find an activation energy.

What earned the marks: a research question naming the variable and its values, the measured quantity and unit, the apparatus and the technique; assumptions (constant density, thermal equilibrium, Newtonian fluid) stated and justified; a range justified physically, stopping below the temperature at which honey breaks down; linearizing to find an activation energy and comparing it with a published range; and specific evaluation points, such as about 1 °C of cooling while transferring the cylinder and the flow possibly becoming non-laminar at the highest temperatures.

Why not higher: the uncertainties were too small to be believable: under 1% on viscosity for a hand-released sphere timed from video, ignoring the much larger spread between trials. The conclusion relied on the trend and on R² instead of on uncertainty, and called the result "close" to the published range when it actually lay below it, without discussing the gap. The evaluation described effects but didn't say how big they were.

Lesson: uncertainty isn't a box to tick during processing; it decides whether your conclusion is justified. Underestimating it costs marks in two criteria at once.

Read Exemplar A My comments on Exemplar A

Exemplar B: spring extension and the damping of a cart (16/24)

A cart oscillating between two pairs of springs on a track, tracked with a motion sensor; the damping coefficient found by fitting a decaying sine curve, for seven spring extensions with three trials each.

What earned the marks: an original question that came from experimenting with the apparatus; a labelled diagram and a method precise enough to repeat; sensible checks that the fit was meaningful (its amplitude matched the 5 cm pull-back and its period was plausible); an honest "no relationship" result, with no attempt to force a fit; and a self-critical evaluation that noticed the damping in the first oscillations was very different from the rest.

Why not higher: the comparison with accepted science was thin, because the report ran out of words at 2,960. The uncertainties were too small to explain the scatter. Significant figures and graph labels didn't follow the conventions. One of the two improvements was vague.

Lesson: a "no relationship" result is a perfectly good IA. Budget your words so the conclusion and evaluation, worth half the marks, get the space they need.

Read Exemplar B Exemplar B with my margin notes My comments on Exemplar B

16. Before you hand it in

Three gates

If any of these fails, don't submit yet: each one caps your mark on its own.

1. My conclusion compares my result with something outside my own data (an accepted value, relationship, manufacturer data or a reasoned plausible range) with a traceable source.
2. My final answer has an uncertainty range, and I have said what that range means for my confidence in the result.
3. Every weakness in my evaluation is specific to my experiment, and says how it affected my result.

Point to the evidence

For each strand, find the page and line where an examiner will see the evidence. If you can't point to it, it isn't there.

• Research Design: the question names both variables, with range and units, in a specific physics context; range, interval, repeats, precision and each control are explained with reasons; the method and diagram are detailed enough to repeat.
• Data Analysis: table headings have quantity, symbol, unit and uncertainty; significant figures and decimal places are consistent; every uncertainty is justified and propagated; the graph answers the question, with a fit justified by theory, bars and gradient lines.
• Conclusion: a number with its uncertainty, the relationship named correctly, gradient and intercept explained, and a comparison with a source.
• Evaluation: specific weaknesses, ranked, each with its effect on the data and the conclusion, and a targeted improvement.

Format check

• Title, candidate code(s) and word count at the top of page 1.
• Page numbers on every page.
• No cover page, contents page, personal engagement section or closing summary.
• Under 3,000 words, ideally nearer 2,200.
• Every table and figure numbered and referred to in the text.
• No R², r or standard deviation.
• A bibliography of only the sources you used, with access dates for websites.
• Any AI use cited, with the tool, the prompt and the date.

17. Check your understanding

18. IA documents

These open in Google Drive and need your school Google account.

• Student IA Workbook

  Your step-by-step guide, from choosing a topic to writing up, with a task at every stage. Start here.

• Draft Self-Review

  A one-hour check of your own draft, criterion by criterion, ending with a ranked list of fixes.

• Final Submission Check

  The last thing you do before handing in: three gates and a format sweep.

• Exemplar A

  Temperature and the viscosity of honey (19/24): a moderated report by a former student.

• Exemplar B

  Spring extension and the damping of a cart (16/24): a moderated report by a former student.

• Exemplar B with margin notes

  The same report with my notes beside each section, showing where it earned and lost marks.

• Comments on Exemplar A

  My strand-by-strand comments on the honey viscosity investigation (19/24).

• Comments on Exemplar B

  My strand-by-strand comments on the damping investigation (16/24).

• IA introduction slides

  The slides from our introduction lessons, with examples of contexts, research questions, variables tables, data and graphs.

• IA criteria clarifications

  Detailed notes on how each criterion is applied, written for teachers and moderators. For when you want to go deeper.

• Physics skills

  Uncertainties, significant figures, propagation, graphs and gradient uncertainties, with worked examples.