If you have ever stared at a scatterplot and felt a small surge of pride because your curve “looks right,” you already understand the most dangerous moment in modeling.
Because in IB, a model that looks right can still be wrong.
Examiners are not only asking whether you can create a regression or function. They are asking whether you can trust it. That is what evaluating model fit is really about: proving, with statistical tools, that your model earns the right to be believed.
This guide shows IB students how to evaluate model fit with the same calm logic you want to have in an exam or in your Math IA. We will use residual analysis, R², MAE, RMSE, correlation, and the “so what?” explanations that turn numbers into marks.
IB student confronted by residuals
A quick checklist IB examiners reward
Before you evaluate model fit, make sure you can do these quickly:
State your model clearly (equation + what variables mean).
Plot data and model together (visual fit check).
Compute residuals and describe their pattern.
Report at least one fit statistic (often R²) and one error statistic (MAE or RMSE).
Interpret results in context (units, scale, reasonableness).
Reflect on limitations: outliers, range, assumptions, overfitting.
Why model fit matters in IB (and why “high R²” is not enough)
In IB Maths, especially in Applications and Interpretation, regression is never treated as magic. It is treated as an approximation--a claim about the world that needs evidence.
A good evaluation of model fit does three things:
It tests reliability: are predictions consistently close to observed values?
It checks bias: does the model systematically overestimate or underestimate?
R² measures how much variation in the dependent variable is explained by the model:
R² = 1 - (SS_res / SS_tot)
R² = 1 means perfect fit (rare in real data).
R² close to 0 means the model explains little.
In IB writing, the key is not just stating “R² is high.” The key is saying what it implies.
Examiner-ready interpretation template:
“An R² of 0.89 suggests the model explains about 89% of the variability in ____, so the overall trend is strong within the observed range. However, residual analysis is still required to check for systematic bias.”
MAE is IB-friendly because it is intuitive. It stays in the same units as your data, so you can interpret it in plain language.
How to explain MAE in IB:
“The MAE is 2.3 cm, meaning the model’s predictions are typically about 2.3 cm away from the measured values.”
Student tries to interpret MAE
Root Mean Square Error (RMSE)
RMSE is similar to MAE but penalizes large errors more:
RMSE = sqrt( Σ(observed - predicted)² / n )
Use RMSE when you want a metric that reacts strongly to outliers or occasional big misses.
IB interpretation hint: If RMSE is much larger than MAE, that often means a few large errors are pulling it up. That is a great moment to talk about influential points or measurement noise.
Correlation coefficient (r)
Pearson’s r measures strength and direction of a linear relationship:
r close to 1: strong positive linear association
r close to -1: strong negative linear association
r close to 0: little linear association
In IB, correlation helps, but it never replaces residual analysis, and it never proves causation.
How to combine tools into one strong IB evaluation paragraph
A reliable structure is:
Visual fit: does the curve track the data broadly?
Residual check: any patterns or bias?
Fit statistic: R² for explained variation.
Error statistic: MAE or RMSE for typical prediction error.
Reflection: limitations and what you would try next.
Example paragraph you can adapt:
“The quadratic regression captures the overall trend in the dataset, particularly in the mid-range of x. Residuals are mostly scattered around 0 with no strong pattern, suggesting the model is not systematically biased, although two high-x values show larger positive residuals. The R² value of 0.91 indicates the model explains approximately 91% of the variation in y. The MAE of 1.8 units suggests predictions are typically within 1.8 units of observed values, which is reasonable given the measurement precision. However, the larger residuals at the upper end imply the relationship may change beyond the observed range, so extrapolation would be unreliable.”
That is the IB balance: confidence without overclaiming.
Common IB mistakes when evaluating model fit
Treating R² like a final grade
A high R² can still hide a curved residual pattern. IB rewards students who say, “R² is strong, and here is what residuals show.”
Ignoring units when reporting error
MAE = 2.3 is meaningless unless you say 2.3 what. IB clarity is mathematical communication.
Overfitting to chase perfection
If you add complexity just to raise R², you may create a model you cannot justify. IB prefers a model you can explain and defend.
Simple vs overfit model joke
Copying calculator output without interpretation
Calculator numbers do not earn marks. Your explanation does.
How many statistics should I use to evaluate model fit in IB?
In IB, one metric rarely tells the whole story, so using a small set is smarter than using a long list. A strong minimum is residual analysis plus R² plus one error measure such as MAE or RMSE. Residual analysis shows whether your model is biased, while R² summarizes explained variation, and MAE/RMSE describe typical prediction error. This combination lets you discuss both the shape of the errors and the size of the errors. Examiners tend to reward students who triangulate evidence rather than relying on one impressive-looking number. If you have room, comparing two models with the same toolkit can also show depth.
What should I say if my model fit is not great?
A weaker model fit is not automatically a problem in IB, especially if your reflection is strong. Start by describing the evidence honestly: residual patterns, low R², or large MAE/RMSE. Then explain plausible reasons in context, such as measurement limitations, missing variables, or a relationship that changes across ranges. After that, propose a realistic improvement: a different model family, restricting the domain, collecting more data, or treating an outlier separately with justification. This shows mathematical maturity, because real modeling is rarely perfect. In many IB contexts, a careful evaluation of why a model fails can score better than an unexamined “perfect” fit.
How do I write model fit evaluation in an exam-style response?
In an IB exam, you want short sentences that do three jobs: state the statistic, interpret it, and mention a limitation. For example, report R² and interpret what percentage of variation is explained, then add a note about residuals or the observed range. If you mention correlation, specify direction and strength and remind that it is not causation if context invites that pitfall. If you mention MAE or RMSE, interpret it using units and say whether that error is reasonable relative to typical y-values. The best exam responses avoid vague words like “good” unless they are tied to evidence. Practice this phrasing with timed sets in the Questionbank so it becomes automatic.
Conclusion: IB model fit is trust, not decoration
A model is a story about reality. In IB, you do not get marks for telling the story confidently. You get marks for proving the story holds up when tested.
Evaluate model fit with residual analysis, R², MAE/RMSE, and clear interpretation. Make the model earn your belief. Then write that judgment with calm precision.
When you want to practice the full loop--learn the tools, apply them to exam questions, and get feedback that sounds like an IB examiner--RevisionDojo brings it together with Study Notes, Flashcards, AI Chat, the Questionbank, Grading tools, Mock Exams, Predicted Papers, the Coursework Library, and Tutors.
IB students do not need more math panic. They need a system. RevisionDojo is that system.
· 2 min read
How to Sleep Properly During the IB Exam Period
Learn how to sleep properly during the IB exam period to improve focus, memory, and exam performance.
