2.6 · SLPaper 1Paper 2

Fitting and testing models

In a minute

Modelling is a cycle: choose a model type, find its , test it against the facts, then use it to predict. Most marks go to the reasons: why this model, which , and how far you can trust a prediction.

You'll be able to

  • Follow the modelling process to choose, fit, test and use a model from 2.5.
  • Justify a model from the shape of the data, the properties of its graph or the context.
  • Find parameters from an , from given points, or from up to three linear equations solved on your GDC.
  • State a reasonable domain for a model in context.
  • Test a model against data and comment on whether it is appropriate and reasonable.
  • Make predictions from a model and explain the dangers of .
Before you start3 quick questions on what this builds on
Check yourself
What is the equation of the horizontal asymptote of y=8+50(0.7)xy = 8 + 50(0.7)^{x}?
Check yourself
Solve the system x+y=14x + y = 14 and 3x+y=303x + y = 30. What is the value of xx?
Check yourself
A model h(t)=30−6th(t) = 30 - 6t gives the height in cm of a melting ice sculpture after tt hours. Which is a sensible domain?

Revise first: 1.8 Solving equations with technology, 2.2 Function concepts, domain and range and 2.5 Modelling with functions

The modelling process

A is a function that describes a real situation, approximately. You build one in four steps, and you go round again if the test fails.

1. Choose a model type 2. Fit the parameters 3. Test against data 4. Use to predict not reasonable? choose again or change the domain
The modelling cycle.

Step 1 uses the families from 2.5. Choose from the context, the shape of the data, or both.

Same change each step, such as a fixed charge per hour: linear. Rises to a peak and falls back symmetrically (or falls to a lowest point and rises again): quadratic.

Same percentage change each step, or changing fastest at the start and then levelling off towards a value: exponential. Repeats at regular intervals: sinusoidal. Different rules on different intervals: piecewise.

With a table of data, test the pattern. Equal differences between yy-values (for equal steps in xx) suggest linear. Equal ratios suggest exponential.

Worked exampleChoose a type of model for each situation, and give a reason. (a) A plumber charges a call-out fee plus a fixed amount per hour. (b) A pie taken out of the oven cools towards the kitchen's temperature of 22 ∘C22\,^{\circ}\mathrm{C}. (c) The water depth at a harbour over two days.

Try it first, then reveal one step at a time.

  1. (a) Each extra hour adds the same amount, so the cost is linear: C=mt+cC = mt + c. cc is the call-out fee and mm the rate per hour.
Worked exampleThe area, A m2A\text{ m}^{2}, of a pond covered by water lilies is measured each week. Weeks ww: 0,1,2,30, 1, 2, 3. Areas: 20,26,33.8,43.9420, 26, 33.8, 43.94. (a) Show that a linear model is not suitable. (b) Justify an exponential model, and find it.

Try it first, then reveal one step at a time.

  1. (a) Differences: 26−20=626 - 20 = 6, 33.8−26=7.833.8 - 26 = 7.8, 43.94−33.8=10.1443.94 - 33.8 = 10.14. They grow, so the data is not linear.
Check yourself
A table gives y=7,11,15,19y = 7, 11, 15, 19 for x=0,1,2,3x = 0, 1, 2, 3. Which type of model fits it?
Check yourself
The percentage of a city's homes with solar panels is 4%4\% now. It grows quickly at first, but can never pass 100%100\%. Which model suits this best?

Finding the parameters

Once you've chosen the type, each unknown parameter needs one fact about the situation. There are three ways to use those facts.

1. Initial condition: put t=0t = 0. In kat+cka^{t} + c this gives k+ck + c, since a0=1a^{0} = 1.

2. Substitute a point: a known value, such as f(10)=95f(10) = 95, gives one equation. Solve it on your GDC, writing the equation down first.

3. Simultaneous equations: if several parameters are unknown, substitute one point per parameter. For ax2+bx+cax^{2} + bx + c each point gives a linear equation in aa, bb and cc. Solve the on your GDC.

Worked exampleA loaf of bread comes out of the oven at 180 ∘C180\,^{\circ}\mathrm{C} into a kitchen at 22 ∘C22\,^{\circ}\mathrm{C}. Its temperature after tt minutes is modelled by T(t)=22+ke−rtT(t) = 22 + k\mathrm{e}^{-rt}. After 1515 minutes it is 95 ∘C95\,^{\circ}\mathrm{C}. (a) Find kk. (b) Find rr, to 3 s.f.

Try it first, then reveal one step at a time.

  1. (a) Initial condition: T(0)=180T(0) = 180. Since e0=1\mathrm{e}^{0} = 1, 22+k=18022 + k = 180, so k=158k = 158.
Worked exampleA basketball's path is modelled by h=ax2+bx+ch = ax^{2} + bx + c, where hh m is its height when it is xx m horizontally from the player. It passes through (1,2.6)(1, 2.6), (3,4.2)(3, 4.2) and (5,3.8)(5, 3.8). (a) Use your GDC to find aa, bb and cc. (b) Find the greatest height of the ball.

Try it first, then reveal one step at a time.

  1. Substitute each point: a+b+c=2.6a + b + c = 2.6, 9a+3b+c=4.29a + 3b + c = 4.2 and 25a+5b+c=3.825a + 5b + c = 3.8.
Check yourself
Membership of a new climbing club, tt months after it opens, is modelled by M(t)=a−b(0.6)tM(t) = a - b(0.6)^{t}. It opens with 4040 members, and after 22 months it has 104104. Find aa, the number the membership levels off at.
Check yourself
A plant's height in cm, tt days after planting, is modelled by H(t)=k(1.08)t+3H(t) = k(1.08)^{t} + 3. It is 55 cm tall when planted. Find H(10)H(10), to 3 s.f.

A reasonable domain

A formula may give an output for any xx, but the situation does not allow every xx. The of a model is the set of inputs where it makes sense.

Ask three questions. Where does it start (often t≥0t \ge 0)? Where does it stop making sense (a height or a price reaches 00, a tank is empty)? Must the input be a whole number (people, tickets, items)?

Worked exampleA stone is thrown upwards from a cliff. Its height above the sea, hh m, tt seconds later is modelled by h(t)=10+5t−5t2h(t) = 10 + 5t - 5t^{2}. State a reasonable domain.

Try it first, then reveal one step at a time.

  1. The model starts when the stone is thrown, so t≥0t \ge 0.
Worked exampleA theatre sets the ticket price at pp dollars. It expects to sell 800−40p800 - 40p tickets, so its revenue is R(p)=p(800−40p)R(p) = p(800 - 40p) dollars. (a) State a reasonable domain for pp. (b) Comment on using the model for p=2.35p = 2.35.

Try it first, then reveal one step at a time.

  1. (a) A price can't be negative, so p≥0p \ge 0.
Check yourself
A pool holds 54 00054\,000 litres. When it is drained, the volume left after tt minutes is V(t)=54 000−1800tV(t) = 54\,000 - 1800t. Which is a reasonable domain?
Check yourself
A bakery's daily profit is modelled by P(n)=2.4n−90P(n) = 2.4n - 90 dollars, where nn is the number of loaves sold. Its oven bakes at most 160160 loaves a day. Which describes a reasonable domain?

Testing a model

A model is appropriate if it matches the data and behaves sensibly where the data stops. Test both.

1. Test against data: compare the model's values vAv_{A} with the measured values vEv_{E}. A small percentage error ε=∣vA−vEvE∣×100%\varepsilon = \left|\frac{v_{A} - v_{E}}{v_{E}}\right| \times 100\%, from 1.6, means a close fit.

2. Test against the context: check the graph's features. Does it go negative when it can't? Does it keep rising when it must level off? Does an asymptote or a maximum match a real limit?

Worked exampleA town's yearly water use, in megalitres, tt years after 2016 is modelled by W(t)=420(1.035)tW(t) = 420(1.035)^{t}. The actual use in 2024 was 545545 megalitres. (a) Find the percentage error of the model for 2024. (b) Comment on the model.

Try it first, then reveal one step at a time.

  1. (a) 2024 is t=8t = 8. W(8)=420(1.035)8=553.059…W(8) = 420(1.035)^{8} = 553.059\ldots megalitres.
data: weeks 1 to 5 extrapolation S (thousand copies) w (weeks) −10 10 20 30 0 2 4 6 8 10 12 S = 30 model A model B
Both models fit the data. Only one makes sense beyond it.
Worked exampleTotal sales of a new video game, SS thousand copies, after ww weeks: weeks 11 to 55 give 12,20,25,27,2812, 20, 25, 27, 28. Two models are suggested: A: S=−1.25w2+11.5w+1.5S = -1.25w^{2} + 11.5w + 1.5; B: S=30−30(0.55)wS = 30 - 30(0.55)^{w}. Both fit the data closely. Find each model's prediction for week 1212, and justify which model is more appropriate.

Try it first, then reveal one step at a time.

  1. Model A: S(12)=−1.25(144)+11.5(12)+1.5=−40.5S(12) = -1.25(144) + 11.5(12) + 1.5 = -40.5.
Check yourself
The percentage of homes in a town with fast broadband, tt years from now, is modelled by P=100−70(0.6)tP = 100 - 70(0.6)^{t}. Which is a mathematical reason the model never reaches 100%100\%?
Check yourself
A museum's model predicts 12401240 visitors on a Saturday. The actual number is 13101310. Find the percentage error of the model, to 3 s.f.

Predictions and extrapolation

Using a model means substituting into it, or solving it for an input. Then say what the answer means in context, with units.

Exponential growth is the classic danger. Food, space or money runs out, so real growth slows. The further you extrapolate, the less you should trust the answer.

Worked exampleRabbits are introduced to an island. Counts for years 00 to 44 fit the model N(t)=60(1.4)tN(t) = 60(1.4)^{t}, where tt is in years. (a) Estimate the number of rabbits after 2.52.5 years. (b) Use the model to predict the number after 1515 years. (c) Comment on both predictions.

Try it first, then reveal one step at a time.

  1. (a) N(2.5)=60(1.4)2.5=139.14…N(2.5) = 60(1.4)^{2.5} = 139.14\ldots, so about 139139 rabbits.
Worked exampleAn app's total downloads, DD thousand, tt months after launch are modelled by D(t)=2.5×1.6tD(t) = 2.5 \times 1.6^{t}. The model was built from data for the first 44 months. (a) Use your GDC to find when downloads reach 6262 thousand. (b) Comment on the reliability of your answer.

Try it first, then reveal one step at a time.

  1. (a) Write the equation: 2.5×1.6t=622.5 \times 1.6^{t} = 62.
Check yourself
A model for the temperature of an oven was built from readings taken every minute from t=2t = 2 to t=10t = 10 minutes. Which prediction is interpolation?
Check yourself
The number of fish in a lake tt years after a survey is modelled by F(t)=1800(0.92)tF(t) = 1800(0.92)^{t}. Use the model to estimate the number of fish after 55 years, to the nearest integer.

Common mistakes

  • Rounding a parameter, then using the rounded value. Store the GDC's full value and round only the final answer.
  • Giving a 3 s.f. decimal when the exact value of a parameter is asked for. Write the exact value.
  • Accepting an impossible prediction, such as negative sales or more people than live in a country. Say the model fails there.
  • Giving a context reason when a mathematical one is asked for. Name a feature of the graph, such as the asymptote.
  • Leaving a count as a decimal when the nearest integer is asked for. Round once, at the end.

Exam tips

  • "Find" a parameter: write the equation you substitute into, then solve it. The written equation earns the method mark even if the GDC answer slips.
  • "Comment" on a model: give a judgement (reasonable or not) and a reason from the data or the context.
  • "Justify" a model choice: point to the data's pattern (differences, ratios, levelling off) or a property of the curve that matches the context.

Key takeaways

Learn thisModelling process: choose a model type, fit its parameters, test it against data and context, then use it. If the test fails, change the model or its domain.
Learn thisFor equal steps in xx: equal differences in yy suggest a linear model; equal ratios suggest an exponential model.
Learn thisParameters: an initial condition (t=0t = 0) gives one equation, and each known point gives another. A quadratic through three points with different xx-values gives three linear equations in aa, bb, cc: solve them on the GDC.
Learn thisA reasonable domain starts where the situation starts, stops where the model stops making sense (such as a height reaching 00), and uses whole numbers for counts.
Learn thisTest a model: compare it with data (percentage error ε=∣vA−vEvE∣×100%\varepsilon = \left|\frac{v_{A} - v_{E}}{v_{E}}\right| \times 100\%, with vAv_{A} the model value and vEv_{E} the actual value) and check its graph's features, such as asymptotes and maxima, against the context.
Learn thisInterpolation predicts inside the data range; extrapolation predicts outside it and is less reliable, because the trend may not continue.

Practise this

15 past-paper questions on Modelling with exponential functions in the AI HL bank, each with a mark scheme and a worked explanation.

Practise Modelling with exponential functions →