JC2 H2 Maths Bukit Timah | Regression Transforms

Did you know? A graphing calculator might tell you that the best straight-line fit is ln y = 1 + 0.2x. If the question asks you to estimate y when x = 5, the answer is not 2. The expression 1+0.2(5)=2 belongs to ln y, the transformed variable. To return to the original measured quantity, you must undo the logarithm: y = e² ≈ 7.389. Reporting 2 would put a mathematically valid regression result on the wrong scale.

This JC2 H2 Mathematics Tutor Bukit Timah article focuses on one important A-Level syllabus 9758 decision: after fitting a straight line to a square, reciprocal or logarithmic transformation, how do we return a prediction to the original units? The guide starts with a quick conversion table, provides worked examples and then tests students on changed models and limits. It deliberately avoids turning a focused reader question into an unfocused list of every regression formula.

For families supporting JC2 H2 Mathematics learning in Bukit Timah, the practical question is what the variables mean, not just whether a calculator can find a and b. Regression is useful when the original relationship is curved but a suitable transformation produces an approximately linear pattern. The 2027 Singapore–Cambridge H2 syllabus includes square, reciprocal and logarithmic transformations to achieve linearity. The same syllabus excludes deriving the general regression formulae and hypothesis tests for correlation or regression. This guide keeps those boundaries visible.

The quick answer: undo the transformation of the quantity you predict

Fitted straight line	Original-scale prediction	Check before using
ln y = a + bx	y = e^(a+bx)	y must be positive; use exp on the fitted logarithm
y² = a + bx	y = √(a+bx) when y ≥ 0	Check the model’s sign and a+bx ≥ 0
1/y = a + bx	y = 1/(a+bx)	Check a+bx ≠ 0; respect y’s sign
y = a + b ln x	y = a + b ln x	Only x is transformed; do not exponentiate y
y = a + b/x	y = a + b/x	Only x is transformed; keep x ≠ 0
ln y = a + b ln x	y = e^a × x^b (x > 0)	Both axes are logarithmic; predict y on its original scale

The column label is the decision tool. If the fitted response is ln y, your calculator predicts a logarithm; exponentiate to obtain y. If the fitted response is y², take the square root on the appropriate sign branch. If it is 1/y, take the reciprocal. But if the transformation was made only to the predictor x, you have already fitted y in its original scale and should not undo a transformation of y that never occurred.

Identify the response: what is being predicted — y, ln y, y² or 1/y?
Choose the correct regression direction: use the fitted response-on-predictor model for the target requested.
Substitute before untransforming: calculate the transformed response at the given predictor value.
Undo exactly once: exp for ln, a square root for a positive squared response, and reciprocal for 1/y.
Inspect limits: predictions far outside the data range are extrapolations and may be unreliable.
Write units: a prediction of ln y is not automatically a y value in the original unit.

The first complete worked example: exponential-looking observations

Suppose a fictional dataset suggests the linear relationship ln y = 1 + 0.2x. At x = 0, the model gives ln y = 1 and so y = e ≈ 2.718. At x = 5, ln y = 2, giving y = e² ≈ 7.389. At x = 10, ln y = 3, giving y = e³ ≈ 20.086. The original y values grow along a curved pattern, even though the transformed logarithms 1, 2 and 3 fall on a straight line.

Notice what the slope 0.2 means. It says that when x increases by one unit, the fitted ln y increases by 0.2. On the original y scale, the fitted y is multiplied by e^0.2 ≈ 1.2214 for each unit increase in x. That is a multiplicative change, not a constant additive increase of 0.2 units in y. Replacing the logarithmic interpretation with a raw y slope would misstate the model.

If the question asks whether the prediction at x=20 is trustworthy but the observed x-values only covered 0 to 10, the formula still produces a number: y=e^5≈148.413. Yet that number is an extrapolation. The assumption that the transformed linear relationship continues beyond the observed range has not been verified. A good answer can calculate the formal prediction and explain its limitation without pretending that the model is a natural law.

Example two: squaring the response to achieve linearity

Imagine a fictional non-negative response y for which the transformed data fit y² = 4 + 3x. At x=4, first calculate y²=4+12=16, then return to the original scale: y=4. The result 16 is a predicted squared response, not the original response. Similarly, at x=1, y=√7≈2.646. The square-root step is necessary even if the regression equation looks simple.

Why insist on “non-negative response”? Because y²=16 has two algebraic real solutions, y=±4. A squared-response fit by itself does not select a unique sign. If the context specifies a length, mass or another quantity that must be non-negative, choose the positive branch. If y may be negative, the model and further context need to resolve the ambiguity. Blindly taking the positive square root is not universally justified by the algebra alone.

The transformed line 4+3x also becomes negative when x<−4/3. A real squared response cannot equal a negative number, so those x values do not yield a valid real y under the model. This is a stronger diagnostic than claiming “the curve might become less accurate”: it identifies a mathematical domain restriction. A learner should not force a real-valued square root of a negative prediction merely because the regression equation was obtained by calculator.

Example three: a reciprocal response

Suppose the fitted relation is 1/y = 0.1 + 0.02x, with y positive and the relevant x-values non-negative. At x=5, the fitted reciprocal is 0.1+0.1=0.2, so the original-scale prediction is y=1/0.2=5. At x=15, the reciprocal is 0.4 and the predicted y is 2.5. The model captures a decreasing response without fitting a straight line directly to y.

The most common error is to report 0.2 instead of 5, or to interpret the fitted slope 0.02 as an increase of y by 0.02 per unit x. The slope belongs to 1/y, not to y. The original response actually declines with x for this example. Whenever the transformed variable is a reciprocal, do an obvious direction check at two x-values: the reciprocal can increase while the original positive quantity decreases.

A second caution concerns the denominator. If 0.1+0.02x=0, the predicted reciprocal is zero, which cannot correspond to a finite nonzero y through y=1/(0.1+0.02x). At x=−5 the model has a singularity. Predictions across that point should not be smoothed into a continuous finite outcome. Check both the mathematical domain and the observed data range.

Example four: the predictor is transformed, not the response

Now reverse the direction of the transformation, not the direction of the regression. Consider the fitted model y=3+2 ln x, where x>0. When x=e², ln x=2, so y=3+4=7. No exponentiation of y is required. The equation already predicts the original response y. The only transformed quantity is ln x on the input side.

For a reciprocal predictor, use y=5+12/x, x≠0. At x=4, y=5+3=8. If you invert 8, you would have undone a transformation of y that did not exist. The labels on the graph axes are therefore essential: “plot y against 1/x” and “plot 1/y against x” may both produce straight lines, but they produce different original-scale prediction formulae.

One more example: a log-log straight line

In an example where both variables are positive and the plot of ln y against ln x is straight, write ln y = 0.5 + 1.5 ln x. At x=4, ln y=0.5+1.5 ln 4. Returning to the original scale gives y=e^0.5×4^1.5=8e^0.5≈13.1898. This is a power-law-looking response despite being linear in the two logarithms.

Notice the different roles of the transformations. We compute ln x because it is the transformed predictor, then exponentiate the fitted ln y because it is the transformed response. A learner who takes only one of those steps gets a result on the wrong scale. When reading a question, write both transformed axis labels explicitly before manipulating the equation.

How transformed data form a straight line

A transformation is useful when the scatter of raw observations is curved but a transformed plot is plausibly linear. Consider a fictional set of three exact model points from y=e^(1+0.2x) at x=0, 5 and 10. Their y values are about 2.718, 7.389 and 20.086, which do not rise by a fixed additive amount. But their ln y values are 1, 2 and 3, which increase by one for each five-unit change in x. The logarithmic plot has slope 0.2.

In a real dataset, there is usually scatter around the fitted straight line. The fact that a transformation can make a pattern appear straighter does not establish a perfect fit, prove causation or remove measurement error. A student should examine the transformed scatter diagram and be cautious if the points fan out, bend further or contain a highly influential outlier. Mathematical correctness of back-transformation cannot rescue a poorly justified fitted relationship.

An important interpretation limit: a back-transformed fit is not always a raw-scale mean

When observations have scatter in ln y, exponentiating a fitted logarithmic line gives a typical model-implied original-scale value, but it need not equal the conditional arithmetic mean of y. This is because exponentiation is nonlinear. The distinction goes beyond most routine H2 numerical prediction questions, which ordinarily ask you to calculate the transformed-line estimate and convert back, but it is worth keeping conceptually honest: a fitted curve is not the same thing as an exact population law.

In an H2 answer, follow the question’s stated model and wording, perform the transformation and label the estimate. Avoid adding unsupported confidence claims such as “this will certainly be the next measurement”. A useful prediction is a numerical estimate with clearly identified variables, assumptions and range of validity.

The one-minute diagnostic

Give a learner four equations on a sheet: ln y = 2+0.1x; y²=9+x; 1/y=0.1+0.05x; and y=4+2 ln x. Ask, “Which expression must I undo before writing y?” The answers are: exponentiate the first, take a context-appropriate square root of the second, invert the third, and leave the fourth response alone. Do not show any calculator screens initially.

Then supply x=2. The first predicts y=e^2.2≈9.025, the second predicts y=√11≈3.317 for y≥0, the third predicts y=5, and the fourth predicts y=4+2 ln 2≈5.386. These changed-case answers probe scale identification, not arithmetic speed. A learner who consistently labels “predicted ln y”, “predicted y²” and “predicted 1/y” before conversion has built a reliable working habit.

Thirty-six original model-reading clinics

All 36 clinics below use hypothetical mathematical models, not actual local experimental measurements, student grade records or claims about real groups. Each calculation is shown on the scale requested by the fictional question. Work three or four from each family, then return a day later to another family; the aim is to recognise the transformed response from the equation, not memorise a particular list of model coefficients.

1. Log-response clinic 1

Try the invented model. A transformed scatter plot has led to the fitted equation ln y = 0.6 + 0.08x. Estimate the original-scale y when x=2. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. ln y = 0.6 + 0.08(2) = 0.76. Exponentiate the fitted log response: y=e^(0.76)≈2.13828. The original y must be positive, as required by a real logarithm. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

Transfer and audit. If x increases by one, the fitted y is multiplied by e raised to the slope, not increased by the slope in raw units. Change the predictor slightly and predict the direction of the response before calculating again. Identify the range of observed data that would be needed to decide whether the new substitution is interpolation or extrapolation. A fitted equation alone cannot certify that a distant prediction is reliable.

2. Square-response clinic 1

Try the invented model. A transformed scatter plot has led to the fitted equation y² = 9 + 2x. Estimate the original-scale y when x=2. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. y² = 9+2(2)=13. Use the non-negative root, y=√13≈3.60555, not 13. The exercise declares y to be non-negative; otherwise a squared value would not determine its sign. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

Transfer and audit. If the fitted squared response decreases below zero at another x, it cannot produce a real non-negative y. Change the predictor slightly and predict the direction of the response before calculating again. Identify the range of observed data that would be needed to decide whether the new substitution is interpolation or extrapolation. A fitted equation alone cannot certify that a distant prediction is reliable.

3. Reciprocal-response clinic 1

Try the invented model. A transformed scatter plot has led to the fitted equation 1/y = 0.12 + 0.015x. Estimate the original-scale y when x=2. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. 1/y = 0.12+0.015(2)=0.15. Take the reciprocal: y=1/0.15≈6.66667. The fitted reciprocal must be nonzero; the chosen x gives a positive original response. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

Transfer and audit. As the fitted reciprocal grows with positive x in this example, its positive original response y becomes smaller. Change the predictor slightly and predict the direction of the response before calculating again. Identify the range of observed data that would be needed to decide whether the new substitution is interpolation or extrapolation. A fitted equation alone cannot certify that a distant prediction is reliable.

4. Log-predictor clinic 1

Try the invented model. A transformed scatter plot has led to the fitted equation y=2+0.8 ln x. Estimate the original-scale y when x=e^1. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=e^1, ln x=1, so y=2+0.8(1). The predicted y is already 2.8. The transformed predictor ln x requires x>0; the response y is not logarithmically transformed. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

Transfer and audit. Do not exponentiate y after substitution, because that would alter a response already given on its original scale. Change the predictor slightly and predict the direction of the response before calculating again. Identify the range of observed data that would be needed to decide whether the new substitution is interpolation or extrapolation. A fitted equation alone cannot certify that a distant prediction is reliable.

5. Reciprocal-predictor clinic 1

Try the invented model. A transformed scatter plot has led to the fitted equation y=3+12/x. Estimate the original-scale y when x=2. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. At x=2, y=3+12/2. The original-scale result is y=9 without an extra reciprocal. The reciprocal predictor 1/x requires x≠0; the fitted response y is not a reciprocal. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

Transfer and audit. As x increases through positive values, the positive fraction b/x falls, drawing y towards its intercept a. Change the predictor slightly and predict the direction of the response before calculating again. Identify the range of observed data that would be needed to decide whether the new substitution is interpolation or extrapolation. A fitted equation alone cannot certify that a distant prediction is reliable.

6. Log-log response clinic 1

Try the invented model. A transformed scatter plot has led to the fitted equation ln y=0.2+1 ln x. Estimate the original-scale y when x=2. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=2, ln y=0.2+1 ln 2. Return to the original scale using y=e^(0.2)×2^(1)≈2.44281. Both x and y are positive in this log-log model. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

Transfer and audit. This transformed straight line represents a power-law-style curve on the original axes; its raw y slope is not the fitted b. Change the predictor slightly and predict the direction of the response before calculating again. Identify the range of observed data that would be needed to decide whether the new substitution is interpolation or extrapolation. A fitted equation alone cannot certify that a distant prediction is reliable.

7. Log-response clinic 2

Try the invented model. A transformed scatter plot has led to the fitted equation ln y = 0.76 + 0.1x. Estimate the original-scale y when x=4. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. ln y = 0.76 + 0.1(4) = 1.16. Exponentiate the fitted log response: y=e^(1.16)≈3.18993. The original y must be positive, as required by a real logarithm. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

8. Square-response clinic 2

Try the invented model. A transformed scatter plot has led to the fitted equation y² = 13 + 3x. Estimate the original-scale y when x=4. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. y² = 13+3(4)=25. Use the non-negative root, y=√25≈5, not 25. The exercise declares y to be non-negative; otherwise a squared value would not determine its sign. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

9. Reciprocal-response clinic 2

Try the invented model. A transformed scatter plot has led to the fitted equation 1/y = 0.14 + 0.019x. Estimate the original-scale y when x=4. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. 1/y = 0.14+0.019(4)=0.216. Take the reciprocal: y=1/0.216≈4.62963. The fitted reciprocal must be nonzero; the chosen x gives a positive original response. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

10. Log-predictor clinic 2

Try the invented model. A transformed scatter plot has led to the fitted equation y=3+1 ln x. Estimate the original-scale y when x=e^2. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=e^2, ln x=2, so y=3+1(2). The predicted y is already 5. The transformed predictor ln x requires x>0; the response y is not logarithmically transformed. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

11. Reciprocal-predictor clinic 2

Try the invented model. A transformed scatter plot has led to the fitted equation y=4+15/x. Estimate the original-scale y when x=4. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. At x=4, y=4+15/4. The original-scale result is y=7.75 without an extra reciprocal. The reciprocal predictor 1/x requires x≠0; the fitted response y is not a reciprocal. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

12. Log-log response clinic 2

Try the invented model. A transformed scatter plot has led to the fitted equation ln y=0.3+1.25 ln x. Estimate the original-scale y when x=3. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=3, ln y=0.3+1.25 ln 3. Return to the original scale using y=e^(0.3)×3^(1.25)≈5.32954. Both x and y are positive in this log-log model. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

13. Log-response clinic 3

Try the invented model. A transformed scatter plot has led to the fitted equation ln y = 0.92 + 0.12x. Estimate the original-scale y when x=6. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. ln y = 0.92 + 0.12(6) = 1.64. Exponentiate the fitted log response: y=e^(1.64)≈5.15517. The original y must be positive, as required by a real logarithm. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

14. Square-response clinic 3

Try the invented model. A transformed scatter plot has led to the fitted equation y² = 17 + 4x. Estimate the original-scale y when x=6. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. y² = 17+4(6)=41. Use the non-negative root, y=√41≈6.40312, not 41. The exercise declares y to be non-negative; otherwise a squared value would not determine its sign. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

15. Reciprocal-response clinic 3

Try the invented model. A transformed scatter plot has led to the fitted equation 1/y = 0.16 + 0.023x. Estimate the original-scale y when x=6. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. 1/y = 0.16+0.023(6)=0.298. Take the reciprocal: y=1/0.298≈3.3557. The fitted reciprocal must be nonzero; the chosen x gives a positive original response. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

16. Log-predictor clinic 3

Try the invented model. A transformed scatter plot has led to the fitted equation y=4+1.2 ln x. Estimate the original-scale y when x=e^3. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=e^3, ln x=3, so y=4+1.2(3). The predicted y is already 7.6. The transformed predictor ln x requires x>0; the response y is not logarithmically transformed. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

17. Reciprocal-predictor clinic 3

Try the invented model. A transformed scatter plot has led to the fitted equation y=5+18/x. Estimate the original-scale y when x=6. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. At x=6, y=5+18/6. The original-scale result is y=8 without an extra reciprocal. The reciprocal predictor 1/x requires x≠0; the fitted response y is not a reciprocal. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

18. Log-log response clinic 3

Try the invented model. A transformed scatter plot has led to the fitted equation ln y=0.4+1.5 ln x. Estimate the original-scale y when x=4. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=4, ln y=0.4+1.5 ln 4. Return to the original scale using y=e^(0.4)×4^(1.5)≈11.9346. Both x and y are positive in this log-log model. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

19. Log-response clinic 4

Try the invented model. A transformed scatter plot has led to the fitted equation ln y = 1.08 + 0.14x. Estimate the original-scale y when x=8. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. ln y = 1.08 + 0.14(8) = 2.2. Exponentiate the fitted log response: y=e^(2.2)≈9.02501. The original y must be positive, as required by a real logarithm. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

20. Square-response clinic 4

Try the invented model. A transformed scatter plot has led to the fitted equation y² = 21 + 5x. Estimate the original-scale y when x=8. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. y² = 21+5(8)=61. Use the non-negative root, y=√61≈7.81025, not 61. The exercise declares y to be non-negative; otherwise a squared value would not determine its sign. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

21. Reciprocal-response clinic 4

Try the invented model. A transformed scatter plot has led to the fitted equation 1/y = 0.18 + 0.027x. Estimate the original-scale y when x=8. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. 1/y = 0.18+0.027(8)=0.396. Take the reciprocal: y=1/0.396≈2.52525. The fitted reciprocal must be nonzero; the chosen x gives a positive original response. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

22. Log-predictor clinic 4

Try the invented model. A transformed scatter plot has led to the fitted equation y=5+1.4 ln x. Estimate the original-scale y when x=e^4. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=e^4, ln x=4, so y=5+1.4(4). The predicted y is already 10.6. The transformed predictor ln x requires x>0; the response y is not logarithmically transformed. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

23. Reciprocal-predictor clinic 4

Try the invented model. A transformed scatter plot has led to the fitted equation y=6+21/x. Estimate the original-scale y when x=8. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. At x=8, y=6+21/8. The original-scale result is y=8.625 without an extra reciprocal. The reciprocal predictor 1/x requires x≠0; the fitted response y is not a reciprocal. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

24. Log-log response clinic 4

Try the invented model. A transformed scatter plot has led to the fitted equation ln y=0.5+1.75 ln x. Estimate the original-scale y when x=5. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=5, ln y=0.5+1.75 ln 5. Return to the original scale using y=e^(0.5)×5^(1.75)≈27.56416. Both x and y are positive in this log-log model. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

25. Log-response clinic 5

Try the invented model. A transformed scatter plot has led to the fitted equation ln y = 1.24 + 0.16x. Estimate the original-scale y when x=10. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. ln y = 1.24 + 0.16(10) = 2.84. Exponentiate the fitted log response: y=e^(2.84)≈17.11577. The original y must be positive, as required by a real logarithm. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

26. Square-response clinic 5

Try the invented model. A transformed scatter plot has led to the fitted equation y² = 25 + 6x. Estimate the original-scale y when x=10. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. y² = 25+6(10)=85. Use the non-negative root, y=√85≈9.21954, not 85. The exercise declares y to be non-negative; otherwise a squared value would not determine its sign. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

27. Reciprocal-response clinic 5

Try the invented model. A transformed scatter plot has led to the fitted equation 1/y = 0.2 + 0.031x. Estimate the original-scale y when x=10. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. 1/y = 0.2+0.031(10)=0.51. Take the reciprocal: y=1/0.51≈1.96078. The fitted reciprocal must be nonzero; the chosen x gives a positive original response. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

28. Log-predictor clinic 5

Try the invented model. A transformed scatter plot has led to the fitted equation y=6+1.6 ln x. Estimate the original-scale y when x=e^5. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=e^5, ln x=5, so y=6+1.6(5). The predicted y is already 14. The transformed predictor ln x requires x>0; the response y is not logarithmically transformed. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

29. Reciprocal-predictor clinic 5

Try the invented model. A transformed scatter plot has led to the fitted equation y=7+24/x. Estimate the original-scale y when x=10. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. At x=10, y=7+24/10. The original-scale result is y=9.4 without an extra reciprocal. The reciprocal predictor 1/x requires x≠0; the fitted response y is not a reciprocal. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

30. Log-log response clinic 5

Try the invented model. A transformed scatter plot has led to the fitted equation ln y=0.6+2 ln x. Estimate the original-scale y when x=6. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=6, ln y=0.6+2 ln 6. Return to the original scale using y=e^(0.6)×6^(2)≈65.59628. Both x and y are positive in this log-log model. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

31. Log-response clinic 6

Try the invented model. A transformed scatter plot has led to the fitted equation ln y = 1.4 + 0.18x. Estimate the original-scale y when x=12. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. ln y = 1.4 + 0.18(12) = 3.56. Exponentiate the fitted log response: y=e^(3.56)≈35.1632. The original y must be positive, as required by a real logarithm. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

32. Square-response clinic 6

Try the invented model. A transformed scatter plot has led to the fitted equation y² = 29 + 7x. Estimate the original-scale y when x=12. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. y² = 29+7(12)=113. Use the non-negative root, y=√113≈10.63015, not 113. The exercise declares y to be non-negative; otherwise a squared value would not determine its sign. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

33. Reciprocal-response clinic 6

Try the invented model. A transformed scatter plot has led to the fitted equation 1/y = 0.22 + 0.035x. Estimate the original-scale y when x=12. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. 1/y = 0.22+0.035(12)=0.64. Take the reciprocal: y=1/0.64≈1.5625. The fitted reciprocal must be nonzero; the chosen x gives a positive original response. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

34. Log-predictor clinic 6

Try the invented model. A transformed scatter plot has led to the fitted equation y=7+1.8 ln x. Estimate the original-scale y when x=e^6. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=e^6, ln x=6, so y=7+1.8(6). The predicted y is already 17.8. The transformed predictor ln x requires x>0; the response y is not logarithmically transformed. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

35. Reciprocal-predictor clinic 6

Try the invented model. A transformed scatter plot has led to the fitted equation y=8+27/x. Estimate the original-scale y when x=12. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. At x=12, y=8+27/12. The original-scale result is y=10.25 without an extra reciprocal. The reciprocal predictor 1/x requires x≠0; the fitted response y is not a reciprocal. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

36. Log-log response clinic 6

Try the invented model. A transformed scatter plot has led to the fitted equation ln y=0.7+2.25 ln x. Estimate the original-scale y when x=7. Before calculating, underline the transformed variable and state whether the final result should be exponential, a square root, a reciprocal or unchanged. The variable names, not the appearance of a convenient calculator menu, determine your method.

Complete working. For x=7, ln y=0.7+2.25 ln 7. Return to the original scale using y=e^(0.7)×7^(2.25)≈160.50062. Both x and y are positive in this log-log model. Report the number as a fitted estimate from this hypothetical model, not as a measured outcome. A result on the transformed scale would answer a different mathematical question.

Six more checkable regression traps

Fitted ln y versus actual y: 1 + 0.2(5) = 2 is the fitted logarithm, so y=e², not 2.
Fitted y² versus signed y: y²=16 means y=±4 algebraically; positive y requires a domain or contextual assumption.
Fitted reciprocal versus original y: 1/y=0.25 means y=4, not 0.25.
Transforming the wrong axis: y=3+2 ln x predicts y directly; its fitted 3+2 ln x must not be exponentiated.
Crossing a singularity: 1/y=a+bx cannot give a finite original prediction when a+bx=0.
Extrapolation without warning: a formula can return a number outside the data range without proving that the underlying relationship holds there.

How to explain a transformed model in one answer paragraph

A clear examination-style interpretation might say: “The fitted transformed regression equation is ln y = 1 + 0.2x. When x=5, the predicted logarithm is 2; converting back gives a modelled y of e²≈7.389 original units. This estimate relies on the fitted relationship and should be treated cautiously if x=5 lies outside the observed predictor range.” That passage identifies the response, shows the conversion and states the appropriate limitation without introducing unnecessary technical jargon.

If the transformed fit uses a squared response, the wording should include why the positive or negative branch is chosen. If it uses a reciprocal, state the denominator restriction. These details are not decorative caveats: they determine whether a numerical prediction is mathematically meaningful. An elegant regression line on the wrong response scale is an incorrect answer to the reader’s actual question.

A practical four-week revision routine

Week 1: mark x and y axes for four transformed scatter diagrams and practise deciding which response is fitted. Week 2: rebuild the original-scale equations for logarithmic, square and reciprocal response transformations, using small numbers that can be audited mentally. Week 3: vary the predictor and compare interpolation with extrapolation. Week 4: solve mixed questions where the transformation is not announced in the title; include one case with an invalid square root, one invalid reciprocal and one positive log-response model.

In a tutorial of up to three learners, one student can explain the graph axes, another can perform the fitted-line substitution and the third can attempt the back-transformation and plausibility checks. Rotating those roles changes the activity from calculator button memorisation into shared mathematical reasoning. The goal is not to insist every curve has a neat transformation, but to understand what a proposed transformation implies and where its interpretation breaks down.

Frequently asked questions

Why is 2 not the answer when ln y=2?

Because 2 is the value of the transformed response ln y. Exponentiate both sides to obtain y=e².

If a fitted equation predicts y²=25, is y always 5?

Only when the context or model specifies y≥0. Algebraically, y could be ±5 unless additional information chooses a branch.

What if 1/y=0?

There is no finite nonzero real y whose reciprocal is exactly zero. The fitted reciprocal form cannot be inverted to a finite value at that predictor.

Does a transformed linear relationship prove causation?

No. It describes an association fitted to the available data. Causation would require additional design and evidence.

Why can two regression directions give different predictions?

Predicting y from x and predicting x from y generally involve different least-squares fits. The separate regression direction tutorial covers that question; this guide covers returning a transformed response to its original scale.

Can I trust a predicted y outside the original x-range?

Not automatically. Such a value is an extrapolation and depends on an unverified continuation of the fitted relationship.

Does the syllabus include logarithmic and reciprocal transformations?

Yes. The published 2027 H2 Mathematics 9758 content specifies square, reciprocal and logarithmic transformations in correlation and regression.

Is deriving the regression formula required?

The 2027 H2 syllabus explicitly excludes problems involving derivation of the regression formulae. Learners still need to select, interpret and use appropriate fitted lines.

Internal learning links and curriculum authority

Complex Conjugation or Multiplying by i?
Regression Transforms Back to Original Units — this guide
Critical Region or p-Value?
Recovering n and p from Binomial Mean and Variance

For wider background, read Bukit Timah H2 correlation and regression and the JC2 Bukit Timah tuition hub. The eduKateSG small-group tutoring reference in Clementi describes the teaching format at another level and location; it is not an authority for statistical models.

Sources and editorial limits

Official curriculum: SEAB H2 Mathematics (9758) 2027 syllabus, Section 6.6 Correlation and Linear Regression. Mathematical functions, coefficients, exercise datasets and scenarios above are fabricated teaching examples rather than official examination questions, real pupil records or measured outcomes. A correct numerical fitted estimate is not a promise of what a future observation or examination grade will be.

JC2 H2 Mathematics Tutor Bukit Timah | Regression Transforms Back to Original Units