THE VR SCHOOL
  • Schools & Districts
    • License the academy
    • Free program for your students
    • Request a demo
  • Catalog
    • AI & Data
    • Cybersecurity & Web
    • Programming
    • Health Science
    • All Guilds
    • All courses
  • Families
  • Accreditation
  • Log In
  • Request a demo
  • Home
  • Courses

RLP Statistics A — Exploring Data, Study Design and Probability

Curriculum

  • 4 Sections
  • 20 Lessons
  • 10 Weeks
Expand all sectionsCollapse all sections
  • Unit 1: Exploring One-Variable Data
    5
    • 1.1
      Distributions and the Statistical Process
      50 mins
    • 1.2
      Measuring Center and Spread
      50 mins
    • 1.3
      Comparing Distributions Across Missouri
      50 mins
    • 1.4
      Standardizing, Transforming and Reporting
      50 mins
    • 1.5
      Performance Task — Climate Station Profile
      150 mins
  • Unit 2: Two-Variable Data and Regression
    5
    • 2.1
      Scatterplots and Correlation
      50 mins
    • 2.2
      Least-Squares Regression
      50 mins
    • 2.3
      Residuals, Influence and Causation
      50 mins
    • 2.4
      Exponential Models and the Limits of Extrapolation
      50 mins
    • 2.5
      Performance Task — Bivariate Investigation Report
      150 mins
  • Unit 3: Designing Studies: Samples and Experiments
    5
    • 3.1
      Populations, Samples and Bias
      50 mins
    • 3.2
      Simulating Sampling Variability and the Margin of Error
      50 mins
    • 3.3
      Experiments, Observational Studies and Confounding
      50 mins
    • 3.4
      Comparing Treatments with a Randomization Test
      50 mins
    • 3.5
      Performance Task — Design, Run and Report a Study
      150 mins
  • Unit 4: Probability, Random Variables and the Binomial Model
    5
    • 4.1
      Probability Rules and Two-Way Tables
      50 mins
    • 4.2
      The Multiplication Rule, Trees and Counting
      50 mins
    • 4.3
      Random Variables, Expected Value and Variance
      50 mins
    • 4.4
      The Binomial Model
      50 mins
    • 4.5
      Performance Task — Expected-Value Decision Analysis
      150 mins

Exponential Models and the Limits of Extrapolation

Unit 2  ·  Modeling & Reporting Workshop  ·  Lesson 9 of 20

Exponential Models and the Limits of Extrapolation

A1.DS.A.5a / A1.DS.A.5bRLP-STATSA
Student compares straight and exponential curves, with the rising curve extending dangerously beyond observed data.
By the end of this lesson I can…

fit an exponential model to data by taking logarithms, compare it with a linear model using residuals, and explain the risks of extrapolation.

Instruction

Not every relationship is linear. A1.DS.A.5a / A1.DS.A.5b asks you to choose a function suggested by the context: linear, quadratic or exponential. Population growth is the classic exponential context, because each decade’s growth is roughly proportional to the population already there.

The data. U.S. Census Bureau decennial resident-population counts for Missouri (millions): 1910: 3.293, 1920: 3.404, 1930: 3.629, 1940: 3.785, 1950: 3.955, 1960: 4.320, 1970: 4.677. Let t = years since 1910.

Model 1: a line. Least squares gives ŷ = 3.1904 + 0.02252t (millions), with r = 0.9814. That looks strong. But the residuals, in order, are +0.103, −0.012, −0.011, −0.081, −0.137, +0.003, +0.135: positive, then four negatives in a row, then positive. That curved pattern means a line is the wrong form.

Model 2: an exponential, fitted by logarithms. If P = a·bt, then log P = log a + t·log b, which is linear in t. So take log10 of each population (in people, not millions) and fit a line to (t, log P):

log10 P = 6.5092 + 0.002504t,    r = 0.9894

Undo the logarithm (Algebra 2): a = 106.5092 ≈ 3,229,751 and b = 100.002504 ≈ 1.0058, so

P ≈ 3,229,751 × 1.0058t

The model says Missouri grew about 0.58% per year from 1910 to 1970, a factor of 1.005810 ≈ 1.059 per decade, doubling roughly every ln 2 / ln 1.0058 ≈ 120 years. The residuals of the log model (0.008, −0.002, 0.001, −0.006, −0.012, 0.001, 0.011) are small, within about 3% of the count, and no run of more than two residuals has the same sign. The exponential form fits the period better than the line does.

Extrapolation. Now use the model outside its data. For 2020 (t = 110): 3,229,751 × 1.0058110 ≈ 6.09 million. The 2020 census counted 6.155 million, so this time the exponential model came within 1.1% (the line from Model 1 predicts 5.67 million, 7.9% too low). Extrapolation can work, and a form chosen with residuals did better than the one they rejected. But it is not a guarantee. The decade-to-decade growth factor in Missouri’s counts bounced around instead of holding steady: 1.034 in the 1910s, 1.066 in the 1920s, 1.092 in the 1950s, 1.041 in the 1980s, 1.093 in the 1990s, and only 1.028 in the 2010s, against 1.059 per decade in the model.

The same method fails elsewhere. Fit the same kind of exponential to the United States counts for 1910-1970 and it grows 1.30% a year; extrapolated to 2020 it predicts 384 million, against a census count of 331.4 million, about 16% too high. For 2050 the Missouri model predicts 7.24 million. Nobody should take that as a forecast, because it assumes that births, deaths and migration keep their 1910-1970 average pattern for another 80 years. No curve fitted to past data knows when the conditions change.

Choosing among models. Use the context first (what process generates the data?), then the scatterplot, then the residual plot. Compare r2 only between models of the same response. The log model’s r is for log P, not P. A quadratic, a + bt + ct2, suits processes that accelerate at a constant rate, such as distance under constant acceleration. Spreadsheet trendlines offer all three, which makes it easy to fit a model the context does not justify.

Reporting a model. State the equation with units, the range of data used to fit it, what the parameters mean (here, the annual growth rate), the residual evidence, and the range over which you would trust a prediction.

📖 Read for this lesson

The reading opens over the lesson; close it to come back to exactly where you were. The checks below draw on it.

  • Missouri’s Census Counts With Their Logarithms: Three Fits and the Extrapolation TestU.S. Census Bureau, 2020 Census apportionment results file (resident population of Missouri and the United States, 1910-2020); log-linear fits computed for this course; retrieved November 10, 2026 · about 4 min

    The twelve Missouri census counts with their logarithms and decade ratios, the lesson’s two fits reproduced, the 1970-2020 fit the task asks you to find (with its residuals), and the U.S. counts as a case where the method fails.

Vocabulary in context

  • exponential model — A model of the form y = a·bˣ, in which y changes by a constant factor b for each unit increase in x.
  • logarithmic transformation — Replacing y with log y so that an exponential relationship becomes linear and can be fitted by least squares.
  • growth factor — The multiplier b per unit of x; b − 1 is the percent growth rate.
  • doubling time — The time for an exponentially growing quantity to double: ln 2 / ln b.

Formative check

Work through these before moving on. They are not graded — they tell you whether the standard below has landed yet.

+50 XP

The linear model of Missouri population 1910-1970 has r = 0.981, but its residuals go +, −, −, −, −, +, +. What does this mean?

A systematic residual pattern shows the form is wrong; r cannot detect this.
Fill in the blank

If log₁₀ P = 6.5092 + 0.002504t, then the growth factor per year is 10 to the power 0.002504 ≈ , which is about % growth per year.

Myth or Fact?

Because the exponential model fits 1910-1970 with r = 0.9894 and came within about 1% of the 2020 count, its prediction of about 7.2 million Missourians in 2050 must be reliable.

A good extrapolation once does not make the next one reliable. Missouri’s growth factor fell to 1.028 in the 2010s, and the same method overshot the 2020 U.S. count by about 16%. A model is trustworthy only near the data used to fit it, and only while the conditions that produced those data last.

Using the Census counts in the lesson (and 1970: 4.677, 1980: 4.917, 1990: 5.117, 2000: 5.595, 2010: 5.989, 2020: 6.155 million), fit an exponential model to 1970-2020 by logarithms. Report the annual growth rate, show the residuals, and write a paragraph comparing it with the 1910-1970 model. Should anyone use either model to predict 2050? Explain.

0 words
Quick self-check

How confident are you that you can fit an exponential model with logarithms, judge it with residuals, and explain the risk of extrapolation?

Not yetVery confident

Practice

Work these on paper or in your notebook, then open Check your answer. Aim for all of Fluency and Application; try at least one Challenge.

Printable version: this unit’s practice workbook (PDF)

Fluency

Build speed and accuracy with the core skill.

  1. A quantity starts at 15,000 and grows by 8% each year. Write a model and find its value after 9 years, to the nearest whole number.
    Check your answer
    Answer: A = 15,000(1.08)t; after 9 years ≈ 29,985
    The growth factor is 1 + 0.08 = 1.08.
  2. A quantity starts at 15,000 and grows by 12% each year. Write a model and find its value after 5 years, to the nearest whole number.
    Check your answer
    Answer: A = 15,000(1.12)t; after 5 years ≈ 26,435
    The growth factor is 1 + 0.12 = 1.12.
  3. A quantity starts at 1,200 and grows by 12% each year. Write a model and find its value after 12 years, to the nearest whole number.
    Check your answer
    Answer: A = 1,200(1.12)t; after 12 years ≈ 4,675
    The growth factor is 1 + 0.12 = 1.12.
  4. Evaluate: log3 81
    Check your answer
    Answer: 4
    Ask: 3 to what power gives 81?
  5. Evaluate: log2 (1/4)
    Check your answer
    Answer: −2
    Ask: 2 to what power gives 1/4?
  6. A log fit gives log₁₀ y = 2.5 + 0.02x. Write the exponential model y = a·bˣ.
    Check your answer
    Answer: a = 10^2.5 ≈ 316.2; b = 10^0.02 ≈ 1.0471; y ≈ 316.2·1.0471ˣ.
  7. An exponential model has growth factor 1.03 per year. Find the doubling time.
    Check your answer
    Answer: ln 2 / ln 1.03 ≈ 23.4 years.

Application

Use the skill in context. Show your reasoning.

  1. Missouri’s census population (millions): 1970 4.677, 1980 4.917, 1990 5.117, 2000 5.595, 2010 5.989, 2020 6.155. Fit an exponential model by regressing log₁₀ P on t = years since 1970 (P in people). Report the annual growth rate.
    Check your answer
    Answer: log₁₀ P ≈ 6.6671 + 0.002550t, so P ≈ 4.65 million × 1.00589^t: about 0.59% growth per year.
    Undo the log: a = 10^intercept, b = 10^slope.
  2. Compare that rate with the 1910-1970 model (about 0.58% per year) and with the same two fits for the whole United States, and say what the comparison shows.
    Check your answer
    Answer: Missouri’s rate barely changed (0.58% a year before 1970, 0.59% after), so the 1910-1970 model happened to extrapolate well to 2020. The U.S. rate fell from 1.30% to 1.00% a year, which is why the same method overshot the 2020 U.S. count by about 16%.
  3. A linear fit to the same 1970-2020 Missouri data gives residuals (millions) of +0.060, −0.016, −0.133, +0.029, +0.106, −0.045. What do they suggest?
    Check your answer
    Answer: Mixed signs with no long run: the line fits the six counts about as well as the exponential does (r = 0.9900 for the line, 0.9911 for the log fit). The largest misses are 1990 (−0.133) and 2010 (+0.106), because Missouri grew 4.1% in the 1980s and 9.3% in the 1990s: growth came in uneven decades, and neither smooth model can follow that.
    Read the pattern of signs, then compare the sizes with the growth factor for each decade.

Challenge

Stretch problems. Expect to think before you write.

  1. Use your 1970-2020 exponential model to predict 2050, then argue whether the prediction should be published.
    Check your answer
    Answer: P(80) ≈ 7.43 million (the 1970-2020 line gives 7.15 million). It should not be published as a forecast: it assumes the 1970-2020 average rate continues, while the most recent decade grew only 2.8% (the model expects about 6.0% per decade). Demographers use models of births, deaths and migration instead.

Review

Keep earlier skills sharp.

  1. (Unit 1) A data set has mean 40 and SD 5. Every value is multiplied by 2 and then 10 is added. Find the new mean and SD.
    Check your answer
    Answer: Mean 90; SD 10.
  2. (Lesson 2.2) State the formula for the least-squares slope in terms of r and the SDs.
    Check your answer
    Answer: b = r·s_y / s_x.

Missouri Learning Standards — Mathematics (data and probability) addressed: A1.DS.A.5a, A1.DS.A.5b, A1.DS.A.5, MP.4, MP.7

Instructional focus: Fitting and evaluating nonlinear models using logarithmic transformation

Reading

Missouri’s Census Counts With Their Logarithms: Three Fits and the Extrapolation Test

U.S. Census Bureau, 2020 Census apportionment results file (resident population of Missouri and the United States, 1910-2020); log-linear fits computed for this course; retrieved November 10, 2026 · Public domain (works of the United States government), with SLS worked calculations

Why this reading. The workshop fits an exponential model by taking logarithms, and its last task asks you to do the same for 1970-2020 and compare. The counts, their logarithms and their decade ratios are printed here so the fit can be done by hand or checked in a spreadsheet, and the fits are worked to the digits the lesson quotes, with the 1970-2020 answer that the task expects you to find. The U.S. counts from the same file are included so you can see a case where the same method fails.

The counts

Resident population of Missouri at each decennial census, 1910-2020, as printed in the Census Bureau’s 2020 apportionment file. t is years since 1910.

Censust (since 1910)CountMillionslog10(count)Ratio to previous
191003,293,3353.2936.5176—
1920103,404,0553.4046.53201.034
1930203,629,3673.6296.55981.066
1940303,784,6643.7856.57801.043
1950403,954,6533.9556.59711.045
1960504,319,8134.3206.63551.092
1970604,676,5014.6776.66991.083
1980704,916,6864.9176.69171.051
1990805,117,0735.1176.70901.041
2000905,595,2115.5956.74781.093
20101005,988,9275.9896.77731.070
20201106,154,9136.1556.78921.028

Fit 1: a line through 1910-1970 (millions against t)

P̂ = 3.1904 + 0.02252t, r = 0.9814. Residuals in order: +0.103, -0.012, -0.011, -0.081, -0.137, +0.003, +0.135. Positive, then four negatives, then positive: the form is wrong.

Fit 2: a line through 1910-1970 in log scale

log10P̂ = 6.5092 + 0.002504t, r = 0.9894. Residuals: +0.0085, -0.0022, +0.0006, -0.0063, -0.0122, +0.0011, +0.0105. No run of more than two of one sign. Undoing the logarithm: a = 106.5092 = 3,229,751 and b = 100.002504 = 1.0058, a growth factor of 1.0058 per year, 1.059 per decade. Extrapolated to t = 110 the model gives 6.09 million for 2020 against a count of 6.155 million (a miss of 1.1%); the line from Fit 1 gives 5.67 million. For 2050 (t = 140) the exponential gives 7.24 million.

Fit 3: the same method on 1970-2020 (t now counts years since 1970)

log10P̂ = 6.6671 + 0.002550t, r = 0.9911. The growth factor is 100.002550 = 1.0059 per year, about 0.59% a year against 0.58% for 1910-1970. Residuals: +0.0028, -0.0009, -0.0091, +0.0042, +0.0083, -0.0053. Read them in order: the signs are positive, negative, negative, positive, positive, negative. The ratio column shows why no smooth curve does better: Missouri grew 4.1% in the 1980s, 9.3% in the 1990s, 7.0% in the 2000s and only 2.8% in the 2010s. Extrapolated to 2050 (t = 80) the model gives 7.43 million; the least-squares line through the same six counts (millions = 4.6163 + 0.03168t, r = 0.9900) gives 7.15 million. If the 2010s decade factor (1.028) were to continue, 2050 would be about 6.68 million; that is an arithmetic scenario, not a forecast.

A case where the method fails: the whole United States

The same file has the national resident-population counts.

CensusUnited States countRatio to previous
191092,228,531—
1920106,021,5681.150
1930123,202,6601.162
1940132,165,1291.073
1950151,325,7981.145
1960179,323,1751.185
1970203,211,9261.133
1980226,545,8051.115
1990248,709,8731.098
2000281,421,9061.132
2010308,745,5381.097
2020331,449,2811.074

Fitting log10 of the U.S. counts for 1910-1970 gives log10P̂ = 7.9660 + 0.005625t, r = 0.9970, a growth factor of 1.0130 per year (1.30%). Extrapolated to 2020 it predicts 384.4 million against a count of 331.4 million, 16% too high. Fitting 1970-2020 instead gives 1.00% a year. Missouri’s rate stayed about the same across the two periods and the nation’s did not, so the 1910-1970 Missouri model extrapolated well without anything in the fit having told you it would.

What to do

  1. Reproduce Fit 3 with =SLOPE and =INTERCEPT on t and log10(count), or by hand using the deviation columns. Report the annual growth rate.
  2. Compute and plot its residuals. Say whether the pattern is random, and use the ratio column to explain what you see.
  3. Write the paragraph the task asks for: the two models compared, and whether either should be used for 2050. Quote the ratio column as your evidence.

Sources: U.S. Census Bureau, 2020 Census Apportionment Results, apportionment.csv (Resident Population, 1910-2020) (www2.census.gov)

Up NextComplete & Continue →

Leave a Reply Cancel reply

Residuals, Influence and Causation
Prev
Performance Task — Bivariate Investigation Report
Next

© 2026 The VR School | Career content by Wiley | Powered by Sejal Learning Systems.

Join free · Accreditation · Privacy Policy · Terms of Service

Modal title

Main Content