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 Algebra 1A — Data, Linear Equations and Systems

Curriculum

  • 4 Sections
  • 20 Lessons
  • Lifetime
Expand all sectionsCollapse all sections
  • Unit 1: One-Variable Statistics
    5
    • 1.1
      Seeing the Shape of Data: Dot Plots, Histograms and Box Plots
      50 mins
    • 1.2
      Measuring Center and Variability: Mean, Median and MAD
      50 mins
    • 1.3
      Outliers, Skew and Standard Deviation: Which Summary Tells the Truth?
      50 mins
    • 1.4
      Comparing Groups in Context: Units, Scale and Precision
      100 mins
    • 1.5
      Performance Task — Data Story
      150 mins
  • Unit 2: Linear Equations and Inequalities in One Variable
    5
    • 2.1
      Expressions That Tell a Story: Terms, Factors and Coefficients
      50 mins
    • 2.2
      Solving Linear Equations and Justifying Every Step
      50 mins
    • 2.3
      Where Did It Go Wrong? Inequalities and Rearranged Formulas
      50 mins
    • 2.4
      Absolute Value as Distance: Tolerances and Alarms
      100 mins
    • 2.5
      Performance Task — Transit Fare Model
      150 mins
  • Unit 3: Linear Relationships, Inequalities and Systems in Two Variables
    5
    • 3.1
      A Line Is a Set of Solutions: Equations in Two Variables
      50 mins
    • 3.2
      Solving Systems: Graphing, Substitution and Elimination
      50 mins
    • 3.3
      Why Elimination Works: A Proof About Systems
      50 mins
    • 3.4
      Constraints and Half-Planes: Systems of Linear Inequalities
      100 mins
    • 3.5
      Performance Task — Community Planning Proposal
      150 mins
  • Unit 4: Two-Variable Statistics
    5
    • 4.1
      Two-Way Tables: Is There an Association?
      50 mins
    • 4.2
      Scatter Plots and Lines of Fit
      50 mins
    • 4.3
      Residuals and Correlation: How Good Is the Fit?
      50 mins
    • 4.4
      Correlation Is Not Causation: Screens, Sleep and Hidden Variables
      100 mins
    • 4.5
      Performance Task — Bivariate Data Study
      150 mins

Correlation Is Not Causation: Screens, Sleep and Hidden Variables

Unit 4  ·  Modeling & Application Workshop  ·  Lesson 19 of 20

Correlation Is Not Causation: Screens, Sleep and Hidden Variables

A1.DS.A.8RLP-ALG1A
Teen studies beside screens while tangled cords connect homework, screen use, and sleeplessness.
By the end of this lesson I can…

interpret a linear model and its correlation in context, and explain why a strong correlation alone does not establish that one variable causes the other.

Instruction

Headlines love causal claims: “Screens are wrecking teen sleep.” “Students who eat breakfast get better grades.” This workshop teaches you to ask what the data can and cannot support.

A data set (illustrative). Ten students recorded screen time after 9 p.m. (x, hours) and sleep that night (y, hours):

x: 0.5, 1, 1, 1.5, 2, 2, 2.5, 3, 3.5, 4
y: 8.6, 8.1, 8.4, 7.9, 7.6, 7.2, 7.4, 6.8, 6.9, 6.2

Technology gives the least-squares line predicted sleep ≈ 8.85 − 0.64x, with r ≈ −0.97. The residuals show no pattern, so a linear model fits well.

Interpreting the model. Slope: for each additional hour of late screen time, predicted sleep is about 0.64 hour (about 38 minutes) less. Intercept: a student with no late screen time is predicted to sleep about 8.85 hours; here zero screen time is realistic and close to the data, so the intercept means something. Prediction: at 3 hours of screen time, 8.85 − 0.64(3) ≈ 6.9 hours of sleep. The correlation says the negative linear association is strong.

But did screens cause less sleep? Maybe. But the data cannot tell you, for three reasons.

  • Confounding variables. A third variable could drive both. A student with a heavy homework night may be on a laptop late and sleep less because of the homework. Stress, a late job shift or a noisy home could do the same.
  • Reverse direction. A student who cannot fall asleep may pick up a phone because they are awake. Then poor sleep leads to screen time, not the other way round.
  • Coincidence and selection. Ten volunteers are a small, unrepresentative group.

A classic example. If you collected data on wildfires, you would expect the number of fire engines sent to be strongly correlated with the acres burned, because larger fires get more engines. Nobody would conclude that fire engines cause burning. The size of the fire causes both. Once you see the confounder, the correlation makes sense without any causal link between the two variables you measured.

What would establish cause? A randomized experiment: randomly assign students to a “screens off at 9 p.m.” group and a usual-habits group, then compare sleep. Random assignment spreads confounders like homework and stress evenly between the groups on average, so a difference in sleep can be attributed to the treatment. Observational data, however strong the correlation, only show that the variables move together.

How to talk about it. Use “is associated with”, “tends to” and “is predicted to” for observational data. Reserve “causes”, “leads to” and “improves” for well-designed experiments. That language is not fussiness; it is the difference between an accurate claim and a misleading one.

Workshop. Choose a real headline from a news source (or one provided) that makes a causal claim from data. Identify the two variables, state what kind of study produced the data, name at least two possible confounders, and design a randomized experiment that could test the claim, including what you would measure and how you would assign groups.

📖 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.

  • Three Causal Claims, Each With the Data Behind ItBLS, The Employment Situation — August 2026, Table A-4 (released September 4, 2026); NOAA NCEI Climate at a Glance city time series for Fresno (USW00093193) and San Diego (USW00023188), July average maximum temperature 1995–2025; retrieved September 23, 2026 · about 5 min

    Three causal claims with the federal data behind them, including one where reverse causation is impossible and the claim is still not safe.

Vocabulary in context

  • causation — A relationship in which a change in one variable produces a change in another.
  • confounding variable — A third variable associated with both variables studied, which can create or hide an association.
  • observational study — A study that records data without assigning treatments.
  • randomized experiment — A study that randomly assigns subjects to treatments, so differences can be attributed to the treatment.

Formative check

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

Make a prediction

Cities with more libraries have more crimes reported. What is the most likely explanation?

Bigger cities have more of almost everything, including libraries and reported crimes. Population is a confounding variable; the correlation is not evidence that either causes the other.
+50 XP

The model predicted sleep ≈ 8.85 − 0.64x has slope −0.64. What does it mean?

Slope is the predicted change in y per unit of x. Observational data support association, not cause.
Fill in the blank

For 3 hours of late screen time the model predicts about hours of sleep; the correlation r ≈ −0.97 shows a strong linear association.

Take the headline “Students who eat breakfast get better grades.” Name two possible confounding variables, explain how each could create the association, and describe a randomized experiment that could test whether breakfast causes better grades.

0 words
Quick self-check

How confident are you that you can explain why a strong correlation does not by itself show causation, and describe how to test a causal claim?

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. Use technology to find the least-squares line and the correlation coefficient for (1, 7), (3, 17), (5, 23), (6, 22), (10, 33), (11, 41), (12, 41).
    Check your answer
    Answer: ŷ = 5.989 + 2.960x; r = 0.985
    Interpret the slope in context: the predicted change in y for each one-unit increase in x.
  2. Use technology to find the least-squares line and the correlation coefficient for (7, 21), (9, 29), (10, 30), (12, 36), (13, 42), (14, 45).
    Check your answer
    Answer: ŷ = −2.502 + 3.354x; r = 0.991
    Interpret the slope in context: the predicted change in y for each one-unit increase in x.
  3. Use technology to find the least-squares line and the correlation coefficient for (4, 18), (5, 20), (7, 22), (9, 29), (10, 27), (11, 31), (14, 38).
    Check your answer
    Answer: ŷ = 9.651 + 1.957x; r = 0.981
    Interpret the slope in context: the predicted change in y for each one-unit increase in x.
  4. Rewrite “Each hour of screens causes 0.64 hour less sleep” in association language.
    Check your answer
    Answer: Each additional hour of late screen time is associated with about 0.64 hour less predicted sleep.
  5. Is a study that surveys students about breakfast and grades observational or experimental?
    Check your answer
    Answer: Observational: nothing is assigned.
  6. What does random assignment do in an experiment?
    Check your answer
    Answer: It spreads confounding variables evenly between groups on average.

Application

Use the skill in context. Show your reasoning.

  1. Use predicted sleep ≈ 8.85 − 0.64x to predict sleep for 2.5 hours of late screen time.
    Check your answer
    Answer: About 7.25 hours
  2. Across lake resort towns, daily ice cream sales and sunburn cases are strongly correlated. Name a confounding variable.
    Check your answer
    Answer: Hot, sunny weather drives both.
  3. Design an experiment to test whether a 9 p.m. phone curfew increases sleep.
    Check your answer
    Answer: Randomly assign volunteers to curfew or usual habits for two weeks and compare mean sleep.

Challenge

Stretch problems. Expect to think before you write.

  1. Use the sleep model to predict sleep for 14 hours of late screen time, and explain why the prediction is not trustworthy.
    Check your answer
    Answer: −0.1 hours, a negative value: 14 hours is far outside the data (0.5 to 4 hours).
    Extrapolation.

Review

Keep earlier skills sharp.

  1. Solve and graph on a number line: 4x − 6 ≥ −14
    Check your answer
    Answer: x ≥ −2
    Add 6, then divide by 4.
  2. For the data 7, 8, 11, 17, 18, 37, 39: find the mean, median, range and interquartile range. (Find the quartiles as the medians of the lower and upper halves, leaving out the overall median when the count is odd.)
    Check your answer
    Answer: mean = 19.57; median = 17.0; range = 32; IQR = 29.0 (Q1 = 8.0, Q3 = 37.0)

Missouri Learning Standards — Mathematics addressed: A1.DS.A.8, A1.DS.A.6, A1.DS.A.7, A1.DS.A.5, MP.4

Instructional focus: Reasoning about causal claims from data

Reading

Three Causal Claims, Each With the Data Behind It

BLS, The Employment Situation — August 2026, Table A-4 (released September 4, 2026); NOAA NCEI Climate at a Glance city time series for Fresno (USW00093193) and San Diego (USW00023188), July average maximum temperature 1995–2025; retrieved September 23, 2026 · Public domain (works of the United States government); worked models SLS original

Why this reading. The workshop asks you to take a headline that makes a causal claim from data, name the variables, name the confounders and design a study that could test it — and then leaves you to go and find a headline. A headline is not enough anyway; without the underlying numbers you cannot say what the data would and would not support. Here are three claims with their data attached, all from federal agencies.

Claim 1 — “A college degree cuts your risk of unemployment nearly in half”

The data. BLS, Employment Situation Table A-4, August 2026, seasonally adjusted, for the civilian population aged 25 and over. Unemployment rate by highest level of education: less than a high school diploma 4.7%; high school graduates, no college 4.4%; some college or associate degree 3.7%; bachelor’s degree and higher 2.7%. Everyone 25 and over: 3.4%.

What the data support. A strong association across groups, in a consistent direction, at national scale, in one month. 4.7 ÷ 2.7 = 1.74, so “nearly in half” is a fair reading of the ratio.

What the data cannot support. The claim is about your risk, and the table is about four groups. The people in those groups differ in far more than schooling. Confounders: family income and wealth before anyone enrolled anywhere; region and local industry; age, since degree-holding is much more common in some cohorts; health; and field of study. Direction: reverse causation is not impossible here — a person with a secure job and a stable income is better placed to finish a degree. Selection: the table counts only people in the labor force, and the share of adults who are in it differs sharply by education, so the comparison is between differently selected populations.

What would settle it. You cannot randomly assign people to get degrees, which is exactly why this argument has lasted decades. Researchers reach for the next best things: comparisons of siblings, of people just above and just below an admissions cutoff, or of cohorts affected by a policy change they did not choose. Naming why the experiment is impossible is part of a good answer, not a dodge.

Claim 2 — “Julys in Fresno are getting hotter”

The data. NOAA NCEI, July average maximum temperature at Fresno Yosemite International Airport, 1995–2025, as (year, °F) pairs: (1995, 94.7), (1996, 100.8), (1997, 95.5), (1998, 96.7), (1999, 95.6), (2000, 93.5), (2001, 96.0), (2002, 99.2), (2003, 101.3), (2004, 98.5), (2005, 101.7), (2006, 103.0), (2007, 98.2), (2008, 98.6), (2009, 100.7), (2010, 98.5), (2011, 96.5), (2012, 97.6), (2013, 101.0), (2014, 99.9), (2015, 96.8), (2016, 100.0), (2017, 101.9), (2018, 102.8), (2019, 99.0), (2020, 99.9), (2021, 103.3), (2022, 100.8), (2023, 101.8), (2024, 105.2), (2025, 97.0).

The model. x̄ = 2010, ȳ = 99.226, Sxx = 2480, Sxy = 410.2, Syy = 235.84. Slope = 410.2 ÷ 2480 = 0.1654 °F per year, intercept −233.23, and r = 410.2 ÷ √(2480 × 235.84) = 0.536. The fitted line rises from 96.7 °F in 1995 to 101.7 °F in 2025, about 1.65 °F a decade.

What is different about this one. Reverse causation is impossible: nothing about July temperature changes what year it is. But that does not make the claim safe. r = 0.536 means the line accounts for less than a third of the variation, and the residuals are large — 2006 sits 4.4 °F above the line and 2025 sits 4.7 °F below it. Confounders here are about the measurement, not the cause: a station on an airfield surrounded by a city that has grown for thirty years, changes of instrument, and changes of siting all push a thermometer’s readings around. “This station recorded a warming trend of about 1.7 °F a decade in July, with wide year-to-year variation” is defensible from this table. “Fresno is getting hotter” is a bigger claim than one station supports.

Claim 3 — “Hot Julys in San Diego make Julys hot in Fresno”

The data. The same thirty-one Julys, paired by year, as (San Diego °F, Fresno °F): (73.3, 94.7), (73.7, 100.8), (72.2, 95.5), (73.0, 96.7), (72.4, 95.6), (73.2, 93.5), (72.8, 96.0), (71.6, 99.2), (74.1, 101.3), (76.2, 98.5), (74.2, 101.7), (80.9, 103.0), (73.8, 98.2), (73.4, 98.6), (75.3, 100.7), (70.1, 98.5), (74.1, 96.5), (72.7, 97.6), (74.2, 101.0), (76.9, 99.9), (75.7, 96.8), (74.4, 100.0), (76.0, 101.9), (78.6, 102.8), (73.3, 99.0), (74.4, 99.9), (76.0, 103.3), (72.2, 100.8), (74.1, 101.8), (76.7, 105.2), (73.3, 97.0).

The model. Sxx = 136.56, Sxy = 102.53, Syy = 235.84. Slope = 0.751, intercept 43.45, r = 0.571. A moderate positive association: warm Julys on the coast do tend to come in the same years as warm Julys inland.

Nobody believes San Diego’s weather causes Fresno’s. The two cities are 300 miles apart and neither one is upstream of the other in any useful sense. A third thing — the large-scale weather and ocean pattern over the eastern Pacific in a given summer — moves both. This is the fire-engine example from the mini-lecture with real numbers attached, and it is worth noticing that the correlation is only moderate: the confounder is real but partial, because the coast and the interior also have weather of their own.

A second lesson hiding in the same data. Fit the line to just the last seven Julys (2019–2025) and you get r = 0.774 instead of 0.571. Same cities, same measurement, a much stronger-looking result — because with seven points, a correlation wanders. If you ever find yourself choosing which years to include after seeing the answer, that is the moment to stop.

Writing about any of the three

Use “is associated with,” “tends to” and “is predicted to” for all of it, because all of it is observational. Save “causes,” “raises” and “improves” for a study that randomly assigned something. Then do the part most reports skip: say what would have to be true for the causal claim to be right, and what evidence would show it was wrong.

Sources: BLS, Employment Situation Table A-4 (employment status by educational attainment) (bls.gov); NOAA NCEI, Climate at a Glance: City Time Series (Fresno) (ncei.noaa.gov); NOAA NCEI, Climate at a Glance: City Time Series (San Diego) (ncei.noaa.gov)

Up NextComplete & Continue →

Leave a Reply Cancel reply

Residuals and Correlation: How Good Is the Fit?
Prev
Performance Task — Bivariate Data Study
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