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 B — Probability Models and Statistical Inference

Curriculum

  • 4 Sections
  • 20 Lessons
  • 10 Weeks
Expand all sectionsCollapse all sections
  • Unit 1: Continuous Distributions and Sampling Distributions
    5
    • 1.1
      Continuous Random Variables and Density Curves
      50 mins
    • 1.2
      The Normal Distribution
      50 mins
    • 1.3
      When Is a Normal Model Appropriate?
      50 mins
    • 1.4
      Sampling Distributions and the Central Limit Theorem
      50 mins
    • 1.5
      Performance Task — Sampling Distribution Simulation Study
      150 mins
  • Unit 2: Estimating with Confidence
    5
    • 2.1
      Estimating a Proportion with Confidence
      50 mins
    • 2.2
      Confidence Intervals for a Mean
      50 mins
    • 2.3
      What Confidence Means, and What It Does Not
      50 mins
    • 2.4
      Comparing Two Groups with Confidence Intervals
      50 mins
    • 2.5
      Performance Task — Estimation Study
      150 mins
  • Unit 3: Testing Claims: Significance Tests
    5
    • 3.1
      The Logic of a Significance Test
      50 mins
    • 3.2
      t-Tests for Means and Paired Data
      50 mins
    • 3.3
      Errors, Power and Significance in the Real World
      50 mins
    • 3.4
      Two-Sample Tests from Experiments and Observational Data
      50 mins
    • 3.5
      Performance Task — Significance Test Investigation
      150 mins
  • Unit 4: Chi-Square, Regression Inference and the Capstone Study
    5
    • 4.1
      Chi-Square Goodness of Fit
      50 mins
    • 4.2
      Chi-Square Tests for Two-Way Tables
      50 mins
    • 4.3
      Inference for the Slope of a Regression Line
      50 mins
    • 4.4
      Choosing the Right Procedure and Writing It Up
      50 mins
    • 4.5
      Capstone — Full Statistical Inference Study
      200 mins

Capstone — Full Statistical Inference Study

Unit 4  ·  Culminating Performance Task  ·  Lesson 20 of 20

Capstone — Full Statistical Inference Study

A2.DS.A.1RLP-STATSB
Student presents collected data and an inference graphic to peers during a capstone research defense.
By the end of this lesson I can…

pose an original question, obtain or produce data, choose and justify an inference procedure, and write and defend a complete statistical report.

Instruction

The capstone. This is the course’s final statement of what you can do. Pose a question that matters to you or to a Missouri community, answer it with data and formal inference, and defend your reasoning to an audience.

Good capstone questions are specific, answerable with available data, and call for inference:

  • Climate: has the number of 90°F days per summer at a Missouri station (Kansas City, Columbia, Springfield) increased? Is winter rain persistence different at Springfield than at St. Louis? (NOAA GHCN-Daily and Global Summary of the Year.)
  • Health: has Missouri’s age-adjusted death rate for a cause changed at a different rate than the national rate? (CDC NCHS, data.cdc.gov bi63-dtpu.)
  • Work and economy: how did Missouri’s unemployment rate behave before, during and after 2020 compared with a neighboring state? (BLS Local Area Unemployment Statistics, data.bls.gov.) Time series call for care: say what inference is and is not meaningful.
  • Communities: how does an ACS estimate for your county compare with the state, given the margins of error? (Census ACS, data.census.gov.)
  • Experiments: a randomized experiment you design, under the ethics rules from RLP Statistics A.

The report follows the structure from Lesson 4.4: question; data and how they were produced; exploration; method and conditions; results; conclusion with scope; limitations; and a reproducibility appendix. It must include at least one formal inference procedure. Most strong capstones combine an interval and a test, or two procedures that look at the question from different angles.

What distinguishes an excellent capstone.

  • Honesty about conditions. Real data rarely meet every condition. Excellent reports say which conditions are shaky, which direction that pushes the result, and whether the conclusion survives. The Lesson 4.3 discussion of autocorrelation is the model.
  • Effect sizes in context. “The rate fell 6.6 deaths per 100,000 per year” means more than “P < 0.001”.
  • Correct scope. Every conclusion sits in the right box of the scope-of-inference table.
  • Alternative explanations. At least one plausible alternative is named and weighed.
  • Reproducibility. A careful reader following your appendix gets your numbers.

Accuracy check. Before your final draft, recompute every number in the report from the data, as this course’s authors did for every figure in these lessons. Most errors in published reports are transcription errors, not conceptual ones.

The defense. Six minutes: question, data, one key display, the inference result with its effect size, the scope, and one limitation. Then answer questions from a parent or guardian and, where possible, other invited adults: family members, mentors, or professionals who use data. A recording with written answers to questions you collect in advance is an equal-credit alternative.

Privacy. No capstone requires personal disclosure. Public datasets are always available, and any data you collect must be anonymous, voluntary and approved.

📖 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.

  • Thirty-Six Months in Four States: A Time Series for the Capstone, and What Inference It AllowsBLS Local Area Unemployment Statistics, statewide seasonally adjusted unemployment rates, series LASST290000000000003, LASST200000000000003, LASST170000000000003 and LASST050000000000003, January 2019 to December 2021, via the BLS Public Data API v1; retrieved 2026-11-10 · about 5 min

    The BLS series the capstone names, hosted: 36 months in Missouri and three neighbors, with a worked account of which inferences a time series allows and a sketched Stage 1 proposal to measure yours against.

  • Nineteen Years of Missouri Death Rates: The CDC Table Behind Three LessonsCDC NCHS, NCHS – Leading Causes of Death: United States, data.cdc.gov dataset bi63-dtpu, Missouri rows, 1999-2017, age-adjusted death rates per 100,000 and deaths; retrieved 2026-11-10 · about 5 min

    The CDC health source for a capstone, complete for Missouri, with two procedures worked.

  • The Census Bureau’s Own Margins: Broadband Subscriptions in Thirteen Missouri AreasU.S. Census Bureau, American Community Survey 2023 1-year estimates, detailed table B28002 (lines 1 and 4) for Missouri and twelve of its counties and county-equivalents, from the ACS Summary File (table-based format, acsdt1y2023-b28002.dat); the claims to judge are SLS original framings; retrieved 2026-11-10 · about 5 min

    The Census ACS source for a capstone: county estimates with published margins, and the comparison arithmetic the ‘my county against the state’ question needs.

Vocabulary in context

  • capstone — A culminating project that applies the whole course to an original question.
  • reproducibility appendix — A step-by-step record of data sources, cleaning and computations that lets others reproduce the results.
  • alternative explanation — A different account of the same result, such as confounding, bias or a change in measurement.

Formative check

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

+50 XP

A capstone uses every day of NOAA data at one station for 2000-2024 to ask ‘what fraction of these days were rainy?’ What inference is needed?

Inference generalizes beyond the data. If the question is only about the days you have, compute the parameter; ask a question about a larger process if you want inference.
Myth or Fact?

An excellent report hides shaky conditions so the conclusion looks stronger.

Excellent reports name shaky conditions, say which way they push the result, and judge whether the conclusion survives.

Stage 1 proposal: write your question, the data source with URL, how the data were produced (sampling, assignment or neither), the procedure(s) you plan to use with the conditions you will check, one alternative explanation you will need to consider, and (if collecting data) your ethics plan.

0 words
Quick self-check

How confident are you that you can design, carry out, write up and defend a complete statistical inference study?

Not yetVery confident

Unit 4 checkpoint

Do this before you start the task. It covers the whole unit and tells you what to review first.

Fill in the blank

A goodness-of-fit test with 7 categories and no estimated parameters has df = .

+50 XP

χ² = 4.18 with df = 6 gives P ≈ 0.65. The conclusion is…

Fail to reject, never ‘prove’.
Myth or Fact?

A binomial goodness-of-fit test in which p was estimated from the data loses one extra degree of freedom.

df = categories − 1 − estimated parameters.
Fill in the blank

The expected count for a cell with row total 688, column total 690 and grand total 2,256 is about .

+50 XP

For a 2 × 2 table, χ² = 68.8. The two-proportion z statistic is about…

χ² = z².
Myth or Fact?

A chi-square statistic can be computed from percentages instead of counts.

It needs counts.
Fill in the blank

A slope of 0.0303 with SE 0.00713 gives t ≈ .

+50 XP

The Springfield temperature slope test has P < 0.001 and r² = 0.198. Which is true?

Significance ≠ strength.
Myth or Fact?

Positively autocorrelated residuals make the slope test’s P-value too small.

The SE is underestimated.
+50 XP

Which question needs no inference?

The whole population is in hand.
+50 XP

Same 30 years at two stations: which procedure?

Paired by year.
Myth or Fact?

A reproducibility appendix lets a careful reader obtain your numbers from the same data by following your steps.

That is its purpose.

Submit: Capstone inference study (1,800-2,000 words + reproducibility appendix) + six-minute defense

Task: Carry out an original statistical study using public federal data (where to look: NOAA https://www.ncei.noaa.gov/, CDC https://data.cdc.gov/, Census https://data.census.gov/, BLS https://data.bls.gov/), or one of the hosted data sets in this lesson’s readings, which is a fully equal route: BLS statewide unemployment, Missouri and three neighbors by month, 2019-2021 (n = 36 months per state); CDC/NCHS Missouri age-adjusted death rates, 1999-2017 (n = 19 years, eight series); Census ACS 2023 broadband subscriptions for Missouri and 12 counties (n = 13 areas, with published margins); each with its source and retrieval date, 2026-11-10. NOAA temperature and rainfall tables (St. Louis and Springfield annual means, 75 years each; the St. Louis 200-day sample) are hosted in the Statistics B Lesson 1.2, 1.5 and 2.5 readings. Or use an approved randomized experiment or random-sample survey, with at least one formal inference procedure from the course. Tools: CODAP, spreadsheet, GeoGebra Probability, or optionally Python in JupyterLite. Every step must be reproducible from your appendix.

Submission workflow

  1. Stage 1 — Proposal. Question, data source with URL, how the data were produced, the planned procedure(s) with conditions to check, and (for collected data) the ethics plan; a parent or guardian’s approval required.
  2. Stage 2 — Self-checked draft. Reproduce two of your own numbers from your appendix as if you were a stranger (or ask a parent or guardian to), and review your conditions and scope; submit those notes and your revision plan.
  3. Stage 3 — Final report. 1,800-2,000 words plus the reproducibility appendix, following the Lesson 4.4 structure.
  4. Stage 4 — Defense. Six minutes, recorded or live, followed by questions from a parent or guardian and other invited adults.

How it is scored

CriterionWhat proficient looks like
Question and data production (A2.DS.A.1, A2.DS.A.3)Specific question; data source and production described; population or process defined.
Procedure choice and conditions (PS.17, PS.18, PS.19, PS.12)Appropriate procedure justified; conditions checked honestly with evidence.
Results and effect size (MP.4)Correct computations by hand and technology; effect sizes interpreted in context.
Scope and alternatives (A2.DS.A.7)Conclusion placed in the correct scope; alternative explanations and limitations weighed.
Reproducibility (MP.1)A reader can reproduce the numbers from the appendix.
Defense (MP.3)Clear presentation; questions answered with evidence and reasoning.

Missouri Learning Standards — Mathematics (data and probability) addressed: A2.DS.A.1, A2.DS.A.7, A2.DS.A.3, A2.DS.A.4, A1.DS.A.5a, MP.1, MP.3, MP.4

Extends beyond Missouri Statistics (enrichment): content Missouri places after Statistics

Instructional focus: Independent statistical investigation with formal inference, written and defended

Reading

Thirty-Six Months in Four States: A Time Series for the Capstone, and What Inference It Allows

BLS Local Area Unemployment Statistics, statewide seasonally adjusted unemployment rates, series LASST290000000000003, LASST200000000000003, LASST170000000000003 and LASST050000000000003, January 2019 to December 2021, via the BLS Public Data API v1; retrieved 2026-11-10 · Public domain (works of the United States government), with SLS original commentary

Why this reading. The capstone lists four federal sources and says of one of them, the Bureau of Labor Statistics, that “time series call for care: say what inference is and is not meaningful”. It does not show such a series. Here is one, hosted so that a capstone in this direction can be planned and its exploration begun without the portal, and so that the care the lesson asks for can be practiced on numbers you can see. Missouri is here with three of its neighbors.

Table 1. Unemployment rate, seasonally adjusted, Missouri and three neighbors, 2019–2021

BLS Local Area Unemployment Statistics, statewide series LASST290000000000003 (Missouri), LASST200000000000003 (Kansas), LASST170000000000003 (Illinois) and LASST050000000000003 (Arkansas), percent of the labor force, retrieved from the BLS Public Data API on 2026-11-10. These are model-based estimates that BLS revises annually; the figures are as published on the retrieval date.

MonthMissouriKansasIllinoisArkansasMO − KS
Jan 20193.33.34.63.7+0.0
Feb 20193.33.34.53.6+0.0
Mar 20193.23.24.43.5+0.0
Apr 20193.13.24.23.4−0.1
May 20193.13.14.03.4+0.0
Jun 20193.13.13.93.4+0.0
Jul 20193.13.13.83.4+0.0
Aug 20193.23.13.73.5+0.1
Sep 20193.33.13.73.5+0.2
Oct 20193.33.13.73.5+0.2
Nov 20193.43.13.63.5+0.3
Dec 20193.43.13.63.5+0.3
Jan 20203.43.13.73.5+0.3
Feb 20203.43.13.83.6+0.3
Mar 20203.63.15.14.8+0.5
Apr 202011.612.118.310.1−0.5
May 20209.69.214.78.9+0.4
Jun 20208.07.512.67.9+0.5
Jul 20207.26.711.47.3+0.5
Aug 20206.45.89.76.4+0.6
Sep 20205.95.49.15.9+0.5
Oct 20205.44.98.25.4+0.5
Nov 20205.14.67.85.1+0.5
Dec 20205.04.47.65.0+0.6
Jan 20214.84.17.24.9+0.7
Feb 20214.83.97.04.8+0.9
Mar 20214.73.86.84.7+0.9
Apr 20214.63.76.74.6+0.9
May 20214.43.76.64.4+0.7
Jun 20214.33.76.44.2+0.6
Jul 20214.13.56.24.0+0.6
Aug 20213.93.45.93.7+0.5
Sep 20213.73.15.53.5+0.6
Oct 20213.52.95.23.3+0.6
Nov 20213.42.84.93.2+0.6
Dec 20213.22.64.83.2+0.6

Peaks: Missouri 11.6% in Apr 2020; Kansas 12.1% in Apr 2020; Illinois 18.3% in Apr 2020; Arkansas 10.1% in Apr 2020.

What is, and is not, a meaningful inference here

Not meaningful: a two-sample t-test of “Missouri’s 36 monthly rates against Kansas’s 36”. The months are not independent observations; each is almost the same as the one before it, so 36 months carry perhaps a handful of months’ worth of independent information, and any P-value from such a test is wildly too small. The rates are also estimates from a survey and a model, not counts, with their own published errors. And the question “was unemployment higher in Missouri than in Kansas in 2020?” needs no inference at all: the table answers it.

Meaningful, with care: a question about a relationship or a change that the month-to-month dependence does not manufacture. For example: did the gap between Missouri and Kansas change from before the pandemic to after it? In 2019 the MO − KS difference averaged +0.08 points (SD 0.13); in 2021, +0.68 (SD 0.14). Treating each year’s twelve differences as a sample of that year’s regime is still generous, and a report should say that the effective number of independent months is smaller than twelve; but the direction and size of the shift, and how far it exceeds the within-year wobble, are worth stating with an interval labeled as approximate. Or: does a state’s recovery slope (rate against month, from the peak to December 2021) differ between states, with the Lesson 4.3 autocorrelation caution attached? Or, more modestly and more defensibly: describe, compare, and put the numbers in context, with no P-value at all, and say why.

The question to write down first. For any time series, write the process you imagine generating it before you choose a procedure. If the answer is “the actual economy of two actual states in three actual years”, you have a census of that process and a description is the honest product. If the answer is “each month’s rate is a stable underlying rate plus noise”, then say what the noise is and whether the months are exchangeable, and inference may follow. Most capstones that go wrong with time series go wrong at this sentence.

A Stage 1 proposal, sketched from this table

ElementSketch
QuestionDid Missouri’s unemployment rate rise as much as Kansas’s in the 2020 shock, and had the gap between the two states returned to its 2019 level by late 2021?
Data and productionBLS LAUS statewide seasonally adjusted rates, 2019-2021, model-based estimates from the Current Population Survey and administrative records; neither sampling nor assignment by the analyst
Planned procedureDescription and displays first (both series on one axis; the MO − KS difference over time). Then an approximate interval for the change in the mean difference between 2019 and 2021, with the dependence caveat, or none if the caveat is judged fatal
Conditions to checkPlot of the differences against month for trend and runs; the BLS revision history; the published error ranges for state estimates
Alternative explanationThe two states’ industry mix differs (Missouri’s leisure, hospitality and manufacturing jobs against Kansas’s aircraft and agriculture); a gap that widened and closed may reflect industry mix rather than policy or geography
EthicsPublic aggregate data; no personal information

To extend the table beyond 2021 or to other states, the BLS API takes any statewide series ID of the form LASSTss0000000000003, where ss is the two-digit state FIPS code, and returns the most recent years without a key.

Sources: BLS, Local Area Unemployment Statistics: Missouri statewide unemployment rate (series LASST290000000000003) (data.bls.gov); BLS Public Data API (the endpoint used here) (bls.gov)

Reading

Nineteen Years of Missouri Death Rates: The CDC Table Behind Three Lessons

CDC NCHS, NCHS – Leading Causes of Death: United States, data.cdc.gov dataset bi63-dtpu, Missouri rows, 1999-2017, age-adjusted death rates per 100,000 and deaths; retrieved 2026-11-10 · Public domain (works of the United States government), with SLS original worked computation

Why this reading. Three lessons point at one CDC dataset. The significance-test investigation offers “a CDC comparison” of yearly changes in two causes of death, paired by year; the procedure workshop asks whether Missouri’s unintentional-injury death rate trended upward from 2010 to 2017; and the capstone lists the same table as a health data source. The lessons name the dataset (data.cdc.gov, bi63-dtpu) and never print it. Here it is for Missouri: eight series over nineteen years, with two of the analyses worked so that yours can be checked.

What the numbers are

The National Center for Health Statistics compiles death certificates from every state. This dataset gives, for each state and year from 1999 to 2017, the number of deaths from each of the leading causes and the age-adjusted death rate per 100,000 population. Age adjustment matters: Missouri’s population aged over these years, and an older population dies more often from most causes whatever else is happening. The rate is what the state’s deaths would be if its age mix matched a fixed standard population, so rates from different years can be compared. The dataset also holds chronic lower respiratory disease, influenza and pneumonia, and kidney disease; they are in the source and not printed here. Retrieved 2026-11-10.

Table 1. Missouri, age-adjusted deaths per 100,000, 1999–2017

YearAll causesHeart diseaseCancerStrokeUnintentional injuriesAlzheimer’sDiabetesSuicide
1999945.0300.5207.965.643.215.026.312.6
2000920.8286.0205.965.141.917.824.612.5
2001913.4273.4207.162.042.517.825.612.8
2002922.3272.7204.163.145.619.026.812.1
2003907.0264.9202.157.747.620.627.111.8
2004872.7248.5201.156.146.421.923.612.4
2005873.3236.8198.052.947.825.624.812.5
2006859.0228.8195.850.550.125.123.513.5
2007837.7218.9190.949.448.825.522.413.5
2008861.3219.8189.549.248.930.020.413.0
2009815.3206.1185.744.947.525.519.814.3
2010819.5201.8185.644.047.928.821.214.0
2011812.0197.4179.443.450.527.420.815.3
2012803.0193.4182.342.247.825.919.614.9
2013807.7194.7179.440.647.027.520.515.6
2014807.0194.7177.741.048.727.419.416.3
2015816.9197.9173.440.850.928.719.717.1
2016808.2192.1167.040.457.030.020.118.4
2017820.1191.1167.341.058.832.321.018.5

For scale, the 2017 counts behind the rates:

Cause (2017)All causesHeart diseaseCancerStrokeUnintentional injuriesAlzheimer’s diseaseDiabetesSuicide
Deaths61,87614,82012,9713,1593,7762,5451,6051,151
Age-adjusted rate820.1191.1167.341.058.832.321.018.5

Worked: do two causes decline at different rates? (paired by year)

Cancer and heart disease both fell over these 19 years (heart disease from 300.5 to 191.1, cancer from 207.9 to 167.3). The question is whether the yearly change differs between them. Compute each cause’s change from one year to the next (18 changes), then the difference between the two changes for the same interval, which pairs them by year:

IntervalCancer changeHeart disease changeDifference (cancer − heart)
1999–2000−2.0−14.5+12.5
2000–2001+1.2−12.6+13.8
2001–2002−3.0−0.7−2.3
2002–2003−2.0−7.8+5.8
2003–2004−1.0−16.4+15.4
2004–2005−3.1−11.7+8.6
2005–2006−2.2−8.0+5.8
2006–2007−4.9−9.9+5.0
2007–2008−1.4+0.9−2.3
2008–2009−3.8−13.7+9.9
2009–2010−0.1−4.3+4.2
2010–2011−6.2−4.4−1.8
2011–2012+2.9−4.0+6.9
2012–2013−2.9+1.3−4.2
2013–2014−1.7+0.0−1.7
2014–2015−4.3+3.2−7.5
2015–2016−6.4−5.8−0.6
2016–2017+0.3−1.0+1.3

The 18 paired differences have mean d̄ = 3.822 and sd = 6.611. H0: μd = 0. SE = 6.611/√18 = 1.558; t = 3.822/1.558 = 2.45, df = 17. From the df-17 row of the critical-values reading, the tail beyond 2.0 is 0.0309 and beyond 2.5 is 0.0115, so 0.023 < P < 0.062, and technology gives 0.0253. Interval: 3.82 ± 2.110(1.558) = 0.53 to 7.11 deaths per 100,000 per year. Heart disease fell faster than cancer by about 3.8 deaths per 100,000 each year, on average.

Now the conditions, honestly. The changes are not a random sample; they are the entire record. The heart-disease changes are more variable than the cancer changes (sd 6.0 against 2.4), and a dot plot of the differences shows large values at 2003–2004 and 2000–2001 that drive the result. Consecutive changes may be related (a big drop is often followed by a small one). All of that makes the P-value an approximation. The size of the interval is the more reliable part of the answer.

Worked: did the unintentional-injury rate trend upward, 2010–2017?

Eight points: 47.9, 50.5, 47.8, 47.0, 48.7, 50.9, 57.0, 58.8. Least squares against year gives slope b = 1.426 per 100,000 per year, SEb = 0.458, t = 3.12 on 6 df, P = 0.0207, r = 0.786. The rate rose by about 1.4 deaths per 100,000 each year. Lesson 4.3 gives the method and its conditions; with eight consecutive years the residuals are not independent, so treat the P-value as an approximation and the direction as the finding. (The national rise in these deaths over the same years is documented by CDC; this table lets you compare Missouri’s rate with the United States rate by downloading the “United States” rows of the same dataset.)

When is inference meaningful here at all?

These are essentially complete counts of a whole state, not samples. Lesson 4.4 says: if the question is about these years in this state, compute and describe; there is nothing to infer. Inference becomes meaningful when the question is about a process: is the underlying risk changing, as opposed to the recorded rate wobbling? A test then treats each year’s rate as the process plus year-to-year noise, and asks whether the trend or difference is larger than that noise. State which question you are asking before you choose a procedure.

Sources: CDC NCHS, Leading Causes of Death: United States (dataset bi63-dtpu) (data.cdc.gov); data.cdc.gov API (the query used here) (data.cdc.gov)

Reading

The Census Bureau’s Own Margins: Broadband Subscriptions in Thirteen Missouri Areas

U.S. Census Bureau, American Community Survey 2023 1-year estimates, detailed table B28002 (lines 1 and 4) for Missouri and twelve of its counties and county-equivalents, from the ACS Summary File (table-based format, acsdt1y2023-b28002.dat); the claims to judge are SLS original framings; retrieved 2026-11-10 · Public domain (works of the United States government), with SLS original worked comparisons

Why this reading. The discourse task says to bring an estimate with a margin of error, “such as an ACS table for your county”. Here is one: the Census Bureau’s 2023 American Community Survey count of households with a broadband subscription, for Missouri and twelve of its counties and county-equivalents, with the Bureau’s own published margins of error. It is exactly the object the lesson teaches you to read: a 90% margin, an estimate that is a count divided by a count, and a comparison you may be tempted to make by eye. Three comparisons are worked, and three claims are set up to judge.

The survey

The American Community Survey samples about 3.5 million addresses a year across the United States and asks, among many other things, whether the household has an internet subscription and of what type. The Bureau publishes each estimate with a margin of error at the 90% confidence level. The figures below are from detailed table B28002, Presence and Types of Internet Subscriptions in Household, 2023 1-year estimates, line 1 (all households) and line 4 (households with a broadband subscription of any type), taken from the ACS Summary File. One-year estimates exist only for areas of 65,000 or more people. Retrieved 2026-11-10.

Table 1. Households with a broadband subscription, ACS 2023 1-year

AreaHouseholdsMOE (90%)With broadbandMOE (90%)PercentMOE of percent (90%)MOE at 95%
Missouri (state)2,556,271±10,8042,319,607±14,20390.7%±0.40±0.48
Jackson County310,868±3,159290,926±3,59893.6%±0.66±0.79
St. Louis County419,348±3,213394,648±4,62094.1%±0.83±0.99
St. Louis city150,404±2,942136,373±3,57690.7%±1.58±1.89
St. Charles County163,272±1,908155,874±2,70695.5%±1.23±1.46
Greene County132,232±1,656115,952±2,54287.7%±1.58±1.88
Clay County103,995±1,73694,469±2,34890.8%±1.67±1.99
Jefferson County88,836±1,55682,463±1,94992.8%±1.47±1.76
Boone County76,762±1,57370,452±2,25191.8%±2.25±2.68
Jasper County52,000±1,09346,490±1,59389.4%±2.42±2.88
Christian County35,654±85131,843±1,45389.3%±3.47±4.14
Platte County44,751±88042,035±1,09993.9%±1.62±1.93
St. Francois County25,489±1,02622,481±1,13688.2%±2.69±3.21

The Bureau publishes the two counts and their margins; the percent and its margin are derived. For a proportion p = Y/X whose numerator is a subset of its denominator, the ACS documentation gives MOE(p) = (1/X)√(MOEY2 − p2MOEX2). For Greene County: p = 115,952/132,232 = 0.8769; MOE = (1/132,232)√(2,5422 − 0.87692 × 1,6562) = 0.01578, or ±1.58 points. The last column converts to 95% by multiplying by 1.96/1.645 = 1.19.

Three comparisons, worked

The standard error of each estimate is MOE/1.645; the SE of a difference between two areas is √(SE12 + SE22); the 90% margin on the difference is 1.645 times that.

ComparisonDifference (points)SE of difference90% margin on differenceVerdict
Jackson minus Greene5.9√(0.4012 + 0.9592) = 1.040±1.71difference exceeds its margin
St. Louis minus St. Louis city3.4√(0.5062 + 0.9632) = 1.088±1.79difference exceeds its margin
Jasper minus Christian0.1√(1.4712 + 2.1112) = 2.573±4.23not distinguishable

Look at the Jasper–Christian row against the naive method. Jasper’s 90% interval is 87.0 to 91.8; Christian’s is 85.8 to 92.8. They overlap almost completely, and the difference is also not significant, so here the two methods agree. They do not always: the lesson’s fifth misreading is that overlapping intervals prove no difference, and the difference’s own margin, √(SE12 + SE22), is always smaller than SE1 + SE2, which is what overlap tests. Try the St. Louis County–St. Louis city row: the two 90% intervals (93.3 to 94.9 and 89.1 to 92.3) do not overlap, and the difference of 3.4 points is larger than its own margin of ±1.79.

Claims to judge (SLS original framings of these real figures)

A. “St. Louis city trails the county around it in broadband: 90.7% of households against 94.1%.” The arithmetic holds: the 3.4-point gap is about 3.2 standard errors. Now ask what a margin of error does not cover. The ACS counts households, not people, and excludes group quarters (dormitories, care homes, shelters). A city with many one-person households, many renters who move often, or many residents in group quarters can differ from the county in ways the subscription question does not capture, and a household may “have a subscription” that it cannot afford to keep.

B. “Only 88% of Greene County households have broadband. Springfield’s county has the worst connectivity in Missouri.” Greene’s estimate is 87.7% ± 1.58, and the gap to the state, 3.1 points, is about 3.1 standard errors: real beyond doubt. But “worst in the state” compares Greene with 114 other jurisdictions (113 counties and St. Louis city), 98 of which are not in the 1-year file at all, and the 16 that are have margins from under one point to about 3.5 points. Buchanan County (St. Joseph), also in the 1-year file, has an estimate of 87.2%, below Greene’s, and its margin is not small relative to the gap. A ranking built from a pile of estimates, each with its own margin, is a much weaker claim than any one comparison.

C. “The margin of error is ±0.40, so the statewide figure of 90.7% is known almost exactly.” Yes for sampling error: with about 2.6 million households and a large sample, the interval is narrow. But the 9.3% without broadband is roughly 236,664 households, and the ACS’s own nonresponse follow-up, imputation of missing answers and the wording of the question all bear on that figure in ways the ±0.40 does not include. A tiny margin of error describes precision, not accuracy.

Bring one of these to the discourse lab, or bring your own county’s row. To retrieve a county not printed here, search data.census.gov for table B28002 and choose the 2023 1-year or, for counties under 65,000 people, the 5-year estimate, whose margins will be different again.

Sources: U.S. Census Bureau, ACS table B28002 for Missouri counties (data.census.gov) (data.census.gov); U.S. Census Bureau, ACS Summary File (the file used here: 2023 1-year, acsdt1y2023-b28002.dat) (www2.census.gov); U.S. Census Bureau, ACS Accuracy of the Data (2023), including the derived-estimate formulas (census.gov); U.S. Census Bureau, guidance on using ACS estimates and margins of error (census.gov)

Final StepComplete Final Lesson →

Leave a Reply Cancel reply

Choosing the Right Procedure and Writing It Up
Prev

© 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