Direct answer
How can a business simulation teach statistics?
Treat each completed run as one observation, not as proof. Students hold the scenario and most decisions steady, repeat a baseline strategy, repeat one alternative, and record the same outcome each time. The two small distributions reveal whether a difference is large compared with variation inside each strategy.
Mean describes arithmetic center, median identifies the middle ordered result, and range gives a quick view of spread. When these summaries disagree or the distributions overlap, students inspect individual values before making a claim instead of selecting whichever strategy had the single best run.
Investigation menu
Six questions with measurable outcomes
| Simulation | Compare | Primary statistic | Guardrail | Worksheet |
| Lemonade stand | Two prices or batch sizes | Profit or cups sold | Waste or unmet demand | Open |
| Restaurant | Two staffing levels | Customers served | Rating or labor cost | Open |
| Coffee shop | Two price-quality plans | Transactions or profit | Queue or waste | Open |
| Bakery | Two production quantities | Units sold or profit | Stockouts or waste | Open |
| Fitness studio | Two membership plans | Members or profit | Churn or utilization | Open |
| Car wash | Two package or labor plans | Cars served or profit | Wait or quality | Open |
Use one primary outcome for the statistical comparison and one guardrail to catch a harmful tradeoff. Keep units and reporting periods consistent.
Ready-to-use sequence
50-minute repeated-trials lesson
Learning goal: compare two small distributions using center, spread, individual values, and a qualified business interpretation.
- Frame the comparison — 6 minutes. Choose one decision, one primary outcome, and one guardrail. Write a directional prediction and list fixed settings.
- Define an observation — 4 minutes. Agree when a run starts and ends, which dashboard value to copy, and its unit.
- Collect strategy A — 9 minutes. Run the baseline three times under comparable conditions. Record every result, including inconvenient values.
- Collect strategy B — 9 minutes. Change only the selected decision and complete three runs without tuning between trials.
- Summarize both distributions — 10 minutes. Order values and calculate mean, median, and range. Compare centers, spreads, overlap, and guardrails.
- Audit the claim — 6 minutes. Check for a possible outlier and explain how the recommendation changes if it is included or excluded.
- Recommend cautiously — 6 minutes. Select A, B, or more testing. Cite center and spread, name a tradeoff, and limit the conclusion to the model.
Printable student record
Repeated-trial evidence table
Check core calculations
Business and scenario: ______________________________________________
Question: How does changing __________________ affect __________________?
Prediction and reason: ______________________________________________
Strategy A: __________________ Strategy B: __________________
Held constant: ______________________________________________________
Primary outcome and unit: __________________ Guardrail and unit: __________________
| Measure | A1 | A2 | A3 | B1 | B2 | B3 |
| Primary outcome | | | | | | |
|---|
| Guardrail outcome | | | | | | |
|---|
| Unexpected condition | | | | | | |
| Summary | Strategy A | Strategy B | Interpretation |
| Ordered values | | | Show values first. |
|---|
| Mean | | | B − A = ______ |
|---|
| Median | | | B − A = ______ |
|---|
| Range | | | Which varied more? |
|---|
| Possible outlier | | | Keep, investigate, or exclude—with reason. |
|---|
| Guardrail | | | Did the gain create a tradeoff? |
Calculation guide
Mean: sum the values and divide by the number of trials.
Median: order values and select the middle; with an even count, average the two middle values.
Range: maximum − minimum. It gives a quick view of spread but not every distribution feature.
Percentage-point difference: subtract one rate from another. From 72% to 78% is 6 percentage points.
Relative percent change: (new − original) ÷ |original| × 100%. State the baseline and avoid it when the baseline is zero.
Read beyond the average
- Do mean and median favor the same strategy?
- Are ranges narrow, wide, or strongly overlapping?
- Does one value pull the mean away from the median?
- Was it a recording error, changed condition, or possible outcome?
- Would the recommendation survive one additional trial?
- Does the guardrail reveal a cost, capacity, quality, or stakeholder tradeoff?
Never remove a result only because it weakens a preferred claim. Record the rule and reason, then report whether the conclusion depends on it.
Write a statistically responsible recommendation
- Conditions: name the simulator, strategies, outcome, unit, and trial count.
- Center: compare both means and medians.
- Spread: compare ranges, overlap, and unusual values.
- Tradeoff: cite the guardrail rather than optimizing one measure.
- Decision: recommend A, B, or more trials and state confidence.
- Boundary: explain why simulated data cannot represent real customers or forecast a business.
Sentence frame: Across ___ trials, strategy ___ had a mean ___ of ___ versus ___, while medians were ___ and ___. The ranges ___, suggesting ___. We recommend ___ inside this model, but would collect ___ before a real decision.
12-point statistics investigation rubric
| Criterion | 3 — strong | 2 — developing | 1 — beginning | 0 — missing |
| Comparable data | Clear strategies, repeated trials, constant conditions, complete units. | Mostly comparable with a minor gap. | Few trials or inconsistent units. | No usable comparison. |
|---|
| Calculations | Mean, median, range, and differences are accurate and shown. | One minor error. | Several errors or missing work. | No summaries. |
|---|
| Distribution reasoning | Uses values, center, spread, outlier, overlap, and tradeoff. | Interprets center and some variation. | Relies on one average or best run. | No interpretation. |
|---|
| Recommendation | Evidence-based, qualified, model-bounded, with a next test. | Plausible with partial qualification. | Overclaims or weak evidence link. | No supported decision. |
Grade reasoning and transparent uncertainty, not the largest profit. The full assessment rubric adds feedback codes and group-work guidance.
Teacher notes, support, and extension
Prevent false precision
Require units and sensible rounding. Three trials with a mean of $143.333 do not justify a tenth of a cent. Ask what precision the inputs and decision support.
Support learners
Provide two approved strategies, a six-value dataset, and a calculation organizer. Let students explain center and spread verbally before writing.
Extend the analysis
Collect five trials, graph dot plots, calculate interquartile range, compare class groups, or predefine an outlier rule. More simulated trials reduce some uncertainty without making the model representative.
Keep inference honest, private, and safe
These observations come from simplified educational rules, not sampled customers, employees, competitors, or markets. More runs can describe the model more precisely but cannot correct missing variables or establish real-world validity.
Do not collect student personal information or invent survey responses. Real research needs permission, privacy protection, truthful methods, current legal and safety requirements, reliable source data, and stakeholder input. Simulator results are not financial, legal, operational, or market advice.
Frequently asked questions
Which simulator works best?
Lemonade Stand, Restaurant, Coffee Shop, and Bakery offer accessible measures. Any simulator works with comparable repeated conditions.
What statistics do students calculate?
Mean, median, range, differences, and possible outliers, followed by a distribution-based interpretation.
How many repeated trials are needed?
Three per strategy fit this lesson. Five or more show variation better, but remain a small model-generated sample.
Do students need accounts or personal information?
No account, name, email, download, or personal information is required.
Can results predict a real business?
No. They describe a simplified model, not a representative real-world sample.