Statistics in Everyday Life: 8 Worked Examples

Eight everyday questions you can answer more carefully

Statistical thinking starts with concrete questions: What is being counted? Compared with what? How uncertain is the result? The following hypothetical examples show the arithmetic behind familiar claims. They are teaching examples, not advice about a real product, treatment, election, or investment.

1. “50% lower risk”: relative versus absolute change

Suppose an unwanted outcome occurs in 20 of 1,000 people under condition A and 10 of 1,000 under B.

• A’s rate: 2%.
• B’s rate: 1%.
• Absolute decrease: 1 percentage point.
• Relative decrease: (2% − 1%)/2% = 50%.

Both descriptions are arithmetically correct, but the relative percentage sounds much larger. To evaluate a real claim, ask about the study design, population, outcome definition, follow-up, uncertainty, and harms or costs. A comparison alone does not establish that the condition caused the difference.

2. A positive test: do not reverse the conditional probability

Consider a fictional screening model for 10,000 people. Assume prevalence 1%, sensitivity 90%, and specificity 95%.

Model group	Positive	Negative	Total
Has condition	90	10	100
Does not have condition	495	9,405	9,900
Total	585	9,415	10,000

Among positives, the modeled fraction with the condition is 90/585 ≈ 15.38%, despite 90% sensitivity. Most people in the model do not have the condition, so even a 5% false-positive rate contributes many positive results.

Sensitivity is P(positive | condition), not P(condition | positive). Real interpretation depends on the relevant prior probability and test context. Explore the arithmetic in the Bayes calculator or the conditional probability lesson.

3. Reviews: more observations do not remove selection bias

A shop has 4.8 stars from 10 reviewers; another has 4.5 stars from 1,000. The second mean is based on more observations, but neither set is automatically representative of all customers.

People choose whether to leave reviews. Incentives, moderation, duplicate accounts, product changes, and different customer populations can affect the result. More data can reduce sampling variability under a model while leaving systematic bias untouched.

Ask about the rating distribution, recent versus older reviews, and how reviewers were selected. A difference in means is a prompt for investigation, not a universal ranking rule.

4. Waiting times: the mean can hide the typical experience

For waits of 2, 3, 3, 4, and 8 minutes, the mean is 4 and the median is 3. If the long wait becomes 18, the mean becomes 6 while the median stays 3.

For capacity planning, total waiting time per customer may make the mean useful. For the middle customer’s experience, the median is useful. For reliability, an upper percentile can matter more than either. Report the summary that serves the question, and include spread.

Try the values in the descriptive statistics calculator.

5. A poll’s margin of error is not every kind of error

Suppose an illustrative simple random sample finds 48% support with an approximate 95% interval of 45%–51%. The interval reflects uncertainty under the sampling and estimation method. It is not a hard boundary on every possible value and does not automatically include nonresponse, coverage, weighting, or measurement error.

Separate candidates’ overlapping intervals are not a formal test of their difference; estimates from the same sample can be dependent. Ask who was sampled, when, how the question was phrased, and which population the estimate concerns.

Read confidence intervals to distinguish coverage from certainty about one fixed interval.

6. A website experiment: count people consistently

Variant A gets 30 signups from 100 independently assigned users; B gets 45 from 100. The observed increase is 15 percentage points, or 50% relative to A.

A larger conversion count is not sufficient if B received more users. Repeated visits by the same user are not necessarily independent trials. Repeatedly checking a fixed-sample test and stopping when p becomes small changes its behavior.

Define assignment, the unit of analysis, the primary outcome, and the stopping rule before comparing. The two-sample lesson works through the illustrative proportion test and interval.

7. “Above average”: know the reference group

A score of 85 on a test with mean 70 and SD 10 has z = 1.5. A score of 90 on a different test with mean 80 and SD 5 has z = 2.

The second score is farther above its own reference mean in SD units. That does not establish greater knowledge if the tests or populations differ. Also, z = 2 converts to roughly the 97.7th percentile only under a suitable normal model; standardization alone does not make the distribution normal.

Use the z-score lesson to separate relative position from normal-percentile conversion.

8. A rain probability: probability of which event?

A precipitation forecast refers to a specified event, location, and period. A 30% probability does not mean rain for 30% of the day or a universal statement about 30% of a town.

The US National Weather Service’s explanation defines probability of precipitation for measurable precipitation at a point in the forecast area during the specified period. Read the forecast’s location, time window, and qualifying event before translating it into a decision.

A forecast is useful partly through calibration: across comparable events assigned a probability around 30%, the event should occur around 30% of the time. One rainy day does not prove a 30% forecast was wrong.

A reusable checklist for a numerical claim

1. What are the numerator and denominator?
2. Which population, time period, and outcome are described?
3. Is the claim absolute, relative, or standardized?
4. How were observations selected or assigned?
5. What uncertainty and possible bias remain?
6. Does the evidence support description, association, or causation?

You do not need an advanced model to ask these questions. Often they reveal the most important limitation before any calculation begins.

Keep learning

Start with statistics foundations, learn means and medians, or follow a guided reading path.