DATA1001 Chap.3 The Normal Distribution
The Normal Distribution
The Normal model connects numerical summaries with probability. Its centre is the mean and its scale is the standard deviation. Standard units express a value as its signed distance from the mean in SDs: subtract the mean and divide by SD. Positive z values lie above the mean and negative values below it. The 68, 95, 99.7 rule gives quick areas for approximately Normal data, while R functions support more precise questions.
The function pnorm returns a cumulative lower tail at a supplied value. Upper tails use one minus that result, and between probabilities subtract two cumulative areas. The function qnorm reverses the direction, taking a cumulative probability and returning a cutoff. Canvas Topic 4, Normal Model, also links Normal reasoning to measurement error.
Repeated independent errors that are centred at zero can be averaged to reduce chance variation, with the standard error of the average shrinking by the square root of the number of readings. Averaging does not remove a systematic calibration bias. A z-score can always be computed when mean and SD are available, but a Normal probability requires a defensible model for the distribution.
Inspect shape and context before treating the bell curve as exact. In exam work, draw a small labelled curve, mark the raw cutoff, translate to standard units, shade the requested area and only then select a function or rule. Finish by stating whether the answer is a probability, a percentile cutoff or a standardised score. This prevents correct arithmetic from being attached to the wrong tail or wrong unit.
What this chapter covers
- 01
Topic 4: Normal Model
- 02
Standardise values with z scores
- 03
Use the empirical rule for quick checks
- 04
Read lower, upper and between areas
- 05
Invert cumulative area with qnorm
- 06
Separate chance error from systematic bias
Worked example · free
Move between a raw value and standard units
- 1Subtract the mean: 50 minus 40 equals 10.
- 1Divide by SD: 10 divided by 5 equals z of 2.
- 1The lower tail is
pnorm(50, 40, 5), equivalentlypnorm(2). - 1About 97.7 percent of modelled values lie at or below 50.
Key terms
- Normal model
- A symmetric bell shaped probability model determined by its mean and standard deviation.
- Standard unit
- A signed distance from the mean measured in standard deviations.
- Z score
- The observed value minus its mean divided by its standard deviation.
- Percentile
- A cutoff at or below which a stated percentage of the distribution lies.
- Chance error
- Random measurement variation that is centred around the exact value under the model.
- Systematic bias
- A persistent directional measurement error that averaging does not remove.
The Normal Distribution FAQ
When may I use a Normal model?
Use it when the distribution is reasonably bell shaped for the task or when a justified sampling approximation applies. Computing a z score alone does not prove Normality. Inspect a histogram or other diagnostics and be cautious with strong skew, multiple modes or influential outliers.
How do pnorm and qnorm differ?
Use pnorm when you know a value and want cumulative area at or below it. Use qnorm when you know a cumulative area and want the corresponding cutoff. Upper tails and middle areas require complements or differences, so sketch the region first.
What does a negative z score mean?
A negative z score means the value lies below the mean. Its magnitude gives the distance in standard deviations. It does not mean the raw value itself is negative, and it does not by itself say whether the value is unusual without a reference model.
Why average repeated measurements?
When independent chance errors have mean zero, their average remains centred correctly while its standard error shrinks with the square root of the number of readings. Averaging improves precision, but a miscalibrated instrument can remain biased no matter how many readings are taken.
How do I check a Normal answer?
Use symmetry and the 68, 95, 99.7 rule. A cutoff two SDs above the mean should have a lower tail near 97.5 percent and a small upper tail. Check that a percentile answer has measurement units and a probability answer lies between zero and one.
Exam move
Make standardisation automatic by practising both directions. From raw value to z, subtract the mean and divide by SD. From z to raw value, multiply by SD and add the mean. Draw and shade a tiny bell curve for every probability question before selecting a command. Work lower tails, upper tails, between areas and percentile cutoffs as distinct question types.
Use the empirical rule as a check rather than a substitute for careful tail direction. Read short R calls aloud: function, cutoff or probability, mean and SD. Add measurement error drills where you calculate how averaging four, nine or sixteen independent readings changes standard error, then state why systematic bias remains. In your error log, separate wrong arithmetic from wrong interpretation.
Finish each answer by naming whether the result is a probability, a z score or a raw cutoff with units. Review the exam lens only after you can explain each command without running R.