COMM5000 Chap.2 Probability and Normal Models
Probability and Normal Models
Translate business uncertainty into events, conditional probabilities, complements and normal-model tail areas while keeping denominators, assumptions and direction visible. Start by defining the experiment, sample space and event in ordinary language. A joint event asks whether two conditions occur together; a conditional event restricts attention to a named group. This distinction changes the denominator.
Contingency tables are useful because they display intersections, marginal totals and conditional groups at once. The complement rule simplifies at-least-one questions, while the addition rule prevents overlapping events from being counted twice. Multiplying event probabilities requires a defensible independence assumption or an explicit conditional probability.
Random variables translate outcomes into numerical values and support probability models. For a normal model, standardise a threshold with z = (x minus the mean) divided by the standard deviation, then match the requested event to a left tail, right tail or interval. The sign of z identifies direction and should agree with the business wording. Normality is an approximation to evaluate, not a label supplied by the calculator.
Inspect shape, hard bounds and outliers, and consider whether the process combines groups with different distributions. A probability conclusion is conditional on the model and data-generating process. It describes uncertainty under stated assumptions; it does not by itself explain why the event occurs or establish an intervention effect.
What this chapter covers
- 01
Events and sample spaces
- 02
Conditional probability
- 03
Independence
- 04
Random variables
- 05
Normal standardisation
Worked example · free
Conditional dispute rates
- +1The four marks used here are not an official university marking scheme. International rate = 12/50 = 24%.
- +1Domestic orders = 150 and domestic disputes = 8.
- +1Domestic rate = 8/150 = 5.33%.
- +1The reviewed international rate is 4.50 times the domestic rate, an association requiring investigation.
Key terms
- Conditional probability
- The probability of an event within a restricted condition or group.
- Independence
- A relationship where learning one event occurred does not change the probability of another.
- Z-score
- A standardised distance from a mean measured in standard deviations.
Probability and Normal Models FAQ
How do I choose the denominator?
For conditional probability, restrict the data to the group named after given and divide by that group total. For a joint probability, use the full relevant sample.
Which normal tail should I use?
Translate the event into an inequality before using a table or software. Above is a right tail, below is a left tail, and between two thresholds is an interval area.
How can I check whether a probability answer is plausible?
Write the event and denominator before calculating, then check that the result lies between zero and one. A conditional probability must use the conditioned group total, while a joint probability uses the relevant full sample. For a normal model, mark the threshold relative to the mean and confirm that a high positive cutoff has a small right-tail area.
Use complements to verify at-least-one results and make sure complementary probabilities sum to one. Finally, question independence or normality when repeated outcomes share a cause, the distribution has hard bounds, or the data show strong skew.
Assessment move
Sketch each event and write the denominator or tail before calculating. Read conditional notation aloud, because the words after given identify the restricted group. Rebuild a two-by-two table from totals and intersections, then calculate joint, marginal and conditional probabilities from the same cells. Use complements for at-least-one problems and explain why direct addition would double-count overlapping outcomes.
When independence is claimed, write what process feature makes one event uninformative about another; if a persistent condition links trials, use a conditional model instead. For normal-model questions, draw the mean and threshold on a line, compute z with its sign, shade the requested area and perform a direction check before reading software output.
Practise translating an area into a sentence about repeated outcomes from the relevant process. Keep individual variation separate from sampling variation, which belongs to later chapters. After every answer, add one assumption check and one possible model failure. In the final report, show the event definition and denominator so the reader can audit the calculation.
Avoid language that turns an observed risk difference into a causal story without examining selection, confounding or process changes. Build a small library of event translations: exactly one, at least one, none, both, either, given and independent. For each phrase, write the symbolic event and a plain-language check.
Compare a joint probability with the corresponding conditional probability using the same contingency table so the change in denominator becomes visible. When a normal approximation is used, record the empirical shape and whether thresholds lie in a region where approximation error could matter.
Test calculator output against simple bounds: a probability must be between zero and one, a right-tail probability beyond a high positive z should be small, and complementary areas should sum to one. If groups have different processes, avoid pooling them into a single probability model without justification.
End the analysis with a proposed operational response that recognises uncertainty and does not attribute the risk difference to a cause that was not tested.
Working through Probability and Normal Models in COMM5000? Sia is AskSia’s AI Statistics tutor — ask any COMM5000 Probability and Normal Models question and get a clear, step-by-step explanation grounded in how COMM5000 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
