Lesson 7 of 8 · 14 min

Estimating probabilities from data

Use relative frequencies from experiments and two-way tables as point estimates of probabilities and conditional probabilities, and as evidence for or against independence.

What you will be able to do

  • Calculate a relative frequency and use it as a point estimate of a probability.
  • Estimate conditional probabilities from a two-way table of data.
  • Compare conditional relative frequencies to judge whether two events might be independent.

What is the probability that a Year 11 student gets at least 8 hours' sleep on a school night? No list of equally likely outcomes can tell you. Instead you collect data and use the proportion of students who do as an estimate. Most probabilities in real life, from insurance to medicine, are estimated this way.

Definition

Relative frequency as an estimate

relative frequency=number of times the event occursnumber of trials or observations\text{relative frequency} = \frac{\text{number of times the event occurs}}{\text{number of trials or observations}}

A relative frequency is a point estimate of the probability: a single best guess from the data. Another sample would give a slightly different estimate, and larger samples tend to give estimates closer to the true probability.

One computer simulation of 500 tosses of a fair coin (247 heads). Early on the relative frequency jumps around; with more tosses it settles close to the probability, 0.5.

A basketballer made 132 of their last 160 free throws. Estimate the probability that they make their next free throw.

Data from a two-way table give conditional relative frequencies too. A Brisbane school surveyed 400 Year 11 students about whether they use their phone in bed after 10 pm (PP) and whether they usually get at least 8 hours' sleep on a school night (SS).

Survey of 400 Year 11 students
StudentsAt least 8 hours (SS)Less than 8 hours (S′S')Total
Phone in bed after 10 pm (PP)66154220
No phone in bed after 10 pm (P′P')9981180
Total165235400

The estimate of P(S)P(S) for all students is 165400≈0.41\frac{165}{400} \approx 0.41. For students who use their phone in bed, restrict to the PP row:

P(S∣P)≈66220=0.30.P(S \mid P) \approx \frac{66}{220} = 0.30.

Students who use their phone in bed were much less likely to get 8 hours' sleep than students in general.

Using the survey table, estimate the probability that a student who does not use their phone in bed after 10 pm gets at least 8 hours' sleep.

Key idea

Judging independence from data

If AA and BB were independent, then P(A∣B)=P(A∣B′)=P(A)P(A \mid B) = P(A \mid B') = P(A). So compare the conditional relative frequencies:

  • Close together: independence is plausible.
  • Clearly different: the events are probably not independent.

Sample estimates need not match exactly, so look for a clear difference, not just any difference, and be more cautious when the sample is small. Here the estimates 0.30 and 0.55 are far apart, so phone use in bed and getting enough sleep do not appear to be independent.

Worked example

Estimating conditional probabilities from rescue data

  • Complex familiar

A Sunshine Coast surf lifesaving club recorded 500 rescues over one summer, noting whether the person rescued was an overseas visitor (VV) and whether they had been swimming outside the flags (OO).

RescuesOutside the flags (OO)Between the flags (O′O')Total
Overseas visitor (VV)11228140
Not an overseas visitor (V′V')198162360
Total310190500

(a) Estimate the probability that a rescued swimmer was outside the flags.

(b) Estimate P(O∣V)P(O \mid V) and P(O∣V′)P(O \mid V').

(c) Comment on whether OO and VV appear to be independent for these rescues.

4 steps

Try it yourself first, then check the working one step at a time.

Method

  1. Relative frequency = count in the event ÷ count in the group.

  2. For a conditional estimate, use only the row or column of the condition.

  3. Compare P(A∣B)P(A \mid B) with P(A∣B′)P(A \mid B'): close suggests independence, clearly different suggests not.

Watch out for

  • Dividing 112 by 500, which estimates P(O∩V)P(O \cap V), not P(O∣V)P(O \mid V).

  • Over-claiming: these are rescues only, so the data can't give the probability that an overseas visitor needs rescuing.

A survey of 300 students recorded whether each student is left-handed and whether they play a ball sport.

StudentsPlays a ball sportDoesn't playTotal
Left-handed181230
Right-handed165105270
Total183117300

Which conclusion is best supported by the data?

Choose one answer

Common mistake

Two errors with data

"Not exactly equal" doesn't mean "not independent". Relative frequencies vary from sample to sample, so small differences are expected even for independent events.

Association isn't causation. The sleep data show that phone use and sleep are related, but they can't show that phones cause less sleep. Students with a lot of homework might do both.

Tip

With technology

For a large table, enter the counts in a spreadsheet and let a formula such as =B2/D2 calculate each conditional relative frequency. A simulation (random numbers on a calculator or spreadsheet) generates long runs of trials, like the coin tosses above, to see how relative frequencies settle.

Summary

  • Relative frequency = number of times the event occurs ÷ number of trials; it is a point estimate of the probability.

  • Larger samples usually give more reliable estimates.

  • Estimate P(A∣B)P(A \mid B) from a table by restricting to the BB row or column.

  • If the conditional relative frequencies P(A∣B)P(A \mid B) and P(A∣B′)P(A \mid B') are close, independence is plausible; if they are clearly different, the events are probably not independent.

  • Data can show that events are related, but not why.

Finished the lesson?

Try the 3 quick checks above and the lesson completes itself, or mark it complete now.

Lesson 7 of 80% read

Save your progress with a free account