Lesson 8 of 8 · 22 min

Quantiles and modelling with the normal distribution

Use the inverse normal to find quantiles and cut-off values, find an unknown mean or standard deviation from given probabilities, and solve multi-step modelling problems with normal distributions.

What you will be able to do

  • Use the inverse normal to find kk such that P(X<k)P(X < k) or P(X>k)P(X > k) is a given value.
  • Find an unknown μ\mu or σ\sigma, or both, from given probabilities using zz-scores.
  • Combine normal probabilities with conditional probability and binomial models.
  • Evaluate whether a normal model and its answers are reasonable in context.

So far you have started with a value and found a probability. Real questions often run the other way. A mango farm in Bowen sells its heaviest 10% as premium fruit: how heavy must a mango be to make the cut? A machine must underfill no more than 1% of bottles: what mean should it be set to? These are inverse problems: the probability is known and a value or parameter is not.

Definition

Quantiles and the inverse normal

The qq quantile of XX is the value kk with P(X<k)=qP(X < k) = q. The 0.9 quantile is the 90th percentile; the 0.5 quantile is the median.

The calculator's inverse normal function takes the left-tail area qq, the mean μ\mu and the standard deviation σ\sigma, and returns kk. For a right-tail condition, convert first: P(X>k)=0.1P(X > k) = 0.1 means P(X<k)=0.9P(X < k) = 0.9.

The premium cut-off kk has 90% of the area to its left and 10% to its right.

Worked example

Grading mangoes with the inverse normal

  • Simple familiar
  • Technology active

The masses of mangoes from a Bowen farm are normally distributed with mean 450 g and standard deviation 40 g. The heaviest 10% are sold as premium and the lightest 5% are sent for juice.

(a) Find the minimum mass of a premium mango.

(b) Find the maximum mass of a mango sent for juice.

Give answers correct to 2 decimal places.

5 steps

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

Method

  1. Write the condition as P(X<k)=qP(X < k) = q.

  2. Inverse normal with qq, μ\mu and σ\sigma.

  3. Check the side of the mean and the size of zz.

Watch out for

  • Entering 0.1 for the top 10%, which gives the bottom 10% cut-off, 398.74 g.

Scores on an aptitude test are normally distributed with mean 80 and standard deviation 6. A prize goes to the top 5% of scores. Find the minimum score needed, correct to 2 decimal places.

Round to 2 decimal places.

If the mean or standard deviation is the unknown, the inverse normal cannot be used on XX directly, because you don't know its parameters. Work on the standard normal instead: find the zz-score with the given tail area from Z∼N(0,1)Z \sim N(0, 1), then set it equal to x−μσ\frac{x - \mu}{\sigma} and solve.

Finding an unknown parameter

  1. Write the probability statement

    For example P(X>260)=0.1P(X > 260) = 0.1.

  2. Find the matching z-score with the inverse normal on N(0, 1)

    P(Z<z)=0.9P(Z < z) = 0.9 gives z≈1.2816z \approx 1.2816.

  3. Equate to the standardised value

    260−μσ=1.2816\frac{260 - \mu}{\sigma} = 1.2816

  4. Solve

    One unknown: rearrange. Two unknowns: write a second equation from the second condition and solve simultaneously.

The lengths of steel rods are normally distributed with mean 250 mm. 10% of rods are longer than 260 mm. Find the standard deviation, correct to 2 decimal places.

in mm

Round to 2 decimal places.

Worked example

Finding the mean and standard deviation from two percentages

  • Complex familiar
  • Technology active

The life of a phone battery on one charge is normally distributed. 15% of batteries last less than 20 hours and 10% last more than 32 hours. Find the mean and standard deviation of battery life, correct to 2 decimal places.

5 steps

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

Method

  1. Turn each condition into P(Z<z)=qP(Z < z) = q and find zz.

  2. Write μ+zσ=x\mu + z\sigma = x for each.

  3. Solve simultaneously, then check.

Watch out for

  • Using +1.03643+1.03643 for the lower value: a value below the mean has a negative zz-score.

  • Rounding zz to 2 decimal places, which shifts the answers in the second decimal place.

Some inverse problems need no technology at all, because symmetry does the work. If X∼N(40,σ2)X \sim N(40, \sigma^2) and P(X<30)=0.2P(X < 30) = 0.2, then 30 is 10 below the mean, and by symmetry P(X>50)=0.2P(X > 50) = 0.2 too, even though you never find σ\sigma.

X∼N(40,σ2)X \sim N(40, \sigma^2) and P(X<30)=0.2P(X < 30) = 0.2. Without technology, find P(40<X<50)P(40 < X < 50).

Modelling problems often chain ideas together. A normal model gives the probability pp that one item has a property; if items are independent, the number of such items in a sample of nn is then Bin(n,p)\mathrm{Bin}(n, p). And "given that" still means conditional probability: P(A∣B)=P(A∩B)P(B)P(A \mid B) = \frac{P(A \cap B)}{P(B)}, where for nested events like X<745X < 745 inside X<750X < 750 the intersection is just the smaller event.

Worked example

Underfilled bottles at a bottling plant

  • Complex unfamiliar
  • Technology active

A machine fills bottles labelled 750 mL. The volume XX mL in a bottle is normally distributed with mean 752 mL and standard deviation 4 mL. Bottles are sold in packs of 6, and fills are independent.

(a) Find the probability that a bottle is underfilled (contains less than 750 mL), correct to 4 decimal places.

(b) Find the probability that a pack contains at least 2 underfilled bottles, correct to 3 decimal places.

(c) Given that a bottle is underfilled, find the probability that it contains less than 745 mL, correct to 4 decimal places.

(d) The company wants only 1% of bottles underfilled, without changing σ\sigma. Find the mean the machine should be set to, correct to 2 decimal places, and comment on the result.

7 steps

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

Method

  1. Normal model for one item; binomial for a count of independent items.

  2. P(A∣B)=P(A∩B)P(B)P(A \mid B) = \frac{P(A \cap B)}{P(B)}; for nested events the intersection is the smaller event.

  3. For an unknown mean, find zz from N(0,1)N(0, 1) and solve z=x−μσz = \frac{x - \mu}{\sigma}.

Watch out for

  • Rounding pp to 0.31 before the binomial step, which gives 0.601 instead of 0.598.

  • Writing P(X<745∣X<750)=P(X<745)P(X < 745 \mid X < 750) = P(X < 745): the condition changes the denominator.

Common mistake

Rounding too early

Keep the unrounded zz-score or probability in your calculator memory for the next step. Rounding zz to 2 decimal places, or a probability to 2 decimal places before it goes into a binomial calculation, can change the final answer in the last place, and markers check it.

Exam tip

Evaluate the model

Questions often end with "comment on the reasonableness" or "evaluate the model". Useful points: does the context suit a symmetric bell-shaped model? Is P(X<0)P(X < 0) negligible for a quantity that can't be negative (it should be more than about 3σ3\sigma below the mean)? Are the items independent for a binomial step? Does the answer make sense in size and direction?

Summary

  • The inverse normal turns a left-tail area into a value: P(X<k)=q⇒kP(X < k) = q \Rightarrow k. Convert right tails first.

  • For an unknown μ\mu or σ\sigma, find zz from N(0,1)N(0, 1) and solve z=x−μσz = \frac{x - \mu}{\sigma}; two unknowns need two equations.

  • Symmetry can answer some inverse questions without technology.

  • A normal probability pp per item feeds a binomial model Bin(n,p)\mathrm{Bin}(n, p) for counts of independent items.

  • Keep full accuracy between steps, and evaluate the model in context.

Finished the lesson?

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

Lesson 8 of 80% read

Save your progress with a free account