The Normal Distribution

Convert any bell curve to the standard one, and a single table answers every probability.

Before this Lesson

A ready-mix plant's cylinders average 4,600 psi, with a standard deviation of 400 psi, and the spec's floor is 4,000 psi. Of the next 500 cylinders, how many should you expect to come in low? Two numbers don't sound like enough to answer that. For a bell curve they are. The answer is about 33, and it takes one line of arithmetic and one look at a table.

Area is probability

Why a bell? A cylinder's strength comes out of hundreds of small nudges: a little more water in this batch, a warmer afternoon, a lazier job of rodding. Each one pushes the result up or down a bit. Most of the pushes cancel, so most cylinders land near the middle. Landing far out takes many more of them pushing one way than the other, and that's rare. Quantities built like this tend to follow the normal distribution. Not everything does (rainfall is skewed), so a question will tell you when a quantity is normal.

Two numbers set the whole curve. The mean μ\mu is where the peak sits, and the standard deviation σ\sigma says how wide it is. The Handbook's Normal Distribution section puts σ\sigma right on the drawing: it's the distance from the peak to each point of inflection, where the curve stops bending downward and starts bending upward.

μ a b x σ inflection point area = P(a < X < b) total area under the curve = 1
The normal curve, with one stretch of area marked

As you saw in Probability, a curve like this is a density, so its height isn't a probability. Area is. The chance that a value lands between aa and bb is the area over that stretch, and the whole area under the curve is 1, because the value has to land somewhere. The chance that one cylinder falls below 4,000 psi and the fraction of all cylinders that do are the same area. A single exact value has no width and so no area, which means "below" and "at or below" give the same answer.

Every bell is the same bell

The Handbook prints the curve's equation:

f(x)=1σ2π e−12(x−μσ)2f(x) = \frac{1}{\sigma\sqrt{2\pi}}\, e^{-\frac{1}{2}\left(\frac{x - \mu}{\sigma}\right)^2}

Nobody integrates this by hand, since it has no antiderivative you can write in ordinary functions. But look at the exponent. The xx, μ\mu, and σ\sigma show up there only in one combination, and that combination is the key to this whole Lesson:

z=x−μσz = \frac{x - \mu}{\sigma}

Read it as a recipe. x−μx - \mu is how far the value sits from the mean, and dividing by σ\sigma counts that distance in standard deviations. So z=2z = 2 means two standard deviations above the mean, and z=−1.5z = -1.5 means one and a half below it, whatever the units. Subtracting μ\mu slides the curve until its peak sits at zero, and dividing by σ\sigma stretches or squeezes it to a standard width.

The 4,000 psi floor is 600 psi below the mean, and one standard deviation is 400 psi:

z=4,000−4,600400=−1.5z = \frac{4{,}000 - 4{,}600}{400} = -1.5

A peak-hour volume of 1,200 veh/h, on a road with μ=1,500\mu = 1{,}500 and σ=200\sigma = 200 veh/h, also sits at z=−1.5z = -1.5. Different quantities and different units land on the same spot of the same curve.

−3 −2 −1 0 1 2 3 z −1.5 3,400 3,800 4,200 4,600 5,000 5,400 5,800 Cylinder strength (psi) 4,000 900 1,100 1,300 1,500 1,700 1,900 2,100 Peak-hour volume (veh/h) 1,200 same area on every ruler
One curve, with a different ruler for each quantity

So the areas only need computing once, for the curve with μ=0\mu = 0 and σ=1\sigma = 1, which the Handbook calls the unit normal distribution. Every normal question goes through zz to that one curve.

One table, four columns

The Handbook's Unit Normal Distribution table lists those areas. Its first column is headed xx, but that's where your zz goes: as the Handbook says, F(x)F(x) becomes F(z)F(z). Each of the other columns you'll use gives one area, and the figure shows which.

0 x F(x) area below x 0 x R(x) area above x 0 −x x 2R(x) area in both tails 0 −x x W(x) area within −x to x
The area each column of the table gives
Part of the unit normal table
xxF(x)F(x)R(x)R(x)2R(x)2R(x)W(x)W(x)
0.00.50000.50001.00000.0000
0.50.69150.30850.61710.3829
1.00.84130.15870.31730.6827
1.28160.90000.10000.20000.8000
1.50.93320.06680.13360.8664
1.64490.95000.05000.10000.9000
1.96000.97500.02500.05000.9500
2.00.97720.02280.04550.9545
2.32630.99000.01000.02000.9800
2.50.99380.00620.01240.9876
2.57580.99500.00500.01000.9900
3.00.99870.00130.00270.9973

Adapted from the FE Reference Handbook, Unit Normal Distribution

The full table runs xx from 0.0 to 3.0 in steps of 0.1, then adds a short list of longer values, its fractiles, where the areas come out round. Its f(x)f(x) column gives the curve's height, which you won't need. When zz falls between rows, interpolate, or use your calculator's normal distribution function if it has one.

There are no negative rows, and you don't need any, because the curve is symmetric. The tail below −1.5 is the mirror image of the tail above +1.5. The Handbook writes this as F(−x)=1−F(x)F(-x) = 1 - F(x), and 1−F(x)1 - F(x) is just R(x)R(x):

F(−x)=R(x)F(-x) = R(x)

That settles the cylinders: F(−1.5)=R(1.5)=0.0668F(-1.5) = R(1.5) = 0.0668, the highlighted row. About 6.7% of them come in low.

One, two, and three standard deviations

Read the WW column at three rows:

W(1.0)=0.6827W(2.0)=0.9545W(3.0)=0.9973W(1.0) = 0.6827 \qquad W(2.0) = 0.9545 \qquad W(3.0) = 0.9973

So about 68%, 95%, and 99.7% of any normal population lies within one, two, and three standard deviations of its mean. The Handbook doesn't state this as a rule, but the exam asks for it directly, so learn those three. The 2R2R column holds what's left over, shared between the two tails.

To choose a column, listen to the wording. "Above" or "below" one limit is one tail, so RR or FF. "Outside" a band, or "more than so far from the mean," is two tails, so 2R2R. "Within" a band is WW.

Now make σ\sigma bigger. The curve stretches sideways, but its area has to stay 1, so the peak drops. That's the 1/σ1/\sigma out front in the equation: double σ\sigma and you halve the peak. Meanwhile, a limit that stays put now sits fewer standard deviations from the mean, so its zz shrinks toward zero.

2,200 3,000 3,800 4,600 5,400 6,200 7,000 floor: 4,000 σ = 400 psi, z = −1.5 σ = 800 psi, z = −0.75 Cylinder strength (psi)
The same mean and floor, with σ doubled
Checkpoint
1 of 3

Daily traffic at a count station is normal, with a mean of 12,000 vehicles and a standard deviation of 800. What fraction of days see between 10,400 and 13,600 vehicles?

A.
0.0455
B.
0.6827
C.
0.9545
D.
0.9772

Below, above, between

Every forward question takes the same three moves: sketch the curve and shade what's asked, turn each limit into zz, then match the shading to a column. The sketch takes ten seconds, and it's what catches the wrong column. Keep the table above handy from here on, or better, open the Handbook's own, since that's the one you'll have on the exam.

Back to the cylinders, with μ=4,600\mu = 4{,}600 and σ=400\sigma = 400 psi. Above 5,000 psi, z=+1.0z = +1.0 with the shading to the right, so the fraction is R(1.0)=0.1587R(1.0) = 0.1587. Below 5,000 psi has the same zz with the shading to the left, so it's F(1.0)=0.8413F(1.0) = 0.8413. Those two add to 1, as they must.

Between 4,000 and 5,000 psi is a lopsided patch that no column shades directly. So take everything below the upper limit, and remove everything below the lower one:

F(1.0)−F(−1.5)=0.8413−0.0668=0.7745F(1.0) - F(-1.5) = 0.8413 - 0.0668 = 0.7745
1.0 F(1.0) = 0.8413 −1.5 F(−1.5) = 0.0668 −1.5 1.0 0.7745 − =
A between area is one below area minus another
Worked Example Peak-hour volumes in a range
1 of 4

Peak-hour volumes on a road are normal, with a mean of 1,500 veh/h and a standard deviation of 200 veh/h. Find the probability that a day's volume falls between 1,200 and 1,900 veh/h.

Running the table backward

Now flip the question: what strength do 90% of cylinders exceed? You have an area and want a value, so run the chain in reverse, from area to zz to xx. Solving the zz equation for xx gives

x=μ+zσx = \mu + z\sigma

Find the area in the column it belongs to, then read zz from the first column. The fractile rows are there for round areas like 0.10 and 0.05. The table holds only positive values, so the sign is on you: a value below the mean gets a negative zz, and the sketch tells you which side you're on.

Worked Example The strength that 90% of cylinders exceed
1 of 3

Cylinder strengths are normal, with μ=4,600\mu = 4{,}600 psi and σ=400\sigma = 400 psi. Find the strength that 90% of cylinders exceed.

Moving the curve to meet a limit

Sometimes you can't move the limit. A bag marked 94 lb has to hold 94 lb. What you can change is the curve: slide it away from the limit, or tighten it, until the tail past the limit is as small as the spec allows.

μ limit Start tail too big μ limit Raise the mean 5% z = −1.6449 μ limit Tighten the spread 5% z = −1.6449
Two ways to shrink the tail past a fixed limit

Both fixes end with the limit at the same zz, because the target fraction sets zz. So x=μ+zσx = \mu + z\sigma is the same equation as before, now with xx known, zz from the table, and either μ\mu or σ\sigma left to find. The answer is a boundary. Slide the curve any farther or tighten it any more, and you still meet the spec with room to spare, which is why these questions ask for the smallest such mean or the largest allowable σ\sigma.

Worked Example Setting a bag filler
1 of 3

A machine fills 94 lb bags of cement, with a standard deviation of 0.5 lb wherever it's set. At what mean should it run so that only 5% of bags come in underweight?

From a fraction to a count

A fraction of a batch becomes a count once you multiply by the batch size, which is the expected value from Probability. Back to the 500 cylinders: (500)(0.0668)=33.4(500)(0.0668) = 33.4, so expect about 33 to come in low.

With both a lower and an upper limit, no value can fail both ways at once. The two tails are mutually exclusive, so their fractions add. The limits are usually different distances from the mean, so work each tail on its own. If they're the same distance, 2R2R gives both at once.

Worked Example Rejected bolts
1 of 3

A plant makes 2,000 bolts with normally distributed diameters, with a mean of 20.00 mm and a standard deviation of 0.10 mm. Bolts outside 19.85 to 20.20 mm are rejected. How many rejects should it expect?

Recap

  • Area under a normal curve is probability. z=(x−μ)/σz = (x - \mu)/\sigma counts standard deviations from the mean, which carries any normal quantity onto the one curve the Handbook's table covers.
  • FF is below, RR is above, 2R2R is both tails, and WW is within. For a negative zz, F(−x)=R(x)F(-x) = R(x). A between area is one FF minus another.
  • About 68%, 95%, and 99.7% lie within one, two, and three standard deviations. A larger σ\sigma flattens and widens the curve and grows the tail past a fixed limit.
  • Backward, go from area to zz to x=μ+zσx = \mu + z\sigma, with the sign from your sketch. That same equation finds the mean or the largest σ\sigma that meets a limit.
  • For a batch, add the tail fractions, then multiply by the batch size.