Term 2 · Module 1 of 4

Normal Distributions

Business Statistics for Entrepreneurs II

Practise this module in Claude, ChatGPT or Gemini

Set it up free
Mind map

Normal distribution: recap and probability calculations

The normal distribution models a continuous quantity that varies around a central value, with deviations becoming less common farther from the center. It is described by its mean μ\mu and standard deviation σ\sigma; its variance is σ2\sigma^2. A standard normal variable ZZ has mean 00 and standard deviation 11. Standardizing converts a value of X∼N(μ,σ2)X\sim N(\mu,\sigma^2) to the number of standard deviations it lies from the mean:

Z=X−μσ,X=μ+σZ.Z=\frac{X-\mu}{\sigma},\qquad X=\mu+\sigma Z.

The standard normal table or a distribution template can then provide cumulative probabilities, tail probabilities and inverse percentiles. For an interval, subtract cumulative probabilities: P(a<X<b)=P(X<b)−P(X≤a)P(a<X<b)=P(X<b)-P(X\le a). For a top-tail cutoff, use the upper-tail probability directly or use symmetry.

Ramesh's pawn-shop counts

Let XX be the daily number of people bringing goods to pawn brokers. The mean is 658658, and the bottom 3%3\% of days have at most 610610 visitors: P(X≤610)=0.03P(X\le610)=0.03. The corresponding standard-normal quantile is approximately −1.88-1.88.

610−658σ=−1.88⟹σ=481.88≈25.52.\frac{610-658}{\sigma}=-1.88\quad\Longrightarrow\quad \sigma=\frac{48}{1.88}\approx25.52.

Thus X∼N(658,25.522)X\sim N(658,25.52^2). For the chance of between 600600 and 700700 visitors, standardize each endpoint:

z600=600−65825.52≈−2.27,z700=700−65825.52≈1.65.z_{600}=\frac{600-658}{25.52}\approx-2.27,\qquad z_{700}=\frac{700-658}{25.52}\approx1.65.

Using the table values P(Z≤1.65)=0.9505P(Z\le1.65)=0.9505 and P(Z≤−2.27)=0.0116P(Z\le-2.27)=0.0116:

P(600≤X≤700)=0.9505−0.0116=0.9389.P(600\le X\le700)=0.9505-0.0116=0.9389.

For the busiest 3%3\% of days, the upper-tail cutoff is z=1.88z=1.88:

x=658+1.88(25.52)≈706.x=658+1.88(25.52)\approx706.

About 706706 or more visitors therefore marks the busiest 3%3\% of days, using the rounded values in the calculation.

Suresh's bottle fill

Let XX be the fill amount in ounces, with adjustable mean μ\mu and fixed standard deviation 0.050.05. To have at least 99.5%99.5\% of bottles contain at least 11.911.9 ounces, set P(X≥11.9)=0.995P(X\ge11.9)=0.995. The standard-normal value with right-tail area 0.9950.995 is −2.57-2.57:

11.9−μ0.05=−2.57⟹μ=11.9+2.57(0.05)=12.0285≈12.03 oz.\frac{11.9-\mu}{0.05}=-2.57\quad\Longrightarrow\quad\mu=11.9+2.57(0.05)=12.0285\approx12.03\text{ oz}.

At this mean, the probability of at least 1212 ounces is

P(X≥12)=P ⁣(Z≥12−12.030.05)=P(Z≥−0.6)=0.726.P(X\ge12)=P\!\left(Z\ge\frac{12-12.03}{0.05}\right)=P(Z\ge-0.6)=0.726.

So the probability is approximately 72.6%72.6\%.

Key takeaways

  • Standardize with Z=(X−μ)/σZ=(X-\mu)/\sigma; reverse with X=μ+σZX=\mu+\sigma Z.
  • Use cumulative-probability differences for intervals and inverse quantiles for cutoffs.
  • Upper-tail cutoffs use positive zz values; lower-tail cutoffs use negative values by symmetry.
  • Track whether a table reports a cumulative area or a tail area.

3.5 Linear Combinations of Normal Random Variables

A weighted sum represents combined uncertainty: revenue from several products, total demand over several weeks, or the difference between two measurements. If the inputs are jointly normal, any linear combination remains normal. Its mean follows the weights directly; its variance depends on both the squared weights and dependence between inputs.

For Y=∑i=1naiXi+bY=\sum_{i=1}^{n}a_iX_i+b,

E[Y]=∑i=1naiμi+b.E[Y]=\sum_{i=1}^{n}a_i\mu_i+b.

If the XiX_i are independent,

Var⁡(Y)=∑i=1nai2σi2.\operatorname{Var}(Y)=\sum_{i=1}^{n}a_i^2\sigma_i^2.

Without independence, covariance terms must be included:

Var⁡(Y)=∑iai2σi2+2∑i<jaiajCov⁡(Xi,Xj).\operatorname{Var}(Y)=\sum_i a_i^2\sigma_i^2+2\sum_{i<j}a_ia_j\operatorname{Cov}(X_i,X_j).

For jointly normal inputs, YY is normal even when the inputs are dependent; dependence changes its variance. Independence is the condition that lets the covariance terms vanish.

Transforming one normal variable

If X∼N(μ,σ2)X\sim N(\mu,\sigma^2) and Y=aX+bY=aX+b, then

Y∼N(aμ+b,a2σ2).Y\sim N(a\mu+b,a^2\sigma^2).

Choosing a=1/σa=1/\sigma and b=−μ/σb=-\mu/\sigma gives Z=(X−μ)/σZ=(X-\mu)/\sigma, the standard normal transformation.

Sums and averages of independent, identically distributed normals

For independent Xi∼N(μ,σ2)X_i\sim N(\mu,\sigma^2), the sum has mean nμn\mu and variance nσ2n\sigma^2; its standard deviation is n σ\sqrt n\,\sigma, not nσn\sigma. The average Xˉ=1n∑iXi\bar X=\frac1n\sum_iX_i has mean μ\mu and variance σ2/n\sigma^2/n, so its standard deviation is σ/n\sigma/\sqrt n. Larger samples reduce the variability of the average.

Piston and cylinder fit

Let X1X_1 be piston radius and X2X_2 cylinder inside radius. The given independent normal measurements are X1X_1 with mean 3030 mm and standard deviation 0.050.05 mm, and X2X_2 with mean 30.2530.25 mm and standard deviation 0.060.06 mm. Define the clearance Y=X2−X1Y=X_2-X_1. Then

E[Y]=30.25−30=0.25 mm,E[Y]=30.25-30=0.25\text{ mm}, Var⁡(Y)=0.062+(−1)2(0.05)2=0.0036+0.0025=0.0061 mm2,\operatorname{Var}(Y)=0.06^2+(-1)^2(0.05)^2=0.0036+0.0025=0.0061\text{ mm}^2, σY=0.0061≈0.0781 mm.\sigma_Y=\sqrt{0.0061}\approx0.0781\text{ mm}.

Thus Y∼N(0.25,0.0061)Y\sim N(0.25,0.0061). A non-fitting piston corresponds to Y≤0Y\le0; the stated normal-template result is P(Y≤0)=0.0007P(Y\le0)=0.0007. If optimal performance requires a clearance from 0.10.1 to 0.350.35 mm, the stated result is

P(0.1≤Y≤0.35)=0.872.P(0.1\le Y\le0.35)=0.872.

The worked probabilities use the rounded standard deviation 0.0780.078 mm.

Average cotton-plant height

Each of 2020 plant heights is modeled as independent and identically distributed normal with mean 29.429.4 cm and standard deviation 2.12.1 cm. For their average Xˉ\bar X:

E[Xˉ]=29.4,Var⁡(Xˉ)=2.1220=4.4120=0.2205,E[\bar X]=29.4,\qquad \operatorname{Var}(\bar X)=\frac{2.1^2}{20}=\frac{4.41}{20}=0.2205, σXˉ=0.2205≈0.4696 cm.\sigma_{\bar X}=\sqrt{0.2205}\approx0.4696\text{ cm}.

The reported 95th percentile is 30.1730.17 cm. The reported lower-tail statement says the average is above 28.6328.63 cm 95%95\% of the time (the 5th percentile). A central 95%95\% interval is reported as 28.4728.47 to 30.3230.32 cm. The reduced spread of the sample average (≈0.47\approx0.47 cm versus 2.12.1 cm for one plant) illustrates the standard-error effect.

Comparing two stock returns

Company A returns are normal with mean 8%8\% and standard deviation 1.51.5 percentage points. Company B returns are normal with mean 9.5%9.5\% and standard deviation 22 percentage points. Assume independence. An A return from 5%5\% through 10%10\% is called moderate; the stated template probability is 0.8860.886. A B return of at least 10%10\% is excellent; the stated upper-tail probability is 0.40130.4013.

To compare their returns, define Y=XB−XAY=X_B-X_A. Independence gives

E[Y]=9.5−8=1.5 percentage points,E[Y]=9.5-8=1.5\text{ percentage points}, Var⁡(Y)=22+(−1)2(1.5)2=4+2.25=6.25,\operatorname{Var}(Y)=2^2+(-1)^2(1.5)^2=4+2.25=6.25, σY=2.5 percentage points.\sigma_Y=2.5\text{ percentage points}.

Therefore Y∼N(1.5,6.25)Y\sim N(1.5,6.25). The reported probabilities are

P(XB>XA)=P(Y≥0)=0.726,P(X_B>X_A)=P(Y\ge0)=0.726, P(XB−XA≥2)=P(Y≥2)=0.421.P(X_B-X_A\ge2)=P(Y\ge2)=0.421.

The latter is the chance that B exceeds A by at least two percentage points.

Key takeaways

  • Linear combinations of jointly normal variables are normal.
  • Multiply means by coefficients; multiply variances by squared coefficients.
  • Add covariance terms unless independence is given.
  • Summing increases spread by n\sqrt n; averaging reduces standard deviation by n\sqrt n.
  • For a difference, define the order explicitly (for example, cylinder minus piston) before interpreting the sign.

Examples with sums of normal variables

The same rules solve portfolio totals and cumulative demand. In each case, first define the aggregate random variable; then combine means and variances; then standardize the requested threshold.

Nilesh's nine investments

Nine independent investments each have normal returns with mean 50,00050{,}000 and standard deviation 10,00010{,}000. If Y=∑i=19XiY=\sum_{i=1}^9X_i, then

E[Y]=9(50,000)=450,000,Var⁡(Y)=9(10,0002),E[Y]=9(50{,}000)=450{,}000,\qquad \operatorname{Var}(Y)=9(10{,}000^2), σY=9(10,000)=30,000.\sigma_Y=\sqrt9(10{,}000)=30{,}000.

The question asks for the value below which total return falls just 1%1\% of the time. The lower-tail standard-normal quantile is approximately −1.28-1.28:

y=450,000+(−1.28)(30,000)=411,600.y=450{,}000+(-1.28)(30{,}000)=411{,}600.

Using the transcript's less-rounded calculation, the reported figure is 411,533411{,}533; with z=−1.28z=-1.28 exactly as stated, the arithmetic gives 411,600411{,}600 (about 411.5411.5 thousand). For the interval 400,000400{,}000 to 520,000520{,}000:

zL=400,000−450,00030,000≈−1.67,zU=520,000−450,00030,000≈2.33.z_L=\frac{400{,}000-450{,}000}{30{,}000}\approx-1.67,\qquad z_U=\frac{520{,}000-450{,}000}{30{,}000}\approx2.33.

Using table values Φ(2.33)=0.9901\Phi(2.33)=0.9901 and Φ(−1.67)=0.0475\Phi(-1.67)=0.0475,

P(400,000≤Y≤520,000)=0.9901−0.0475=0.9426.P(400{,}000\le Y\le520{,}000)=0.9901-0.0475=0.9426.

Naman's three-week inventory

Weekly demand is independent and normal: means 50,45,6550,45,65 tons and standard deviations 10,5,1510,5,15 tons. For total demand Y=X1+X2+X3Y=X_1+X_2+X_3:

μY=50+45+65=160 tons,\mu_Y=50+45+65=160\text{ tons}, σY2=102+52+152=100+25+225=350,σY=350≈18.71 tons.\sigma_Y^2=10^2+5^2+15^2=100+25+225=350,\qquad \sigma_Y=\sqrt{350}\approx18.71\text{ tons}.

With 180180 tons in stock, a stockout occurs when Y>180Y>180:

z=180−16018.71≈1.07,P(Y>180)=P(Z>1.07)≈0.142.z=\frac{180-160}{18.71}\approx1.07,\qquad P(Y>180)=P(Z>1.07)\approx0.142.

So stockout risk is about 14%14\% (roughly 86%86\% chance of no stockout). For a 98%98\% service probability, use Φ(z)=0.98\Phi(z)=0.98, for which z≈2.05z\approx2.05:

S=160+2.05(18.71)≈198.4 tons,S=160+2.05(18.71)\approx198.4\text{ tons},

or plan for about 199199 tons.

Key takeaways

  • Add independent normal returns or demands by summing their means and variances.
  • Standard deviation is the square root of the combined variance.
  • For a lower percentile use a negative quantile; for a high service-level stock target use a positive quantile.
  • State and question independence assumptions: correlated investments or demands require covariance terms.

3.6 Chi-square distribution

Squaring a standard-normal observation removes its sign and makes large deviations contribute strongly. The sum of squared independent standard normals produces a chi-square distribution. Its degrees of freedom count how many squared standard normals are summed.

Z∼N(0,1)⇒Z2∼χ12,X=∑i=1kZi2∼χk2(Zi independent).Z\sim N(0,1)\Rightarrow Z^2\sim\chi^2_1,\qquad X=\sum_{i=1}^{k}Z_i^2\sim\chi^2_k\quad( Z_i\text{ independent}).

The distribution is supported on positive values, has one parameter kk, mean kk and variance 2k2k. As kk grows it becomes less skewed and more bell-shaped, approaching a normal shape. Applications named in the material include goodness-of-fit and other nonparametric hypothesis tests, feature selection/classification, and testing independence in logistic-regression contexts.

Critical values and table reading

The upper-tail critical value χα,k2\chi^2_{\alpha,k} is defined by

P(X≥χα,k2)=α.P(X\ge\chi^2_{\alpha,k})=\alpha.

Thus 95%95\% of observations lie at or below the critical value when α=0.05\alpha=0.05. The chi-square table is indexed by degrees of freedom and upper-tail probability. If a requested degree of freedom or tail area is absent, the table permits an approximation by interpolation; a distribution template can calculate the quantity directly. The main-module template examples give these right-tail cutoffs (columns correspond to α=0.2,0.1,0.05,0.01\alpha=0.2,0.1,0.05,0.01):

Degrees of freedom kk 0.20.2 0.10.1 0.050.05 0.010.01
5 7.3 9.2 11.1 15.1
10 13.4 16.0 18.3 23.2
15 19.3 22.3 25 37.6
20 25 28.4 31.4 37.6

The values generally rise with degrees of freedom, while skewness falls. The separate table exercises illustrate how to read exact upper-tail entries:

  1. k=5, α=0.05k=5,\ \alpha=0.05: row 55, column 0.050.05 gives 11.0711.07. Hence P(X>11.07)=0.05P(X>11.07)=0.05 and P(X≤11.07)=0.95P(X\le11.07)=0.95.
  2. k=15, α=0.025k=15,\ \alpha=0.025: row 1515, column 0.0250.025 gives 27.4927.49; the lower cumulative probability is 0.9750.975.
  3. k=20, α=0.975k=20,\ \alpha=0.975: row 2020, column 0.9750.975 gives 9.599.59.
  4. k=10, α=0.1k=10,\ \alpha=0.1: row 1010, column 0.10.1 gives 15.9915.99.
  5. k=18, α=0.95k=18,\ \alpha=0.95: row 1818, column 0.950.95 gives 9.399.39.

For the last three, the unusual large upper-tail areas are read as stated: the critical value is low because most probability lies to its right.

Key takeaways

  • A chi-square variable is a sum of squared independent standard normals; kk counts the terms.
  • Its support is positive, mean is kk, and variance is 2k2k.
  • Chi-square critical-value notation here uses right-tail area; read the correct kk row and α\alpha column.
  • Table gaps call for an approximation; do not confuse the upper tail with the CDF.

3.7 Student's tt distribution

The Student's tt distribution is a symmetric, zero-centered distribution with heavier tails than the standard normal at finite degrees of freedom. It arises from dividing an independent standard normal by the square root of an independent chi-square variable per degree of freedom:

T=ZX/k,Z∼N(0,1),X∼χk2,Z⊥X.T=\frac{Z}{\sqrt{X/k}},\qquad Z\sim N(0,1),\quad X\sim\chi^2_k,\quad Z\perp X.

It is used for hypothesis tests about population means, especially with small samples, for comparing means, and in linear-regression hypothesis testing and diagnostics. Its mean is 00 for k>1k>1 (the lecture states two or more); its variance is k/(k−2)k/(k-2) for k>2k>2. As kk grows, it approaches the standard normal. William Sealy Gosset, a statistician at Guinness studying barley and hops, published under the pseudonym “Student” in 1908 because the brewery restricted employee publication.

Critical values and central areas

The t table rows give degrees of freedom and the columns give right-tail areas. Symmetry means a right-tail area at +t+t equals the left-tail area at −t-t. For central area 1−α1-\alpha:

P(−tα/2,k≤T≤tα/2,k)=1−α.P(-t_{\alpha/2,k}\le T\le t_{\alpha/2,k})=1-\alpha.

The main-module template reports upper-tail critical values for α=0.1,0.05,0.025,0.005\alpha=0.1,0.05,0.025,0.005:

kk 0.10.1 0.050.05 0.0250.025 0.0050.005
5 1.48 2.02 2.57 4.03
10 1.37 1.81 2.23 3.17
15 1.34 1.75 2.13 2.95
20 1.33 1.73 2.09 2.85

With 1616 degrees of freedom, table lookups and symmetry give:

  1. P(T>2.12)=0.025P(T>2.12)=0.025.
  2. P(T<1.337)=1−0.10=0.90P(T<1.337)=1-0.10=0.90.
  3. P(T<−1.746)=P(T>1.746)=0.05P(T<-1.746)=P(T>1.746)=0.05.
  4. P(T>2.583)=0.01P(T>2.583)=0.01.
  5. P(−2.12<T<2.12)=1−2(0.025)=0.95P(-2.12<T<2.12)=1-2(0.025)=0.95.
  6. P(−1.746<T<1.746)=1−2(0.05)=0.90P(-1.746<T<1.746)=1-2(0.05)=0.90.

If a value such as 2.22.2 is beyond the table entry 2.122.12 in the same row, its right-tail probability is less than 0.0250.025; if the value is smaller, such as 1.91.9, its right-tail probability is greater than 0.0250.025. A table may bound but not give the exact probability.

For inverse lookups:

  1. k=12k=12, upper tail 0.0250.025: t=2.179t=2.179.
  2. k=50k=50, lower tail 0.050.05: symmetry gives t=−1.676t=-1.676.
  3. k=30k=30, upper tail 0.10.1: t=1.31t=1.31.
  4. Central 90%90\% with k=25k=25: each tail is 0.050.05, so the interval is [−1.708,1.708][-1.708,1.708].
  5. Central 95%95\% with k=45k=45: each tail is 0.0250.025, so the interval is [−2.01,2.01][-2.01,2.01].

Key takeaways

  • The t distribution is symmetric and has heavier tails than the standard normal for finite kk.
  • Degrees of freedom matter: critical values shrink toward standard-normal values as kk increases.
  • For central intervals, split the excluded probability equally between the two tails.
  • A lower-tail quantile is negative when the corresponding table gives the positive right-tail value.

3.8 FF distribution

The FF distribution (Fisher distribution, named for Ronald Fisher) compares two independent chi-square quantities after scaling each by its degrees of freedom:

Fk1,k2=X1/k1X2/k2,X1∼χk12,X2∼χk22,X1⊥X2.F_{k_1,k_2}=\frac{X_1/k_1}{X_2/k_2},\qquad X_1\sim\chi^2_{k_1},\quad X_2\sim\chi^2_{k_2},\quad X_1\perp X_2.

k1k_1 is the numerator degrees of freedom and k2k_2 the denominator degrees of freedom. Swapping them generally changes the distribution. The variable is positive, unimodal, right-skewed and has a long right tail. The stated mean is k2/(k2−2)k_2/(k_2-2) for k2≥3k_2\ge3 (the transcript says k2≥2k_2\ge2, but at k2=2k_2=2 the mean is undefined). Its standard deviation decreases as k2k_2 increases; as both degrees of freedom grow the density concentrates near 11. Uses include analysis of variance, feature selection and testing overall multiple-regression fit.

Upper-tail critical values

The critical value Fα,k1,k2F_{\alpha,k_1,k_2} satisfies P(F≥Fα,k1,k2)=αP(F\ge F_{\alpha,k_1,k_2})=\alpha. In the F tables, denominator degrees of freedom index rows, numerator degrees of freedom index columns, and each table is for a particular upper-tail area. Preserve numerator/denominator order. The main-module template values below use columns α=0.2,0.1,0.05,0.01\alpha=0.2,0.1,0.05,0.01:

(k1,k2)(k_1,k_2) 0.20.2 0.10.1 0.050.05 0.010.01
(5, 5) 2.23 3.45 5.05 10.97
(5, 10) 1.80 2.52 3.33 5.64
(10, 5) 2.19 3.30 4.74 10.05
(20, 20) 1.47 1.79 2.21 2.94

Three table lookups demonstrate the row/column convention:

  1. (k1,k2)=(5,10)(k_1,k_2)=(5,10), upper tail 0.050.05: use denominator row 1010, numerator column 55, giving 3.333.33.
  2. (20,15)(20,15), upper tail 0.0250.025: the corresponding table entry is 2.762.76.
  3. (8,12)(8,12), upper tail 0.010.01: the corresponding table entry is 4.54.5.

For the first lookup, P(F5,10>3.33)=0.05P(F_{5,10}>3.33)=0.05, or 5%5\%; percent notation matters. Where a needed tail area or degree of freedom is absent, approximate by interpolation or use a distribution template.

Key takeaways

  • An F variable is a ratio of scaled independent chi-square variables and is positive/right-skewed.
  • Always keep numerator k1k_1 and denominator k2k_2 in the right positions.
  • Its critical values depend on both degrees of freedom and the right-tail area.
  • Larger degrees of freedom generally concentrate the distribution closer to 11.

Connecting the distributions

The normal family and its related distributions form a useful ladder: standardize a normal observation; square standard normals and add them to obtain chi-square; combine a standard normal with an independent scaled chi-square to obtain tt; and take a ratio of two independently scaled chi-squares to obtain FF. These distributions support later inference about means, fit, independence, variance comparisons and model significance. Their density formulas can be complicated; the central practical tasks are identifying the right distribution, its degrees of freedom, the requested tail, and the correct critical value.

Key takeaways

  • Normal linear combinations make probability calculations for sums, averages and differences tractable.
  • Chi-square, tt and FF distributions are related through standard-normal and chi-square variables.
  • Table conventions differ: check symmetry for tt, and carefully track right-tail area and degrees of freedom for all three.
  • Critical-value tables may require interpolation when they do not list the requested combination.

Test yourself on this module

Questions from these notes, one at a time, with feedback. Your weak topics are saved.

Set it up free Takes about a minute. Free plans work.