Term 1 · Module 3 of 4

Discrete Probability Distribution

Business Statistics for Entrepreneurs I

Practise this module in Claude, ChatGPT or Gemini

Set it up free
Mind map

Discrete random variables and the Bernoulli model

A discrete random variable assigns numerical values to outcomes that can be counted, such as defects, late flights, or customer purchases. Its probability mass function (PMF) gives the probability of each value; the cumulative distribution function (CDF) adds probabilities up through a value. These let us describe both individual outcomes and ranges of outcomes.

f(x)=P(X=x),F(x)=P(X≤x)=∑u≤xf(u)f(x)=P(X=x),\qquad F(x)=P(X\le x)=\sum_{u\le x}f(u)

The expectation is the probability-weighted average, and the variance is the probability-weighted squared distance from that average. The standard deviation is the square root of variance.

E[X]=∑xxf(x),Var⁡(X)=∑x(x−E[X])2f(x),σX=Var⁡(X)E[X]=\sum_x x f(x),\qquad \operatorname{Var}(X)=\sum_x (x-E[X])^2f(x),\qquad \sigma_X=\sqrt{\operatorname{Var}(X)}

A Bernoulli random variable represents one yes/no outcome. Code the event of interest as success, Y=1Y=1, with probability pp; its complement is failure, Y=0Y=0, with probability q=1−pq=1-p. “Success” is a label, not necessarily a desirable outcome: it can mean a purchase, satisfaction, a star-sales day, or any chosen event.

yy Meaning Probability
11 Success pp
00 Failure q=1−pq=1-p

E[Y]=1p+0q=pE[Y]=1p+0q=p

Var⁡(Y)=(1−p)2p+(0−p)2q=pq(p+q)=pq,σY=pq\operatorname{Var}(Y)=(1-p)^2p+(0-p)^2q=pq(p+q)=pq,\qquad \sigma_Y=\sqrt{pq}

Discrete random variables and the Bernoulli model

Defining success from a business outcome

  • A fair die: success is a result of 22 or less, so p=1/3p=1/3 and q=2/3q=2/3.
  • A biased die: success is a result of 55 or more, so p=1/6p=1/6 and q=5/6q=5/6.
  • Pumps: success is at most one defective pump in a day, p=0.65p=0.65; failure is two or more, q=0.35q=0.35.
  • Airport: success is at most two late departures in the evening slot, p=0.6p=0.6; failure is three or more, q=0.4q=0.4.
  • A sales call is a Bernoulli trial if its outcome is purchase/no purchase. A service survey can be coded satisfied/not satisfied. Daily store demand can be grouped as above 10 versus at most 10.

For the fair-die definition, E[Y]=1/3E[Y]=1/3, Var⁡(Y)=(1/3)(2/3)=2/9\operatorname{Var}(Y)=(1/3)(2/3)=2/9, and σY=2/3\sigma_Y=\sqrt{2}/3. For the biased die, the corresponding values are 1/61/6, 5/365/36, and 5/6\sqrt{5}/6. For pumps they are 0.650.65, 0.65(0.35)=0.2275≈0.230.65(0.35)=0.2275\approx0.23, and 0.2275≈0.477\sqrt{0.2275}\approx0.477. For late flights they are 0.60.6, 0.240.24, and 0.24≈0.490\sqrt{0.24}\approx0.490.

If an insurance call converts with probability p=0.2p=0.2, then q=0.8q=0.8, E[Y]=0.2E[Y]=0.2, Var⁡(Y)=0.16\operatorname{Var}(Y)=0.16, and σY=0.4\sigma_Y=0.4. The expectation means 20% of calls convert in the long-run average; it does not predict a fraction of a purchase on any one call. For a “star day” defined as sales above 10,000, p=0.3p=0.3: the mean is 0.30.3, variance 0.210.21, and standard deviation 0.21≈0.46\sqrt{0.21}\approx0.46.

Key takeaways

  • A PMF describes exact values; a CDF describes all values up to a threshold.
  • Bernoulli coding reduces any chosen two-outcome event to 11 and 00.
  • The mean and variance are pp and pqpq; define the success event before calculating.

Binomial distribution

A binomial experiment repeats the same Bernoulli trial a fixed number of times. It requires a fixed nn, exactly two outcomes per trial, independent trials, and the same success probability pp on every trial. The binomial random variable YY counts successes and takes values 0,1,…,n0,1,\ldots,n. It counts how many successes occur, not their positions in the sequence.

Y∼Binomial⁡(n,p),q=1−pY\sim\operatorname{Binomial}(n,p),\qquad q=1-p

For exactly yy successes, one particular arrangement has probability pyqn−yp^yq^{n-y}. There are (ny)\binom{n}{y} arrangements of those successes among nn trials, so:

P(Y=y)=f(y)=(ny)pyqn−y=n!y!(n−y)!pyqn−y,y=0,…,nP(Y=y)=f(y)=\binom{n}{y}p^yq^{n-y}=\frac{n!}{y!(n-y)!}p^yq^{n-y},\quad y=0,\ldots,n

The mean adds the expected successes across trials; independence makes the variances add as well.

E[Y]=np,Var⁡(Y)=npq,σY=npqE[Y]=np,\qquad \operatorname{Var}(Y)=npq,\qquad \sigma_Y=\sqrt{npq}

Binomial distribution

Computing exact, cumulative, and tail probabilities

Translate wording carefully: “at most kk” means P(Y≤k)P(Y\le k); “at least kk” means P(Y≥k)P(Y\ge k); “more than kk” means P(Y≥k+1)P(Y\ge k+1); “between aa and bb, inclusive” means ∑y=abP(Y=y)\sum_{y=a}^{b}P(Y=y). Complements often shorten a calculation: P(Y≥k)=1−P(Y≤k−1)P(Y\ge k)=1-P(Y\le k-1).

For spreadsheet work, a binomial PMF function takes the target yy, trial count nn, probability pp, and a cumulative toggle: non-cumulative returns P(Y=y)P(Y=y); cumulative returns P(Y≤y)P(Y\le y). A table of PMF, CDF, and tail values makes ranges and parameter changes easier to inspect. An inverse probability problem reverses the usual direction: find a parameter such as pp that makes a specified probability reach a target.

Worked examples: biased die and defective pumps

Ten biased-die tosses. Success is 55 or more, so n=10n=10, p=1/6p=1/6, q=5/6q=5/6.

E[Y]=10(16)=1.67,Var⁡(Y)=10(16)(56)=2518≈1.39,σY≈1.18E[Y]=10\left(\frac16\right)=1.67,\quad \operatorname{Var}(Y)=10\left(\frac16\right)\left(\frac56\right)=\frac{25}{18}\approx1.39,\quad \sigma_Y\approx1.18

For at least three successes, add the probabilities for y=3y=3 through 1010; the reported result is P(Y≥3)=0.225P(Y\ge3)=0.225. To raise this probability to at least 0.50.5, changing pp gives P(Y≥3)=0.263P(Y\ge3)=0.263 at p=0.18p=0.18, 0.3220.322 at p=0.20p=0.20, 0.440.44 at p=0.24p=0.24, 0.4740.474 at p=0.25p=0.25, and about 0.5010.501 at p=0.259p=0.259. Thus p≈0.259p\approx0.259 meets the target in the reported calculation.

Fourteen production days. A success day has at most one defective pump, with p=0.65p=0.65, q=0.35q=0.35, and n=14n=14.

E[Y]=14(0.65)=9.1,σY=14(0.65)(0.35)≈1.78E[Y]=14(0.65)=9.1,\qquad \sigma_Y=\sqrt{14(0.65)(0.35)}\approx1.78

Adding P(Y=y)P(Y=y) for y=10,11,12,13,14y=10,11,12,13,14 (reported endpoint values include P(Y=10)=0.202P(Y=10)=0.202, P(Y=11)=0.137P(Y=11)=0.137, and P(Y=14)=0.002P(Y=14)=0.002) gives P(Y≥10)=0.423P(Y\ge10)=0.423. Raising pp to 0.660.66, 0.670.67, and 0.680.68 gives reported tail probabilities 0.4540.454, 0.4860.486, and 0.5190.519; at p=0.675p=0.675 the probability is just over 0.50.5.

Worked examples: airport delays, sales calls, and star days

Thirty airport days. A success is a day with at most two late flights, p=0.6p=0.6, q=0.4q=0.4, n=30n=30. The reported P(Y≥18)=0.578P(Y\ge18)=0.578. At most 10 failure days is equivalent to at least 20 success days, so P(Y≥20)=0.291P(Y\ge20)=0.291. For 15 through 25 successes inclusive, sum those eleven PMF values: the reported total is 0.9010.901.

Twelve insurance calls. With p=0.2p=0.2, q=0.8q=0.8, and n=12n=12:

E[Y]=12(0.2)=2.4,σY=12(0.2)(0.8)≈1.39E[Y]=12(0.2)=2.4,\qquad \sigma_Y=\sqrt{12(0.2)(0.8)}\approx1.39

The source first asks for at least 10 purchases but then calculates at least 6; its numerical result is for the latter: P(Y≥6)=0.019P(Y\ge6)=0.019 (using the reported P(Y=6)=0.016P(Y=6)=0.016, P(Y=7)=0.003P(Y=7)=0.003, plus the smaller remaining terms). At most three purchases sums the reported PMF values 0.069+0.206+0.283+0.236=0.7940.069+0.206+0.283+0.236=0.794, reported as 0.7950.795 (rounding).

Ten store days. A star day means sales above 10,000; n=10n=10, p=0.3p=0.3, q=0.7q=0.7. The expected count is 10(0.3)=310(0.3)=3. At most three star days:

P(Y≤3)=0.028+0.121+0.233+0.267=0.649≈0.65P(Y\le3)=0.028+0.121+0.233+0.267=0.649\approx0.65

Between three and six inclusive, summing the probabilities for 3, 4, 5, and 6 gives a reported 0.6070.607. The source also describes CDF subtraction, but its stated values do not match its result: 0.989−0.233=0.7560.989-0.233=0.756, not the reported 0.6060.606. The direct inclusive sum is retained as reported; the subtraction figures are inconsistent.

Choosing the event and parameters from context

These examples emphasize that changing the event changes pp, while changing the observation period changes nn:

Question nn Success event and pp Required probability Reported result
Pump days with no defects, at least 7 of 14 14 No defects, p=0.4p=0.4 P(Y≥7)P(Y\ge7) 0.3080.308
Days with at least 3 pump defects, at most 3 of 7 7 At least 3 defects, p=0.15p=0.15 P(Y≤3)P(Y\le3) 0.9880.988
At most one pump defect on all 14 days 14 At most one defect, p=0.4+0.25=0.65p=0.4+0.25=0.65 P(Y=14)P(Y=14) 0.0020.002
Airport days with at most one late flight, at least 15 of 30 30 At most one late flight, p=0.2+0.1=0.3p=0.2+0.1=0.3 P(Y≥15)P(Y\ge15) 0.0170.017
Airport days with exactly 4 late flights, at most 5 of 10 10 Exactly 4 late, p=0.2p=0.2 P(Y≤5)P(Y\le5) 0.9440.944
Airport days with at most 2 late flights, exactly 15 of 30 30 At most 2 late, p=0.6p=0.6 P(Y=15)P(Y=15) 0.0780.078

The pump distribution implies P(0 defects)=0.4P(0\text{ defects})=0.4, P(1 defect)=0.25P(1\text{ defect})=0.25, and P(3+ defects)=0.15P(3+\text{ defects})=0.15 in these examples. “No defects” is not the same success event as “at most one defect.” Likewise, “exactly four late flights” is different from “four or fewer.”

The model is useful only while its assumptions are credible. If trial outcomes depend on earlier outcomes or success probabilities vary, the sequence is not binomial; another model is needed. The binomial assumptions are an abstraction, so assess whether they are reasonable for the application.

Key takeaways

  • The binomial counts successes in nn independent, identical Bernoulli trials with fixed pp.
  • P(Y=y)=(ny)pyqn−yP(Y=y)=\binom{n}{y}p^yq^{n-y}, with mean npnp and variance npqnpq.
  • Wording determines the event: inclusive ranges, complements, and strict versus inclusive thresholds matter.
  • Correctly defining success and the trial unit is the main business-translation task.

Poisson distribution and process

A Poisson random variable counts how many times an event occurs in a specified interval of time or space. It takes non-negative integer values 0,1,2,…0,1,2,\ldots. Examples include arrivals in a time window, road defects along a distance, or flagged terms in a conversation. Its single parameter λ\lambda is the average count for the interval under consideration.

Y∼Poisson⁡(λ),P(Y=y)=f(y)=e−λλyy!,y=0,1,2,…Y\sim\operatorname{Poisson}(\lambda),\qquad P(Y=y)=f(y)=\frac{e^{-\lambda}\lambda^y}{y!},\quad y=0,1,2,\ldots

The mean and variance are both λ\lambda, so σY=λ\sigma_Y=\sqrt{\lambda}. The model assumes the occurrence probability is the same across equal-length intervals and occurrences in separate intervals are independent. These assumptions distinguish the model from a changing-rate process.

Poisson distribution and process

Poisson process and scaling the interval

A Poisson process describes event counts through time. The intervals between successive events are independent and identically distributed exponential random variables. If the event rate is λ0\lambda_0 per unit and the observation length is tt, the expected count for that interval is:

Λ=λ0t\Lambda=\lambda_0t

For example, at 10 arrivals per hour, the expected count over two hours is 10(2)=2010(2)=20; over half an hour it is 10(0.5)=510(0.5)=5. The rate is not itself the mean for every interval; rescale the mean whenever the window changes.

A spreadsheet table can calculate the PMF, cumulative sums, and tail sums. For instance, P(Y≤y)=P(Y≤y−1)+P(Y=y)P(Y\le y)=P(Y\le y-1)+P(Y=y), and P(Y≥y)=P(Y≥y−1)−P(Y=y−1)P(Y\ge y)=P(Y\ge y-1)-P(Y=y-1). As with the binomial, carefully distinguish P(Y>k)=P(Y≥k+1)P(Y>k)=P(Y\ge k+1) from P(Y≥k)P(Y\ge k).

Time-based examples

Eatery arrivals. With an average of 10 customers per hour, λ=10\lambda=10. The reported probability of exactly eight arrivals is P(Y=8)=e−10108/8!≈0.113P(Y=8)=e^{-10}10^8/8!\approx0.113. More than ten means P(Y≥11)P(Y\ge11), reported as 0.4170.417. In 30 minutes, λ=5\lambda=5; exactly four arrivals has probability e−554/4!≈0.175e^{-5}5^4/4!\approx0.175. At most 10 in 30 minutes is P(Y≤10)=0.986P(Y\le10)=0.986.

Social-media likes. Average weekly likes are 14, so for a week λ=14\lambda=14. At least 10 is reported as P(Y≥10)=0.891P(Y\ge10)=0.891. The inclusive range 10–20 sums the eleven PMF terms and is reported as 0.8430.843. A day has mean 14/7=214/7=2: at least five likes is P(Y≥5)=0.053P(Y\ge5)=0.053, and zero likes is P(Y=0)=e−2≈0.135P(Y=0)=e^{-2}\approx0.135.

Call-center arrivals and a break. Six calls per 15 minutes gives a rate of 6/15=0.46/15=0.4 calls per minute. In five minutes, λ=2\lambda=2, so exactly three calls is reported as 0.180.18. In 15 minutes, λ=6\lambda=6, and exactly 10 calls is reported as 0.0410.041. In a three-minute break, λ=6(3/15)=1.2\lambda=6(3/15)=1.2; the probability of no interruption is P(Y=0)=e−1.2≈0.301P(Y=0)=e^{-1.2}\approx0.301.

Space-based examples and interpreting a fixed range

Mall entry window. A rate of 180 visitors per hour gives a ten-minute mean of 180(10/60)=30180(10/60)=30. The probability of 10–20 arrivals inclusive is reported as 0.5260.526. At rates of 150 and 210 per hour, the ten-minute means are 25 and 35; the same range has reported probabilities 0.7300.730 and 0.2250.225. The probability falls as the mean moves to the right, away from the fixed 10–20 window. It is the location of the window relative to the distribution—not simply whether the rate rose or fell—that explains the result.

Road defects. At three defects per kilometre, a one-kilometre stretch has λ=3\lambda=3. At least three defects is reported as P(Y≥3)=0.577P(Y\ge3)=0.577; at most five is P(Y≤5)=0.916P(Y\le5)=0.916. An eight-kilometre stretch has mean 8(3)=248(3)=24. Thus at most 40 defects is P(Y≤40)=0.999P(Y\le40)=0.999, and at least 20 is P(Y≥20)=0.82P(Y\ge20)=0.82. Raising the one-kilometre probability P(Y≤5)P(Y\le5) to the fifth power would impose the much narrower condition that every one of five sections contain at most five defects; that is not equivalent to the total across eight kilometres being at most 40.

Flags across conversations. With an average of five flags per conversation, λ=5\lambda=5: 2–8 flags inclusive is reported as 0.8910.891, and “more than 10” means P(Y≥11)P(Y\ge11); the source’s table lookup of 0.0320.032 corresponds to P(Y≥10)P(Y\ge10), so its event threshold and reported lookup do not agree. For two conversations the mean is 1010: 4–16 inclusive is reported as 0.9630.963. Fewer than four means P(Y≤3)=0.010P(Y\le3)=0.010, not P(Y≤4)P(Y\le4).

Key takeaways

  • A Poisson model counts occurrences over a defined time or space interval.
  • The parameter is the interval’s expected count; multiply rate by interval length to rescale it.
  • The mean and variance equal λ\lambda; the standard deviation is λ\sqrt{\lambda}.
  • Use CDFs, tails, or sums of PMF terms according to exact event wording.

Additional worked applications: risk, portfolios, and sampling

Insurance claims: expected cost, pricing, and risk protection

An insurer’s claim payment XX is discrete, with values described as ₹0, ₹500, ₹1,000, ₹2,500, ₹5,000, ₹7,500, and ₹10,000. The listed probabilities are said to sum to one. The individual probabilities for several amounts are not all reproduced in the source; it reports the resulting expectation as ₹1,235. The expectation calculation is the weighted sum:

E[X]=∑xxP(X=x)=0(0.60)+500(0.15)+1000(0.06)+⋯+10000(0.04)=Rs. 1,235E[X]=\sum_x xP(X=x)=0(0.60)+500(0.15)+1000(0.06)+\cdots+10000(0.04)=\text{Rs. }1{,}235

For variance, square each deviation from ₹1,235, weight by its probability, and sum. The reported result and standard deviation are:

Var⁡(X)=∑x(x−1235)2P(X=x)=6,447,275,σX=6,447,275≈Rs. 2,539.15\operatorname{Var}(X)=\sum_x(x-1235)^2P(X=x)=6{,}447{,}275,\qquad \sigma_X=\sqrt{6{,}447{,}275}\approx\text{Rs. }2{,}539.15

The reported break-even premium is the expected claim payment, ₹1,235, before any other costs are considered. At an annual premium of ₹3,000, the policyholder’s expected net value, defined here as expected claims minus premium, is:

Rs. 1,235−Rs. 3,000=−Rs. 1,765\text{Rs. }1{,}235-\text{Rs. }3{,}000=-\text{Rs. }1{,}765

The policy can still be valuable as protection against a large repair bill: expected value alone does not capture the value of avoiding a severe loss.

Two-stock portfolio: covariance and diversification

Let XX and YY be returns on stocks A and B. Their expected returns are 6% and 5%; standard deviations are 4% and 2%. The stated covariance is −3-3 in the source’s percentage-point-squared calculation. Correlation standardizes covariance:

ρXY=Cov⁡(X,Y)σXσY=−3(4)(2)=−0.375\rho_{XY}=\frac{\operatorname{Cov}(X,Y)}{\sigma_X\sigma_Y}=\frac{-3}{(4)(2)}=-0.375

This negative relationship means the returns tend to move in opposite directions, which can reduce portfolio variability. For an investment of ₹5,000 in stock A alone, expected gain is Rs. 5,000(0.06)=Rs. 300\text{Rs. }5{,}000(0.06)=\text{Rs. }300, expected ending value is ₹5,300, and the standard deviation of the investment value is Rs. 5,000(0.04)=Rs. 200\text{Rs. }5{,}000(0.04)=\text{Rs. }200.

For portfolio weights wA,wBw_A,w_B, the mean is the weighted mean and variance includes the covariance term:

E[Z]=wAE[X]+wBE[Y]E[Z]=w_AE[X]+w_BE[Y]

Var⁡(Z)=wA2Var⁡(X)+wB2Var⁡(Y)+2wAwBCov⁡(X,Y)\operatorname{Var}(Z)=w_A^2\operatorname{Var}(X)+w_B^2\operatorname{Var}(Y)+2w_Aw_B\operatorname{Cov}(X,Y)

Equal investment. With wA=wB=0.5w_A=w_B=0.5, expected return is:

E[Z]=0.5(6%)+0.5(5%)=5.5%E[Z]=0.5(6\%)+0.5(5\%)=5.5\%

Using the source’s squared percentage-point units, Var⁡(X)=42=16\operatorname{Var}(X)=4^2=16, Var⁡(Y)=22=4\operatorname{Var}(Y)=2^2=4, and covariance =−3=-3:

Var⁡(Z)=(0.5)2(16)+(0.5)2(4)+2(0.5)(0.5)(−3)=4+1−1.5=3.5\operatorname{Var}(Z)=(0.5)^2(16)+(0.5)^2(4)+2(0.5)(0.5)(-3)=4+1-1.5=3.5

σZ=3.5≈1.87%\sigma_Z=\sqrt{3.5}\approx1.87\%

The reported equal-weight portfolio has expected return 5.5% and standard deviation 1.87%, below either individual stock’s standard deviation.

70% A and 30% B.

E[Z]=0.7(6%)+0.3(5%)=5.7%E[Z]=0.7(6\%)+0.3(5\%)=5.7\%

Var⁡(Z)=(0.7)2(16)+(0.3)2(4)+2(0.7)(0.3)(−3)=7.84+0.36−1.26=6.94\operatorname{Var}(Z)=(0.7)^2(16)+(0.3)^2(4)+2(0.7)(0.3)(-3)=7.84+0.36-1.26=6.94

σZ=6.94≈2.63%\sigma_Z=\sqrt{6.94}\approx2.63\%

This portfolio has greater expected return and risk than the equal-weight portfolio; the choice depends on risk appetite. The covariance units are not explicitly stated, so the calculations above retain the source’s convention of percentage-point units.

Retirement savings: binomial probabilities from a CDF table

Suppose each randomly selected adult independently has probability p=0.3p=0.3 of saving nothing for retirement. With n=15n=15, let success mean “saves nothing.” An exact probability can be recovered from a cumulative table by subtraction:

P(Y=4)=P(Y≤4)−P(Y≤3)P(Y=4)=P(Y\le4)-P(Y\le3)

The transcript instead reports subtracting the cumulative value at four from that at five, with values 0.72160.7216 and 0.51550.5155, and gives P(Y=4)=0.2186P(Y=4)=0.2186. The displayed subtraction is 0.7216−0.5155=0.20610.7216-0.5155=0.2061, so the source’s stated operands and result do not agree; retain this as an unresolved table-index or transcription error. The other reported results are P(Y≤2)=0.1286P(Y\le2)=0.1286 and:

P(Y≥8)=1−P(Y≤7)=1−0.95=0.05P(Y\ge8)=1-P(Y\le7)=1-0.95=0.05

Stock-market participation by age group

The source gives participation probabilities of 0.30.3 for ages 18–29 and 0.60.6 for ages 50–64. For a sample of 25 people aged 18–29, the expected number investing is:

E[Y]=np=25(0.3)=7.5E[Y]=np=25(0.3)=7.5

The reported probability that at most 10 invest is P(Y≤10)=0.9022P(Y\le10)=0.9022. For 15 people aged 50–64:

E[Y]=15(0.6)=9E[Y]=15(0.6)=9

The reported probability at least 10 invest is 1−P(Y≤9)=1−0.5968=0.40321-P(Y\le9)=1-0.5968=0.4032.

Bank arrivals: Poisson counts over longer windows

With an average of two customers per minute, the one-minute count has λ=2\lambda=2. The reported probability of at most two arrivals is P(Y≤2)=0.6767P(Y\le2)=0.6767. Over two minutes, λ=2(2)=4\lambda=2(2)=4; at least four arrivals is:

P(Y≥4)=1−P(Y≤3)=1−0.4335=0.5665P(Y\ge4)=1-P(Y\le3)=1-0.4335=0.5665

Over a five-minute break, λ=2(5)=10\lambda=2(5)=10. Exactly ten arrivals is obtained by differencing adjacent CDF values:

P(Y=10)=P(Y≤10)−P(Y≤9)=0.6968−0.583≈0.1138P(Y=10)=P(Y\le10)-P(Y\le9)=0.6968-0.583\approx0.1138

The transcript reports 0.12510.1251 for this calculation, which does not equal the subtraction of its stated values; its CDF operand or result is unclear.

Key takeaways

  • Discrete distributions support expected cost and risk calculations as well as event counts.
  • Portfolio variance depends on covariance; negative covariance can reduce combined risk.
  • CDF tables answer exact-value questions by differencing adjacent cumulative probabilities.
  • Check arithmetic and table indices when reported values conflict with their stated operands.

Putting the three distributions together

Model What is counted Parameters Mean Variance
Bernoulli Success in one trial pp pp p(1−p)p(1-p)
Binomial Successes in nn trials n,pn,p npnp np(1−p)np(1-p)
Poisson Occurrences in an interval λ\lambda λ\lambda λ\lambda

The Poisson and binomial PMFs, means, and variances connect when pp is small and nn is large enough that npnp remains the relevant average: then 1−p1-p is near one, making binomial mean and variance both close to npnp. This explains why the distributions can look similar in that setting. Discrete distributions describe countable values; continuous random variables can also take fractional values and require adjustments to the probability analysis.

Key takeaways

  • Bernoulli is one two-outcome trial; binomial aggregates a fixed number of such trials.
  • Poisson models event counts over an interval with a rate-based expected count.
  • Always identify the modeled quantity, interval or trial count, parameters, and requested event before calculating.
Revise before your examAll 1 exam tip from this module in one list
  1. Write down the unit of a trial, the precise success event, nn, pp, and the target event before opening a table. This prevents confusing “at least” with “more than,” or a count of successes with the number of trials.

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.