5.3. The Poisson distribution. Introduction. Prerequisites. Learning Outcomes. Learning Style

Transcription

1 The Poisson distribution 5.3 Introduction In this block we introduce a probability model which can be used when the outcome of an experiment is a random variable taking on positive integer values and where the only information available is a measurement of its average value. This has widespread applications in analysing traffic flow, in fault prediction on electric cables, in the prediction of randomly occurring accidents etc. Prerequisites Before starting this Block you should... Learning Outcomes After completing this Block you should be able to... recognise and use the formula for probabilities calculated from the Poisson model 1 understand the concepts of probability and probability distributions. Learning Style To achieve what is expected of you... allocate sufficient study time use the recurrence relation to generate a succession of probabilities use the Poisson model to obtain approximate values for binomial probabilities briefly revise the prerequisite material attempt every guided exercise and most of the other exercises

2 1. The Poisson distribution Suppose that it has been observed that, on average, 180 cars per hour pass a specified point on a particular road in the morning rush hour. Due to impending road works it is estimated that congestion will occur closer to the city centre if more than 5 cars pass the point in any one minute. What is the probability of this latter event occurring? We cannot use the binomial model here since there is no fixed number of cars which will pass the point and we have no estimate of p. We must use a different probability model - the Poisson distribution - to handle situations where the only information is an average rate. If the occurrences of an event are relatively rare, we may assume that the number of occurrences in any small interval (of time or space as appropriate) is independent of the number of occurrences in any other small interval. Under these assumptions the Poisson model can be developed. (We omit the derivation). If X is the random variable Key Point The Poisson Probabilities number of occurrences in a given interval for which the average rate of occurrences is λ then, according to the Poisson model, the probability of r occurrences in that interval is given by λ λr P (X = r) =e r! Now do this exercise Using the Poisson distribution write down the formulae for P (X =0),P(X =1),P(X = 2), P(X =6), noting that 0! = 1. Answer We have calculated P (X =0)toP (X = 5) when λ = 2 and presented the results to 4d.p. in Table 1. r P (X = r) Table 1 Notice how the values for P (X = r) increase and then decrease relatively rapidly (due to the decreasing exponential term e λ ). In this example two of the probabilities are equal and this will always be the case when λ is an integer. We need not calculate each probability directly. We can use the following relations (which can be checked from the formulae for P (X = r)): P (X =1)= λ 1 P (X =0), P(X =2)=λ 2 P (X =1) P (X =3)=λ 3 P (X =2), etc. In general, for ease of calculation we use the recurrence relationship P (X = r) = λ r P (X = r 1) for r 1. Engineering Mathematics: Open Learning Unit Level 1 2

3 Now do this exercise Returning to the car example, referred to at the beginning of this section, let X be the random variable number of cars arriving in any minute. Calculate the probability that more than 5 cars arrive in one minute. First we convert the information on the average rate (cars per hour) to obtain a value for λ (cars per minute). Then we calculate P (X =0)toP (X = 5) and finally the required probability. This is best done by noting that, Thus P (more than 5 cars in one minute) = 1 P (5 cars or less arrive in one minute) P (X >5) = 1 P (X 5) = 1 P (X =0) P (X =1) P (X =2) P (X =3) P (X =4) P (X =5) Each probability can be calculated by the recurrence relation above, starting from P (X = 0). Calculate the individual probabilities to 5 d.p. and the final number to 4 d.p. Answer Relation to the binomial distribution It can be shown that the Poisson distribution may be obtained from the binomial distribution by taking a particular limit. When the number of trials n in the binomial model is increased without bound and p (the probability of a success on an individual trial) tends to zero in such a way that np remains at a value (= λ) say then the binomial model becomes more like the Poisson model introduced here. We shall see later that the Poisson model can then be used as a convenient way of approximating the probabilities arising in the binomial model. Note that if X is a random variable whose probabilities are given by the Poisson model then, because of its relation to the binomial distribution (in which E(X) =np = λ and V (X) = npq = λ(1 p), we find: E(X) =λ and V (X) =λ Clearly, then, the standard deviation is λ. 2. Poisson approximation to the binomial distribution We have already noted that the Poisson model is obtained by taking the special limit (p 0, n ) of the binomial distribution and putting λ = np. Remember that if X is a random variable following a binomial distribution then E(X) =np and V (X) =npq. If p is quite small then q 1 and npq np. In some problems for which the binomial model is appropriate, the calculation of the (binomial) probabilities can be tedious and inaccurate. Since V (X) E(X) it is possible to use a Poisson model to generate probabilities approximately equal to their binomial counterparts. The decision is either to use an accurate model with tedious (and probably inaccurate) calculations or to use an approximate model with more easily-calculated probabilities which will be approximations only. We cannot give a hard-and-fast rule as to how 3 Engineering Mathematics: Open Learning Unit Level 1

4 small p should be for the Poisson approximation to be reasonable; we can simply say the smaller the better so long as np remains of a sensible size. As an example, suppose mass-produced items are packed in boxes of It is believed that 1 item in 2000 on average is substandard. What is the probability that a box contains more than 2 defectives? The correct model is the binomial distribution with n = 1000, p= (and q = ). Now do this exercise Using the binomial distribution write down P (X =0),P(X = 1) and calculate P (more than 2 defectives). Hint: don t evaluate the final probability until you have a compact expression for it. Answer Now do this exercise Now choose a suitable value for λ in order to use a Poisson model to approximate the probabilities. Answer Now do this exercise Now recalculate the probability that there are more than 2 defectives using the Poisson distribution with λ = 1 2. Answer We have obtained the same answer to 4 d.p., as the exact binomial calculation, essentially because p was so small. We shall not always be so lucky. Engineering Mathematics: Open Learning Unit Level 1 4

5 More exercises for you to try 1. Large sheets of metal have faults in random positions but on average have 1 fault per 10m 2. What is the probability that a sheet 5m 8m will have at most one fault. 2. If 250 litres of water are known to be polluted with 10 6 bacteria what is the probability that a sample of 1cc of the water contains no bacteria? 3. Suppose vehicles arrive at a signalised road intersection at an average rate of 360 per hour and the cycle of the traffic light is set at 40 seconds. In what percentage of cycles will the number of vehicles arriving be (a) exactly 5, (b) less than 5? If, after the lights change to green, there is time to clear only 5 vehicles before the signal changes to red again, what is the probability that waiting vehicles are not cleared in one cycle? 4. Previous results indicate that 1 in 1000 transistors are defective on average (a) Find the probability that there are 4 defective transistors in a batch of (b) What is the largest number, N, of transistors that can be put in a box so that the probability of no defectives is at least 1/2. 5. A manufacturer sells a certain article in batches of By agreement with a customer the following method of inspection is adopted: A sample of 100 items is drawn at random from each batch and inspected. If the sample contains 4 or fewer defective items, then the batch is accepted by the customer. If more than 4 defectives are found, every item in the batch is inspected. If inspection costs are 75p per hundred articles, and the manufacturer normally produces 2% of defective articles, find the average inspection costs per batch. Answer 5 Engineering Mathematics: Open Learning Unit Level 1

6 End of Block 5.3 Engineering Mathematics: Open Learning Unit Level 1 6

7 P (X =0) = e λ λ0 0! = e λ 1 1 e λ P (X =1)=e λ λ 1! = λe λ P (X =2) = e λ λ2 2! = λ2 2 e λ P (X =6)=e λ λ6 6! = λ6 720 e λ Back to the theory 7 Engineering Mathematics: Open Learning Unit Level 1

8 λ = 3 cars/minute. We present the results in a table. First, P (X =0)=e 3 r Sum P (X = r) Then P (more than 5) = = = (4 d.p). Back to the theory Engineering Mathematics: Open Learning Unit Level 1 8

9 ( ) P (X =0) = 2000 ( ) P (X = 1) = ( ) 1 = ( ) P (X =0)+P (X =1)= ( ) 999 ( ) = 2 Hence P (more than 2 defectives) = Back to the theory ( ) (4d.p.) Engineering Mathematics: Open Learning Unit Level 1

10 λ = np = = 1 2 Back to the theory Engineering Mathematics: Open Learning Unit Level 1 10

11 P (X =0)=e 1 2, P(X =1)= e 2 P (X =0)+P (X =1)= 3 2 e 1 2 = (4 d.p.) Hence P (more than 2 defectives) = Back to the theory 11 Engineering Mathematics: Open Learning Unit Level 1

12 1. Poisson Process. In a sheet size 40m 2 we expect 4 facults λ =4 P (X = r) =λ r e λ /r! P (X 1) = P (X =0)+P (X =1)=e 4 +4e 4 = In 1cc we expect 4 bacteria(= 10 6 /250000) λ =4 P (X =0)=e 4 = In 40 seconds we expect 4 vehicles λ =4 (a) P (exactly 5) = λ 5 e λ /5!= i.e. in 15.6% of cycles [ ] (b) P (less than 5) = e λ 1+λ 4 + λ2 2! + λ3 3! + λ4 4! [ = e ] = Vehicles will not be cleared if more than 5 are waiting. P (greater than 5) = 1 P (exactly 5) P (less than 5) = = (a) Poisson approximation to Binomial 1 λ = np = =2 P (X =4)=λ 4 e λ /4!=16e 2 /24 = (b) λ = Np = N/1000; P (X =0)= λ0 e λ 0! = e λ = e N/1000 e N/1000 N = = ln(0.5) N = choose N = 693 or less. Engineering Mathematics: Open Learning Unit Level 1 12

13 5. P (defective) = Poisson approx. to Binomial λ = np = 100(0.02)=2 P (4 or fewer defectives in sample of 100) = P (X =0)+P (X =1)+P (X =2)+P (X =3)+P (X =4) = e 2 +2e e ! e ! e 2 = Inspection costs Cost c P (X = c) E(Cost) = 75( ) (0.0526) = 268.5p Back to the theory 13 Engineering Mathematics: Open Learning Unit Level 1