3 Pages

PoissonAndExponential

Course Number: MATH 218, Spring 2000

College/University: USC

Word Count: 903

Rating:

Document Preview

Poisson and Exponential, Connected The Poisson distribution counts the number of discrete events in a fixed time period; it is closely connected to the exponential distribution, which (among other applications) measures the time between arrivals of the events. The Poisson distribution is a discrete distribution; the random variable can only take nonnegative integer values. The exponential distribution can take any...

Unformatted Document Excerpt
Coursehero >> California >> USC >> MATH 218

Course Hero has millions of student submitted documents similar to the one
below including study guides, practice problems, reference materials, practice exams, textbook help and tutor support.

Course Hero has millions of student submitted documents similar to the one below including study guides, practice problems, reference materials, practice exams, textbook help and tutor support.

and Poisson Exponential, Connected The Poisson distribution counts the number of discrete events in a fixed time period; it is closely connected to the exponential distribution, which (among other applications) measures the time between arrivals of the events. The Poisson distribution is a discrete distribution; the random variable can only take nonnegative integer values. The exponential distribution can take any (nonnegative) real value. The Poisson process is completely described by a parameter , the average number of arrivals per unit time. If X denotes the number of events which happen in unit time, then k - e . k! This is connected to the random variable Y , the time between successive events, by the fact that Y is exponentially distributed with density function e-x , and therefore P (X = k) = t (1) (2) P (Y t) = 0 e-x dx = 1 - e-t , P (Y > t) = 1 - P (Y t) = e-t . Sometimes we may wish to speak about the number of arrivals in a time interval T which is different from the unit time interval. This is done by scaling: multiply (average number of arrivals per unit time) by T (number of units of time) to get T , the average number of arrivals in time T . (So if an average of six babies arrive per day in a hospital, then an average of 0.25 babies arrive per hour.) The probability that exactly k events will occur in time interval T is therefore (3) (T )k -T e . k! This makes the connection between the Poisson and exponential distributions transparent: for the event Y > T means that the first arrival takes more than T units of time; which means that there are 0 arrivals in the first T units of time. Taking k = 0 in (3), this has probability e-T , i.e. P (Y > T ) = e-T . This was exactly what we claimed for the exponential distribution in (2). For an example, we will model the arrival of babies at a lying-in hospital. These are Poisson events, because they satisfy the three characteristics of a Poisson model: Arrivals of babies are independent events. (The mothers don't conspire together.) The probability of babies being born almost simultaneously is negligible. The probability of a baby being born in a time interval only depends on the length of the time interval, not on when it is during the day. For many years the hospital has been averaging 6 births per day. That means that the average time between births will be 1/6 of a day. Therefore if X denotes the number of babies born in a day, then 6k P (X = k) = e-6 , k! 1 2 while if Y denotes the time between births, then P (Y > t) = e-6t . The probability density function for Y is 6e-6x . Recall that the exponential which distribution has PDF e-x has mean 1/, so this fits: Y should have mean 1/6 (days/birth). Also note how easy it is to connect these parameters through "dimensional analysis": X, the number of births, is measured by babies/day; Y , the time between births, is measured in days/baby. So 6 babies/day is equivalent to 1/6 days/baby, or 4 hours/baby. For example, in the above scenario, given that a baby has just been born, the probability that the next baby will take one or more hours to arrive is found by converting hours to days, 1 hour = and computng P (Y > Surely time enough for a coffee break! Consistency of the Poisson model. How far can we trust the Poisson model? Actually, pretty far. We'll give an example to illustrate how consistent it is. It illustrates a more general fact (the sum of two independent Poisson RV's with mean and is a Poisson RV with mean + ). This is quite reasonable: think of two hospitals, one giving birth to an average of babies per day, the other giving birth to babies per day. If you combine the hospitals (say, a health conglomerate buys them both) then the conditions for a Poisson RV are satisfied for births at the combined hospitals, which would average + per day.) Consider a hospital which has an average of 6 babies born each day. There are two shifts for the nursing staff, so the administration prefers to use 12-hour periods, with an average of 3 babies born each 12-hour period. Does using different scales like this make a difference in the probabilities? No, if we keep the time intervals strai...