85 Sampling Distribution of 8? 25 formula ‘The probability that a random sample produces a x* value greater than some specified value is equal to the area under the eurve to the right of this value. It is ‘customary to let x2 represent the x? value above which we find an area of a. This is illustrated by the shaded region in Figure 8.7, ° z , Figure 8.7: The chi-squared distribution Table A.5 gives valuos of x2 for various values of «and v. The areas, «, are the column headings the degrees of freedom, w, ae given in the left column and the table entries are the x2 values. Henee, the x? value with 7 degrees of freedom, leaving an area of 0.05 to the right, x25 = 1.067. Owing to lack of symmetry, wwe must also use the tables to find 2.9, = 2.167 for Exactly 95% of a chi-squared distribution lies between x3 on, and x3.oa5- A x? value falling to the right of x3 og, is not likely to occur unless our assumed value of 2? is too small. Similarly, a x* value falling to the let of x2 yy, is unlikely unless ‘our assumed value of 0? is too large. In other words, it is possible to have a x? value to the left of x2.75 OF to the right of x2 oy when 0” is correct, but if this should occur, it is more probable that the assumed value of 0 is in error. Example 8.7: A manufacturer of car batteries guarantees that the batteries will last, on average, 3 years with a standard deviation of 1 year. If ive of these batteries have lifetimes of 19, 24, 80, 3.5, and 4.2 years, should the manufacturer still be convinced that ‘the batteries have a standard deviation of 1 year? Assume that te battery lifetime follows a normal distribution Solution: We first find the sample variance using Theorem 8.1, 2 _ (5)(48.26) ~ (15)? (4) = 0.815. ‘Then (4)(0.815) 7 3.26246 Chapter § Fundamental Sampling Distributions and Data Descriptions is a value from a chi-squared distribution with 4 degrees of freedom. Since 95% of the x? values with 4 dogrees of froedom fall between 0.484 and 11.145, the computed value with ¢? = 1 is reasonable, and therefore the manufactuser has no zeason to suspect that the standard deviation is other than 1 year. 4 Degrees of Freedom as a Measure of Sample Information 8.6 Recall from Corollary 7.1 in Seetion 7.3 that > (X= 0? has a x7-distribution with n degrees of freedom. Note also Theorem 8.4, which indicates that the random variable (wast _ ys =X) ‘has a x?distribution with n—1 degrees of freedom, ‘The reader may also recall that the term degrees of freedom, used in this identical context, is discussed in Chapter 1 As we indicated earlier, the proof of Theorem 8.4 will not be given. However, ‘the reader can view Theorem 8.4 as indicating that when jis not known and one considers the distribution of there is 1 less degree of freedom, or a degree of freedom is lost in the estimation ofp (ie,, when pis replaced by 2). In other words, there are n degrees of fkee- dom, or independent pieces of infermation, in the random samaple from the normal distribution, When the data (the values in the sample) are used to compute the ‘mean, there is 1 less degree of freedom im the information used to estimate 0? t-Distribution In Section 8.4, we discussed the utility of the Central Limit Theorem. Its applica- ‘ious revolve around inferences on a population mean or the difference between two population means, Use of the Central Limit Theorem and the normal distribution is certainly helpful in this context. However, it was assumed that the population standard deviation is known. This assumption may not be unreasonable in situ- ations where the engineer is quite familiar with the system or process. However, in many experimental scenarios, knowledge of o is certainly no more reasonable than knowledge of the population mean ,., Often, in fact, an estimate of o must ‘be supplied by the same sample information that produced the sample average 2. As a result, a natural statistic to consider to deal with inferences on jis86 t-Distribution 27 since $ is the sample analog to 0. If the sample size is small, the values of S? flue- ‘uate considerably from sample to sample (sce Exercise 8.43 on page 259) and the distribution of T deviates appreciably from that of a standard normal distribution. If the sample size is large enough, say n > 30, the distribution of T does not differ considerably from the standazd normal. However, for n < 30, itis useful to deal with the exact distribution of T. In developing the sampling distribution of T, ‘we shall assume that our random sample was selected from a normal population. ‘We can then waite (X=wioiva)_ Vs /V/(n—1)" where hhas a chi-squared distribution with v = n—1 degrees of freedom, In sampling from normal populations, we can show that X and S? are independent, and consequently so are Z and V. The following theorem gives the definition of a random variable T as a function of Z (standard normal) and x, For completeness, the density function of the t-distribution is given, Let Z be a standard normal random variable and V a chi-squared random variable with v degrees of freedom. If Z and V are independent, then the distribution of| the random variable T, where ‘Theorem 8.5: Zz To WR Js given by the density function Tie +1/2) By ene MED Ys crew This is known as the tdistribution with » degrees of freedom. ay From the foregoing and the theorem above we have the following corollary.24s. Chapter § Fundamental Sampling Distributions and Data Descriptions Tet X;,Xz,...,X, be independent random variables that are all nommal with] mean j2 and standard deviation 0. Let Corollary 8.1: ty, mat mi he he ain ase T= if ba atin wh 9 = 1 dee ‘The probability distribution of T was first published in 1908 in a paper written by W. S. Gosset. At the time, Gosset was employed by an Irish brewery that prohibited publication of research by members of its staff. To circumvent this re- striction, he published his work secretly under the name "Student,” Consequently, the distribution of T'is usually called the Student ¢-distribution or simply the distribution, In deriving the equation of this distribution, Gosset assumed that the samples were selected from a normal population. Although this would seem to be a very restrictive assumption, it can be shown that nonnermal populations possess- ing nearly bell-shaped distributions will still provide values of T that approximate the t-distribution very closely. ‘What Does the t-Distribution Look Like? ‘The distribution of Tis similar to the distribution of Z in that they both are symmetric about a mean of zero. Both distributions are bell shaped, but the t- distribution is more variable, owing to the fact that the T-values depend on the fluctuations of two quantities, X and S®, whereas the Z-values depend only on the changes in X from sample to sample. The distribution of 7 differs from that of Z in that the variance of T depends on the sample size n and is always greater than 1. Only when the sample size n > o0 will the two distributions become the same. In Figure 8.8, we show the relationship between a standard normal distribution (v = ce) and t-distributions with 2 and 5 dogrees of freedom. The percentage points of the (distribution are given in Table A Figure 8.8: The (distribution curves for v = 2,5, Figure 8.9: Symmetry property (about 0) of the and 2. t-distribution,86 t-Distribution 249 It is customary to let fq, represent the ¢-value above which we find an area equal toa. Hence, the t-value with 10 degrees of freedora leaving an azea of 0.025 to the right is ¢= 2.298, Since the t-distzibution is symmetric about a mean of zero, we have (4 = ~ta; that is, the C-value leaving an area of 1 ~ a to the right and ‘therefore an avea of a to the left is equal to the negative t-value that leaves an area, of ein the right tail of the distuibution (sce Figure 8.9). That is, toes = —toos, too = —to.1, and so forth. Example 8.8: The l-value with v = 14 degrees of freedom that leaves an area of 0.025 to the left, and therefore an area of 0.975 to the right, is toons = —to.oas = ~2.148. 4 Example 8.9: Find P(—too2s <7 < toon): Solution: Since tos leaves an area of 0.05 to the right, and —to o2s leaves an area of 0.025 to the left, we find a total area of 1 0.085 ~ 0.025 = 0.925 ‘between —ty oar and ty os. Hence P(to gas <7 < tos) = 0.925, 4 Example 8.10: Find k such that P(k <7 < -1.761) selected from a normal distribution and =. 0,045 for a random sample of size 15 Teo ° Figure 8.10: The t-values for Example 8.10. Solution: From Table A.A we note that 1.761 corresponds to ty o5 when v = 14. Therefore, “too; = 1761. Since k in the original probability statement is to the left of ~toor = -1.761, lot k= —ta. Thon, from Figure 8.10, we have 0.045 = 0.05 — a, or a = 0.005, Hence, from Table Ad with » = 14, k= ~lo.905 = ~2.977 and P(-2.977 < T < -L.761) = 0.045. 4250 Chapter § Fundamental Sampling Distributions and Data Descriptions Exactly 95% of the values of a t-distribution with v = n—1 degrees of freedom, lie between ~fo.o25 and fo.025. Of course, there are other values that contain 95% of the distribution, such as —to.2 and to, but these values do not appear in Table A.A, and furthermore, the shortest possible interval is obtained by choosing t-values ‘that leave exactly the same area in the two tails of our distribution. A t-value that falls below —to 925 oF above to 25 would tend to make us believe either that a very rare event has taken place or that our assumption about jis in error. Should this hhappen, we shall make the the decision that our assumed value of jr is in error In fact, a f-value falling below —to o: or above ty o1 would provide even stronger evidence that our assumed value of 1 is quite unlikely. General procedures for testing claims concerning the value of the parameter will be treated in Chapter 10, A preliminary look into the foundation of these procedue is illustrated by the following example. Hxample 8.11: A chemical engineer claims that the population mean yield of a certain batch ‘process is 500 grams per milliliter of raw material. To check this claim he samples 25 batches each month. If the computed (-value falls between —fo.05 and lo 05, he js satisfied with this claim. What conclusion should he draw from a sample that hhas a mean 2 = 518 grams per milliliter and a sample standard deviation s = 40 grams? Assume the distribution of yields to be approximately normal. Solution: From Table AA we find that (95 = 1.711 for 24 dogrees of freedom. Therefore, the ‘engineer can be satisfied with his claim if a sample of 25 batches yields a t-value ‘between —1711 and 1.711. If y= 500, then 18 ~ 500 0/85 1 value well above 1.711. The probability of obtaining a évalue, with v = 24, equal to or greater than 2.25 is approximately 0.02. If w > 500, the value of ¢ computed from the sample is more reasonable. Hence, the engineer is likely to conclude that the process produces a better product than he thought. 4 = 225, ‘What Is the tDistribution Used For? ‘The L-distribution is used extensively in problems that deal with inference about ‘the population mean (as illustrated in Example 8.11) or in problems that involve comparative samples (i., in cases where one is trying to determine if means from ‘two samples are significantly different). ‘The use of the distribution will be extended, im Chapters 9, 10, 11, and 12, The reader should note that use of the ¢-distribution for the statistic x Sivn roquires that Xy,X2,...,Xp be normal. The use of the (distribution and the Theorem. ‘The use sample size consideration do not relate to the Central Limit, of the standard normal distribution rather than for n> 30 merely implies that S is a sufficiently good estimator of & in this case. In chapters that follow the Ledistribution finds extonsive usage.87 P-Distribution 251 8.7 F-Distribution We have motivated the t-distribution in part by its application to problems in which there is comparative sampling (i.e., a comparison between two sample means). For example, some of our examples in future chapters will take a more formal approach, chemical engineer collects data on two catalysts, biologist collects data on two growth media, or chemist gathers data on two methods of coating material to inhibit corrosion. While it is of interest to let sample information shed light ‘on two population means, it is often the ease that a comparison of variability is ‘equally important, if not more so. ‘The F-distribution finds enormous application in comparing sample variances. Applications of the F-distribution are found in problems involving two or more samples. ‘The statistic F is defined to be the ratio of two independent chi-squared random. variables, each divided by its number of degrees of freedom. Hence, we can write _ Ul ~ Vin" U and V are independent random variables having chi-squared distributions ith v1 and vz degrees of freedom, respectively. We shall now state the sampling distribution of F. Tet U and V be two independent random variables having chi-squared distributions} with vy and v2 degrees of freedom, respectively. ‘Then the distribution of the random variable F ‘Theorem 8.6: 7}e ia given by the density function Nlestuyaie ya? por. ap) = (RE — Terra I> fo This is known as the Fidistribution with vy and vy degrees of freedom (4.6). We will make considerable use of the random variable F in future chapters. Ho- ‘ever, the density function will not be used and is given only for completeness. The curve of the F-distribution depends not only on the two parameters v and vs but also on the order in which we state them. Once these two values aze given, we can identify the curve. Typical F-distributions are shown in Figure 8.11 Let fo be the f-value above which we find an arca equal to a. This is illustrated by the shaded region in Figure 8.12. Table A.6 gives values of fa only for a = 0.05 and a = 0.01 for various combinations of the degrees of freedom and v2. Hence, the f-value with 6 and 10 dogrecs of freedom, leaving an area of 0.05 to the right, is foos = 3.22. By means of the folowing theorem, Table A.6 can also be used to find values of foe and fo99. The proof islet for the reader