Scribd
Upload
60 views18 pages

SSCH 4

Uploaded by

yogi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
60 views18 pages

SSCH 4

Uploaded by

yogi
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 18
43 80 Chapter Simple Random Sampling raRLE 4.1 Patient record tobe included inthe samp 104 289 510 as 23 098 ow 103 s2t on 70 Estimation of a Population Mean and Total ‘We have stated previously thatthe objective of survey sampling is to draw inferences bout a population from information contained in a sample. One way to make infer. fences is to estimate certain population parameters by using the sample information, The objective of a sample survey is often to estimate a population mean, den: by 1, ora population (otal, denoted by 7. Thus, the auditor in Example 4.1 might be interested in the mean dollar value for the accounts receivable or the total dollar famount in these accounts. Hence, we consider the estimation of the two population meters, gt and 7, in this section, ‘Suppose that a simple random sample of n accounts is drawn, and we are to the mean value per account for the total population of hospital records. Intuitively, we employ the sample average by 5-2 to estimate wu. This estimator can be justified as being an appropriate one by referring to the discussion and example in Section 3.3. Simple random sampling is equivalent torandom sampling without replacement and with equal weights, as described there. For such a sampling scheme 77,= n/N and the unbiased estimator of the population total, 7, is given by Because the population mean is related to the total by the equation 7/N ple mean will be an unbiased estimator of the population mean. That is, the sam- BQ) = # Ofcourse, a single value of F tells us very litle about the population mean 4, tess we are able to evaluate the goodness of our estimator. Hence, in addition to est mating 2, we would like to place a bound on the error of estimation, To accomplish -aw inferences to make infer- sle information, mean, denoted dle 4.1 might be the total dollar two population and we are to \ospital records. one by referring ing is equivalent described there of the population JN= the sam- tis, ition mean #, un 1 addition to est 43 Evimation of a Population Mean and Total 81 this we need the variance of the estimator; fora simple random samp out replacement from a population of size N, vg) =£ (44) Considering the sample variance ‘hosen with- it ean be shown (see Appendix A, Derivations) that »_ Ny A) = se so that V (3) can be unbiasedly estimated from the sample by wn Sao) = (7) +2(Z) #2(4) «4(2) = (S)av2ea44 dan) =20 random sampling of @ finite population, n Probability of| sample, equal weights vo) 6 6 0.105 6 8 0500 6 10 Ls v6 10 0.125 V6 2 0500 6 4 0.125 ter-4 Simple Random Sampling Lo - HP) se-as9(8)+0-2 reanas(t) | Because each of these sample means can occur with probability 1/6, we can comPute E(5)and V(3). From our definition of expected valve EG) = DyO) ~us(2) sen(2)+e9(2) +2960) von) 09(2) = 250 =H (summed over all values of Y) | and VG) = EG ~ Wy as=259(!) = 20- 2s7(2) +-257(6) es(t)-§ Recalling that for this example n) - sio4=2) 2 \a=1/ “iG)oir ‘Considering the sample variances, we have (05 +20 +454 05+ N JA 105 + L125 + 0.125 + 05 + 0.125) 6 Also, pute 43 Estimation of a Population Mean and Total 83 Thus, we have demonstrated that a(t) and that (5) is an unbiased estimator of V(). The key results of this section are summarized next. Estimator of the population mean je: an (42) 43) ‘The quantity 1 ~n/N is called the finite population correction (pe). Note that this correction factor differs slightly from the one encountered in the true variance of y ‘When tematins smal relative to the population size the fp i elose to unity, Pract, cally speaking, the fpc can be ignored if | ~ n/N =.95, or equivalently, n = (1/20)N. In that case, the estimated variance of y is the more familiar quantity #/n. In many ceases, the population size is not clearly defined or is unknown. Suppose very small laboratory specimens are selected from a large bulk tank of raw sugar in order to ‘measure pure sugar content, How NV will be determined is unclear, but it ean gener- ally be assumed to be quite large, Hence, the fpe can be ignored. Ifa sample of vot- sis selected from the population of a state, to obtain a precise NV for that point ia time is generally impossible, Again, Wis assumed large andthe fpe is ignored Some texts present the fpe as (VW — n)/N; we prefer I — n/N because it highlights the role ofthe sampling fraction n/N. The sampling fraction is often denoted by £=n/N, in which ease the fpe can be represented as | — f. Later inthis book, when the formulas get more complicated, and brevity of formulas therefore more attractive, ‘we will switch to this latter notation {mn theory, ifa two-standard deviation bound on the error (often called a margin of crvor) is subtracted from and added to the sample mean, the resulting confidence i terval has approximately a 95% chance of capturing the population mean within its boundaries. This result is built on a theory that requires the sample mean in question {0 have approximately a normal distribution. To illustrate how this works we return to the brain weight data in Seton 3.4 Figure 3.4 shows tha, or data on the original Seale, the sampling distribution for the mean of samples of size 5is highly skewed. Figure 3.6 shows that, for data on the logarithmic seale, the sampling distribution of ‘ample means is quite normal looking, How is this behavior of sampling distributions imple Random Sampling FIG 4a Confidence intervals for samp of brain weights, = 5 GURE Confidence intervals for samples of log brain weights, = Sample aurber Iwerval reflected in the performance of confidence intervals? Figure 4.1 shows 50 confidence intervals constructed from ranclom samples of size 5 with a two-standard deviation tround on the error using the original population data for brain weights. Only 28 of tae inervaly (56%) caver the true population mean of 394.5; many of the intervals are too short and lie too far tothe left Using the same method on the log-transformed data results in the intervals in Figure 42. Here, 48 of the calculated intervals (96%) cover the population mean @) SOLUTION BXAMPLE 4.3 @) sows 50 confident standard deviation eights. Only 28 of ny of the intervals sin the imervals if xe population meat 43 Esty won of ¢ Population Mean and Total 85 of 2.98. Quite a difference! Not only that the intervals in Figure 4.2 are also more “uniform in length. The message to be learned is thatthe results of this section will not work well unless there is reasonable assurance that the sample means being studied have sampling distributions that are not too far from normal. Of course, that can be ‘essentially guaranteed ifthe sample size is large. You can study for yourself sampling the original and the log-transformed population using a sampling tool found in elec- tronic Section 4.0. Bootstrapping (Section 11.9) is another option; look in electronic Section 11.0 for a single-sample bootstrapping tool. Refer to the hospital audit in Example 4.1 and suppose that a random sample of 200 accounts is selected from the total of N'= 1000, The sample mean of the accounts is found to be 5 = $94.22, and the sample vatiance is s* = 445.21. Esti- imate j, the average due for all 1000 hospital accounts, and place a bound on the error of estimation. We use ¥ = $94.22 to estimate . A bound on the error of estimation can be found by using Ea. (4.3): = f,_ a\st 200 \ 445.21 =al(- I Es ») ay( N ) nV ( 1000) 200) = 2VTTROR = $2.67 Thus, we estimate the mean value per account, 1, to be ¥ = $94.22. Because m is large, the sample mean should possess approximately a normal distribution, so that $94.22 + $2.67 is approximately a 95% confidence interval for the population mean. [A simple random sample of n= 9 hospital records is drawn to estimate the average ‘amount of money due on N= 484 open accounts. The sample values for these nine records are listed in Table 4.3, Estimate, the average amount outstanding, and place 1 bound on your error of estimation. TABLE 4.3 Amount of money owed » 33.50 % 32.00 % 52.00 % 8.00 % 40.00 Ys 41.00 ” 45.00 Ys 4250 » 39.00 SOLUTION 86 Chapter 4 Simple Random Sampling FIGURE 4.3 Box plot ofthe hospital account data [As good data analysis suggests, itis always wise to make an appropriate plot of to see if anything unusual appears. Figure 4.3 shows the box plot of these data, Which suggest that two of the observations are a bit unusual, compared to the rest, one on the high side and one on the low side, Perhaps the investigator should Took at these two accounts again to see if they represent something different from the rest of the accounts in the sample. Our estimate of pis 368.00 SO — 4089 To find a bound on error of estimation, we must compute 5 = 3567 and use Eq, (4.3) to obtain the bound on the error of estimation: w= 8)e- V0 ia) ao = 2V/3800 = 3.944 = $3.94 To summarize, the estimate of the mean amount of money owed per account, #4 is ¥ = $40.89. Although we cannot be certain how close y iso 1, we are reasonably ‘confident that the error of estimation is less than $3.94. ‘As we have already seen, many sample surveys are conducted to obtain information about a population total. The federal auditor in Example 4.1 would probably be inter- «sted in verifying the computer figure for the total accounts receivable (in dollars) forthe N= 1000 open accounts, The population toal is denoted by the symbol r. Because Nor iate plot of the of these data, red to the rest, should look at rom the rest of a ) sr account, H » are reasonably stain information robably be inter- in dollars) forthe ‘17. Because | EXAMPLE 44 © SOLUTION 43 Buimation ofa Population Mean and Total 87 Wwe know that the estimator of ris times the estimator of y. It is also true that the margin of error for estimating a total is N times the margin of error for estimating the Estimator of the population total 7: xB» (a4) Estimated variance of +: (8) = IWS) = m( 2)(£) 4s) Bound on the error of estimation: ‘An industrial firm is concerned about the time spent each week by scientists on certain trivial tasks. The time-log sheets of a simple random sample of n = 50 employees show the average amount of time spent on these tasks is 10.31 hours, with a sample variance s?=2.25. The company employs = 750 scientists. Estimate the total num- ber of worker-hours lost each week on trivial tasks and place a bound on the error of estimation. ‘We know the population consists of N= 750 employees from which a random sam- ple of n = 50 time-log sheets was obtained. The average amount of time lost for the ‘50 employees was ¥ = 10.31 hours/week. Therefore, the estimate of 7 is = NY = 750(10.31) = 7732.5 hours To place a bound on the error of estimation, we apply Eq. (4.7) to obtain a = 2VIE6B = 307.4 hours avi ‘Thus, the estimate of total time lost is # = 7732.5 hous. We are reasonably confi dent thatthe eror of estimation isles than 307.4 hours. 102 ch pler 4 Simple Random Sar Exercises Some ofthe exercises are relatively das-intensive; look in the electvonie Section 4.0 for links to those data in Excel files. | 4A Listall possible blation (0, 1,2, 3, 4). Caleulate & forthe population and V() for the sample random sampes of size n= 2 that can be selected from the pop: ulate for 42. For the simple random samples generated in Exercise 4.1, cl th sample ‘Show numerically that Ee?) 43° Suppose you want to estimate the numb ‘What is the population, struct a frame? How would you select a simple random sample? Ifa sampling unit is an area such as a square yard, does the size chosen for a sampling unit affect the of weed clusters ofa certain type ina field 1d what would you use for sampling units? Hlow would you vy of the results? What considerations go into our choice of size of sampling unit? 44 In which of the following situations can you reasonably genes the population? ‘a. You use your sitisies clas to get an estimate of the percentage of students in your seboo! who study at Teast 0 hours @ night b. You use the average annual income of the ambassadors to the United Nations to get a estimate of average per-capita income forthe world asa whole mn 1996, a Gallup poll sampied 235 U.S. residents ages 18 to 29, o estimate the per centage of all US. residents ages 18 to 29 who favored cuts in social spending ie from the sample to 45 Describe the type of sample selection bias that would result rom each of these sampling methods. 4, Astudent wants to determine the aver some rie randomly on @ ms in a county in Towa, He drops of the county and uses the farms bit by grains of rie as the sample bs. Twa study about whether valedictorias “sueceed big in life,” a professor “traveled cross Tinois, tending high school graduations and selecting 81 students t0 participate, ... He picked students from the most diverse communities possi from litle rural schools to rich suburban schools near Chicago to city schools. Source: Michael Ryan, “Do Vi May 17, 1998, pages 14-15, ¢. Toestimate the percentage of students who passed the frst Advanced Placement Sta tities exam, a teacher onan Inernet discussion ist for teachers of AP Statistics asked students took the test and how ictorians Succeed Big in Life?” Parade Magazine, teachers on the listo report to him how many ofthe many passed. 4. To ind the average length of string ina bag, a student reaches in, mixes up the strings, selects one, mixes them up again, se In 1984, Ann Landers conducted a poll on the marital happiness of women by asking ats another, and $000. ‘women to write to her. 4.6 Suppose you want to know what per age of 13 living at home, Each weekday, from 90S, your poll takers call the households time they reach a person in the sample, they ask, “Do you have tage of U.S, households have children under the in your sample ‘ion 4.0 for links 2d from the pop- sample, ‘or each sample type ina field. How would you fa sampling unit sg unit affect the size of sampling bm the sample to students in your ced Nations to get estimate the per- al spending of these sampling in Towa, He drops yy grains of rice as fessor “traveled vg 81 students to vanities possible, to city schools.” Parade Magazine, cd Placement Ste AP Statistics asked the test and how ‘xes up the strings, ‘women by asking children under the call the households ask, “Do you have 47 48. 49 410 4a Exercises 103 children under the age of 13 living in your household?” E households that cannot be reached, Will your estimate of the percentage of U.S. households that have children under the age of 13 living in them prbably be t00 low, (00 high, or about right? i. How does this example help explain why poll takers ae likely to call at Ainnertime? ‘You want 10 estimate the average number of U.S. states that people living in the United States have visited. Ifyou asked only those at least 40 years old, would you expect the {estimate tobe 10 high ot low? What kind of bias might you expect if you selet your sample ftom only those living in Rhode Island? fentually you give up on the ‘Toestimate the average numberof children per family in the city where you live, you use {your statistics class asa convenience sample. You ask each student in the sample how many children there are in his or her family. Do you expect the sample average to be higher or lower than the population average? Explain why. “Television today is more offensive than ever, say the overwhelming majority —929- ‘of readers who took part in USA Weekend's third survey measuring attitudes toward the smal sreen.” More than 21,600 people responded to this write-in survey. [Source: USA Weekend, May 16-18, 1997, page 20} ‘a, Do you trust the results ofthe survey? Why or why not? b. What percentage of the entire U.S. TV-watching public do you think would say that “today’s shows are more offensive than ever": moze than 929%, quite a bit less than ‘92%, or just approximately 9296? Why do you think that? Suppose you wish to estimate the average sizeof English classes on your campus. Com- pare the merit of these two sampling methods. a, You get a list ofall students enrolled in English clases, take a random sample of ‘those stadens, end find out how many students are enrolled in each sampled student's English class. b. You get a list of al English classes, take a random sample of those classes, and find ‘out how many students are enrolled in each sampled class. Decide ifthe following sampling methods produce e simple random sample of stents from a class of 30 students. I'not, explain why. 4, Select the fist six students on the class roll shes. ’, Pick a digit at random and select those students whose phone numbers end in tht digit, ‘Ifthe classroom has six rows of chats with five seats in each row, choose a row at random and select ll students in that ro If the class consists of 15 boys and 15 girls, assign the boys the numbers from 11015, ‘and the girls the numbers from 16 to 30. Then use a random digit table to select six ‘numbers from 1 {0 30, Select the students assigned those numbers in your sample , Ifthe class consists of 1S boys and 15 girls, assign the boys the numbers from 11015, ‘and the gies the numbers from 16 to 30, Then use a random digit table to select three ‘numbers from | 1015 and three numbers from 16 to 30. Select the students assigned those numbers in your sample. Randomly choose a letter from the English alphabet and select forthe sample those students whose last names begin with that leer. If no last name begins with that letter, candomly choose another letter from the alphabet a 412 According tothe 2000 Statistical Abstract of the United States, approxitnatly 60% o 18 and 19-year-olds are enrolled ins 20OIpubs/sttabysecO4 pa (Table 245), AF you take a random sample of 40 randomly hosen 18 and 19-year-oks, would itbe reasonably likely to find that 32 were in school? ol. Source: itpiwwwcensus. gowlprady 4a The data set USPOP in Appencix C list resident popolation figures per state from the 2000, census, Selecta simple random sample of five states. Use the 2000 population figures forthe Sampled states to estimate the total U.S. population and place a bound on the error of tstimation, Des your terval answer include the total population figure given in the table? Do you think every possible 95% confidence interval based on samples of size $ would calculate the percent nde the true total? Compare your interval 1o those of other members ofthe class and ‘of observed intervals that actually capture the true total 4.14 state park officals were interested inthe proportion of eampers who consider the camp. site spacing adequate in particular campground, They decided to take a simple random Sample of n = 30 rom te first N= 300 camping pats that visited the campground, Let y; = 0 ifthe head ofthe th party sampled does not think the campsite spacing i a equate and yy = Lif he does ( 30), Use the data in the accompanying table 1 estimate p, the proportion of campers who consider the campsite spacing adequate Place @ bound on the error af estimation, Camper sampled 29 ' 30 1 4.15 Use the data in bound onthe exor of estimation of m Exercise 4.14 to determine the sample size required to estimate p with a nitude B = 0.05, 4.6 A simple random sample of 100 water meters within «community is monitored to est nate the average daily water consumption per household over a specified dry spell. The sample mean and sample variance are found to be¥ = 125 and s?= 1252, 1 we assume that there are W/— 10,000 households within the community, estimate , the true mean daily consumption 4.17 Using data in Bxereise 4.16, estimate the total numberof gellons of water, used daly during the dry spel. Place bound on the ertorof estimation 4.18 Resource managers of forest game lands were concerned about the size of the deer and rabbit populations dusing the winter months in a particular forest. As an estimate of population size, they proposed using the average numberof pellet groups for rabbits and de man aerial photograph, the forest was divided 0,000 30-foot-square grids. A simple random sample of 1 = $00 plots was per 30-footsquare plots ¥ nately 60% of sus gov/prod! 40 randomly vere in school? 2 from the 2000 figures for he on the error of leninthe table? of size $ would ofthe class and we total sider the eam simple random se campground. ep mpanying table acing adequate ing is a estimate p with a nonitored toes ied dry spel. The 152. I we assur the trve meat ater, 7, used ze ofthe deer and 4s an estimate of soups for rabbits forest was divided 1 = 500 plots 5 Exercises 105 taken, and the numberof pellet groups was observed for rabbits and for deer. The results companying table. imiberof pellet groups for deer and rabbit errors of estimation, For the two animals, this stody are summarized inthe Estimate 1, and yy, the average n Lively, per 30-foocsquare-plots, Place bounds on & size of pellet proups per plot respec 'b, Estimate the difference inthe mes with an appropriate margin of error. Deer Rabbits ‘Sample mean = 2.30 Sample mean = 4.52 Sample variance =0.65 Sample variance = 0.97 4.19 A dentist was imeresed inthe effectiveness of a new toothpaste. A group of N= 1000 schoolchildren participated in «study. Prestudy records showed there was an average of 2.2 cavities every six months for the group. After three months ofthe study, the dentist imple n

You might also like