Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
LECTURE 3 SAMPLING THEORY EPSY 640 Texas A&M University POPULATIONS • finite population consists of the actual group of objects or persons, which we know is potentially countable and finite. • infinite population population is a mathematical abstraction that is useful because the properties of the population are assumed or defined carefully, , POPULATIONS • Parameter = characteristic of the population. • If a sample is drawn and the characteristic computed, it will be a statistic for the sample. POPULATIONS • Accessible vs. Target Populations. • Target Population, the population we wish to represent. • Instead, we might be able to draw from all public school grade 3 students in class during a particular week in the school year. This is our Accessible Population, the population we have access to. Accessible population Sample Target population Figure 4.1: Inferences from sample to Populations Sampling Methods • RANDOM SAMPLING – – – – SIMPLE STRATIFIED MULTISTAGE CLUSTER • SYSTEMATIC SAMPLING • CONVENIENCE (NONRANDOM) SAMPLING RANDOM SAMPLING • If every member of a population has an equal chance of being selected • involves being able to define and count the population. • can then use a process called randomization to select the sample Table of Random Numbers Location RN Location RN 234 75 308 01 ….. 235 13 309 26 ….. 236 95 310 31 ….. 237 22 311 69 ….. 238 46 312 29 ….. 239 86 313 98 ….. 240 55 314 34 ….. 241 59 315 17 ….. In selecting a sample of 20 students from a list of of 75, a random start point was selected as shown above. The ad hoc rule was to go down the column to the bottom and up the next. Thus, children with identifiers 75 1, 13, 26, 69, 22, 46, 29, 55, 34, 59, 59, and17 have been selected within this section of the random number table. The location value allows checking and replication of a random sample selection process. finite population correction • fpc= 1- n/N = 1-f where n= # in sample N=# in population Finding survey sample size (z/d)2 n = ________________________ 1 + (1/N)(z/d)2 z = z-score for probability for confidence interval required (usually 1.96 for .05 or 2.59 for .01) = SD of distribution (can be 1.0 for arbitrary units) d = desired degree of error in SD units Finding survey sample sizeexample Alpha=.05, N=1,000,000 d=.1 , = 1 (1.96/.1)2 n = ________________________ 1 + (1/1000000)(1.96/.1)2 = 19.62 = 384.16 Population SizeSample Size Required for d= .1 20 30 40 50 75 100 125 150 175 200 225 250 275 300 350 400 500 600 750 1000 1500 2000 2500 5000 7500 10000 100000 1000000 for = .05 for = .01 19 19 28 29 36 38 44 46 62 67 79 87 94 105 108 122 120 138 132 154 142 168 151 182 160 194 168 207 183 229 196 250 217 285 234 315 254 352 278 399 306 460 322 498 333 525 357 586 365 610 370 623 383 660 384 663 Table 4.2: Sample sizes required for various population sizes for 95% and 99% confidence intervals Mean and standard deviation for simple random sampling • • • • • (x.) = (sample mean estimates population mean unbiasedly) V(x.) = (1/n) s2(1-f) (variance must be corrected) _____ s x. = V(x.) = standard error of the mean =sm -1.96sm -sm sm Mean from a particular sample 1.96sm Original Data Distribution Distribution of Means -1.96sm -sm sm Mean from a particular sample 1.96sm Confidence interval • Mean zsx. • z = # SDs of normal distribution for some probability of confidence, usually .01 or .05 • for real data: x. 1.96s x gives a confidence interval around the mean: – Interpretation: in 95 of 100 times we do the study, the population mean will be in the interval we construct. Distribution of Means Confidence interval -1.96sm -sm sm 1.96sm Mean from a particular sample Interpretation: in one event is either IN or OUT of the confidence interval; for 100 intervals, it should be IN 95 times on average. Stratified random sampling • subpopulations, called strata. • We then use simple random sampling for each stratum. • We can decide to sample proportionately or disproportionately. Stratified random sampling • Proportionate sampling: percentage in sample is same as in population • or disproportionate sampling: percentage in sample is different from that of population • Example Males and Females (50% in pop.). – Proportional: 50 males, 50 females – Disproportional: 75 males, 25 females Stratified random sampling • Example: Ethnicity of students in District: 80% Anglo, 10% Hispanic, 5% African American, 5% Native American • Proportional for 200 student sample: – 160 Anglo, 20 Hispanic, 10 African-American, 10 Native American • Disproportional: – 50 Anglo, 50 Hispanic, 50 African-American, 50 Native American Stratified random sampling • Example: Ethnicity of students in District: 80% Anglo, 10% Hispanic, 5% African American, 5% Native American • Proportional for 200 student sample: – 160 Anglo, 20 Hispanic, 10 African-American, 10 Native American – May give poor estimates for H, AA, NA samples • Disproportional: – 50 Anglo, 50 Hispanic, 50 African-American, 50 Native American – Will give estimates with similar confidence intervals for all groups – may need fpc for some groups Mean for stratified random sample. s x..est = ( Ni xi.)/N i=1 Where Ni = numer of cases in the population stratum i, N = total number of cases in the entire population, and s = number of strata. Mean for stratified random sample- example 3 strata, N1=1000, N2=2000, N3=3000 X1 = 70, X2 = 80, X3 = 90 s x..est = ( Ni xi.)/N i=1 = [(1000 x 70) + (2000 x 80) + (3000 x 90) ] / 6000 = 83.33 SD for stratified random sample. • • • s V(x..est) = Ni2 s2x ./N2 • x. = V(x..est) , i i=1 • where s2x .= V(xi.), the variance error of the mean using the simple random sample formula i SD for stratified random sample. SUBPOPULATION NI ni X. si sm. A 77 50 10 5 .419 B 229 50 11 6 .751 C 738 50 12 7 .956 X..est = (77 x 10 + 229 x 11 + 738 x 12)/1044 = 11.63 V(X..est) = (772 x (.419)2 + 2292 x (.751)2 + 7382 x (.956)2 ) /10442 = .485 s(X..est) = .696 Table 4.3: Calculation of stratified sample mean and variance error of the mean SD for stratified random sample. SUBPOPULATION NI ni X. si sm. A 77 50 10 5 .419 B 229 50 11 6 .751 C 738 s2m = (1/ni)si2 (1-fi) 50 12 7 .956 .4192 = (1/50)52 (1-50/77) .7512 = (1/50)62 (1-50/229) .9562 = (1/50)72 (1-50/738)