Estimating the Population Mean Income of Lexus Owners • Sample Mean + Margin of Error • Called a Confidence Interval • To Compute Margin of Error, One of Two Conditions Must Be True: • The Distribution of the Population of Incomes Must Be Normal, or • The Distribution of Sample Means Must Be Normal. A Side-Trip Before Constructing Confidence Intervals What is a Population Distribution? What is a Distribution of the Sample Mean? How Does Distribution of Sample Mean Differ From a Population Distribution? What is the Central Limit Theorem? Assume Small Population of Lexus Owners’ Incomes (N = 200) 190 149 187 361 97 76 302 234 371 276 165 162 330 174 247 272 349 275 256 199 105 89 182 91 135 353 263 314 353 295 316 316 242 101 287 153 361 353 354 185 254 85 348 287 94 105 127 317 346 250 284 312 337 368 195 305 253 211 235 206 345 124 215 320 183 152 196 154 238 121 156 278 207 161 108 285 242 125 115 174 340 141 203 367 221 308 241 128 225 343 346 302 140 235 344 333 365 94 100 138 363 80 166 215 228 279 288 335 277 188 87 360 333 197 191 76 220 77 248 297 79 161 368 165 187 318 242 109 222 135 288 261 159 374 104 279 152 237 324 232 197 178 317 300 371 292 129 93 119 137 211 292 286 343 308 354 320 260 95 344 334 241 372 180 87 101 366 303 86 175 230 365 287 318 310 88 224 223 156 256 117 182 152 308 144 115 281 297 314 173 152 186 188 348 275 230 330 249 285 232 Distribution of N = 200 Incomes Mean 30 75 125 175 225 275 325 Constructing a Distribution of Samples of Size 5 from N = 200 Owners Obs 1 362 80 166 214 228 278 288 335 277 188 86 359 332 Obs 2 Obs 3 79 160 367 165 187 317 241 109 221 135 287 260 159 197 177 316 300 370 292 129 92 118 136 211 291 285 Obs 4 333 241 372 180 87 100 366 303 86 175 229 365 287 x Obs 5 Mean 116 182 151 307 144 114 281 296 313 172 151 185 187 217.4 168.0 274.4 233.2 203.2 220.2 261.0 227.0 203.0 161.2 192.8 292.0 250.0 Distribution of Sample Mean Incomes (Column #7) 25 21 Estimated Std. Error 20 15 8 10 6 3 5 2 0 125 175 225 275 325 Distribution of Sample Means Near Normal! ( X) Central Limit Theorem Even if Distribution of Population is Not Normal, Distribution of Sample Mean Will Be Near Normal Provided You Select Sample of Five or Ten or Greater From the Population. For a Sample Sizes of 30 or More, Dist. of the Sample Mean Will Be Normal, with mean of sample means = population mean, and standard error = [population deviation] / [sqrt(n)] Thus Can Use Expression: X MOE Why Does Central Limit Theorem Work? As Sample Size Increases: Most Sample Means will be Close to Population Mean, Some Sample Means will be Either Relatively Far Above or Below Population Mean. A Few Sample Means will be Either Very Far Above or Below Population Mean. Impact of Side-Trip on MOE MOE = (Confidenc e factor) Spread in Xs Determine Confidence, or Reliability, Factor. Distribution of Sample Mean Normal from Central Limit Theorem. Use a “Normal-Like Table” to Obtain Confidence Factor. Determine Spread in Sample Means (Without Taking Repeated Samples) Drawing Conclusions about a Pop. Mean Using a Sample Mean Select Simple Random Sample Compute Sample Mean and Std. Dev. For n < 10, Sample Bell-Shaped? For n >10 CLT Ensures Dist of x Normal x MOE Draw Conclusion about Population Mean, m Federal Aid Problem Suppose a census tract with 5000 families is eligible for aid under program HR-247 if average income of families of 4 is between $7500 and $8500 (those lower than 7500 are eligible in a different program). A random sample of 12 families yields data on the next page. Federal Aid Study Calculations Representative Sample 7,300 7,700 8,100 8,400 7,800 8,300 8,500 7,600 7,400 7,800 8,300 8,600 x $7,983 s $441 ( 7300 7983)2 ..(8600 7983)2 12 1 x MOE 7,983 MOE Estimated Standard Error Measures Variation Among the Sample Means If We Took Repeated Samples. But We Only Have One Sample! How Can We Compute Estimated Standard Error? Based on Constructing Distribution of Sample Mean Slide, Will Estimated Standard Error Be Smaller or Larger Than Sample Standard Deviation (s)? Estimated Std. Error ______ than s. Estimated Standard Error Expression sample standard deviation sample size For Federal Aid Study s n Confidence Factor for MOE: Appendix 5 Df = n-1 2-Sided 90% 2-Sided 95% 2-Sided 99% 1.86 2.306 3.355 11 1.796 2.201 3.106 14 1.761 2.145 2.977 17 1.740 2.110 2.898 8 Can Use t-Table Provided Distribution of Sample Mean is Normal 95% Confidence Interval X Confidence Est. S tan dard Error 441 7, 983 ( 2. 201) 12 $7, 703 $8, 263 Interpretation of Confidence Interval 95% Confident that Interval $7,983 + $280 Contains Unknown Population (Not Sample) Mean Income. If We Selected 1,000 Samples of Size 12 and Constructed 1,000 Confidence Intervals, about 950 Would Contain Unknown Population Mean and 50 Would Not. So Is Tract Eligible for Aid??? Would Tract Be Eligible? X Conf Factor Spread in Means Situation A: 7,700+ 150 7,550 7,850 Situation B: 8,250+ 150 8,100 8,400 Situation C: 8,050+ 150 7,900 8,200 Width versus Meaningfulness of Two-Sided Confidence Intervals For n = 1000 60 % CI WIDTH s MOE 0.842 n CONFIDENCE LOW Ideal: 99 % CI MOE HIGH _________ Level of Confidence and _________ Confidence Interval . How Obtain? Chapter Summary Why Must We Estimate Population Mean? Why Would You Want to Reduce MOE? How Can MOE Be Reduced Without Lowering Confidence Level?