Statistics and Data Analysis Professor William Greene Stern School of Business IOMS Department Department of Economics 1/18 Part 15: Hypothesis Tests Statistics and Data Analysis Part 15 – Hypothesis Tests: Part 3 2/18 Part 15: Hypothesis Tests A Test of Independence 3/18 In the credit card example, are Own/Rent and Accept/Reject independent? Hypothesis: Prob(Ownership) and Prob(Acceptance) are independent Formal hypothesis, based only on the laws of probability: Prob(Own,Accept) = Prob(Own)Prob(Accept) (and likewise for the other three possibilities. Rejection region: Joint frequencies that do not look like the products of the marginal frequencies. Part 15: Hypothesis Tests A Contingency Table Analysis 4/18 Part 15: Hypothesis Tests Independence Test Step 2: Expected proportions assuming independence: If the factors are independent, then the joint proportions should equal the product of the marginal proportions. [Rent,Reject] [Rent,Accept] [Own,Reject] [Own,Accept] 5/18 Hypothetical 0.54404 x 0.21906 = 0.11918 0.54404 x 0.78094 = 0.42486 0.45596 x 0.21906 = 0.09988 0.45596 x 0.78094 = 0.35606 (Actual) (.13724) (.40680) (.08182) (.37414) Part 15: Hypothesis Tests Comparing Actual to Expected The statistic is N times the sum over the four cells (Observed-Expected)2 = N × Rows Columns Expected If this is large (because the observed proportions don't 2 look like the expected ones) then rej ect the hypothesis. (This is a "chi squared statistic.") (0.13724 0.11918)2 (0.40680 0.42486) 2 0.11918 0.42486 2 13,444 2 2 (0.08182 0.09988) (0.37414 0.35608) 0.09988 0.35608 = 103.33013 6/18 Part 15: Hypothesis Tests When is Chi Squared Large? For a 2x2 table, the critical chi squared value for α = 0.05 is 3.84. (Not a coincidence, 3.84 = 1.962) Our 103.33 is large, so the hypothesis of independence between the acceptance decision and the own/rent status is rejected. 7/18 Part 15: Hypothesis Tests Computing the Critical Value For an R by C Table, D.F. = (R-1)(C-1) CalcProbability Distributions Chisquare The value reported is 3.84146. 8/18 Part 15: Hypothesis Tests Analyzing Default 9/18 Do renters default more often (at a different rate) than owners? To investigate, we study the cardholders (only) We have the raw observations in the data set. OWNRENT 0 DEFAULT 0 1 All 4854 615 5469 46.23 5.86 52.09 1 4649 44.28 381 3.63 5030 47.91 All 9503 90.51 996 9.49 10499 100.00 Part 15: Hypothesis Tests 10/18 Part 15: Hypothesis Tests 11/18 Part 15: Hypothesis Tests [.4623 (.9051 .5209)]2 [.0586 (.0949 .5209)]2 (.9051 .5209) (.0949 .5209) 2 10499 [.4428 (.9051 .4791)]2 [.0363 (.0949 .4791)]2 (.9051 .4791) (.0949 .4791) 12/18 Part 15: Hypothesis Tests Hypothesis Test 13/18 Part 15: Hypothesis Tests In my sample of 210 travelers between Sydney and Melbourne, it appears that there is a relationship between income and the decision whether to fly or not. Do the data suggest that the mode choice and income are independent? 14/18 Part 15: Hypothesis Tests Treatment Effects in Clinical Trials Does Phenogyrabluthefentanoel (Zorgrab) work? Investigate: Carry out a clinical trial. N+0 = “The placebo effect” N+T – N+0 = “The treatment effect” Is N+T > N+0 (significantly)? Placebo 15/18 Drug Treatment No Effect N00 N0T Positive Effect N+0 N+T Part 15: Hypothesis Tests 16/18 Part 15: Hypothesis Tests Confounding Effects 17/18 Part 15: Hypothesis Tests What About Confounding Effects? Normal Weight Obese Nonsmoker Smoker Age and Sex are usually relevant as well. How can all these factors be accounted for at the same time? 18/18 Part 15: Hypothesis Tests