1 a T-stats allow you to remove at most one-variable at a time from a model.
2 a Distance and time are likely to be highly corelated. 3 b Use P(A and B) = P(A | B) P(B), where A is the event "does not fall by more than 3%". 0.3 * 0.2 = 0.06. 4 e The complement of the event is loose all three, with probability = 0.973^3. Subtract this from 1. 5 b 10000 * (0 * 0.7 + 1 * 0.2 + 2 * 0.1) = 4000 4000 6 d This is the parallel lines regression model. 7 e These measures all contain the units of Y. 8 e Using the Empirical Rule 57.1 - 1.645 * 10.8 = 39.33 39.334 9 a 57.1 +/- 2 * 10.8/SQRT(180) 55.49003 58.70997 10 d Using the "s" from the question: ((2 * 10.8)/(1))^2 = 467. 466.56 11 a The p-value is 0.102 which is not significant. 12 b We fail to reject, so the potential error is a Type II error. 13 e Confidence is a property of the process of creating confidence intervals. None of these statements capture this idea. 14 c 0 * 0.42 + 10000 * 0.49 + 20000 * 0.09 = 6700 6700 15 b SQRT(((0 - 6700)^2 * 0.42 + (10000 - 6700)^2 * 0.49 + (20000 - 6700)^2 * 0.09)) 6333.246 16 c 6700 - 5000 17 b This is 0.42^2, so they are assuming independence. 0.1764 18 c All three of these aspects could show up in the scatterplot matrix. 19 c The t-stat for Num recs is not significant (p-value > 0.05). This is the difference between marginal and partial association. It is because of correlation between the x-s that partial 20 d can be different from marginal. This is equivalent to saying that the slope for commercials is -1. But that value is not in the 95% CI for commercials, and 21 b the slope is more extreme than -1. 22 b The Classical coefficient is both positive and significant. 23 b Low RMSE is good. It means narrower prediction intervals. 24 d 25 a The prediction comes from the LS Means table and from there 62.905+/- 2 * 6.3837 50.1376 75.6724 26 c This is the definition of interaction.