Stat401E Fall 2010 Lab 10 1. Demand theory states that demand for a product depends (among other things) on the price of the product and on consumers' incomes. As the price of a product increases, demand for it will decrease. As consumers' incomes increase, the demand for the product will increase. You have data on the cigarette purchases and the incomes of a random sample of 78 U.S. smokers of Marlboro cigarettes. Your data were collected at random times over the last 30 years--years during which (even after adjusting for inflation) the price of a pack of Marlboro cigarettes has varied considerably. Your data set has the following three variables: DEMAND the average number of cigarette packages purchased daily INCOME annual income (adjusted for inflation) PRICE price of a package of Marlboro cigarettes (adjusted for inflation) Correlations among these three variables are as follows: DEMAND INCOME PRICE DEMAND 1.00 .45 -.35 INCOME .45 1.00 .15 -.35 .15 1.00 PRICE (Make sure you read these data correctly. For example, the zero-order correlation between DEMAND and INCOME is .45 .) a. What proportion of the variance in DEMAND is explained by the linear effects of INCOME and PRICE? (Hint: The explained proportion should include both variance explained separately by INCOME and PRICE and variance explained in common by INCOME and PRICE.) b. Of "the variance in DEMAND that is not explained by INCOME," what proportion of this unexplained variance is explained by PRICE? c. Find a 95% confidence interval for the first-order partial correlation between DEMAND and PRICE while controlling for INCOME. 1 2. Let X = j + kW , where j and k are positive constants. a. Show algebraically that the correlation between Y and X equals the correlation between Y and W. (Hints: Begin with the formula for r XY [Lecture Notes, p. 144]. Substitute "j + k W " for X and "j + k W " for X in this formula. Then simplify.) b. Show algebraically that when k is negative, r XY = – r WY . (Hints: 2 2 See hints for part a, plus remember k k , but instead k = k .) c. Finally, show that r XY = r Z Z , where Z X and Z Y are standardized X Y forms of X and Y (respectively). (Hints: Begin with the formula for . Then substitute Z X and Z Y in this formula with each's expression in terms of an unstandardized variable, its mean, and rZ XZ Y its standard deviation. [See page 137 in the Lecture Notes for such an expression.] Then simplify.) d. Extra Credit: If the regression equation for the regression of Y ˆ = â + bˆ W , derive the regression coefficients for the on W equals Y 1 1 regression of Y on X in terms of j, k, â 1 , and bˆ 1 . (Hints: Begin with formulas for â 2 and bˆ 2 from the regression of Y on X. Substitute "j + k W " for X and "j + k W " for X in these formulas. Simplify into expressions including only j, k, â 1 , and bˆ 1 .) 3. The warden of a prison believes that his rehabilitation program does not work, since there is a zero-order correlation of r = .71 between how well a prisoner does in the program and the number of crimes committed after a prisoner leaves prison. However, you discover that the first-order partial correlation between the two variables equals -.71 when you adjust for the effects of race (0 = African-American and 1 = white). Explain how this is possible in the following ways: a. First of all, show that it is mathematically possible to have a zero-order correlation of .71 and a first-order partial correlation as small as -.71 . You can do this by choosing two zero-order correlations (namely, between race and subsequent crime and between race and performance in program) such that the partial correlation is smaller than -.71 . (Hint: In doing this problem use the 2 computation formula for a first-order partial correlation on page 173 of the Lecture Notes.) b. Next, make a rough sketch of how the relation between (1) prisoners' performance in the program and (2) their later criminal record may differ according to (3) race. (Hints: Instead of drawing a sketch by hand, you may prefer to obtain a plot and data set using the following URL: http://www.public.iastate.edu/~carlos/401/lectures/applets/partial.html Your browser’s print capability can be used to print the 2nd part of this web page. If you print this part in black-and-white, please indicate on the sketch which points are for African Americans and which are for whites. In working on this problem, one strategy is to place points on the graph [no more than 5 data points per race and keeping data points at least a quarter-inch apart, please] by trial and error until you get a positive zero-order correlation between performance and crimes, but a negative partial correlation between the two variables when controlling for race. Then erase a point. If the former correlation becomes larger and the latter becomes smaller, leave it erased, otherwise replace it. Then add a point. If the former correlation becomes larger and the latter becomes smaller, leave it on the plot, otherwise erase it. As the former correlation approaches 1 and the latter approaches -1, a clear pattern will emerge.) c. Finally explain in words a possible (that is, a theoretical) reason why African-American and white prisoners may react differently to the rehabilitation program. Be sure to use an explanation that is consistent with the sketch obtained in part b. (Hints: Your explanation should account for the association represented by each correlation. First, why do you theorize that performance and race are associated as they are in the sketch? Second, why do you theorize that crimes and race are associated as they are in the sketch? Third, assuming that the sketch and these two theoretical arguments are accurate, how could the rehabilitation program be effective despite the fact that those who perform well in it are more likely to commit crimes after leaving prison than those who performed poorly in it?) 3