For a bivariate dataset, r = 0.6, Sy = 15, and Sx = 5. If a new dataset is created by doubling all x-values (and keeping y the same), what happens to the slope and the correlation?
- AThe slope is halved; r remains 0.6.Correct
- BThe slope remains the same; r doubles.
- CBoth the slope and r double.
- DThe slope doubles; r remains 0.6.
Why A is right
Doubling all x-values multiplies Sx by 2. New slope: b1_new = r*(Sy/Sx_new) = r*(Sy/(2*Sx)) = (1/2)*r*(Sy/Sx) = (1/2)*b1. So the slope is halved. Correlation r is scale-invariant — multiplying x by a positive constant does not change r. So r remains 0.6.
Why the other 3 choices miss
- B
- Incorrect because the slope does change when x is rescaled. Multiplying x by 2 changes Sx, which changes b1. The claim that the slope remains unchanged is false.
- C
- Incorrect because r is invariant to positive linear rescaling of x. Doubling x does not change r. Only the slope changes (is halved).
- D
- Incorrect because doubling x doubles Sx, and since slope = r*(Sy/Sx), a larger Sx means a smaller slope, not a larger one. The slope is halved, not doubled.
For the LSRL equation y-hat = b0 + b1*x, which of the following correctly identifies how r^2 changes when all values of the response variable y are divided by a constant k (where k > 1)?
- Ar^2 decreases because the variability in y decreases.
- Br^2 increases because the y-values are now smaller.
- Cr^2 becomes 0 because the y-values are rescaled.
- Dr^2 remains unchanged because dividing by a constant is a linear transformation that does not affect the proportion of variation explained.Correct
Why D is right
Dividing all y-values by a constant k is a linear transformation: y_new = y/k. Correlation (and hence r^2) is invariant under linear transformations of either variable. The relative positions and the proportion of variation explained remain unchanged; both Sy and the scale of residuals change by the same factor k, leaving r^2 = SSR/SST unchanged.
Why the other 3 choices miss
- A
- Incorrect because while Sy does decrease by factor k, SST and SSE both decrease by the same factor (k^2), so their ratio (r^2 = 1 - SSE/SST) remains constant.
- B
- Incorrect because r^2 does not respond to the magnitude of y-values; it responds to the relative variation. Smaller y-values with proportionally smaller residuals yield the same r^2.
- C
- Incorrect because r^2 = 0 would mean the model explains none of the variation. A linear rescaling of y preserves the linear relationship and therefore does not change r^2.
The correlation between x and y in a dataset of 50 observations is r = 0.50. The mean of x is 10, the mean of y is 20, Sx = 3, and Sy = 6. A student correctly computes the LSRL and then realizes that the y-axis scale was entered incorrectly — all y-values should be multiplied by 2. What is the new slope of the LSRL?
- A1.0
- B0.5
- C4.0
- D2.0Correct
Why D is right
Original slope: b1 = r*(Sy/Sx) = 0.5*(6/3) = 0.5*2 = 1.0. After multiplying all y-values by 2, the new Sy = 2*6 = 12 (and r remains 0.5). New slope: b1_new = r*(Sy_new/Sx) = 0.5*(12/3) = 0.5*4 = 2.0.
Why the other 3 choices miss
- A
- Incorrect because 1.0 is the original slope before the correction. After multiplying y by 2, the slope must also double to 2.0.
- B
- Incorrect because 0.5 is the correlation coefficient, not the new slope. Multiplying y by 2 doubles the slope, not halves it.
- C
- Incorrect because 4.0 = Sy_new/Sx = 12/3, which is the ratio of standard deviations without multiplying by r. The slope = r*(Sy_new/Sx) = 0.5*4 = 2.0.