In: Statistics and Probability
Almost all U.S. light-rail systems use electric cars that run on
tracks built at street level. The Federal Transit Administration
claims light-rail is one of the safest modes of travel, with an
accident rate of .99 accidents per million passenger miles as
compared to 2.29 for buses. The following data show the miles of
track and the weekday ridership in thousands of passengers for six
light-rail systems.
City | Miles of Track | Ridership (1000s) |
Cleveland | 14 | 17 |
Denver | 16 | 37 |
Portland | 37 | 83 |
Sacramento | 20 | 33 |
San Diego | 46 | 77 |
San Jose | 30 | 32 |
St. Louis | 33 | 44 |
SSE | = |
SST | = |
SSR | = |
MSE | = |
X | Y | XY | X² | Y² |
14 | 17 | 238 | 196 | 289 |
16 | 37 | 592 | 256 | 1369 |
37 | 83 | 3071 | 1369 | 6889 |
20 | 33 | 660 | 400 | 1089 |
46 | 77 | 3542 | 2116 | 5929 |
30 | 32 | 960 | 900 | 1024 |
33 | 44 | 1452 | 1089 | 1936 |
Sample size, n = | 7 |
Ʃ x = | 196 |
Ʃ y = | 323 |
Ʃ xy = | 10515 |
Ʃ x² = | 6326 |
Ʃ y² = | 18525 |
x̅ = | 28 |
y̅ = | 46.14285714 |
SSxx = Ʃx² - (Ʃx)²/n = | 838 |
SSyy = Ʃy² - (Ʃy)²/n = | 3620.857143 |
SSxy = Ʃxy - (Ʃx)(Ʃy)/n = | 1471 |
a) b1 = SSxy/SSxx = 1.75537
= 1.76
b0 = y̅ -b1* x̅ = -3.0075 = -3.01
Regression equation:
ŷ = -3.01 + 1.76 x
b)
SSE = SSyy -b*SSxy = | 1039.708 = 1038.7 |
SST = SSyy = Ʃy² - (Ʃy)²/n = | 3620.8571 = 3620.9 |
SSR = b*SSxy = | 2582.1492 = 2582.1 |
MSE = SSE/(n-2) = | 207.7 |
c) Coefficient of determination, R² = SSR/SST = 0.713
Yes, it provides a good fit.
------------------
d) Predicted value at X = 30
ŷ = -3.0075 + 1.7554 * 30 = 49.65
Standard error, se = √(SSE/(n-2)) = 14.4132
At α = 0.05 and df = n-2 = 5, critical value, t_c = T.INV.2T(0.05, 5 ) = 2.5706
95%Confidence interval for the mean weekday ridership for all
light-rail systems with 30 miles of track :
Lower limit = ŷ - tc*se*√((1/n) + ((x-x̅)²/(SSxx))) =
35.4
Upper limit = ŷ + tc*se*√( (1/n) + ((x-x̅)²/(SSxx))) =
63.9
------------------
e) 95% prediction interval for the weekday ridership for the Charlotte system:
Lower limit = ŷ - tc*se*√(1 + (1/n) + ((x-x̅)²/(SSxx))) =
10.0
Upper limit = ŷ + tc*se*√(1 + (1/n) + ((x-x̅)²/(SSxx))) =
89.3
------------