In: Math
A professor has a class with four recitation sections. Each section has 16 students (rare, but there are exactly the same number in each class...how convenient for our purposes, yes?). At first glance, the professor has no reason to assume that these exam scores from the first test would not be independent and normally distributed with equal variance. However, the question is whether or not the section choice (different TAs and different days of the week) has any relationship with how students performed on the test.
Group-1 | Group-2 | Group-3 | Group-4 |
---|---|---|---|
73.5 | 76.7 | 75 | 65.7 |
81 | 66.4 | 77.8 | 50.5 |
61.8 | 60.3 | 66.7 | 83 |
69.5 | 81 | 70.3 | 81.4 |
77.4 | 57.9 | 77.7 | 74.9 |
91.2 | 59.2 | 68.1 | 82.9 |
70.6 | 67.9 | 83.5 | 85.4 |
64 | 54.9 | 87.8 | 63.6 |
73.2 | 63.2 | 80.6 | 67.6 |
77.7 | 69.8 | 58.9 | 73.6 |
73.6 | 69.1 | 86.7 | 81.5 |
77.2 | 51.8 | 74.7 | 80.5 |
54 | 60.5 | 66.9 | 71.8 |
65.4 | 55.4 | 76.7 | 68.1 |
77.8 | 68.2 | 76.3 | 55.8 |
81.6 | 64.8 | 69.5 | 70.4 |
First, run an ANOVA with this data and fill in the summary table.
(Report P-values accurate to 4 decimal places and all
other values accurate to 3 decimal places.
Source | SS | df | MS | F-ratio | P-value |
---|---|---|---|---|---|
Between | |||||
Within |
To follow-up, the professor decides to use the Tukey-Kramer method
to test all possible pairwise contrasts.
What is the Q critical value for the Tukey-Kramer critical range
(alpha=0.01)?
Use the website link in your notes
(http://davidmlane.com/hyperstat/sr_table.html) to locate the Q
critical value to 4 decimal places.
Q =
Using the critical value above, compute the critical range and then
determine which pairwise comparisons are statistically
significant?
Group-1 | Group-2 | Group-3 | Group-4 | Total | |
Sum | 1169.5 | 1027.1 | 1197.2 | 1156.7 | 4550.5 |
Count | 16 | 16 | 16 | 16 | 64 |
Mean, sum/n | 73.09375 | 64.19375 | 74.825 | 72.29375 | |
Sum of square, Ʃ(xᵢ-x̅)² | 1195.049 | 941.0294 | 917.31 | 1551.929 |
Null and Alternative Hypothesis:
Ho: µ1 = µ2 = µ3 = µ4
H1: At least one mean is different
Number of treatment, k = 4
Total sample Size, N = 64
df(between) = k-1 =3
df(within) = k(n-1) =60
df(total) = N-1 =63
SS(between) = (Sum1)²/n1 + (Sum2)²/n2 + (Sum3)²/n3 + (Sum4)²/n4 - (Grand Sum)²/ N =1071.552
SS(within) = SS1 + SS2 + SS3 + SS4 =4605.318
SS(total) = SS(between) + SS(within) =5676.870
MS(between) = SS(between)/df(between) =357.184
MS(within) = SS(within)/df(within) =76.755
F = MS(between)/MS(within) =4.654
p-value = F.DIST.RT(4.6535, 3, 60) = 0.0054
Decision:
Reject the null hypothesis.
ANOVA | |||||
Source of Variation | SS | df | MS | F | P-value |
Between Groups | 1071.552 | 3 | 357.184 | 4.654 | 0.0054 |
Within Groups | 4605.318 | 60 | 76.755 | ||
Total | 5676.870 | 63 |
At α = 0.01, k = 4, N-K = 60, Q value = 4.5942
Critical Range, CV = Q*√(MSW/n) = 4.5942*√(76.7553/16) = 10.0625
Comparison | Absolute Diff. = (xi - xj) | Critical Range | Results |
x̅1-x̅2 | 8.9 | 10.0625 | Means are not different |
x̅1-x̅3 | -1.73125 | 10.0625 | Means are not different |
x̅1-x̅4 | 0.8 | 10.0625 | Means are not different |
x̅2-x̅3 | -10.6313 | 10.0625 | Means are different |
x̅2-x̅4 | -8.1 | 10.0625 | Means are not different |
x̅3-x̅4 | 2.53125 | 10.0625 | Means are not different |