In: Statistics and Probability
Suppose that a MBA level stat course is taught using four
different methods of instruction: (1) 100% online; (2) a “half and
half” format where one week the class meets for a lecture, the next
week, material is posted online, etc.; (3) traditional weekly
lecture meetings plus supplementary material posted online; and (4)
traditional weekly lecture meeting with no use of the web.
Twenty students are surveyed from each course
and are asked to estimate the average number of hours per week that
they spent on the course, including time spent attending lectures
if the course had any. The results appear in the included data
file.
(a) Create boxplots for these four sets of data (all
on the same graph). Based on the plots, what do you think about the
ANOVA assumptions of normal populations and equal variances?
(You’ll test these in Part (d), I’m just interested in a visual
interpretation here.) Your answer should include
justification….. don’t just say ‘yup’ or ‘nope.’
(b) Create interval plots for this data (four
intervals on the same plot) so that we can visually compare the
four groups. This is the plot that shows a confidence interval for
each unknown population mean. Based on the plot, do you believe
that the population mean time spent is the same for all groups?
Again, show the reasoning behind your answer.
(c) Let μ1 represent the population mean time spent
for method (1), μ2 the mean time spent for method (2), and so on.
Test the null hypothesis that all means are equal at the 0.05 level
of significance, versus the usual ANOVA alternative.
(d) Is there evidence of violations of the usual ANOVA
assumptions of equal variances and normal populations? Set up and
perform appropriate TESTS at the α = 0.05 level of
significance.
(e) If your answer in Part (c) was to “reject H0” then
perform an appropriate statistical procedure to determine which
means are different from which other means (a visual inspection is
not sufficient.). If differences exist, be sure to
report the ‘direction’ of the d
Time | Method |
6.0 | OnLine |
7.1 | OnLine |
5.9 | OnLine |
8.9 | OnLine |
7.3 | OnLine |
6.1 | OnLine |
7.7 | OnLine |
7.1 | OnLine |
6.3 | OnLine |
8.4 | OnLine |
7.9 | OnLine |
6.9 | OnLine |
6.9 | OnLine |
6.3 | OnLine |
6.7 | OnLine |
6.0 | OnLine |
6.4 | OnLine |
8.3 | OnLine |
7.5 | OnLine |
8.2 | OnLine |
5.0 | Half&Half |
5.9 | Half&Half |
8.1 | Half&Half |
7.9 | Half&Half |
7.1 | Half&Half |
7.9 | Half&Half |
7.6 | Half&Half |
4.7 | Half&Half |
6.8 | Half&Half |
6.2 | Half&Half |
7.4 | Half&Half |
5.0 | Half&Half |
5.8 | Half&Half |
6.9 | Half&Half |
6.9 | Half&Half |
5.7 | Half&Half |
5.8 | Half&Half |
6.7 | Half&Half |
7.4 | Half&Half |
6.9 | Half&Half |
5.7 | LecturePlus |
5.5 | LecturePlus |
7.0 | LecturePlus |
5.9 | LecturePlus |
4.1 | LecturePlus |
7.1 | LecturePlus |
6.6 | LecturePlus |
6.4 | LecturePlus |
5.4 | LecturePlus |
5.7 | LecturePlus |
6.1 | LecturePlus |
4.8 | LecturePlus |
7.2 | LecturePlus |
6.2 | LecturePlus |
4.9 | LecturePlus |
6.3 | LecturePlus |
5.4 | LecturePlus |
6.3 | LecturePlus |
6.1 | LecturePlus |
5.5 | LecturePlus |
5.3 | Lecture |
5.1 | Lecture |
0.0 | Lecture |
6.4 | Lecture |
4.9 | Lecture |
7.3 | Lecture |
5.6 | Lecture |
6.5 | Lecture |
6.2 | Lecture |
4.8 | Lecture |
7.1 | Lecture |
7.1 | Lecture |
5.8 | Lecture |
5.9 | Lecture |
5.9 | Lecture |
5.7 | Lecture |
7.3 | Lecture |
5.2 | Lecture |
5.9 | Lecture |
6.4 | Lecture |
(a)
From the above box plots we see that lengths of boxes are not same so assumption of equal variances does not hold.
From the box plots, we see that third quartile-median is almost equal to median-first quartile for "Lecture" only so the assumption of normality does not hold here.
(b)
From the above plots we see that the population mean time spent is not same for all groups since 95% C.I.s of "LecturePlus" and "Online" are disjoint as well as 95% C.I.s of "Lecture" and "Online" are disjoint.
(c)
One-way ANOVA: Time versus Method
Source DF SS MS F P
Method 3 23.97 7.99 6.51 0.001
Error 76 93.26 1.23
Total 79 117.24
Since p-value<0.05 so we reject H0 at 0.05 level and conclude that population means time spent for 4 methods are not same.
(d)
Test for Equal Variances: Time versus Method
95% Bonferroni confidence intervals for standard deviations
Method N Lower StDev Upper
Half&Half 20 0.72618 1.02458 1.67793
Lecture 20 1.09983 1.55177 2.54130
LecturePlus 20 0.56134 0.79200 1.29704
OnLine 20 0.64324 0.90756 1.48628
Bartlett's Test (Normal Distribution)
Test statistic = 10.26, p-value = 0.016
Levene's Test (Any Continuous Distribution)
Test statistic = 0.43, p-value = 0.730
Under normality assumption, the assumptions of equal variances does
not hold (since p-value<0.05).
Since p-value of AD test<0.05 so normality assumption does not hold here.
(e)
Tukey 95% Simultaneous Confidence Intervals
All Pairwise Comparisons among Levels of Method
Individual confidence level = 98.97%
Method = Half&Half subtracted from:
Method Lower Center Upper
---------+---------+---------+---------+
Lecture -1.786 -0.865 0.056 (-------*------)
LecturePlus -1.596 -0.675 0.246 (------*-------)
OnLine -0.411 0.510 1.431 (------*-------)
---------+---------+---------+---------+
-1.2 0.0 1.2 2.4
Method = Lecture subtracted from:
Method Lower Center Upper
---------+---------+---------+---------+
LecturePlus -0.731 0.190 1.111 (-------*------)
OnLine 0.454 1.375 2.296 (------*-------)
---------+---------+---------+---------+
-1.2 0.0 1.2 2.4
Method = LecturePlus subtracted from:
Method Lower Center Upper
---------+---------+---------+---------+
OnLine 0.264 1.185 2.106 (-------*-------)
---------+---------+---------+---------+
-1.2 0.0 1.2 2.4
From the obove C.I.s we observed that population means of "Lecture"
and "Online" and population means of "LecturePlus" and "Online" are
significantly different since the corresponding C.I.s do not
contain zero.