Increasingly over the past few years students have wanted to do these assignments using Microsoft Excel, which was already installed on their computers. I allowed them to do the assignments in Excel but didn't provide much guidance in class, feeling that it would be too distracting for most students to hear about both packages. When it came time to actually perform essential statistical tasks in Excel, such as constructing histograms and randomly sampling from a column of data, they discovered that Excel wasn't as easy to use for statistical analysis as it was for other types of tasks.

Yet I wanted to help students succeed with Excel because it is so widely available and used. I myself relied extensively on Excel in my operations management class (for which statistics is a prerequisite) and felt that our students would benefit greatly by seeing the same software package in more than one course. Perhaps they would start to think about solving most of their quantitative problems in a spreadsheet environment. Consequently, in the Spring 2001 semester, I took the plunge and created new versions of my assignments that contained detailed instructions in Excel. I developed these partly by consulting statistical texts geared toward Excel such as Berk and Carey (2000).

2. The Spring 2001 Experiment

On the first day of class I explained to students that they could use either package to do the computer assignments, and briefly outlined their main advantages and disadvantages. I said that Excel is probably already installed on their computers, whereas they'd have to go to the college computer lab to use Minitab. Minitab, however, is geared to do exactly the kinds of things we'd be discussing in class, while Excel is not and would require a little more work. In fact, Excel users would first need to make sure that the "Data Analysis Toolpak" had been installed on their machines with Excel before they could perform the required analyses.

All five assignments contained instructions in both Minitab and Excel. During the class in which the first assignment came out, I demonstrated what needed to be done in both environments, which took up quite a bit of time. With subsequent assignments, I reviewed (but didn't demonstrate) the instructions emphasizing important points and differences between the packages. For example, the two packages calculate class frequencies in a histogram differently: Minitab includes values equal to the left class endpoint (but not the right endpoint) whereas Excel does just the opposite. A few students actually did some of the assignments both ways and discovered additional discrepancies between the packages, e.g., they use different terms and symbols for the same statistical concept, and have slightly different formulas for some things such as the first and third quartiles.

Assignment 1 focuses on making several histograms from a dataset with multiple columns and using these to answer questions about the distribution of the variables. The second assignment also requires students to construct histograms for several variables, as well as to examine basic descriptive statistics. Assignment 3 requires simple linear regression, while the fourth demonstrates the Central Limit Theorem by having the students randomly sample repeatedly from a large population of data. The last assignment requires the calculation of confidence intervals for the mean, in addition to performing 1-sample t-tests for the mean.

3. Results

I wanted to investigate whether or not student performance on the assignments depended on which software package they used. Based on the printed output they turned in with their assignments I could easily tell which package students used and recorded their grades in separate columns of my (Excel) grade book. However, I graded the assignments in a random order, i.e., I didn't sort them into two separate groups prior to grading. The mean scores for the two groups of students are contained in Table 1 below, where each assignment was worth a total of 20 points.

The third row of the table shows that as the semester progressed about one-sixth of the Excel users switched to Minitab, under the impression that it was not only faster and easier to work in Minitab, but that their grade would also improve. However, the first row of the table reveals that mean student scores tended to differ by a single point or less, except for the first assignment.

Table 1: Mean scores among Mintab and Excel users, by assignment, along with 2-sample tests for differences in the mean (at a = .05).

	Assignment. 1		Assignment. 2		Assignment. 3		Assignment. 4		Assignment. 5
	Mtb	Excel	Mtb	Excel	Mtb	Excel	Mtb	Excel	Mtb	Excel
Sample Mean	16.65	13.10	16.84	15.97	18.36	17.58	14.71	15.00	17.22	16.14
Sample Variance	5.87	26.60	7.79	7.97	3.38	5.88	7.62	13.39	6.97	12.48
Number of Students	42	35	46	33	45	31	45	34	49	29
Hypothesized Mean Diff.	0		0		0		0		0
Degrees of Freedom	46		69		53		59		47
t Statistic	3.748		1.353		1.505		-0.385		1.436
P-Value	0.0005		0.1804		0.1382		0.7017		0.1576
t critical value (2-tail)	2.013		1.995		2.006		2.001		2.012

The bottom half of Table 1 shows the results of doing 2-sample tests for no difference in the mean scores received by Minitab and Excel users. On assignments 2-5, there was no significant difference at a = .05 in the mean scores of Minitab and Excel users. The first assignment was difficult for Excel users because it required them to construct and accurately interpret histograms, which is definitely a bit tricky on one's first attempt. By the second assignment, which was similar to the first, Excel users had caught on and essentially pulled even with Minitab users.

While I did try to grade both groups of assignments as consistently as possible, the experiment conducted here was neither double nor single blind. Students were allowed to pick their software package, so the results could reflect some kind of self-selection bias, perhaps with the more grade-conscious students electing to use Minitab. However, I know that there were both strong and weak students in each group. Overall, I feel the data indicate that there really was little difference in the mean scores of the two groups of students, except for the first assignment.

Perhaps the more telling statistic is that the variance of the Excel scores on each assignment was higher than that of the Minitab users, reflecting the greater degree of difficulty in getting adjusted to the Excel environment, e.g., accurately reading the variable axis of a histogram. In many settings Excel requires more effort from the user to get the same results easily achieved with Minitab; apparently, some Excel users never quite made the required effort.

4. Conclusion

Some of the strengths and weaknesses of these two packages are summarized in Table 2 below. After weighing these, I'm still not sure what I'll do the next time I teach the course. On the one hand, I liked being able to offer students their choice of software packages. Several students said that they appreciated the opportunity to learn a new package. On the other hand, a fair amount of class time was lost in explaining two packages, when most students really only wanted to know about one of them. Given the limited amount of class time that I have to teach the underlying statistical concepts, I'm leaning toward using just Excel in the course next time around. Though it will require a little more effort from the students, I think it will give the class more focus and the students a solid introduction to an extremely powerful and prevalent analytical tool.

Table 2: Pros and cons of using Minitab and Excel for an introductory statistics class.

Minitab Pro	Excel Con
Easier for the novice to use - can do most tasks with just a few clicks. Uses common statistical terminology. Generates a default histogram without user-specified intervals. Performs almost all inferences discussed in a first statistics course.	Harder for the novice to use - often must enter cell formulas and cell ranges. Occasionally misuses terminology, e.g., checking the box for "confidence level for the mean" yields the CI's margin of error. Harder to get a histogram, e.g., user must set up a column of "bins." Doesnt have built-in procedures for confidence intervals and 1-sample t-tests (although these can be performed).
Excel Pro	Minitab Con
Interface dialogs are consistent and have a modern look and feel. Flexible. Easy to control and modify the location of input data and output. Graphs are dynamically updated, and can be embedded in Word files. Wide use in business and other courses.	Function interfaces are not consistent; many features are buried underneath several layers of dialog boxes. Somewhat inflexible. Can't control location of output; hard to modify graphs. Graphs are static: if the data change, the graphing process must be restarted. Limited use in business, other courses.

References:

Moore, David S., The Basic Practice of Statistics, Second Edition, W. H. Freeman, New York, 2000.

Carver, Robert, Doing Data Analysis with Minitab 12, Duxbury Press, Pacific Grove, CA., 1999.

Berk, Kenneth N. and Patrick Carey, Data Analysis with Microsoftš Excel, Duxbury Press, Pacific Grove, CA., 2000.