Database Reference
In-Depth Information
SIDEBAR: EXPERIMENTAL DESIGN TO OVERCOME THE LIMITS OF
EXCEL—cont'd
age group, 9/28, 4/26, 7/26, 11/23, 9/23). And, let us not forget another Excel restriction mentioned in
Chapter 7. All factors must be ixed factors—although that is usually the case in UX applications, except
for the aforementioned “within-subjects” or “repeated measures” designs discussed in Chapter 7.
The only recourse you have if you are using Excel is to “sort” the data and arrive at the restric-
tive conditions required. For example, if you examine
Table 8.2
, you will note that each age group
has at least four males and four females (age group 2 has only four males). So, you can get to the
conditions needed by Excel by considering eight people in each age group, four males and four
females. You should choose the four males (in the other age groups), and the four females in all the
age groups, randomly. You would now have ive age groups and two genders and each of the 10
combinations (“cells”) contains the same number of people: four.
However, it is worth discussing this point a bit more. The above paragraph assumes
that the data have already been collected. There was an effort to make the sample size
in each age group the same or nearly so (as they are);
no attention was paid to gender
. Indeed,
you were
unlucky
that the minimum number of males in any column was only 4, so that
meeting the requirements for Excel leads to using only 40 (10 age/gender combinations times
4) of the 126 data points. Nobody likes to “waste” data!! Had there been, for example, eight
males in group 2, and that was the minimum number of males in any of the ive age groups,
then you would have been able to use 80 (10 combinations of 8 data points each) of the 126
data points.
That brings us to the issue of “designing” the data set. If you had realized earlier that gender
was important, and had a budget for roughly the same number of data points—126, then perhaps
you could have
chosen
to have 10 combinations of 12 people each, thus utilizing 120 data points. Of
course, this supposes that you have the ability to control the age and gender mix of your participants
in the study, and are OK with the potential added expense of carefully arranging the participant mix
to be balanced in this way. This idea of “designing” your sample is a precursor of the entire ield
of “design of experiments,” a subject that is beyond the scope of this topic, but is a worthy topic. If
you wish to follow up on this topic, we recommend the text,
Experimental Design
,
with Applica-
tions in Management, Engineering and the Sciences
, by Berger and Maurer (2002), referenced at
the end of Chapter 6.
First we'll display our original data of sophistication and age from Chapter 6.
Remember that we actually have 126 rows of data and
Figure 8.1
is showing only
some of them (since they would not all it on one screen shot).
We now add a column of data relecting gender. We will use “0” for male and “1”
for female. (Which two numbers we use, and which gender gets which number, does
not matter, as long as you remember [or write it down!] which number means which
gender.) This is shown in
Figure 8.2
.
Again, rows 20-42 out of the 126 rows are showing. We now pull down “Ana-
lyze” and go again to “General Linear Model” and sub menu “Univariate,” as shown
in
Figure 8.3
(see arrows).
The resulting dialog box, “Univariate,” is shown in
Figure 8.4
.
We now bring “Sophistication,” our dependent measure, over to “Dependent Vari-
able,” and bring both of our factors over to “Fixed Factors.” The age and gender variables
are clearly “ixed factors”—neither will (in fact, can't!!) be extrapolated or interpolated—
there are no other genders, and there are no age groups of interest beside the ive being
studied, and there are no gaps between any of the age groups. This produces
Figure 8.5
.
Search WWH ::
Custom Search