Matrix Data Analysis Chart (MDAC)

The Matrix Data Analysis Chart (MDAC) is used to identify clusters of related
items within a larger group.

• When to use it: Situations when it is useful.

•
How to understand it : Details of how it works.
• Examples : Some examples of usage.
• How to do it : Step-by-step instructions on using it.
• Practical variations : Variants and variation.

Notes

The Matrix Data Analysis Chart is one of the second seven tools according to
Mizuno, although others replace it with the Prioritization Matrix. It is often
abbreviated to MDAC.

The complexity of the statistical method normally associated with this diagram
( Principal Component Analysis) has led to other books not discussing MDAC and
offering an alternative 'seventh' tool. It is included here not just for completeness, but
because practical use can still be made of it.

When to use it

• Use it when investigating factors which affect a number of different items, to

determine common relationships.
• Use it to determine whether or not logically similar items also have similar
factor effects.
• Use it to find groups of logically different items which have similar factor
effects.

Fig. 1. Using MDAC in problem solving

How to understand it

When comparing a large set of items, the complexity of the situation can make it
difficult to determine how different factors relate to one another. In particular, it can
be useful to find groups of items that behave in similar ways. For example, a washing
powder may have different efficiencies at achieving 'softness' and 'stain removal' in

garments made of acrylic, polyester, wool and various fiber mixtures. If similar
affects are found in a group of fibers, then changing the powder ingredients may
affect the whole group in a similar way.

The Matrix Data Analysis Chart (or MDAC) helps classify items by identifying two
major characteristics common to all items and then plotting each item as a point on a
standard x-y chart. This makes it easier to see how the individual items relate both to
the characteristics and to one another, thus:

Fig. 2. MDAC plot

Identifying the best characteristics to measure is an important task, as different sets

of measurements can give very different charts. It would be useful to be able to
compare many characteristics together, such as the density, color, texture, strength,
etc. of cement mixtures, but we are constrained by the two dimensions available on
paper. There are mathematical methods for combining multiple factors, but these are
beyond the scope of this book.

A key interpretation point about an MDAC is to consider how points on the chart
group together or form into clusters (this may be contrasted with the Scatter Diagram,
which looks for linear trends). This interpretation is helped by highlighting significant
groups of points with linear links, as in Fig. 3.
 Fig. 3. Clustering

Typical items of interest on an MDAC include:

• The behavior of logical groups of items, which might be expected to form

close clusters. For example, in a washing powder test, logical groups might be
woolen items, acrylics and mixtures.

• Actual clustering on the chart which might highlight divergence from expected
behavior, and prompt new actions. For example, investigation of an
unexpected cluster of different fiber types might show that they come from
one manufacturer who has developed processes to give different fibers with
similar specifications.

Example

A toy store was aiming to increase sales while improving the satisfaction of its
customers with the toys that it sold. As a part of this, it employed a market research
company to measure both the initial appeal (which related to actual purchase) and the
longer term satisfaction (which related to company image) of a range of toys for boys
aged 5 to 10, both being scored on a one-to-ten scale. This limited sector was chosen
to prevent excessive complexity and confusion in the analysis.

These were plotted on a matrix to identify the best toys to promote and to find
possible ways of improving other toys. The axes were crossed at their mid-points to
form value quadrants, as illustrated below.
 As a result, improved packaging and promotion was sought for the better
construction toys, in order to increase initial appeal, some bottom end toys were
dropped, and the results of the survey were published in a form which customers
could easily understand. The result was an increase in the reputation of the store as
putting customer interests first, as evidenced by the increase in complimentary letters.

Fig. 4. MDAC example

Other examples

• A vineyard, aiming to increase the consistency of the quality of its wines

measures a 'quality rating' along with a range of different other factors, such as
grape, additives, storage, etc. It then uses MDACs to isolate clusters of factors
that contribute to its finest wines.
• A pharmaceutical combine examines the pain-killing drugs of its subsidiaries
in terms of the cost to product and general efficacy. Products which are high
cost but are not of highest efficacy are dropped. Low-cost drugs of reasonable
efficacy are promoted, and high-cost drugs have a project initiated to reduce
production cost.
 • A production unit, looking for alternative materials to build a more durable
gear casing, compares key attributes of available materials using cost and
durability.

How to do it

1. Identify the items which are to be compared, and dec ide on the pri mary
objective of using the Matrix Data Analysis Chart. For example, a restaurant
may compare menu items with the objective of finding out what makes
popular dishes.

2. Identify the measurement units for the horizontal and vertical axes of the
chart. These shoul d be two factors which are the most critical representations
of the objective identified in step 1. For example, the restaurant may use a
survey to find the aspects of eating that their customer most value.

3. Measure the fact ors identified in step 2, aimi ng to get realis tic and unb iased
values. Thus, the restaurant might take an average of customer ratings for
texture and flavor of each menu item over several months.

4. Draw a chart and plot each point on it. Ensure that the scale on the axes results
in the points being spread over the whole chart area.

5. Look for significant clusters of poin ts on the chart, and highlight them by
linking them together into a ring. The appearance and ease of interpretation
are more important than the order of linkage. Groups may be either of:

• Items that have a close logical relationship, e.g. fish dishes.

• Items that form a close physical group on the chart.

6. Interpret the chart and act on the results. Typical activities include the
investigation into and subsequent action on:

• Why items which might be expected to group closely do not.

• Why items unexpectedly form clusters.
• Why individual items are not positioned where they were expected to be on
the chart.
Practical variations

• When multiple measurements are made, such as texture, color, taste, etc., a
number of MDACs can be drawn and then evaluated as a group, looking for
similar clusters in each chart.

This can only practically be done on a limited scale, as multiple MDACs can
be difficult to interpret. For example, with three measurements, A, B and C,
there may be three MDACs, A versus B, B versus C and A versus C. However,
with six measurements there could be fifteen MDACs!

• Another approach when there are multiple measurements is to combine similar

measurement sets. Thus if customers give similar marks to texture and color,
then they may be averaged, and the plotted characteristic now becomes
'texture and color'.

• The vertical and horizontal axes do not have to cross at their zero points. This
can be used to deliberately divide the chart into four areas, as below. If both
axes represent desirable characteristics, then the top right and bottom left
quadrants represent desirable and undesirable regions, respectively. The top
left and bottom right quadrants, where only one characteristic is desirable,
represent areas of opportunity for improvement of the other characteristic.

Fig. 5. Using quadrants