Module 4
Module 4
Data Analysis
Contents
4.1 Introduction.............................................................................................
4.2 Management Problems in Data Analysis .............................................
4.3 Guidelines for Data Analysis ..................................................................
Learning Summary .........................................................................................
Learning Objectives
By the end of this module the reader should know how to analyse data systematical-
ly. The methodology suggested is simple, relying very much on visual interpretation,
but it is suitable for most data analysis problems in management. It carries implica-
tions for the ways information is produced and used.
4.1 Introduction
What constitutes successful data analysis? There is apparently some uncertainty on
this point. If a group of managers are given a table of numbers and asked to analyse
it, most probably they will ‘number pick’. Individual numbers from somewhere in
the middle of the table which look interesting or which support a long-held view
will be selected for discussion. If the data are profit figures, remarks will be made
such as: ‘I see Western region made £220 000 last year. I always said that the new
cost control system would work.’ A quotation from Andrew Lang, a Scottish poet,
could be applied to quite a few managers: ‘He uses statistics as a drunken man uses
lamp posts – for support rather than illumination.’
Real data analysis is concerned with seeking illumination, not support, from a set
of numbers. Analysis is defined as ‘finding the essence’. A successful data analysis
must therefore involve deriving the fundamental patterns and eliciting the real
information contained in the entire table. This must happen before sensible remarks
can be made about individual numbers. To know whether the cost system in the
above example really did work requires the £220 000 to be put in the context of
profit and cost patterns in all regions.
Module 4 / Data Analysis
The purpose of this module is to give some guidelines showing how illumination
might be derived from numbers. The guidelines give five steps to follow in order to
find what real information, if any, a set of numbers contains. They are intended to
provide a framework to help a manager understand the numbers he or she encoun-
ters.
One might have thought that understanding numbers is what the whole subject
of statistics is about, and so it is. But statistics was not developed for use in man-
agement. It was developed in other fields such as the natural sciences. When it is
transferred to management, there is a gap between what is needed and what
statistics can offer. Certainly, many managers, having attended courses or read
books on statistics, feel that something is missing and that the root of their problem
has not been tackled. This and other difficulties involved in the analysis of manage-
ment data will be pursued in the following section, before some examples of the
types of data managers face are examined. Next, the guidelines, which are intended
to help fill the statistics gap, will be described and illustrated. Finally, the implica-
tions of this gap for the producers of statistics will be discussed.
rather like reading. When looking at a business report, a manager will usually
read it carefully, work out exactly what the author is trying to say and then decide
whether it is correct. The process is similar with a table of numbers. The data
have to be sifted, thought about and weighed. To do this, good presentation (as
stressed in Module 3 in the rules for data presentation) may be more important
than sophisticated techniques. Most managers could do excellent data analyses
provided they had the confidence to treat numbers more like words. It is only
because most people are less familiar with numbers than words that the analysis
process needs to be made more explicit (via guidelines such as those in Section
4.3 below) in the case of numbers.
(c) Over-complication by the experts. The attitude of numbers experts (and other
sorts of experts as well) can confuse managers. The experts use jargon, which is
fine when talking to their peers but not when talking to a layperson; they try
sophisticated methods of analysis before simple ones; they communicate results
in a complicated form, paying little regard to the users of the data. For example,
vast and indigestible tables of numbers, all to five decimal places, are often the
output of a management information system. The result can be that the experts
distance themselves from management problems. In some companies specialist
numbers departments have adopted something akin to a research and develop-
ment role, undertaking solely long-term projects. Managers come to believe that
they have not the skills to help themselves while at the same time believing that
no realistic help is available from experts.
Accounting Data
In Module 3 Table 3.12 showed the income statement of a multinational shipping
company. It is difficult to analyse (i.e. it is difficult to say what the significant
features of the company’s business were). Some important happenings are obscured
in Table 3.12, but they were revealed when the table was re-presented in Table 3.13.
Module 4 / Data Analysis
MONTH CUMULATIVE
TERMINAL COSTS
ESTIMATE STANDARD VARIANCE VAR % ESTIMATE STANDARD VARIANCE VAR % BUDGET
LO-LO
STEVEDORING
STRAIGHT TIME - FULL 131 223 143 611 1 288 8.6 1 237 132 1 361 266 124 134 9.1 1 564 896
STRAIGHT TIME - M.T. 13 387 14 651 1 264 8.6 256 991 281 399 24 408 8.7
(UN)LASHING 78 (78) 78 (78)
SHIFTING 801 (801) 11 594 (11 594)
OVERTIME, SHIFT TIME OF
WAITING & DEAD TIME 7 102 (7 102) 190 620 (190 620)
RO-RO
STEVEDORING
TRAILERS
STRAIGHT FULL 20 354 26 136 5 782 22.1 167 159 215 161 48 002 22.3 330 074
STRAIGHT M.T. 178 228 50 21.9 14 846 18 993 4 147 21.8
RO-RO COST PLUS
VOLVO CARGO
ROLLING VEHICLES 14 326 19 515 5 189 26.6 98 210 157 163 58 951 37.5
BLOCKSTONED 29 27 (2) (7.4) 613 674 61 9.1
(UN) LASHING RO-RO 355 (355) 355 (355)
SHIFTING 977 (977) 3 790 (3 790)
OVERTIME, SHIFT TIME OF
WAITING & DEAD TIME 1 417 (1 417) (28 713) (28 713)
HEAVY LIFTS (OFF STANDARD) 2 009 (2 009)
CARS
STEVEDORING
STRAIGHT TIME 6 127 6 403 276 4.3 38 530 35 328 (3 202) (9.1) 168 000
(UN) LASHING 2 (2)
SHIFTING 795 (795) 1 288 (1 288)
OVERTIME, SHIFT TIME OF
WAITING & DEAD TIME 7 573 (7 573)
OTHER SHIPSIDE OF COSTS 3 422 (3 422) 24 473 (24 473)
TOTAL TERMINAL COSTS 200 571 210 571 16 000 4.5 2 083 976 2 069 984 (13 992) (.7) 2 062 970
Module 4 / Data Analysis
Market Research
Table 4.2 indicates what can happen when experts over-complicate an analysis. The
original data came from interviews of 700 television viewers who were asked which
British television programmes they really like to watch. The table is the result of the
analysis of this relatively straightforward data. It is impossible to see what the real
information is, even if one knows what correlation means. However, a later and
simpler analysis of the original data revealed a result of wide-ranging importance in
the field of television research. (See Ehrenberg, 1975, for further comment on this
example.)
In all three examples any message in the data is obscured. They were produced by
accountants, computer scientists and statisticians respectively. What managers
would have the confidence to fly in the face of experts and produce their own
analysis? Even if they had the confidence, how could they attempt an analysis? The
guidelines described below indicate, at a general level, how data might be analysed.
They provide a starting point for data analysis.
Module 4 / Data Analysis
The re-presentation being recommended does not refer just to data that are liter-
ally a random jumble. On the contrary, the assumption is that the data have already
been assembled in a neat table. Neatness is preferable to messiness but the patterns
may still be obscured. When confronted by apparent orderliness one should take
steps to re-present the table in a fashion which makes it easy to see any patterns
contained in it. The ways in which data can be rearranged were explored in detail in
the previous module. Recall that the seven steps were:
(a) Round the numbers to two effective figures.
(b) Put rows and columns in size order.
(c) Interchange rows and columns where necessary.
(d) Use summary measures.
(e) Minimise use of gridlines and white space.
(f) Make the labelling clear and do not allow it to hinder comprehension of the
numbers.
(g) Use a verbal summary.
encapsulate every nuance of reality, but for some time they have been able to
summarise and predict reality to a generally acceptable level of approximation. In
management the objective is usually no more than this.
Only if the simple approach fails are complex methods necessary, and then ex-
pert knowledge may be required. As a last resort, even if the numbers are random
(random means they have no particular pattern or order), this is a model of a sort
and can be useful. For example, the fact that the day-to-day movements in the
returns from quoted shares are random is an important part of modern financial
theory.
comparison. The other results may be from another year, from another company,
from another country or from another analyst. In other words, reference can usually
be made to a wider set of information. In consequence, questions may be prompted:
Why is the sales mix different this year from the previous five? Why do other
companies have less brand switching for their products? Why is productivity higher
in the west of Germany? Making comparisons such as these provides a context in
which to evaluate results and also suggests the consistencies or anomalies which
may in turn lead to appropriate management action.
If the results coincide with others, then this further establishes the model and
may mean that in future fewer data may need to be collected – only enough to see
whether the already established model still holds. This is especially true of manage-
ment information systems where managers receive regular printouts of sets of
numbers and they are looking for changes from what has gone before. It is more
efficient for a manager to carry an established model from one time period to the
next rather than the raw data.
Example: Consumption of Distilled Spirits in the USA
As an example of an analysis of numbers that a manager might have to carry out,
consider Table 4.3 showing the consumption of distilled spirits in different states of the
USA. The objective of the analysis would be to measure the variation in consumption
across the states and to detect any areas where there were distinct differences. How
can the table be analysed and what information can be gleaned from it? The five stages
of the guidelines are followed.
Stage 1: Reduce the data. Many of the data are redundant. Are percentage fig-
ures really necessary when per capita figures are given? It is certainly possible, with
some imaginative effort, to conceive of uses of percentage data, but they are not
central to the purposes of the table. It can be reduced to a fraction of its original
size without any loss of real information.
Stage 2: Re-present. To understand the table more quickly, the numbers can be
rounded to two effective figures. The original table has numbers, in places, to eight
figures. No analyst could possibly make use of this level of specification. What con-
clusion would be affected if an eighth figure were, say, a 7 instead of a 4? In any
event, the data are not accurate to eight figures. If the table were a record docu-
ment (which it is not) then more than two figures may be required, but not eight.
Putting the states in order of decreasing population is more helpful than alphabetical
order. Alphabetical order is useful for finding names in a long list, but it adds nothing
to the analysis process. The new order means that states are just as easy to find.
Most people will know that California has a large population and Alaska a small one,
especially since no one using the table will be totally ignorant of the demographic
attributes of the USA. At the same time, the new order makes it easy to spot states
whose consumption is out of line with their population.
The end result of these changes, together with some of a more cosmetic nature, is
Table 4.4. Contrast this table with the original, Table 4.3.
Module 4 / Data Analysis
Alaska 46 47 1 391 172 1 359 422 2.3 0.33 0.32 3.64 3.86
Arizona 29 30 4 401 883 4 144 521 6.2 1.03 0.98 1.94 1.86
Arkansas 38 38 2 534 826 2 366 429 7.1 0.60 0.56 1.20 1.12
California 1 1 52 529 142 52 054 429 0.9 12.33 12.32 2.44 2.46
Colorado 22 22 6 380 783 6 310 566 1.1 1.50 1.49 2.47 2.49
Connecticut 18 18 7 194 684 7 271 320 (−1.1) 1.69 1.72 2.31 2.35
Delaware 45 43 1 491 652 1 531 688 (−2.6) 0.35 0.36 2.56 2.65
Dist. of 27 27 4 591 448 4 828 422 (−4.9) 1.08 1.14 6.54 6.74
Columbia
Florida 4 4 22 709 209 22 239 555 1.7 5.33 5.28 2.70 2.67
Georgia 13 13 10 717 681 9 944 846 7.8 2.52 2.35 2.16 2.02
Hawaii 41 40 2 023 730 1 970 089 2.7 0.48 0.47 2.28 2.28
Illinois 3 3 26 111 587 26 825 876 (−2.7) 6.13 6.35 2.33 2.41
Indiana 19 20 7 110 382 7 005 511 1.5 1.67 1.66 1.34 1.32
Kansas 35 35 2 913 422 2 935 121 (−0.7) 0.68 0.70 1.26 1.29
Kentucky 26 26 4 857 094 5 006 481 (−3.0) 1.14 1.19 1.42 1.47
Louisiana 21 21 7 073 283 6 699 853 5.6 1.66 1.59 1.84 1.77
Maryland 12 12 10 833 966 10 738 731 0.9 2.54 2.54 2.61 2.62
Massachusetts 10 10 13 950 268 14 272 695 (−2.3) 3.28 3.38 2.40 2.45
Minnesota 15 15 8 528 284 8 425 567 1.2 2.00 1.99 2.15 2.15
Missouri 20 17 7 074 614 7 697 871 (−7.9) 1.66 1.82 1.48 1.61
Nebraska 36 36 2 733 497 2 717 859 0.6 0.64 0.64 1.76 1.76
Nevada 30 31 4 360 172 4 095 910 6.5 1.02 0.97 7.15 6.92
New Jersey 8 8 15 901 587 16 154 975 (−1.6) 3.73 3.82 2.17 2.21
New Mexico 42 41 1 980 372 1 954 139 1.3 0.47 0.46 1.70 1.70
New York 2 2 41 070 005 41 740 341 (−1.6) 9.64 9.88 2.27 2.30
North Dakota 47 46 1 388 475 1 384 311 0.3 0.33 0.33 2.16 2.16
Oklahoma 33 29 3 904 574 4 187 527 (−6.8) 0.92 0.99 1.41 1.54
Rhode Island 39 39 2 073 075 2 131 329 (−2.7) 0.49 .50 2.24 2.30
South 23 25 5 934 427 5 301 054 11.9 1.39 1.26 2.08 1.88
Carolina
South Dakota 48 48 1 312 160 1 242 021 5.6 0.31 0.29 1.91 1.82
Tennessee 24 24 5 618 774 5 357 160 4.9 1.32 1.27 1.33 1.28
Texas 5 6 17 990 532 17 167 560 4.8 4.22 4.06 1.44 1.40
Wisconsin 11 11 10 896 455 10 739 261 1.5 2.56 2.54 2.36 2.33
Total licence 319 583 215 317 874 435 0.5 75.04 75.22 2.13 2.13
Module 4 / Data Analysis
Stage 3: Build a model. The pattern is evident from the transformed table. Con-
sumption varies with the population of the state. Per capita consumption in each
state is about equal to the figure for all licence states with some variation (±30 per
cent) about this level. The pattern a year earlier was the same except that overall
consumption increased slightly (1 per cent) between the two years. Refer back to
Table 4.3 and see if this pattern is evident even when it is known to be there. There
may of course be other patterns but this one is central to the objectives of the anal-
ysis.
Stage 4: Exceptions. The overall pattern of approximately equal per capita con-
sumption in each state allows the exceptions to be seen. From Table 4.4, three
states stand out as having a large deviation from the pattern. The states are District
of Columbia, Nevada and Alaska. These states were exceptions to the pattern in the
earlier year as well. Explanations in the cases of District of Columbia and Nevada are
readily found, probably being to do with the large non-resident populations. People
live, and drink, in these states who are not included in the population figures (diplo-
mats in DC, tourists in Nevada). An explanation for Alaska may be to do with the
lack of leisure opportunities. Whatever the explanations, the analytical method has
done its job. The patterns and exceptions in the data have been found. Explanations
are the responsibility of experts in the marketing of distilled spirits in the USA.
Stage 5: Comparison. A comparison between the two years is provided by the
table. Other comparisons will be relevant to the task of gaining an understanding of
the USA spirits market. The following data would be useful:
(i) earlier years, say, five and ten years before;
(ii) a breakdown of aggregate data into whisky, gin, vodka, etc.;
(iii) other alcoholic beverages: wine, beer, etc.
Once data from these other sources have been collected they would be analysed in
the manner described, but of course the process would be shorter because the
pattern can be anticipated. Care would need to be taken that like was being com-
pared with like. For example, it would have to be checked that an equivalent
definition of consumption was in force ten years earlier.
The second implication is more direct. Data should be presented in forms which
enable them to be analysed speedily and accurately. Much of the reduction and re-
presentation stages of the guidelines could, in most instances, be carried out just as
well by the producer of the data as by the user. It would then need to be done only
once rather than many times by the many users of the data. Unfortunately, when
time is spent thinking about the presentation of statistics, it is usually spent in
making the tables look neat or attractive rather than making them amenable to
analysis.
There is much that the producers of data can do by themselves. For example,
refer back to the extract from a management information system shown in Ta-
ble 4.1: if thought were given to the analysis of the data through the application of
the guidelines, a different presentation would result (see Table 4.5).
(a) Some data might be eliminated. For instance, full details on minor categories of
expenditure may not be necessary. This step has not been taken in Table 4.5
since full consultation with the receivers would be necessary.
(b) The table should be re-presented using the rules of data presentation. In
particular, some rounding is helpful. This is an information document, not an
auditing one, and thus rounding is appropriate. In any case, no different conclu-
sions would be drawn if any of the expenditures were changed by one unit. In
addition, improvement is brought about by use of summary measures and a
clearer distinction between such measures and the detail of the table.
(c) A model derived from previous time periods would indicate when changes were
taking place. There is a good case for including a model or summary of previous
time periods with all MIS data. This has not been done for Table 4.5 since previ-
ous data were not available.
(d) Exceptions can be clearly marked. It is, after all, a prime purpose of budget data
to indicate where there have been deviations from plan. This can be an automat-
ic process. For example, all variances greater than 10 per cent could be marked.
This might even obviate the need for variance figures to be shown.
(e) The making of comparisons is probably not the role of the data producer in this
example, involving as it does the judgement of the receivers in knowing what the
relevant comparisons are. The task of the producer has been to facilitate these
comparisons.
Making the suggested changes does of course have a cost attached in terms of
management time. However, the cost is a small fraction of the cost of setting up and
operating the information system. The changes can transform the system and make
it fully operational. If an existing system is being largely ignored by managers, there
may be no alternative.
Module 4 / Data Analysis
Table 4.5 Budgeting data from an MIS (amended from Table 4.1)
Port: Liverpool OCEAN PORT TERMINAL COSTS – SHIPSIDE OPERATIONS
Period: December (U.S Dollars: Conversion rate 1.60)
MONTH CUMULATIVE
ESTIMATE STANDARD VARIANCE VAR % ESTIMATE STANDARD VARIANCE VAR % BUDGET
LO-LO: Stevedoring (STR-FULL) 131 000 144 000 12 000 9 1 240 000 1 360 000 120 000 9 1 560 000
Stevedoring (STR-MT) 13 400 14 700 1 300 9 257 000 281 000 24 000 9
Unlashing 78 0 −78 * 78 0 −78 *
Shifting 200 0 −200 * 12 000 0 −12 000 *
Overtime, etc. 7 100 0 −7 100 * 191 000 0 −191 000 *
RO-RO: Stevedoring TR (STR-FULL) 20 400 26 100 5 800 22 167 000 215 000 48 000 22 330 000
Stevedoring TR (STR-MT) 180 230 50 22 15 000 19 000 4 100 22
Stevedoring cost plus 0 0 0 0 0 0 0 0
Stevedoring Volvo 0 0 0 0 0 0 0
Stevedoring rolling 14 300 19 500 5 200 27 98 000 157 000 59 000 37
Stevedoring blockstow 29 27 −2 −7 610 670 60 9
Unlashing 350 0 −350 * 350 0 −350 *
Shifting 980 0 −980 * 3 800 0 −3 800 *
Overtime, etc. 1 400 0 −1 400 * 29 000 0 −29 000 *
Heavy lifts 0 0 0 0 2 000 0 −2 000 *
CARS: Stevedoring (STR) 6 100 6 400 280 4 38 000 35 000 −3 000 −9 170 000
Unlashing 0 0 0 0 2 0 −2 *
Shifting 800 0 −800 * 1 300 0 −1 300 *
Overtime, etc. 0 0 0 0 7 600 0 −7 600 *
TOTALS: LO-LO 152 000 158 000 5 900 4 1 700 000 1 640 000 −59 000 4
RO-RO 38 000 46 000 8 300 18 315 000 392 000 77 000 20
CARS 6 900 6 4001 −520 −8 47 000 35 000 −12 000 −34
OTHER 3 400 0 −3 400 * 24 000 0 −24 000
GRAND 201 000 211 000 10 000 5 2 080 000 2 070 000 −14 000 −1 2 060 000
TOTAL
Totals may not agree because of rounding.
*Zero standard cost, therefore variance not calculable.
Module 4 / Data Analysis
Few managers will not admit that there is currently a problem with the provision
and analysis of data, but they rarely tell this to their IT systems manager or whoever
sends them data. Without feedback, inexpensive yet effective changes are never
made. It must be the responsibility of users to criticise constructively the form and
content of the data they receive. The idea that computer scientists/statisticians
always know best, and if they bother to provide data then they must be useful, is
false. The users must make clear their requirements, and even resort to a little
persistence if alterations are not forthcoming.
Learning Summary
Every manager sees the problem of handling numbers differently because each sees
it mainly in the (probably) narrow context with which he or she is familiar in his or
her own work. One manager sees numbers only in the financial area; another sees
them only in production management. The guidelines suggested here are intended
to be generally applicable to the analysis of business data in many different situa-
tions and with a range of different requirements. The key points are:
(a) In most situations managers without statistical backgrounds can carry out
satisfactory analyses themselves.
(b) Simple methods are preferable to complex ones.
(c) Visual inspection of well-arranged data can play a role in coming to understand
them.
(d) Data analysis is like verbal analysis.
(e) The guidelines merely make explicit what comes naturally when dealing with
words.
The need for better skills to turn data into real information in managerial situa-
tions is not new. What has made the need so urgent in recent times is the
exceedingly rapid development of computers and associated management infor-
mation systems. The ability to provide vast amounts of data very quickly has grown
enormously. It has far outstripped the ability of management to make use of the
data. The result has been that in many organisations managers have been swamped
with so-called information which in fact is no more than mere numbers. The
problem of general data analysis is no longer a small one that can be ignored. When
companies are spending large amounts of money on data provision, the question of
how to turn the data into information and use them in decision making is one that
has to be faced.