Capitulo II: Diagnóstico determinación de necesidades 2.1 Diagnóstico y determinación de necesidades.
NIVEL BAJO :
IV. El hombre como agente de cambio y conservación del medio ambiente.
2.4 Validación de los resultados
The previous section reviewed the general OR and simulation literature regarding
learning from the process of building a model. One problem with the empirical
studies in that literature is that it can be difficult to draw out decision maker
learning that occurs primarily from involvement in model building. This section
reviews the literature addressing model use and learning. Perhaps unsurprisingly, it
is easier to identify empirical studies primarily focussed on model use. This would
seem to be due to relative simplicity in the procedure for laboratory based studies
focussed on use compared to building.
This section, firstly, reviews the DES literature on model use. This ranges from
conceptual discussions of the benefits (or problems) associated with visual interac-
tion modelling (Van der Zee and Slomp, 2009; Chwif and Barretto, 2003; Belton
and Elder, 1994; Musselman, 1990) to empirical studies of learning and perceptions
(Tako and Robinson, 2009; Bell and O’Keefe, 1995). Following this is a review of the
SD literature. The SD literature exploring model use is much more closely linked to
the history of simulation gaming (Lane, 1995) and appears to be much more mature
than the field of DES. In fact, there are numerous well known experimental studies
investigating learning from simulation type board games (Sterman, 1989b,a) and simulation models (Paich and Sterman, 1993; Bakken et al., 1994), as well as re-
Rouwette et al., 2004; Gr¨oßler, 2004). The section closes with a summary of the
main points covered in the review.
2.3.1 DES Research
Belton and Elder (1994) argue, based on their combined experience, that use of
Visual Interactive Simulation (VIS) for simulation experimentation is beneficial for
discovery, clarification, change and creation of a client’s views and ideas about sys-
tem management. Indeed these benefits would seem to suggest alternative uses
for DES such as gaming and training (Van der Zee and Slomp, 2009; Chwif and
Barretto, 2003). The reasoning behind this is similar to that expressed by Lane
(1994). Decision makers are able to explore the consequences of their assumptions
and decisions while receiving fast and understandable feedback from the model.
This experience helps makes the tacit views they hold explicit and possibly reveals
inconsistencies or incompatibilities between them (Belton and Elder, 1994).
Bell and O’Keefe (1995) reiterate the argument for visual interactive simulation
and also provide an insight into an opposing view: namely the overemphasis of the
result from a single run and subjective based analysis (Musselman, 1990, quoted in
Bell and O’Keefe, 1995). Bell and O’Keefe point out that most proponents of VIS
will use a combination of batch and interactive experimentation, but also that the
majority of what is known about using VIS is anecdotal.
Hence to explore the proposed benefits of VIS, Bell and O’Keefe (1995) conduct
an experimental study investigating the results users of interactive experimentation
can achieve relative to a statistical analysis. Participants, namely MBA students,
are provided with a model based around a queuing and resource-allocation problem
at a mine. The model is simple and allows for a ‘correct solution’ to be found.
Participants read the case study the night before and provide an initial solution
a revised answer. Statistics on usage are collected throughout.
Findings show that the participant’s solutions were a substantial improvement
on their initial efforts. However, only 39% provided the correct solution. Analysis
of the usage statistics identified that the participants who made most use of the
animated display found the correct results most quickly. Furthermore, participants
that collected multiple data from a scenario (i.e. viewing the results in detail and
drilling into data) before proceeding to the next, performed best.
The most recent study was conducted by Tako and Robinson (2009): an empir-
ical comparison of the use of SD and DES models. Participants in the experiment,
executive MBAs, take the role of a government consulting service and use either
a SD or DES model of the UK prison population to explore and draw conclusions
on how the process could be improved. At the end of the experiment participants
fill in a questionnaire to elicit their perceptions of model understanding, complex-
ity, credibility and interpretation of results. Interestingly, the results showed little
difference between the measures. In fact, only slight differences were found in how
representative of the case study participants found the models (favouring the SD
model) and that the participants found the results of the DES model more difficult
to interpret.
2.3.2 SD Research
The literature exploring and testing learning from using SD models is arguably more
substantial and mature than the DES literature just discussed. The focus is largely
on the use of management flight simulators or SD business games; essentially using
simulation models as games to help the user, who may or may not be a manager,
learn about a specific concept. Several review papers exist to summarise the key
insights that have been discovered (Lane, 1995; Rouwette et al., 2004; Gr¨oßler, 2004).
back to the 1960’s. Business games appear to have had a mixed history with some
early successes, but also failures including an infamous paper by Neuhauser (1976)
declaring that it is the model design process, not use, where the majority of learning
is found. Rouwette et al. (2004) review only the empirical studies of model use found
within the SD literature. They provide an overview of the characteristics of models,
simulators (e.g. a simulation business game) and players that influence decision
making over time. A key finding, already discussed in the section on model building,
was that model transparency has a positive relation to performance. Gr¨oßler (2004)
reviews the methodological issues in using simulation as an educational tool and lists
15 main issues. These include overcomplicated models, accounting for the different
learning styles of participants and the lack of risk for participants’ decision making.
Turning to specific relevant case studies, Bakken et al. (1994) conduct an em-
pirical study concerned with transfer of learning using executive MBA and un-
dergraduate students. In the experiment MBAs, working in groups of two, and
undergraduate students, working individually, attempt to manage an SD simulation
model of either a real estate or oil tanker system. The underlying feedback structure
of these systems is identical. After a certain amount of time the participants use
the other simulation model. To infer transfer of learning Bakken et al. look at the
performance of participants using the second simulation model.
The results show that the students outperform the executive MBAs in the trans-
fer model. This is particularly interesting as a number of the MBAs have substantial
experience in one of the domains used in the experiment. To explain this result
Bakken et al. provide data that shows that undergraduate students are much more
exploratory than the executive MBAs in the first simulation model. In fact, the
undergraduate students make approximately double the amount of decisions com-
pared to the real world managers. This results in an increased bankruptcy rate in
tions. In contrast an underlying assumption of the executive MBAs with real world
experience appeared to be ‘I know a lot about this market; so I just do in the game
what I would do in real life’ (Bakken et al., 1994).
Learning problems are not just limited to managers using simulation games;
students also have difficulties. Paich and Sterman (1993) conduct an empirical
study where postgraduate students, including MBAs, use a SD simulation model to
manage a product from launch through to maturity. The objective of the game is
to minimise the boom and bust dynamics of the market using pricing and capacity
decisions. Participants use the model over five trials and make a total of 200 pricing
and capacity decisions. After each decision stage the simulation model provides
results to the participant on current performance as well as the history of decisions
and performance in the game. Furthermore, participants can take as long as they
need to make decisions.
The results of the study found that on average the participants performed better
over the five trials. However, the participants performed poorly relative to a naive
benchmark: mean performance (profit) was 60% of the benchmark with only 17%
of participants outperforming the benchmark in the final trial. This performance
is explained by the student learning the average demand for replacement of the
product in the market and matching their capacity to this area. However, they
completely ignored other aspects of the market, for example, the growth rate of
demand for new products. In fact their actions worsened the extent of the boom
and bust dynamics.
The results of Paich and Sterman (1993) support the misperception of feedback
hypothesis put forward by Sterman (1989b) that ‘the stronger the feedback process in the environment the worse people do relative to potential’. Paich and Sterman
(1993) conclude that in order for participants to learn when dynamic complexity is
2.3.3 Conclusions on Model Use and Learning
In summary, the literature on learning from the use of simulation models is mixed.
It would seem that there is some benefit from interactively using a simulation model
to improve on decision makers initial solutions. However, the search and results re-
viewing process that a decision maker uses can radically affect this outcome. Indeed,
some studies illustrate that learning can be quite poor relative to potential.
One possible problem with drawing conclusions from this literature review is
the lack of empirical studies using DES models and the prevalence of SD models.
Some confidence can be drawn from the Tako and Robinson (2009) study. This
demonstrated that users’ perceptions of using a specific SD and DES model is largely
the same. One issue may be that participants in the Tako and Robinson study
indicated that the DES results were more difficult to interpret. Hence learning
difficulties may be worse in a complex DES study.
Further to what has been discussed, an important aspect of learning from mod-
elling and model use is the credibility of the study, model and results with the
individuals or group who will actually make the decisions. This is the topic of the
final section of this chapter.