Business Intelligence/Test Causal Link Map

From Wikibooks, open books for an open world
Jump to: navigation, search


Business Intelligence
Framing the strategy Test Causal Link Map Create and disseminate scorecard


  • Purpose: Test the validity of the causal links in the strategy diagrams
  • Input: Strategy Diagrams
  • Activities: Turn the causal links into a theory and test hypotheses deduced from the theory
  • Outputs: Evidence demonstrating the validity of the theory behind the strategy diagrams
  • Documents: Validated Strategy Diagrams

Introduction[edit]

Accordingly, the process of documenting causality at the objective level requires a management team to become explicit about their understandings / beliefs about the reasons why achievement of one objective will influence another: conversely a good challenge to a Balanced Scorecard design is to test the objectives in different perspectives to see if the implied causality is plausible (Cobbold and Lawrie 2000b)

"In arguing that explanations proceed in terms of implicit conceptual models, this essay makes no claim that foreign policy analysts have developed any satisfactory, empirically tested theory. In this essay, the use of the term "model" without qualifiers should be read 'conceptual scheme.'" (Allison 1969)

Allison also makes an important point about unproven models (not empirically tested). He calls these a conceptual scheme. All decision makers bring implicit conceptual models to the table. This does not mean that it is impossible to test these models. We argue that building a Business Intelligence system using a framing approach requires turning implicit assumptions and implicit models into explicit assumptions and explicit models. We used strategy diagrams to make implicit conceptual models explicit. Next, we propose a means to test these models using causal link maps. Proving conceptual models turns implicit conceptual schemas into explicit strategic models.

Strategy diagrams are visual representations of a firm's strategy. They show how decision makers believe the company can manage customer outcomes in order to maximize profitability. The key to this chapter is understanding that the three types of diagrams contain links between activities, themes or investments and that these links are an explicit representation of cause and effect relationships that constitute the strategy. Each link represents a hypothesis. A theory, at its core, is a set of inter-related hypotheses that explain a phenomenon. Therefore, taken as a whole, the strategy diagram is a visual representation of a theory regarding how activities generate profitability.

Turning the diagram(s) into a theory and then testing the theory requires the following activities:

  • Testing a hypothesis
  • Testing a theory
  • Testing the strength of a theory

An untested strategy diagram may not be useful or could be counter-productive. That is why this is perhaps the most important step in building a strategy and therefore a Business Intelligence architecture.

Definitions[edit]

Before demonstrating how the strategy diagram is a theory it is necessary to present some definitions.

Hypothesis[edit]

A hypothesis consists either of a suggested explanation for an observable phenomenon (observable occurrence) or of a reasoned proposal predicting a possible causal correlation among multiple phenomena (different types of observable occurrences) [1]

Hypotheses are usually written in the "if-then form": If X, then Y. Or, X-→Y. (See Hypotheis)

Each causal link in the causal link map can be considered a hypothesis.

Theory[edit]

There are a few acceptable definitions of a theory. This book assumes that a theory is a group of hypotheses that have been supported with repeated testing.[2] Some argue that a theory is ONE tested and accepted hypothesis. However, it is just as acceptable to assume that a theory is a set of hypotheses.

A theory is an analytic structure (set of hypotheses) designed to explain a set of observations. A theory does two things:[ http://en.wikipedia.org/wiki/Theory#Pedagogical_definition]

  1. Identifies a set of distinct observations as a class of phenomena, and
  2. Makes assertions about the underlying reality (hypotheses) that brings about this class

A scientific theory can be thought of as a model of reality, and its statements as axioms (assumptions) of some axiomatic system. [3]

Mechanism[edit]

An assertion about the underlying reality regarding how X→Y. The testing of a hypothesis is a test of the validity of the mechanism.

Measure[edit]

Quantitative or qualitative data about a phenomenon. It is not possible to test underlying mechanisms. Rather, we create measures associated with the mechanism. For instance, it is not possible to test the correlation between two concepts. It is only possible to test the correlation between the measures of concepts. That is why effective measures and clean data are important.

Model[edit]

This book focuses on a scientific theory of strategy represented explicitly as a model (strategy diagram). "A model in is a physical, mathematical, or logical representation of a system of entities, phenomena, or processes. In essence a model is a simplified, abstract view of any complex reality. It may focus on particular views, enforcing the 'divide and conquer' principle for a compound problem. Formally a model is an interpretation which deals with empirical entities, phenomena, and physical processes in a mathematical, or logical way. " [4]

"A business process is a collection of related, structured activities or tasks that produce a specific service or product (serve a particular goal) for a particular customer or customers. It often can be visualized with a flowchart as a sequence of activities." [5]

A model is an abstract visual representation of the processes that link critical activities and resources to customer outcomes.

Strategic Theory[edit]

This section demonstrates that a strategy diagram is a theory.

Again, a strategy diagram is a causal link map (theory) with hypotheses (links) that form a theory (set of hypothesis). For instance, in the figure below, we are saying that more customers (X) leads to higher profitability (Y). Assume that costs (C) remain constant regardless of the number of items produced or purchased (A). Further, assume that each customer purchases the same number of items over a specific period of time (B). The more customers (X) the more items purchased and greater revenue (R), which will generate profits when revenue exceeds costs (Y).

Scorecard


Summarized the point is this (purposefully simplistic):

X→R
IF (R>C) →Y
Assume A and B.
→ means causes

The transition scorecard shows that the strategy map is really also a theory. This is because any item on a strategy map can be turned into a hypothesis. Demonstrating this will prove that the strategy map is a theory and demonstrate an important chapter activity.

Deducing testable hypotheses[edit]

Why is this important? Because the strategy diagrams constitute an important part of the BI frame. Reports allow the decision maker to understand the world by monitoring performance. Specifically, they allow the decision maker to ask:

  1. What happened?: Y1 versus Y2
  2. What may happen?: Xf -->Yf

Reports allow the decision maker to monitor performance by asking and answering these questions. However, the ability to effectively ask and answer these questions is based on testing the links in the causal map. For instance, the decision maker may want to know:

Using the model for monitoring performance requires knowing:

  • What happened?: Y1(actual performance) versus Y2 (prediction or expectation)

However, this assumes that X-->Y1. What if Y1 (actual measure) was lower than Y2 (what we expected)? We would assume that X was a problem and would take corrective measures. What if X-->Y1 was not only untested but false? We would be incorrect and changes to X would not necessarily have any impact on Y1.

Similarly, using the model for prediction requires asking:

  • What may happen?: Xf -->Yf

What if we tried to predict Yf but in reality Xf -->Yf was not only untested but false? When we tried to predict Yf we could be wrong.

An important step in building a BI system is to make sure that the strategy is correct. To do this requires deducing hypotheses in order to test the theory. While we could test X-->Y1 directly it is better to test as much of the theory as possible. Not every hypothesis can and needs to be tested.

Deducing hypotheses for testing requires taking causal links. Assume the strategy map shows X-->Y-->Z. We could use the direct (X-->Y) or indirect (X-->Z) link to formulate a hypothesis. Testing concepts that are closer on the maps is more fruitful but may be impractical, based on time or resource restrictions.

Testing Hypotheses[edit]

We will adopt hypothesis testing modified from an econometrics point of view (Gujarati 1999)

  1. Create a hypothesis
    • Why did it happen?: X-->Y
    • How did it happen?: X1-->Y2
  2. Collect data
  3. Specify the hypothesis, either in terms of a verbal or mathematical model
  4. Specifying the experimental or quasi-experimental model of theory or hypothesis (causal effects or mechanisms, discussed below)
  5. Estimating the parameters of the chosen model
  6. Checking for model adequacy: model specification testing
  7. Testing the hypothesis derived from the model

While this approach is generally more applicable to quantitative methods it does not preclude qualitative measures. Steps 5, 6 and 7 will differ depending on the model of explanation applied to the hypothesis.

This book presents two ways to test hypotheses. The first is the use of causal effects. The second is the use of causal mechanisms.

Before going forward a very short description of the philosophy of social science is in order. One approach to social scientific explanation is the covering-law model (Little 1991). For our purposes this model accepts the explanation of an event or regularity when it can be subsumed under the strategic theory. What we are trying to understand is how a phenomenon or regularity derives from general law. However, we are not working with laws but theories, which will have to do.

Deductive Models and Criticisms[edit]

One version of the covering-law model, called the deductive-nomological (D-N) model, requires us to ask: Why was the phenomenon to be explained necessary in the circumstance? (Little 1991 page 5). This type of explanation is deductive, meaning that the outcome is necessary given the events.

D-N model of explanation (see Little 1991 page 5)

Li (one or more universal laws)
Ci (one or more statements of background circumstances)
____________(deductively entails)
E (statement of the fact of regularity to be explained)

The difficulty with purely deductive models is that, for instance, they founder in the face of quantum mechanics (see George and Bennett). It is also unlikely that such a strict definition of causality would apply to business strategy. So, purely deductive models are not feasible.

Inductive Models and criticisms[edit]

There is an inductive approach of the covering law model. The inductive-statistical model describes a statistical explanation as consisting of one or more statistical generations, one or more statements of particular fact, and an inductive argument to the effect.

I-S model of explanation (see Little 1991 page 6)
Li (one or more statistical laws)
Ci (one or more statements of background circumstances)
==========(makes very likely)
E (statement of the fact of regularity to be explained)

Note the difference between an inductive and deductive argument. A deductive means that the truth of the premises guarantees the truth of the conclusion. An inductive argument means the truth of the premises makes the conclusion more probable but not always certain.

What are the criticisms of the inductive models, such as the I-S or other related explanations? Salmon argued that they put causality in a black box. This is true. So, purely inductive models will not work.

One way to approach the problem of explanation is focusing on the common believe that explanations share the assumption that there is an underlying mechanism by which X produces Y. This book proposes two, mutually supporting ways to test theories related (note that many more exist) that accept mechanisms:

  • Causal effects
  • Causal mechanisms

Causal Mechanisms and Causal Effects[edit]

The approach proposed in this book to both avoid the tedious arguments surrounding causality and provide for a pragmatic, realistic approach to testing causality by focusing on causal mechanisms and causal effects. A causal mechanism approach is similar to the deductive approach in that it attempts to use a theory to explain how X produced Y. However, attempts to demonstrate causal mechanisms generally need to be applied to a wider range of phenomenon than are tested using methods available for studying causal mechanisms. Causal effects, based on statistical methodologies, can help determine if the mechanism is applicable to a wider range of observations. Using causal effects and mechanisms creates a unified methodology where the weaknesses of one are augmented by the strengths of the other. So, a mixed approach is the preferred method.

Why? - Causal Effects[edit]

At the heart of causal effects is the correlation between two variables. Causal effects examines data to determine if the changes in one variable are linked with changes in another variable.

'Statistical correlation is explanatory to the extent that it provides evidence of a credible causal process underlying the variables being analyzed.' (Little 1991 page 159). However, Little also points out that statistical correlation is not acceptable without a causal story indicating the mechanisms through which the observed correlations evolve (1991). Correlation is not causation. It simply demonstrates that one measure of a variable is associate with another measure of a variable, but says nothing about the mechanism by which one variable produced changes in another.

How? - Causal Mechanisms[edit]

Causal mechanism posit how changes in one variable produce changes in another variable. Little defines a causal mechanism as follows:

C is a cause of E = df there is a series of events Ci leading from C to E, and the transition form each Ci to Ci+1 is governed by one or more laws Li.

Testing theories[edit]

These are mutually reinforcing as a means to assure the strength of a theory.

Competing hypotheses[edit]

When facing alternative hypotheses consider [6]:

  • Falsifiability - framed in such a manner that they can be proven false
  • Simplicity - (as in the application of "Occam's razor", discouraging the postulation of excessive numbers of entities)
  • Scope - the apparent application of the hypothesis to multiple cases of phenomena
  • Fruitfulness - the prospect that a hypothesis may explain further phenomena in the future
  • Conservatism - the degree of "fit" with existing recognized knowledge-systems

Conclusion[edit]

In essence a strategy diagram shows how activities and investments generate customer outcomes. Specifically, the causal link map serves as a visual representation of how company activities effect customer outcomes and behavior. Previous chapters demonstrate how to develop strategy diagrams. This chapter demonstrated how a strategy diagram is a formal model of a strategy firm's strategy and how to test the strength of the underlying theory. Testing the strength of the theory helps the company understand how executing the strategy will influence profitability, if at all.