are analytical constructions for representing summary ideas and organizing knowledge. Knowledge scientists usually use such frameworks — knowingly or unknowingly — to derive venture plans, choose machine studying fashions that steadiness varied trade-offs, and current findings and suggestions to stakeholders. This text offers an summary of frequent forms of conceptual frameworks, a easy three-step course of for constructing customized frameworks, and suggestions for efficiently doing so.
Observe: All figures within the following sections have been created by the creator of this text.
Widespread Framework Sorts
Though conceptual frameworks are available many alternative sizes and shapes, 4 primary framework sorts stand out as being particularly frequent in knowledge science tasks: hierarchies, matrices, course of flows, and relational maps. We’ll briefly go over every of those framework sorts under.
Hierarchies
Hierarchical frameworks typically take the type of tree diagrams, beginning with a root node and ending with a number of leaf nodes, as proven in Determine 1. For instance, the basis node might symbolize an overarching idea in a taxonomy or an preliminary binary query in a choice tree. A node’s place within the hierarchy (or tree) offers us useful details about its relationship to different nodes. Though Determine 1 labels the objects within the hierarchy as “ideas,” they are often any sort of entity. Entities could also be impartial (e.g., ideas, subjects, segments) or have some constructive or detrimental valence (e.g., revenues, prices, issues, points). The hierarchical construction can range in depth and breadth.
In visible representations of hierarchies, vertical hyperlinks between two entities are sometimes drawn explicitly and will be non-directional (easy strains) or directional (downward or upward arrows, relying on the stream of the connection). Against this, horizontal hyperlinks between entities on the similar stage of a hierarchy are sometimes not proven explicitly. Identical-level entities could also be topic to a pure ordering (e.g., temporal or spatial), which will be proven by putting them accordingly within the framework. As an example, entities that happen earlier in an ordering must be positioned to the left of entities that happen later. If the entities don’t include a pure ordering, you may nonetheless think about ordering them not directly (e.g., by stage of significance or precedence) to help reasoning. Entities on the similar stage in a hierarchy ought to usually even be on the similar stage of abstraction.
In lots of conditions, it helps if the nodes of a hierarchy are mutually unique and cumulatively exhaustive, or MECE (pronounced “me-see”), to a big extent. Being mutually unique implies that the ideas represented by particular person nodes haven’t any main overlaps (i.e., no redundancies), whereas being cumulatively exhaustive implies that the framework leaves out nothing vital. A MECE hierarchy will be helpful for breaking down a broad idea into sub-concepts (or parts) to establish key drivers of the entire.
Matrices
A matrix is a tabular knowledge construction consisting of n rows and m columns. Knowledge scientists engaged on tabular use circumstances routinely leverage matrices for storing coaching knowledge and mannequin weights. Coaching machine studying fashions can yield high-dimensional matrices of weights that seize advanced relationships between predictors and targets. Low-dimensional matrices just like the one proven in Determine 2 will be helpful for analyzing issues and speaking key insights.
The generic two-by-two matrix proven in Determine 2 compares two completely different dimensions towards one another. Such a matrix naturally yields 4 quadrants. By conference, the bottom-left quadrant (the place each dimensions are “low”) is usually the undesirable area of the matrix, and the top-right quadrant (the place each dimensions are “excessive”) represents the fascinating area. For instance, the market analysis agency Gartner makes use of two-by-two matrices to research the aggressive panorama in varied trade sectors and calls the top-right area of the matrix (the place the market leaders are plotted) the “magic quadrant.”
The scale of a matrix might symbolize steady, ordinal or categorical knowledge sorts. Ideally, these dimensions (or axes) must be vital to the overarching framework goal not directly (e.g., key sub-concepts, issues, or drivers in a given context). The interactions between these dimensions must be of specific curiosity as a supply of perception, since it’s these interactions that matrices can seize effectively.
Generally, the MECE precept additionally applies to the selection of dimensions — they need to collectively cowl the vital sub-concepts or drivers of the issue being investigated and keep away from redundancies. In any other case, trying on the interplay can be no completely different from taking a look at a person dimension. If analyzing the interplay will not be vital, a hierarchical framework could also be extra appropriate. Changing between a matrix framework and its hierarchical analog will be simple. As an example, to remodel the matrix in Determine 2 right into a hierarchy, create a root node that defines the general context, let its baby nodes be Dimensions 1 and a pair of, and let their respective baby nodes be “excessive” and “low.”
Course of Flows
A course of stream defines a sequence of logically ordered actions that work together to attain an overarching goal. As an example, instruments equivalent to Dataiku and KNIME permit customers to assemble knowledge science pipelines as course of flows, going from knowledge ingestion all the best way to modeling and report technology. Determine 3 depicts a generic course of framework.
The entities of the method in Determine 3 are labeled as actions, however these could possibly be steps, levels, operations, and so on. The method begins with an exercise (Exercise 1), ends with an exercise (Exercise 3), and has a number of actions in between (Exercise 2). Some inputs are sometimes fed into the method at the beginning and remodeled over the sequence of actions to yield an output. Observe that inputs and outputs can even enter and depart at intermediate steps throughout the course of.
As with hierarchies and matrices, the MECE precept will be vital in formulating the completely different actions of the method. If two actions have vital conceptual overlap, you may think about both grouping them right into a single exercise or breaking them up right into a extra granular set of distinct actions. As an example, the intermediate actions in Determine 9 might have resulted from this kind of evaluation; Exercise 2 could possibly be the end result of merging some overlapping actions, whereas Actions 2.1-2.3 could possibly be a granular breakdown of a particular subset of these merged actions. If an exercise or a bigger a part of the method repeats, then it may be represented as a cycle, whereby an exercise transitions to a different exercise that has already occurred earlier than.
The transition from one exercise to a different ought to meaningfully rework the inputs of the method (e.g., by growing, lowering, combining or in any other case altering the inputs not directly) with the goal of manufacturing the specified output. If a transition doesn’t change the inputs, then the 2 actions on both aspect of the transition are possible redundant and must be merged or break up up in another way, as mentioned above.
Relational Maps
Relational maps shift the main target from particular person ideas (or entities) to the relationships between them. Knowledge scientists working with data graphs or box-and-arrow “path diagrams” of causal relationships (as proven in Determine 4) can be conversant in this framework sort.
A relationship can usually be any perform that hyperlinks two completely different ideas collectively. 4 forms of relationships are particularly frequent:
- Transactional: A relationship can symbolize a number of transactions between entities. The transactions might contain the stream of tangible issues (e.g., merchandise purchased and offered) or intangible issues (e.g., info, cash). Transactional relationships can incorporate directionality; a transaction can stream from A to B, from B to A, or in each instructions, and every of those circumstances has a distinct that means for the entities (e.g., they might be receivers, senders, or each).
- Causal: Entities A and B could also be causally associated if A is accountable — at the very least partly — for the prevalence or state of B (or vice versa). The character of the causal relationship might range. The function of A is robust if its presence is ample to totally trigger B (though A is probably not the one entity that may totally trigger B). The function of A can be sturdy whether it is essential to trigger B (though A might not be capable to do that alone). Furthermore, if A causes B, it doesn’t essentially comply with that B causes A; the notion of directionality is clearly vital for specifying causal relationships.
- Similarity-based: Entities could also be associated as a result of they’re related or dissimilar not directly. For instance, entities A and B will be related as a result of they have an inclination to look in the identical place or occur on the similar time (and dissimilar if the prevalence of 1 entity tends to preclude the prevalence of the opposite). The notion of correlation is a mathematical formalization typically used to assemble measurable, similarity-based relationships. Observe that, simply because two entities are correlated doesn’t essentially imply that they’re causally associated (though if they’re causally associated, then they might even be correlated).
- Membership-based: Entities will be linked collectively by being members of the identical group, neighborhood, or class. As an example, individuals will be associated by being in the identical neighborhood, grocery objects will be a part of the identical product class, and a set of sub-concepts could also be a part of an overarching idea. Certainly, one may apply a hierarchical framework to drill down into successively deeper ranges of membership inside entities into account.
How you can Construct Your Personal Frameworks
The next three-step course of can be utilized to construct a customized framework:
- Outline the framework’s goal.
- Establish the appropriate constructing blocks (i.e., the framework sort and dimensions).
- Put the constructing blocks collectively in an efficient method to reply the framework’s goal.
Step 1: Outline the Goal
In defining the framework’s goal, ask your self: In what context will the framework be used? What ought to the framework accomplish? Can an current framework be reused — maybe with some minor modifications — or does a brand new one have to be constructed to suit your particular wants?
The development of the framework must be tied to the next aim, such because the supply of a venture, formulation of a choice, or creation of some documentation. As soon as the context has been correctly understood, cautious consideration must be given to what the framework ought to accomplish in concrete phrases. Is the framework supposed as a decision-making device? Is the framework meant to construction the stream of an argument in a report or a presentation?
Simply since you want a framework doesn’t imply that you have to construct one your self. In lots of conditions, current conceptual frameworks will be reused with out vital modification. Spending some effort to take care of a strong, up-to-date overview of related current frameworks avoids downstream prices of “reinventing the wheel.” Reusing current frameworks has advantages past not having to start out from scratch; if the framework has been round for a while, its important options, in addition to its strengths and limitations, could also be well-documented and examined in several settings. Platforms equivalent to In the direction of Knowledge Science are a fantastic supply for conserving abreast of conceptual frameworks associated to knowledge science tasks.
Step 2: Establish the Framework Kind and Dimensions
Having clarified the target of the framework, it’s time to assume extra concretely concerning the development of the framework itself. One of many important difficulties right here is that conceptual frameworks are inherently not as tangible as bodily ones (like molds in a manufacturing unit). We are likely to intuit the hyperlink between kind and performance — the framework and its function — extra simply when the framework and its object are tangible. The hallmark of an excellent conceptual framework is its skill to show a seemingly intangible argument or choice into one thing extra tangible, and the important thing to that is illustration.
Broadly talking, there are two elements that decide the illustration of conceptual frameworks: the sort of the framework and the dimensions of the framework. You’re prone to discover the framework sort first because it determines how the framework seems as an entire. The earlier sections coated the 4 frequent framework sorts. The framework dimensions dictate what the framework can particularly symbolize (e.g., when it comes to granularity and ordering). By adjusting the scale, the identical framework sort will be reused to generate a variety of various insights. Following are three frequent lessons of framework dimensions:
- Categorical: These dimensions encompass a finite set of discrete classes that totally describe the dimension. The classes needn’t be ordered (e.g., a set of merchandise, buyer segments, gender).
- Ordinal: These dimensions are ordered, which implies which you could analyze whether or not one thing is “lower than,” “higher than,” “equal to,” and so forth, in relation to one thing else (e.g., detrimental/constructive, low/medium/excessive).
- Steady: Such dimensions can take the notion of ordinal dimensions to a way more granular stage. Being steady implies that the dimension is numerical and might embrace decimals (e.g., 1.23, -2.718, 3.14159).
Step 3: Put It All Collectively
As soon as the framework sort and dimensions have been recognized, they are often mixed to supply a customized framework. Usually, the identification and mixture steps should not explicitly separated, because you not often do one with out the opposite. However the framework sort and its dimensions — the essential constructing blocks — should not essentially wedded to one another. Some combos might make extra sense than others, and you’ll usually combine and match the constructing blocks in some ways, over a number of iterations, till the framework feels proper. Be capable of spot and exploit this combinatorial flexibility is a vital talent that it is best to begin creating from the outset of your framework-building journey.
Furthermore, there are broadly 4 “pathways of research” that seize the hyperlink between the framework and its goal:
- Descriptive: Approaches the framework’s goal by gathering and organizing previous info (e.g., utilizing visuals equivalent to graphs and tables, or written summaries). Doing so permits us to higher describe and analyze what occurred prior to now, however it might not essentially inform us why one thing occurred, or whether or not it can occur once more.
- Diagnostic: Takes descriptive info of previous occasions and goes a step additional to take a look at why one thing occurred. That is accomplished by drilling down into the info, mining for clues and correlations, and looking for a believable hyperlink between trigger and impact. As with the descriptive pathway, the main target is on the previous.
- Predictive: Differs from the prior two by asking and answering questions concerning the future. The main target is on making an informed guess about what is going to occur sooner or later by counting on a number of sometimes quantitative strategies that vary from the straightforward (e.g., primary chance idea, linear fashions) to the extra advanced (e.g., neural nets).
- Prescriptive: Goes past merely predicting future occasions to recommending methods to cope with them. The main target is on determining the way to make one thing occur — or whether or not it ought to occur — sooner or later. The reasoning for the prescription will be quantitative (e.g., primarily based on statistics or simulation modeling) or qualitative (e.g., primarily based on private expertise).
Framework sorts and dimensions can due to this fact be mixed in several methods to supply customized frameworks that lend themselves to descriptive, diagnostic, predictive, and prescriptive use circumstances.
High Ideas
This part offers 5 suggestions for constructing good conceptual frameworks. The information are not at all an exhaustive record of the factors that it is best to think about, however symbolize a primary set of issues to remember.
Tip 1: Concentrate on the Goal and Viewers
The method of constructing frameworks broadly consists of three steps, specifically defining the target, then figuring out and mixing the constructing blocks (framework sorts and dimensions) accordingly. Whereas step one will, by its nature, emphasize the strategic goal and target market of the framework, the main target within the latter two steps shifts to the nitty-gritty particulars of the framework’s constructing blocks. The deeper you get into the mechanics of the framework, the tougher it may be to take care of visibility of the unique goal. To take care of visibility of the larger image, it might assist to take a step again every now and then in the course of the framework-building course of and remind your self of the strategic goal and target market. It could additionally assist to delay a part of the evaluation till the required knowledge turns into obtainable and to hunt common suggestions from colleagues and the target market of your framework the place potential.
Tip 2: Hold It as Easy as Potential
To paraphrase a quote typically attributed to Albert Einstein — one of the crucial achieved builders of conceptual frameworks of the final century — we are able to say {that a} framework must be made so simple as potential, however not easier. Because the course of inherently entails making an attempt out completely different combos of framework sorts and dimensions, it might generally be tempting to snap an increasing number of items collectively. But sacrificing simplicity can probably diminish the broader worth of the framework in follow. Complicated frameworks will be obscure, apply, consider, and construct — you could have to confirm a number of assumptions and preconditions, and regulate many alternative levers throughout the framework.
Tip 3: Make It MECE
Making certain {that a} framework is MECE has some vital benefits. From a theoretical standpoint, being MECE implies that the sub-concepts comply with a constant, additive part-whole logic; you count on the sub-concepts to “add up” to kind the larger idea. Crucially, this logic means that you can substitute the set of sub-concepts for the larger idea (and vice versa) all through your evaluation. The additive logic of MECE additionally permits you to evaluate completely different ideas in a rigorous method; as a substitute of claiming that two ideas are related, you may state exactly the extent to which they’re related by figuring out the sub-concepts they share. From a sensible perspective, being MECE means which you could “divide and conquer” massive issues effectively and options to some sub-problems could also be reusable. Generally you may even attain the answer of the larger drawback with out fixing all of the sub-problems (e.g., if the larger drawback will be represented as a disjunction of the sub-problems). Bypassing sub-problems additionally works if you find yourself fixing the larger drawback inductively (e.g., as in mathematical induction).
Tip 4: Make It Versatile
Basically, a conceptual framework must be designed to fulfill its general goal, so you could be questioning why flexibility is a vital side to contemplate. In follow, there are at the very least two forms of conditions during which flexibility is usually a massive assist. Within the first scenario, you could be coping with an goal that could be a transferring goal, with some components of the target’s full scope altering (even barely) every now and then; responding to such scope modifications is usually a ache if some flexibility will not be baked into the framework. Within the second scenario, your framework might should bear a number of iterations, during which completely different framework sorts and dimensions are added, modified and eliminated over the course of the framework’s evolution; a versatile design makes it a lot simpler to facilitate such alterations of the framework’s form and content material. Modularity, scalability, robustness, extensibility, and portability — whereas sometimes related to software program engineering and structure — are additionally related design concerns for constructing versatile conceptual frameworks.
Tip 5: Construct It Iteratively
It could be nice for those who may provide you with the right framework in a single go, however it not often works out that manner. A number of elements could make the primary iteration extra of a primary draft, to be adopted by at the very least a couple of extra. The overarching goal — and particularly the operational implications with regards to constructing the framework — is probably not totally clear at first. Over a few iterations, nevertheless, you’ll most likely start to get the cling of which framework sorts and dimensions work and which don’t. Whereas your output after a given iteration could also be removed from good, it may nonetheless quantity to a minimal viable product (MVP) if it yields a viable resolution to the overarching goal with minimal effort and complexity. The MVP will be examined (e.g., with precise knowledge and actual customers) to know its strengths and weaknesses. Every successive iteration can produce an improved MVP by including, eradicating or altering options of the earlier iteration.
To shut off, here’s a video that shares some extra good recommendation on constructing and utilizing conceptual frameworks:
The Wrap
Conceptual frameworks assist us flip summary concepts into concrete, tangible merchandise that different individuals can see, use, and respect. This may be particularly vital for knowledge scientists, or so-called “data employees,” whose jobs contain gathering, analyzing, and deriving conclusions from knowledge. In case you are studying this text, you’re most likely a data employee. To paraphrase well-known administration guru Peter Drucker, “It’s knowledge that allows data employees to do their job,” however it’s the skill to meaningfully arrange this knowledge that results in a job effectively accomplished — and that, in a nutshell, is why the correct use of conceptual frameworks can assist the profitable design and supply of information science tasks.
