Core Concepts #2: Alternatives
The “what” of tradespace exploration
This is part two of a seven-part series of posts on the core concepts of tradespace exploration, designed to help beginners become familiar with terminology and the general structure of tradespace data. Click here to find the other posts in this series, as well as other Tradespace 101 posts.
Alternatives are the “what” of a decision problem: the specific solutions that have been proposed to the problem at hand. When most people picture a “tradespace”, they picture a large cloud of data points — one point for each alternative. Each alternative is something that the decision maker could do (or build, or get, etc.) that could possibly improve their current situation and deliver them value.
When creating a tradespace, the list of alternatives should be as comprehensive as possible in order to represent the full breadth of possible solutions. Specifically, alternatives are defined by the set of choices (alternative variables) under the decision maker’s control. The distinction that alternatives are completely under the control of the decision maker is an important one: if something cannot be controlled, it cannot be chosen. While we do control what alternative is chosen, other variables outside our control may still impact how valuable it ends up being. These other factors will have to be accounted for by other means, typically by some type of uncertainty analysis.
Generating a good list of potential alternatives requires creativity, expert knowledge, and intuition. There are three main steps:
- The alternative variables themselves are typically brainstormed by asking the question: “what is under the decision maker’s control?” Ideally this list includes anything that could feasibly impact the needs/desires of the stakeholders, and any variables that do not are discarded.
- For each variable, allowable levels are then determined. When the choice is numerical, a valid range is established. When the choice is a categorical variable (e.g. color), the available selections must be researched and compiled.
- The individual alternatives are then enumerated as combinations of those levels. For technical applications, this is typically done with a Design of Experiments (DoE) methodology — e.g. full-factorial, Latin Hypercube — in order to evenly represent the levels and minimize interaction effects. Some particularly complex applications will use an optimization such as a genetic algorithm to create a tradespace with more “good” alternatives than an evenly-distributed enumeration. And a simpler application might just involve manually creating alternatives by hand: combining levels that are considered interesting or likely solutions. The full set of alternatives is sometimes called the alternative space.
Let’s walk through an example: when you are at a restaurant for dinner, you make a decision about what to eat. Different combinations of menu items are your alternatives. You might construct a tradespace of dinner alternatives by using alternative variables for Appetizer, Entrée, and Dessert. The levels of each would be the menu items for each category (plus none, if you are watching your weight or your wallet). Your final set of alternatives for dinner are different combinations of Appetizer, Entrée, and Dessert. You probably don’t need to enumerate every possible combination just to have a nice meal, so you manually consider a few promising options. You almost went with {clam chowder, fish and chips, no dessert} but ultimately decided you really wanted a brownie so you picked a smaller entrée with {clam chowder, hot dog, brownie}.
It is not necessary for all combinations of levels to result in a valid alternative. Some alternatives may be found non-functional after evaluation, like an airplane that cannot generate enough lift to take off because its wings are too small. Other times this can be obvious at first glance: “buying” a dinner by choosing to skip all three courses isn’t really dinner at all.
Some tradespaces include multiple alternative concepts; for example, a tradespace of air vehicles may want to include disparate possibilities such as airplanes, helicopters, and zeppelins. These different concepts typically require different alternative variables — there’s no need to choose a wingspan for your blimp! This creates an informal hierarchy where the “concept” alternative variable determines which other alternative variables are needed. The Design of Experiments used to generate alternatives should account for this to avoid redundant or infeasible solutions. The most straightforward method is to do a separate DoE for each concept.
Finally, a quick note on the importance of creativity when developing alternatives. It may seem like “make a list of all the different things I could do” is a straightforward task of data collection, but that’s not the case. The brainstorming of both alternative variables and levels is critically dependent on creativity: projects that devote insufficient time to formulating the alternative space will often end up omitting possible variables/levels, heavily constraining the solutions being compared.
Omitted levels (say, forgetting to include the Daily Specials menu in your dinner tradespace) artificially limit the number of alternatives that can be considered. Omitted variables (What about your drink? What if you want two appetizers?) are even more pernicious as, in addition to limiting alternatives, they are essentially assumptions being ignored by your analysis. For technical problems, make sure to gather appropriate subject matter expertise so that you can be confident all the controllable variables have been considered. Make an effort to think critically about your alternatives the next time you build a tradespace!
For anyone confused by the terminology used for this concept, you aren’t alone — it varies greatly between fields/domains, and many of these words are used interchangeably in different places. “Alternative” is our preferred generic term for the different solutions in the tradespace, while “alternative variables” are the individual choices under decision maker control.
However, “design” and “design variable” have considerable traction, given the history of tradespace exploration and its development on engineering design problems. This usage is very intuitive when appropriate to the problem, however it is less general because “design” has additional meanings. Specifically, calling alternatives “designs” can lead to potential confusion when applied to problems where the boundary between system architecture and design is considered.
Additionally, “choice” is sometimes used to mean “alternative” rather than “alternative variable”; this is particularly common when the alternatives are defined by a single variable, typically something to be bought or selected from a list of options — like a tradespace containing the available choices for buying a used car. This is another example where “design” would be an awkward term, as the colloquial definition of the word does not match the action: does a person really “design” a used car, or just choose one from the dealer?
When talking about a specific decision, people will often refer to alternatives solely with terms specific to that problem. This could be “systems” for engineering, or “portfolios” for finance, or “phones” if you are buying a phone: the “noun” of problem is often used in place of the word “alternative” when speaking about one case and not tradespace exploration in general.
