Models

Frameworks, causal graphs, logic models, coding trees and argument maps. The project’s diagrams.

Models is where the structure of your thinking gets drawn, and none of it is decoration, since a conceptual framework, a causal graph, a logic model, a coding hierarchy or an argument map are the diagrams that end up in a proposal or a thesis and that you argue about long before they get there. This tab is the list of them, with search, rename and delete, and opening one lands you on the canvas, a full editor with its own page. Models need a paid plan, and the tab shows you what it does and what opens it instead of hiding.

Starting from a template

Seven templates, each pre-built with the right shape so you start by editing instead of staring at a blank canvas. Each one carries a line saying what the diagram asserts, which is the thing to read before you pick:

• Conceptual framework. Predictor → mediator → outcome with the direct path, and a moderator on the mediated path. A causal chain with its mechanism named. Reach for it when you can say how the predictor reaches the outcome and what conditions that path.
• Causal DAG. Exposure → outcome with the direct path, a confounder and a mediator made explicit. Your causal assumptions drawn for identification. Reach for it to decide what to adjust for before estimating an effect.
• SEM (latent mediation). Two observed predictors → a latent mediator → an observed outcome, with direct paths, an exogenous covariance, and error terms. A mediation model whose mechanism is a latent construct, which is what you want when the mediator is something you infer rather than measure, and the ellipse, the double-headed arrow and the two disturbances are all doing methodological work rather than decorating.
• Logic / process model. Inputs → activities → outputs → outcomes. A programme’s route from resources to results, for planning or evaluating an intervention rather than proving causation.
• Coding hierarchy. A theme with the codes that roll up into it. The theme is derived from its codes, which is the direction the arrows encode, and it is what you want when synthesising qualitative coding.
• Argument map. A claim with the reasons for and against it, typed as support and contradiction rather than cause, for stress-testing a conclusion before you commit to it.
• PRISMA 2020 flow. The systematic-review flow diagram, counted from your own library: how many records entered the review and how many survived each stage. See the PRISMA 2020 flow, since it works differently from the other six.

Pick the one that matches what you are trying to say, remembering that a causal DAG and a logic model look similar on the page while committing you to completely different claims, which is exactly why drawing one is worth an hour.

The structures are not approximations. Each template has been checked against the convention it is named for, so the conceptual framework draws the direct path alongside the mediated one rather than quietly asserting complete mediation, and the causal DAG does the same, because in a DAG an arrow you did not draw is a claim that there is no direct effect. Delete an edge when you mean to make that claim; that is a decision, and the template leaves it to you.

You can also start from an empty canvas. When AI is on for the project, describe your objective to the advisor on this page and it will create a real editable model around your question and variables, with an optional grounding source.

When a model is worth building

Three moments repay the effort.

Before you write the framework chapter, because if you cannot draw it you cannot write it, and the drawing costs an hour where the writing costs a week.

When two papers disagree and you cannot see why, since drawing both usually reveals that one has a mediator where the other has a moderator, or that they measure different constructs using the same word.

When your committee asks what your study actually claims, where a diagram that survives the conversation does more for you than three pages of prose that does not.

There is a real counter-case, which is that a model is premature if you are still collecting literature and have not formed a position, in which case go and read more.

The list

Every model in the project, searchable by name, each showing when it was created and last updated, with rename and delete available from here.

Spend a moment on names, because “Model 1” and “Model 2” are indistinguishable in six months whereas “Sampling framework, committee version” is not.

Each model has its own URL, so you can link a specific diagram from a note, an email or your own writing.

How models connect to the rest

A model is not an island, since concepts on the canvas can be bound to sources in your library, turning a box into a claim with literature behind it rather than a word, and your advisor can read the model and tell you where the evidence is thin. Models also turn up in ⌘K search, appear in your Journal when built or revised, and can be pulled into an advisor conversation as scope.

All of that happens on the canvas. Next: The model canvas.