---
name: theory-building-and-conceptual-contribution
description: >
  Builds theory rather than only testing it: defines constructs that exclude
  things, specifies relationships with direction, form and mechanism, sets
  boundary conditions, and distinguishes a genuine theoretical contribution from
  an existing theory relabelled. Use for "is this a theory or a framework",
  "how do I define my construct", "write propositions", "boundary conditions",
  "what is the mechanism", "my conceptual model is just boxes and arrows",
  "building theory from qualitative data", "grounded theory contribution",
  "extending versus challenging theory", "is my contribution theoretical".
category: 15 Academic University Research
ref: "15.15"
tier: 3
inherits: [K2, K3, K4, K5]
---

# Theory Building and Conceptual Contribution

## 1. One-line description

A method for developing a theoretical contribution that explains rather than describes: constructs defined tightly enough to exclude things, relationships specified with direction, form and mechanism, boundary conditions that limit the claim, and a stated position on how the contribution relates to the theory that already exists.

## 2. What this skill is used for

**The research problem it solves.** Most work that calls itself theoretical is not. A diagram of labelled boxes joined by unlabelled arrows is a picture of a hypothesis space, not a theory. A list of factors that matter is a checklist. A framework that organises a literature is a filing system. Each is useful and none explains anything, because explanation requires saying *why* one thing produces another and *under what conditions* it stops. Alongside that sits a second failure that examiners catch and authors rarely see: an existing theory, restated in new vocabulary, presented as a new contribution. This skill forces the elements that make a theory a theory, and applies the tests that separate a contribution from a paraphrase.

**Where it sits in the research lifecycle.** Before design, where the theory generates what must be measured and observed; during qualitative analysis, where theory is built from data; and at the discussion stage, where an empirical result is converted into a theoretical claim or fails to be.

**Typical use cases.**
- Turning a conceptual model of boxes and arrows into a specified theory with mechanisms.
- Defining a construct precisely enough to be measured, and distinguishing it from the three adjacent constructs it keeps being confused with.
- Writing propositions and deriving testable hypotheses from them.
- Establishing the boundary conditions of a claim, which is what stops a theory from claiming everything and explaining nothing.
- Building theory from qualitative data using a named approach with its epistemological commitments stated.
- Deciding whether the contribution extends, refines, challenges or builds theory, and assembling the evidence each of those requires.
- Testing whether a proposed contribution is genuinely new or a relabelling.

**Who uses it.** Doctoral candidates whose contribution is wholly or partly theoretical, in any discipline that theorises; researchers writing a conceptual paper; qualitative researchers moving from themes to theory; supervisors testing whether a candidate's model is a theory.

## 3. When to use it

- Your conceptual model consists of boxes and arrows and someone has asked what the arrows mean.
- Your findings are interesting and you cannot say what they contribute theoretically.
- A construct in your work has been used with slightly different meanings in different chapters, or in different papers you cite.
- You need to write propositions and are unsure how they differ from hypotheses.
- You have been told your theory applies too broadly, or you have never been asked where it stops applying.
- You are building theory from qualitative data and need to know what your chosen approach actually commits you to.
- You are claiming a theoretical contribution and need to establish that it is not an existing theory in new words.
- A reviewer or examiner has said the work is descriptive rather than explanatory.

## 4. When NOT to use it

- **The contribution is empirical and does not need to be theoretical.** In many disciplines and many doctorates the contribution is new evidence, and dressing it in a theoretical claim it cannot support weakens it. Establish what kind of contribution you actually have with **15.13 Doctoral Research Positioning and Originality** first. Manufacturing a theoretical contribution from an empirical study is a standard route to an examiner finding the theory unsupported.
- **You need to select and apply an existing theoretical lens, not build one.** Choosing a theory, justifying the choice and building a conceptual framework that links its constructs to your research questions is **15.07 Theoretical and Conceptual Framework Development**. This skill takes over only where the contribution is to the theory itself.
- **The theory would be built on data that does not exist yet.** Theory built ahead of evidence is speculation with a diagram, and it becomes a thesis that argues for a model rather than testing one. Draft the theory as a set of propositions to be examined, and mark it as such (K4 §3.4).
- **The construct is already well defined in the literature and the definition works.** Redefining a functioning construct because a slightly different definition suits your data is a common and damaging move: it makes your work incommensurable with everything before it, and buys nothing. Use the existing definition, cite it, and if it genuinely fails, say specifically where and why.
- **The aim is prediction and explanation is not required.** In some applied and computational contexts a predictive model is the deliverable and the mechanism is not needed. That is legitimate, and it must be described as a predictive model, not a theory. A model that predicts without explaining is a correlation with ambition, and calling it a theory invites an attack it cannot survive.
- **The evidence is a single case and the claim is general.** A mechanism identified in one setting is a hypothesis about mechanism. Theory-building from a single case is possible and respectable, but the claim must be bounded to what the case can carry (K4 §3.3), and the boundary work in Step 6 becomes the most important part of the exercise.
- **The task is synthesis of a literature rather than theorising.** Organising what is known is **15.06 Systematic Literature Review** or **15.14 Advanced Systematic Review and Meta-Analysis**. A synthesis becomes a theoretical contribution only when it produces new explanation, not when it produces a new arrangement.
- **Academic integrity.** Per §12.1: this skill does not write theoretical passages, propositions or definitions for submission as the candidate's own unaided work. It interrogates the candidate's constructs, tests their relationships and returns what is missing. Where an institution prohibits AI assistance for a task, this skill must not be used for it.

## 5. Required inputs

**Required.**
- **The candidate's own statement of the phenomenon to be explained.** Not the topic and not the research question: the thing that happens which is puzzling, and which existing explanations do not account for. Theory answers "why does this happen"; without a stated explanandum, there is nothing to theorise about.
- **The existing theory in the space, read rather than summarised.** A theoretical contribution is defined relative to what exists, and a claim of novelty made against unread theory is untested (K4 §2.4).
- **The evidence or intended evidence base.** What the theory is built from or will be examined against, with its design, so that the claim can be sized to it.

**Optional, and what each one adds.**
- **The candidate's current conceptual diagram.** Provides the raw material for the arrow interrogation in Step 4, which is usually where the missing mechanisms become visible.
- **Existing measures of the constructs.** Reveals whether the construct as defined is measurable, and whether existing measures capture the definition or something adjacent. A construct that cannot be operationalised needs either a better definition or an honest statement that it is not yet measurable.
- **Qualitative data and the analytic approach used.** Determines which theory-building tradition applies and what its commitments are.
- **The discipline's theoretical conventions.** What counts as a theoretical contribution differs sharply between fields, and in some fields propositions are expected while in others they read as pretension.
- **Competing explanations for the same phenomenon.** The strongest theoretical contributions are argued against rivals, and having them named turns the contribution from an assertion into a comparison.
- **Cases that the theory should not explain.** Sharpens the boundary condition work more effectively than any amount of abstract reasoning.

## 6. Questions to ask before starting

1. **What exactly is being explained?** State the explanandum as an observable regularity or puzzle. Default if unanswered: extract it from the research question and confirm, because theorising about an unclear phenomenon produces an unclear theory.
2. **Why is the existing explanation inadequate?** A theoretical contribution is only visible against the account it improves on. Default: assume an existing account exists and find it; the case where nothing has been said is rare and usually means the search was too narrow.
3. **What is the mechanism, in plain language, with no jargon?** If it cannot be said plainly it has usually not been specified. Default: press for the plain version before accepting the technical one.
4. **Where does this stop being true?** Determines the boundary conditions. Default: treat the absence of an answer as a defect to be fixed, not a strength to be preserved.
5. **What observation would refute this?** Determines whether the theory is testable. Default: if nothing would refute it, it is not yet a theory, and say so.
6. **Is this extending, refining, challenging or building?** Each carries a different evidentiary burden. Default: classify from the relationship to existing theory, not from ambition.
7. **What is the unit and level of the theory?** Individual, dyad, team, organisation, field, society. Cross-level claims are where theories most often break, and mixing levels silently is a standard examiner target. Default: state the level and check every proposition sits at it.

## 7. Step-by-step methodology

**Step 1. State the explanandum, and check it is a puzzle rather than a topic.**
Write one sentence: the thing that happens, in what population or setting, that is not adequately explained. A good explanandum has a note of surprise in it, because if the outcome follows obviously from what is already known, there is nothing to explain. Test it three ways. Is it an observable regularity rather than an abstraction? Does an existing theory already account for it, and if so, where specifically does that account fail? Would a competent colleague agree that it is puzzling, or would they say "obviously, because..." and give you the mechanism you were about to propose? The third test kills a substantial proportion of proposed theoretical contributions in about a minute, which is a service. *Correct result: one sentence naming the phenomenon and one naming the inadequacy of the existing account.*

**Step 2. Define each construct so that it excludes things.**
A definition that excludes nothing means nothing. For each construct, write a definition, then immediately write three things that fall outside it but might be mistaken for it and say what distinguishes them. Then name the two or three closest neighbouring constructs in the literature and state the difference between yours and each: if you cannot, you have a synonym rather than a construct, and a new label for an existing construct fragments the literature and will be identified as such. Specify properties: continuous or categorical, state or trait, individual or collective, dimensional or not. Then check stability: write the sentence in which the construct appears in your research question, your propositions and your measurement, and confirm the same meaning survives all three. Definitional drift between chapters is one of the most reliably fatal findings in an examination. *Correct result: a definition, an exclusion set, a differentiation from named neighbours, stated properties, and a demonstration that the meaning is constant across uses.*

**Step 3. Specify each relationship with direction and form, not an arrow.**
For every proposed link, state five things. Which construct is antecedent and which consequent, and what establishes the direction, since an arrow asserts a temporal or causal ordering a cross-sectional design cannot support (K4 §3.2). The functional form: linear, threshold, curvilinear, saturating or step. Whether the relationship is direct, mediated or conditional, and if mediated, what the mediator is and why it mediates rather than merely correlates. The expected sign and, where the discipline permits, a rough magnitude, since a theory predicting only "there will be a relationship" is barely a prediction. And the timescale: immediate, lagged or cumulative, the specification most often omitted and most often responsible for a failed test. *Correct result: a relationship specification table where no cell reads simply "positive relationship".*

**Step 4. Specify the mechanism, which is the actual contribution.**
The mechanism answers "how does A produce B" and it is what separates a theory from a correlation with ambition. Write it as a sequence of steps, each a thing that happens, with an agent where there is one: A occurs, which causes the individual to do or experience X, which changes Y, which produces B. Then test it. Is each step independently plausible and, in principle, observable? Could it be interrupted, at which step, and what would that predict? Does it explain the specific form of the relationship you specified in Step 3, or would it equally predict a different form? Does it explain the boundary conditions you are about to set, or are they arbitrary? And critically: does it generate any prediction a simple association would not? A mechanism predicting exactly what "A is associated with B" predicts is doing no work. The best mechanisms generate a surprising secondary prediction, often the strongest test available in a thesis. *Correct result: a stepwise mechanism, its interruption points, and at least one prediction it makes that a bare association would not.*

**Step 5. Write propositions, then derive hypotheses, and keep them distinct.**
A proposition states a relationship between constructs at the conceptual level: general, not directly testable, naming no measure. A hypothesis states a testable relationship between operationalised variables in a specified population. One proposition can generate several hypotheses, and it should: if it generates only one, it is probably an over-dressed hypothesis. Number them, keep the numbering consistent throughout, and make the derivation visible, because the proposition-to-hypothesis link is where an examiner checks whether the empirical work actually tests the theory. Watch for two errors: hypotheses following from no proposition, which means the empirical work has drifted from the theory, and propositions no hypothesis tests, which means part of the theory is asserted rather than examined and must be presented as such. *Correct result: numbered propositions at construct level, numbered hypotheses at variable level, and an explicit mapping between them with the untested propositions identified.*

**Step 6. Set boundary conditions, because a theory that applies everywhere explains nothing.**
Specify where the theory holds and, more importantly, where it does not, across four dimensions. Population: which units, and what property of them the theory depends on. Setting: which contexts, and what feature of the context is load-bearing. Time: over what horizon, and whether the mechanism needs a period to operate. Assumptions: what must be true for the mechanism to work at all, for example that actors have discretion, information, or an alternative available. Then name a specific case where the theory should fail, and say why. A theory whose author can name its failure cases is credible and testable; one claiming universality is neither, and universality is almost always an unexamined default rather than a considered claim. Boundary conditions are not limitations to be apologised for in a final section; they are part of the theory and belong in its statement. *Correct result: boundaries on all four dimensions, and at least one named case the theory predicts it will not explain.*

**Step 7. Classify the contribution, and meet the burden it carries.**
Four kinds, with different requirements:

| Kind | What it claims | What it must show |
|---|---|---|
| Extending | The theory applies to a domain, population or level it had not covered | The extension is non-obvious, and the mechanism still operates there for a stated reason |
| Refining | The theory is right but under-specified: a moderator, a form, a boundary or a mechanism step was missing | The refinement changes a prediction, not just the wording |
| Challenging | The theory is wrong, or wrong under conditions it claims to cover | Engagement with the theory at its strongest, evidence that the original design would have accepted, and an account of why the original evidence looked supportive |
| Building | A new explanation for a phenomenon existing theory does not account for | That existing theories genuinely fail here, and that the new account is not one of them relabelled |

Challenging is the highest burden and the most common source of an examiner's hostility, because a candidate who challenges a theory they have engaged only through secondary sources will be dismantled. Building carries the relabelling risk in Step 8. *Correct result: a named kind, and the specific evidence assembled for its burden.*

**Step 8. Run the relabelling test.**
Write your theory out using only vocabulary from the existing literature, substituting established terms wherever a plausible equivalent exists, then ask what remains. If the translated version is an existing theory, you have a relabelling, which is the single most common false theoretical contribution: a familiar mechanism dressed in new nouns. If something survives, that residue is your contribution, usually smaller and more precise than the original claim. Then reverse it: take the two closest existing theories and ask what yours predicts that they do not. If there is no differential prediction, the contribution is presentational, and a clearer articulation of an existing theory is a modest contribution in some fields and none in others. `RESEARCHER DECISION REQUIRED` where the residue may be too small to carry a doctoral theoretical claim, since sufficiency is the supervisor's and institution's judgement (K5 §2.7). *Correct result: a translated version, the surviving residue stated explicitly, and at least one differential prediction against the nearest rivals, or an honest statement that there is none.*

**Step 9. Trade parsimony against completeness deliberately.**
Every construct added buys explanatory coverage and costs testability, communicability and the risk of unfalsifiability. Go through the model construct by construct and ask what would be lost by deleting it: if the answer is "some variance", that is not enough, since a theory is not a regression. Delete anything whose only role is to make the diagram look complete. Then check the other direction: is anything essential missing, particularly a condition the mechanism silently assumes? The judgement turns on the theory's purpose. A theory intended to guide action needs enough specificity to say what to change; a theory intended to explain needs enough to identify the mechanism and no more. State which you are building. *Correct result: a construct-by-construct justification, with deletions made and recorded.*

**Step 10. Establish testability and name the refutation.**
For each proposition, write what observation would count against it. If nothing would, the proposition is either a definition in disguise or unfalsifiable, and both must be identified. Watch for the two disguises. Tautology: "organisations that adapt effectively perform better", where effective adaptation is defined by outcomes. And infinite flexibility: a theory whose boundary conditions and moderators can absorb any result, so that a failed test is always explained by an unmet condition. Then specify what an adequate test would require: design, measures, population, timescale and what would need to be observed. Where your own study cannot provide it, say so, and put the required test in future research where it belongs. A theory offered with a specification of what would refute it is treated as serious; one offered without is treated as advocacy. *Correct result: a refutation condition per proposition, and a specification of the test each would need.*

**Step 11. Where theory is built from qualitative data, use a named approach and state its commitments.**
Theory-building traditions differ in ways that matter. The classic grounded-theory position holds that the researcher approaches data without a preformed frame and that categories emerge from constant comparison; a later systematised variant adds coding paradigms and structured procedures, which some regard as forcing data into a template; a constructivist variant rejects emergence altogether and treats theory as co-constructed with participants, which changes what the theory claims to be. Multiple-case theory building uses replication logic and cross-case comparison, closer to an analytic-generalisation position. Abductive and interpretive traditions treat theorising as inference to the best explanation, moving iteratively between data and existing theory rather than bracketing theory out. These are not interchangeable and the choice cannot be made on convenience. Whichever is used, the theory-building steps must be visible: what was compared with what, which cases were sought to test an emerging category, what the negative cases were and how they changed the account, and where the abstraction from data to construct happened. That visible chain is what makes a qualitative theoretical contribution examinable rather than assertable (K2 §2.1). *Correct result: a named approach with its position stated, and an inspectable trail from data to construct to relationship, including the cases that did not fit.*

**Step 12. Write the theory as prose before drawing the diagram.**
If the theory can only be expressed as a diagram, it is not specified. Write it as continuous prose: the phenomenon, the constructs with their definitions, the relationships with their forms, the mechanism as a sequence, the boundaries, and what follows. Then draw the diagram, and check that every element in the diagram appears in the prose and vice versa, since diagrams routinely contain arrows the text never mentions and text routinely contains conditions the diagram omits. Label every arrow with the mechanism, not with a sign. *Correct result: a prose statement that stands alone, and a diagram that adds nothing the prose does not contain.*

## 8. Analytical framework

A theory has five components, and the absence of any one is diagnostic of what kind of object you actually have:

    Constructs (what) → Relationships (how they connect) →
    Mechanism (why the connection exists) → Boundaries (where it holds) →
    Predictions (what follows, and what would refute it)

Applying it as a diagnostic:

| What is present | What you have |
|---|---|
| Constructs only | A vocabulary or a taxonomy |
| Constructs and relationships | A conceptual model or hypothesis set |
| Constructs, relationships, mechanism | A theory without boundaries, which over-claims |
| Constructs, relationships, boundaries, no mechanism | A contingency description; predicts without explaining |
| All five | A theory |

The second structure is the contribution test, and it runs after the theory is specified:

    Translate into existing vocabulary → What survives? → Does the residue
    generate a prediction the nearest rivals do not? → Is that prediction testable?

Only a claim that survives all four steps is a theoretical contribution. A claim that fails at step two is a relabelling; one that fails at step three is a restatement; one that fails at step four is philosophy, which may be valuable but is not the claim usually being made.

## 9. Output format

**1. The phenomenon.** The explanandum in one sentence, and why existing accounts are inadequate.

**2. Construct definitions table.**

| Construct | Definition | What it excludes | Nearest existing constructs and the difference | Properties (level, type, dimensions) | How operationalised |
|---|---|---|---|---|---|

**3. Relationship specification table.**

| Link | Antecedent → consequent | Direction established by | Form | Sign and expected magnitude | Mediated or conditional on | Timescale |
|---|---|---|---|---|---|---|

**4. Mechanism statement.** Stepwise, in prose, with interruption points and the secondary prediction it generates.

**5. Propositions and hypotheses.** Numbered, at their proper levels, with the mapping between them and the untested propositions flagged.

**6. Boundary conditions.** Population, setting, time, assumptions, plus named cases the theory predicts it will not explain.

**7. Contribution classification.** Extending, refining, challenging or building, with the evidence assembled for that burden.

**8. Relabelling test result.** The translated version, the surviving residue, and the differential predictions against the nearest rival theories.

**9. Refutation conditions.** Per proposition, with the test each would require and a note of which the present study can provide.

**10. Theory statement in prose**, then the diagram.

**When the evidence is thin, the format must not force fabrication (K4 §1).** A mechanism that cannot be specified is reported as unspecified, not filled with plausible-sounding steps, because an invented mechanism is the most persuasive and least detectable form of theoretical fabrication. A relationship that the data cannot direct is reported as an association with the direction unestablished. Where the relabelling test returns nothing, the honest output says the contribution is presentational, and the researcher decides what to do with that.

## 10. Quality checks

Run before any theoretical claim is presented. These sit on top of K4 §8.

1. Is the phenomenon stated as an observable puzzle rather than a topic?
2. Does every construct definition exclude something specific, and are the exclusions written down?
3. Is every construct differentiated from its nearest existing neighbours by a stated difference?
4. Does each construct mean the same thing in the research question, the propositions and the measurement?
5. Does every relationship specify direction, form, sign and timescale, with no arrow left unlabelled?
6. Is the direction of each relationship supported by the design, or is it asserted on cross-sectional data?
7. Is the mechanism written as a sequence of things that happen, and does it generate a prediction a bare association would not?
8. Are propositions at construct level and hypotheses at variable level, with the mapping explicit?
9. Are there propositions no hypothesis tests, and are they flagged as untested rather than presented as supported?
10. Are boundary conditions stated on all four dimensions, with at least one named case the theory should not explain?
11. Has the relabelling test been run, and is the surviving residue stated?
12. Does the theory make at least one prediction that the nearest existing theory does not?
13. Is there a stated refutation condition for every proposition, and is any proposition tautological?
14. Is every construct in the model earning its place, or is one there to complete the diagram?
15. For qualitative theory building, is the approach named, its epistemological position stated, and is the trail from data to construct inspectable including negative cases?

## 11. Common failure modes

| Failure | How to recognise it | How to prevent it |
|---|---|---|
| **Boxes and arrows** | Arrows carry no label beyond a sign | Specify direction, form, mechanism, timescale (Steps 3, 4) |
| **The factor list** | The contribution is that several things matter | Require a mechanism connecting them (Step 4) |
| **Relabelling** | New nouns, familiar mechanism | Run the translation test (Step 8) |
| **Construct proliferation** | A new construct where an existing one would serve | Differentiate from named neighbours or use the existing one (Step 2) |
| **Definitional drift** | The construct means slightly different things in different chapters | Check the definition across all three uses (Step 2) |
| **Universal theory** | No boundary conditions anywhere | Name a case it should fail to explain (Step 6) |
| **Unfalsifiable flexibility** | Any contrary result absorbed by an unmet condition | State refutation conditions in advance (Step 10) |
| **Tautology** | The consequent is part of the antecedent's definition | Check whether the proposition could be false (Step 10) |
| **Causal arrows on cross-sectional data** | Direction asserted by diagram convention | Establish direction from design (Step 3, K4 §3.2) |
| **Level mixing** | Individual mechanism producing an organisational outcome with no aggregation account | State the level and check every proposition sits at it |
| **Untested propositions presented as supported** | The discussion treats the whole model as confirmed | Flag the proposition-to-hypothesis gaps (Step 5) |
| **Mechanism invented to fit** | A plausible causal story with no evidentiary basis | Mechanisms are evidenced or labelled as proposed. See §12 |
| **Grounded theory as a synonym for coding** | The approach is named but no theory-building steps appear | Show constant comparison, negative cases, theoretical sampling (Step 11) |
| **Parsimony ignored** | Fourteen constructs, none deletable, none tested | Justify each construct's presence (Step 9) |

## 12. AI guardrails

Skill-specific. The universal prohibitions in K4 apply in full and are not repeated.

1. **Academic integrity, and it binds this whole skill.** These skills assist a researcher's thinking, structure and rigour. They do not produce work to be submitted as the student's own unaided output. The user must comply with their institution's AI use policy and its declaration requirements, which vary by institution and by assessment. Where an institution prohibits AI assistance for a task, this skill must not be used for it. The skill never writes a passage for submission as though the student wrote it; it interrogates, structures, critiques and teaches. Operationally here, and this matters more than in most skills: a candidate whose theory was articulated for them cannot defend it, because a viva on a theoretical contribution consists almost entirely of questions about why this construct and not that one, why this mechanism, why this boundary. The skill returns the questions; the candidate writes the answers.

2. **Never invent a mechanism.** A causal story that fits the data is easy to generate and almost impossible for a reader to distinguish from one that was derived from evidence. Any mechanism the skill articulates is labelled as a candidate mechanism proposed for the researcher's evaluation, with the evidence that would support or refute it named, and it never enters an output as an established explanation (K4 §3.4).

3. **Never attribute a theoretical position to a named scholar or tradition without a verified source.** Theoretical literatures are exactly where model knowledge produces confident, plausible and wrong attributions: the right idea assigned to the wrong author, a position stated in a stronger form than its originator held, or a distinction between traditions that does not exist. If it cannot be sourced from a document that has been opened, state the idea without the attribution and mark the attribution as needing verification (K4 §2.4).

4. **Never present a construct as new without checking the literature for it.** Novelty of a construct is an absence claim and requires a search, including across disciplines, since the same construct is frequently named differently in adjacent fields.

5. **Never resolve the relabelling test in the candidate's favour by default.** Where the residue after translation is small, say it is small. Inflating a presentational contribution into a theoretical one produces a claim that fails in examination.

6. **Never assign a theory-building tradition a position it does not hold, or present these traditions as variants of one method.** Their disagreements are real and substantive, particularly about whether categories emerge or are constructed. Present the options and their commitments; do not choose for the researcher.

7. **Never fill an empty cell in a relationship or construct table.** An unspecified form, timescale or boundary is recorded as unspecified. A plausible guess in a specification table is fabrication with the appearance of rigour.

8. **Cap confidence in any theoretical claim at the level the evidence supports, and treat mechanism claims as one step less certain than the relationships they explain** (K3 §3.6). Establishing that A relates to B is easier than establishing why, and the confidence attached should show it.

## 13. Best-practice principles

- **A theory answers "why", and everything else is preparation.** Descriptions, frameworks and factor lists are useful, and none of them is a theory. Being accurate about which one you have is worth more than claiming the highest available label.
- **Definitions are made by what they exclude.** A definition that admits everything in the neighbourhood is a label, and it will produce a measure that correlates with everything and distinguishes nothing.
- **Do not add a construct because a phenomenon needs a name.** The literature is full of constructs that are one existing construct plus a new adjective, and each of them fragments a field's ability to accumulate evidence.
- **Boundary conditions are the theory, not its limitations.** Knowing where an explanation stops is knowledge, and stating it makes the theory more useful and more credible, not less ambitious.
- **The best mechanisms make a second, surprising prediction.** If your mechanism predicts only what a bare association predicts, it is decoration. If it predicts something odd that turns out to be true, it is a contribution.
- **Write the theory before you draw it.** Diagrams hide unspecified relationships behind arrows and hide missing conditions behind whitespace.
- **Engage the theory you are challenging at its strongest.** Attacking a weak version of an established position is the fastest way to lose an examiner, and it usually indicates the original was read through secondary sources.
- **Parsimony is not aesthetic preference.** Each additional construct multiplies the tests needed to examine the theory, and an untestable theory does not accumulate evidence.
- **Levels of analysis break more theories than any other single thing.** An individual-level mechanism does not automatically produce a collective-level outcome, and the aggregation step needs its own account.
- **A theory that survives translation into existing vocabulary is a real contribution, however small the residue.** Precision about the size of the residue is what separates a defensible claim from an inflated one.
- **Negative cases build theory faster than confirming ones.** In qualitative theory building, the case that does not fit is where the boundary condition and the missing moderator come from, and searching for it deliberately is what distinguishes theory building from theme reporting.

## 14. Worked example

Generic fictional scenario, academic, organisational research in a public sector setting.

**INPUT**

A doctoral candidate has interviewed 34 frontline caseworkers across three public agencies and built a model with five boxes: workload pressure, discretion, peer support, emotional exhaustion and "adaptive rule bending", joined by arrows. Her supervisor has said the model is descriptive. She believes she has a theoretical contribution and cannot state it.

**PROCESS**

*Step 1.* The explanandum is drafted three times. The first version, "caseworkers bend rules", is a topic. The second, "caseworkers under pressure bend rules", is answered obviously by existing discretion literature. The third survives: caseworkers under high pressure bend rules in ways that protect the client in some agencies and in ways that protect themselves in others, and existing accounts of discretion do not predict which.

*Step 2.* "Adaptive rule bending" is examined. It excludes: rule breaking for personal gain, error, and formally sanctioned flexibility. Its nearest neighbours in the literature are street-level discretion and workaround behaviour, and the differentiation is that both existing constructs are defined by deviation from procedure, whereas hers is defined by the beneficiary of the deviation. That is a real distinction and it survives.

*Step 3 and 4.* The arrow from peer support to rule bending is interrogated. It had been drawn as a positive relationship. What is the mechanism? The candidate's first answer, "peer support makes people more confident to bend rules", predicts more bending, which her data does not show. Working from the transcripts, a better mechanism emerges: peer discussion of borderline cases supplies a shared account of what the rule is for, which redirects bending toward the client rather than reducing it. That is a mechanism about the direction of bending, not its amount, and it produces the secondary prediction that in teams without regular case discussion, bending will be equally common but self-protective. Her three agencies differ on exactly this, which she had recorded and not used.

*The judgement call.* The relabelling test. Translated into existing vocabulary, most of her model becomes standard discretion theory plus job demands and resources. The residue is small: the claim that peer sensemaking determines the beneficiary of discretion rather than its volume. The instinct is to keep the full five-box model as the contribution. The decision taken is to keep the residue as the contribution and demote the rest to the framework that positions it. The contribution shrinks and becomes defensible, and the theory now makes a prediction that neither of the two parent theories makes.

*Steps 6 and 10.* Boundary conditions are set. The mechanism requires that caseworkers have discretion at all, that they have a forum where cases are discussed, and that the client's interest and the worker's interest can diverge. She names the case it should fail to explain: fully automated eligibility decisions, where discretion is absent. The refutation condition for the core proposition is stated: a team with regular case discussion whose bending is predominantly self-protective would count against it.

*Step 11.* Her analysis has been abductive rather than classically grounded, moving between transcripts and existing discretion theory, and she had been describing it as grounded theory. The description is corrected, which also removes a line of examiner attack.

**OUTPUT**

A theory statement in prose with one core proposition, three derived hypotheses of which her data tests two, a stepwise mechanism, boundary conditions on four dimensions, a named failure case, and an explicit statement of what her contribution adds to the two nearest existing theories. The five-box diagram survives as a framework figure with every arrow labelled by mechanism, and two arrows deleted because nothing was lost by deleting them.

`RESEARCHER SIGN-OFF REQUIRED` from the supervisor on whether the residual theoretical claim, now considerably smaller, meets the doctoral standard in this discipline (K5 §2.7).

## 15. Advanced usage

**Theorising a null or unexpected result.** An unpredicted finding is a theoretical opportunity if it is treated as a boundary condition problem: the theory predicted X, X did not occur, so either the mechanism does not operate here or a condition it assumes was absent. Working out which, and specifying the condition, is a refinement contribution and frequently the strongest thing in an otherwise disappointing study. The failure is to report the null and move on.

**Competing mechanism designs.** Where two mechanisms predict the same relationship, the valuable study is the one that identifies where they diverge and measures there. Designing for mechanism discrimination rather than relationship confirmation is what turns a hypothesis-testing thesis into a theoretical one. See **15.16 Doctoral Methodology Justification and Rigour** and **15.17 Longitudinal and Multi-Study Design**.

**Theory across levels.** Cross-level theorising requires an explicit account of emergence: how individual-level processes compose into a collective-level property, whether by aggregation, by interaction, or by a structural constraint. Stating the composition model is a contribution in itself and its absence is a standard examiner attack.

**Formal and computational specification.** Where a theory can be expressed formally or simulated, doing so exposes ambiguity that prose conceals: implicit parameters, unstated initial conditions, and relationships that cannot hold simultaneously. This is a strong test even in fields where the final thesis is prose, because a theory that cannot be specified precisely enough to simulate usually has an unspecified step.

**Contribution to a paper rather than a thesis.** A conceptual paper carries the whole theoretical burden without empirical support, so the standards on mechanism, boundary and differential prediction rise sharply, and the positioning against existing theory becomes the paper's spine. See **15.24 Journal Article Development**.

## 16. Skill chain

**Recommended previous skills:**
- **15.13 Doctoral Research Positioning and Originality.** Establishes whether a theoretical contribution is the right claim for this project before the theory is built.
- **15.07 Theoretical and Conceptual Framework Development.** Hands over the theoretical lens and the framework this skill may then contribute to.
- **15.14 Advanced Systematic Review and Meta-Analysis.** Hands over an evidence base whose unexplained heterogeneity is often the phenomenon worth theorising.
- **07.01 Thematic Analysis.** Hands over the themes that theory building works on, and is not itself theory building.

**Recommended next skills:**
- **15.16 Doctoral Methodology Justification and Rigour.** Designs the study capable of testing the propositions, and defends the design against alternatives.
- **15.17 Longitudinal and Multi-Study Design.** Where testing the mechanism requires more than one study or repeated measurement.
- **15.18 Thesis Architecture and Chapter Coherence.** Carries the constructs and propositions consistently across chapters.
- **15.19 Examiner-Proofing and Defence Strategy.** Prepares the theoretical claim for expert challenge.

**Runs well alongside:**
- **02.07 Scale and Measurement Selection**, where a construct definition has to become a measure.
- **15.11 Discussion and Contribution Development**, where an empirical result is converted into a theoretical claim.
- **13.04 Bias Detection**, which checks whether the mechanism was derived from the data or fitted to it.

---
A Yazi Supplied Skill and resource.