6 Theory in the Discussion
Chapters 2 through 5 dealt with the introduction — finding your anchor, applying the filters, writing the funnel, and checking that the argument holds. The Discussion is a different challenge, and in some ways a harder one. Theory overload is most visible in the introduction, where the landscape of competing frameworks is impossible to miss. In the Discussion it does its quieter damage: by the time the reader reaches it, they have invested in your argument, and a Discussion that loses focus is harder to recover from than an introduction that does.
The pattern is predictable. A writer who mentioned four theories in the introduction feels obliged to return to all four in the Discussion. The results came out in favour of Event Segmentation Theory, so that gets a paragraph. Then, because the Event-Indexing Model was mentioned in the introduction, it gets a sentence noting that the results are “also consistent with” it. The Structure Building Framework gets something similar. Narrative Comprehension accounts are noted as offering a “complementary perspective.” By the end, every theory that appeared in the introduction has reappeared in the Discussion, and the reader is no longer sure which one the paper was actually about.
This is the Discussion version of the omnibus paragraph. It has a specific cause: writers confuse two distinct tasks — interpreting results and cataloguing them.
6.1 Two jobs the Discussion actually has
The Discussion has two theoretical jobs, and only two.
Job 1: Explain what the results mean for your anchor theory. Did the results confirm the prediction? If so, what does that tell us about the mechanism? If not — or not cleanly — what does that tell us? This is the primary job. It should take most of the theoretical space in your Discussion.
Job 2: Address theories that passed both filters. If a competing theory earned full treatment in your introduction — because it made a different prediction that your design could adjudicate — then your results say something about it, and the Discussion should say what. Did your results discriminate between the theories? If yes, state that clearly. If no — if the results were consistent with both — say that too, and explain what a future study would need to look like to adjudicate them.
A theory that received only a brief mention in the introduction — because it failed F2 — has not earned a paragraph in the Discussion. Results being “consistent with” a theory is not a reason to discuss it — almost any result is consistent with almost any sufficiently flexible theory. “Consistent with” is not an argument. The next section shows what it actually means.
6.2 The mirror test
Here is a simple diagnostic. Draw a line down the middle of a page. On the left, list every theory that received substantive treatment in your introduction. On the right, list every theory that receives substantive treatment in your Discussion.
The two lists should largely match — and the reason follows directly from the two filters. A theory that passed both filters earned full treatment in your introduction and generated a result your study can speak to; it belongs on both sides. A theory that failed F1 or F2 earned at most a mention in your introduction; it has not earned a paragraph in your Discussion either. The filter decision made in the introduction determines the Discussion’s shape.
| Theory treatment in introduction | Expected Discussion treatment |
|---|---|
| Full treatment (passes F1 and F2) | Full treatment — state what the results showed |
| Mention + explain (passes F1, fails F2) | One sentence — state that the study could not adjudicate |
| Citation only (fails F1) | No Discussion space |
Deviations from symmetry are sometimes legitimate, but each requires an explicit justification. A theory that loses space in the Discussion — because the results were null or ambiguous — still requires a verdict: say what the result means for that theory, even if the answer is that the study cannot adjudicate. A theory that gains space in the Discussion — because an unexpected finding opened up a theoretical connection that wasn’t visible before data collection — must be clearly flagged as post-hoc and exploratory.
For preregistered studies, and especially for registered reports, the mirror test has additional force. The left side of the page was fixed before data collection; it is part of the public record. The Discussion must account for every theory on the left, regardless of what the results showed. Dropping a preregistered competitor from the Discussion because the result was inconvenient violates the theoretical commitments the study was designed to test. Additions on the right are permitted, but must be labelled as unregistered, exploratory, and interpreted with appropriate caution.
6.3 What “consistent with” actually means
When writers say their results are “consistent with” Theory X, they usually mean one of three things: the theory makes no differential prediction for this design; both theories happen to predict the same outcome; or the theory is flexible enough to accommodate almost any result. In the first two cases, “consistent with” is true but uninformative. In the third, it is not even a meaningful claim.
The only case where “consistent with” earns space is when you can follow it with a sentence explaining why the consistency matters — what it adds to the argument that is not already provided by your anchor theory’s confirmation. If you cannot write that sentence, cut the phrase.
6.4 A before and after
Here is the Discussion opening from the bloated first attempt in Chapter 9:
The present results are consistent with Event Segmentation Theory, as predicted. Participants rated surprise as lower at event boundaries than within events, supporting the claim that boundary detection reduces prediction error. These results are also consistent with predictive processing frameworks, which similarly predict reduced prediction error at moments of perceptual transition. Situation Models offer a complementary perspective: the updating of spatial and causal representations at boundaries may contribute to the reduction in surprise we observed. Narrative Comprehension accounts suggest that goal-structure processing at event transitions could additionally account for the boundary effect.
Four theories in five sentences. The anchor result is buried under three sentences of “also consistent with.” The reader cannot tell which confirmation the paper is actually claiming.
Here is the same opening after applying the two-job rule:
Surprise ratings were lower at event boundaries than within events, confirming the prediction of Event Segmentation Theory that boundary detection triggers a model update that reduces prediction error. The effect was robust across all three activity types and held at both the group and individual level.
The present design does not discriminate between EST and predictive processing frameworks, which generate identical directional predictions for this paradigm. What the data do support is the more specific claim within EST that boundary-related prediction error reduction is perceptual rather than narrative in origin: the effect was present for clips of non-goal-directed activities as well as goal-directed ones, which is harder to account for on a goal-structure account.
Two paragraphs. The first states the confirmed prediction and its scope. The second does genuine theoretical work — it uses a feature of the results to narrow the theoretical picture. Theories that received only a brief mention in the introduction do not reappear here. The reader does not miss them.
The same logic that governed your introduction governs your Discussion. The anchor holds at both ends. The next chapter addresses what happens when a reviewer disagrees.