The attention tax is usually asserted in craft literature and rarely argued. It can be argued, and — more interestingly — it can be argued against, from within the same critical tradition. This section assembles both cases, because a writer who understands why a second chain is expensive prunes more intelligently, and a writer who understands why the expense is sometimes the point prunes less.
Seven strands follow: the classical objection and its oldest answer; Levin on the relations between plots; Empson on juxtaposition as a machine for meaning; Chatman on kernels and satellites; Woloch on the distribution of narrative attention, approached from the plot side; the historical conditions that produced the multi-plot novel and then withdrew them; and the cognitive literature on tracking interleaved threads. The section closes by restating the chapter’s first principle and immediately qualifying it.
3.1 Unity of action: the oldest objection, and Forster’s answer
The case against the subplot is as old as poetics.
Aristotle’s Poetics (c. 335 BCE) makes unity of action the organising requirement of tragedy: the plot must imitate one action, complete and whole, and its parts must be so connected that the removal or displacement of any one of them disjoints the whole. The negative formulation is more useful for present purposes than the positive one. Aristotle names the episodic plot — the plot in which episodes succeed one another without probability or necessity — as the worst kind, and his objection is the one the attention tax formalises: material that does not participate in the causal structure of the action does not belong to it.
Two clarifications prevent the standard misuse of this authority.
First, Aristotle’s “double plot” is not a subplot. In Poetics chapter 13 he discusses tragedies with a double ending — the good rewarded and the bad punished — and judges the structure second-best, granting it its popularity only through the weakness of audiences. This is a claim about closure, not about secondary chains, and citing it against subplots is a mistranslation of the argument.
Second, Aristotle permits in epic what he restricts in tragedy. The epic, being longer, may take on epeisodia — episodes that extend the work without belonging to the central action — and he treats this as a formal property of the longer form rather than as a licence for incoherence. The distinction matters for the novel, which is generically epic in scale and dramatic in its inherited critical vocabulary. Much bad advice about subplots consists of applying tragedy’s unity requirement to a form Aristotle would have classed with epic.
The oldest answer to the objection is E. M. Forster’s, in Aspects of the Novel (1927). Forster’s late chapters on pattern and rhythm propose that a novel may be organised by relations other than causality: by shape (his examples include the hour-glass structure of The Ambassadors, in which two characters exchange positions), and by rhythm — repetition with variation, the recurrence of a motif whose reappearance carries an accumulated charge, for which his example is the little phrase of Vinteuil in Proust. Forster’s contribution to this chapter is a permission: a novel may be unified by pattern rather than by plot, and a second chain may be integrated by rhyme rather than by consequence.
Forster also supplies the corresponding caution, and it is honest. He notes that pattern, pursued rigorously, tends to constrain what the novel can contain — that the beautifully shaped book pays for its shape in life. A subplot admitted for pattern alone is subject to this risk, and §8.13 names the resulting failure.
3.2 Levin: the relations between plots
Richard Levin’s The Multiple Plot in English Renaissance Drama (University of Chicago Press, 1971) is the definitive study of how secondary actions relate to primary ones, and it remains the source of the most useful vocabulary available for the problem. Levin’s material is the English stage between roughly 1580 and 1640, a body of drama in which multiple plots were the norm rather than the exception; his method is to ask, of each play, what kind of relation the actions bear to one another, and to resist the assumption that the relation must be causal to be real.
Three of Levin’s categories are load-bearing for this chapter.
Direct linkage. The secondary action affects the primary one causally: its events change what is possible in the main plot, and the two are connected at the level of what happens.
Analogy. The secondary action repeats the primary action’s pattern in a different key — a different social level, a different register, a different set of persons — so that the two illuminate each other by resemblance.
Contrast (Levin). The secondary action inverts the primary one: the same situation resolved the other way, the same choice made differently, the same pressure producing the opposite result.
Levin’s central and most useful argument is that analogical relation can integrate a play whose plots never physically meet. A secondary action that shares no character and no event with the main action may nonetheless be structurally integral, because the relation between the two is the play’s argument. He develops the point against a critical tradition — descending through Dryden and Johnson to the nineteenth century — that treated multiple action as a native English vice, an offence against unity to be excused rather than analysed.
Two refinements of Levin’s model are worth carrying into novel criticism.
The three-level play. Levin observes that Renaissance dramatists frequently constructed three actions rather than two, arranged on descending social levels — courtly, gentle, plebeian — with the relations of analogy running down the ladder. The structure survives in the multi-plot novel: Bleak House has its aristocratic Dedlocks, its professional and middling Chancery suitors, and its Tom-all-Alone’s, and the argument of the book is carried by the fact that a single secret runs through all three.
Relations may be mixed and may change. A thread may begin in analogy and end in direct linkage; the change is itself an event, and in the novel it is usually the event called convergence. This gives the chapter a precise way to say what convergence does: it converts an analogical relation into a causal one, and the reader’s experience of a well-managed convergence is the retrospective recognition that the rhyme was also a mechanism.
A caution about the transfer. Levin’s plays are two to three hours long and their plots are held in a single sitting. A novel’s threads are separated by days of reading. Analogy that is instantly perceptible across an interval of twenty minutes on stage may be imperceptible across an interval of eighty pages, and the novelist must therefore mark analogies that a dramatist could leave implicit. §6.6 gives the mechanism.
3.3 Empson: the double plot as a machine for meaning
William Empson’s essay “Double Plots” — the second chapter of Some Versions of Pastoral (1935) — is the strongest available defence of the second chain, and it argues on grounds the attention tax cannot reach.
Empson’s subject is the Elizabethan and Jacobean habit of running a “low” comic action beside a “high” tragic or heroic one, and his claim is that the pairing is not a concession to the groundlings but a device for producing meaning. The two actions are set beside each other so that each is read through the other; the comic action’s coarse version of the heroic action’s dilemma is a comment on it, and the heroic action’s elevation is a comment on the comic one. Neither comment is stated. Both are produced by juxtaposition, and — this is Empson’s essential move — the meaning generated belongs to neither plot alone.
The consequences for this chapter are large enough to state as a list.
- The double plot generates a surplus. If the relation between two actions yields something neither carries by itself, then the second chain is not simply a deduction from the first’s page budget. It is an investment that pays in a currency the single-plot novel does not have. This is the most serious argument against the attention tax available in the critical literature, and the chapter accepts it.
- The “low” plot is not decorative. Empson’s reading of the Falstaff material in the Henry IV plays makes the tavern action a sustained interrogation of the honour the main action asserts. The comic thread is not relief from the argument; it is the other half of it. §8.7 builds its failure pattern on the difference between these two things.
- Juxtaposition is a technique with its own craft. Meaning by proximity requires the proximity to be managed: the two actions must be near enough in the reader’s memory for the comparison to occur. Empson’s dramatists could rely on a single sitting; the novelist cannot, and §6.6 converts this into a rule about dormancy.
- The relation may be felt without being formulable. Empson is careful that the effect works below the level of paraphrase — that an audience feels the two actions commenting on one another without being able to say what the comment is. Craft advice that demands a subplot’s thematic contribution be statable in a sentence is therefore demanding something the mechanism does not require. §7.1, test 4 is worded to allow for this.
Where Empson is vulnerable. His readings are frankly ingenious, and it is possible to construct an Empsonian defence of almost any two things placed side by side — a fact his critics have not been slow to observe. The defence “these threads comment on each other” is therefore unfalsifiable as stated, and a writer using it should be required to say what the comment is doing that a single plot could not do. That requirement is the whole content of §8.13, the thematic alibi.
3.4 Chatman: kernels, satellites, and what a subplot is
The sharpest instrument in this chapter comes from Seymour Chatman’s Story and Discourse (1978), and it is worth setting out with care because it converts several of the chapter’s intuitions into something that can be checked.
Chatman, developing Barthes’s distinction between cardinal functions and catalyses, divides narrative events into two classes.
Kernel. A narrative moment that raises and answers a question about the direction of events — a branching point. Kernels cannot be deleted without destroying the narrative logic: remove one and the story’s causal structure fails.
Satellite. A narrative moment that does not open an alternative. It elaborates, delays, textures or amplifies. Satellites can be deleted without destroying the logic, though the discourse is impoverished by the deletion.
Chatman is explicit that satellites are not worthless. Their deletion damages the aesthetic of the narrative while leaving its skeleton intact; a novel consisting only of kernels would be a synopsis. The distinction is between two kinds of necessity, not between the necessary and the disposable.
Applied to the subplot, the distinction yields the most precise definition this chapter can offer, and it is worth setting out as a proposition.
A subplot is a chain of events that are kernels within its own system and satellites within the main plot’s — until convergence, which is the moment its kernels become the main plot’s kernels too.
Unpack the proposition in three parts.
Kernels within its own system. This is the chain requirement of §2.1 restated formally. Inside the subplot, its turning points must be branching points: this thread must be capable of going one way or another, and its scenes must be the moments where that is settled. A thread whose “turning points” are all satellites even inside its own system is not a subplot at all — it is texture. This is what §7.1, test 2 measures.
Satellites within the main plot’s. This is the attention tax, formalised. Before convergence, the subplot’s events can be deleted from the main plot’s causal structure without breaking it. That is what makes them cost something: they are, from the primary line’s point of view, elaboration.
Until convergence. And this is why convergence is the pivot of the whole chapter. Convergence is the moment at which the subplot’s kernels are promoted into the main plot’s kernel set — the moment when deleting the thread would break the main story’s logic and not merely thin it. Before that moment the removal test (§7.1, test 9) is a question about impoverishment. After it, the removal test is a question about coherence.
Three practical uses follow immediately.
| Use | Procedure | What it reveals |
|---|---|---|
| Locating convergence | Find the earliest point at which deleting the thread breaks the main plot’s causal logic | Where convergence actually is, as opposed to where the writer believes it is |
| Diagnosing the vestigial thread | Ask whether the thread’s own turning points are kernels inside it | A thread of pure satellites has no chain and cannot converge |
| Measuring the tax | Count the pages spent on main-plot satellites before convergence | The exact size of the loan the convergence must repay |
A caution. Chatman’s categories are cleaner in theory than in application, and the boundary between kernel and satellite is frequently arguable — a fact he acknowledges. Two competent readers can disagree about whether a given scene opens an alternative. The instrument is therefore diagnostic rather than decisive: its value is in forcing the question what would break if this were gone?, which is a question drafts are rarely asked.
3.5 Woloch from the plot side: which lives get plots
Chapter 13 (The Viewpoint Shift) uses Alex Woloch’s The One vs. the Many (2003) to theorise the empathy budget, working from the character side: character-space, the compression of the minor character, and the distortion that compression produces. That material is not repeated here. What follows approaches the same book from the other direction, and the direction changes what the theory says.
Woloch’s core claim is that a novel’s characters form a character-system — an arrangement of delimited character-spaces in which attention given to one is attention withheld from another. The claim this chapter takes up is a corollary Woloch does not need to state, because his object is characterisation:
A character-system is also a plot-system. The same finitude that determines how much space a character receives determines whether they receive a chain. Some characters are granted goals, obstacles and reversals; others are granted appearances. The distribution of chains across a cast is a formal structure in the same sense that the distribution of interiority is.
Three consequences follow, and they are the reason this chapter needs Woloch as well as Levin and Empson.
First: having a plot is a stronger grant than having a viewpoint. A character may be focalised for a chapter and still have no chain — the reader inhabits them and learns nothing about what they want. Conversely, a character never focalised may possess a complete chain that the reader reconstructs from outside: Dov, in this book’s running example, is observed only from outside until a single bounded chapter near the climax, and has a full goal–conflict–disaster chain throughout. Viewpoint confers access; a plot confers consequence. The second is the scarcer and the more consequential grant, and it is not tracked by the vocabulary of viewpoint at all.
Second: the distribution of chains carries an argument. Woloch’s insistence that the character-system is a political fact transfers directly and, arguably, more sharply. To give a character a plot is to say that their wanting produces effects in the world — that their life has causal weight. To deny them one is to say that their life is a condition rather than a story. A novel in which the servants appear constantly and never want anything is making a claim about servants, whether or not its author intended one. A novel about a strike in which the owners have a plot and the workers have a crowd is making a claim about who acts.
Third: the tax and the grant are the same fact seen from two sides. The attention tax says: this thread costs the main line. Woloch’s model says: the main line’s centrality is produced by what the other threads are denied. The protagonist is not naturally central; centrality is manufactured out of the compression of everyone else. Held together, the two claims produce the chapter’s least comfortable question, which is set as a seminar topic at §12.2 and asked as a diagnostic at §7.1, test 10:
The promotion question. If this thread were the main plot, would the book be better?
A writer who cannot answer that question has not decided which book they are writing. A writer who answers yes has learned something a hundred pages of line-editing would not have told them.
A caution. Woloch’s model is calibrated to the nineteenth-century realist novel — large casts, a strong protagonist convention, an omniscient habit. It describes the multi-plot novel superbly and a two-hander poorly. Chapter 13 §3.3 enters the same caveat from the character side, and it applies here with equal force.
3.6 The historical strand: what produced the multi-plot novel, and what withdrew it
Contemporary suspicion of subplots is presented in craft literature as a discovery about narrative. It is substantially an artefact of book length, and book length is an artefact of the market. Establishing this is not a debunking exercise; it is the difference between knowing a convention and mistaking it for a law.
The Renaissance stage. Multiple plotting was the English theatrical norm for two generations, and Levin’s study documents the extent: double and triple actions in the majority of surviving plays. The conditions were a repertory system, a mixed audience, and an aesthetic with no allegiance to the neoclassical unities — an allegiance that arrived later, in the Restoration and through French influence, and brought with it the critical embarrassment about Shakespeare’s double actions that Levin’s book was written to dispel.
The Victorian three-decker and the monthly number. The nineteenth-century English multi-plot novel was underwritten by two material arrangements that no longer exist.
| Condition | Mechanism | Effect on plot structure |
|---|---|---|
| Serial publication in parts | Novels issued in monthly (or weekly) instalments over one to two years | Each part had to contain variety; alternation between threads was the simplest way to supply it |
| The circulating library | Mudie’s and its rivals bought in bulk and lent by subscription, and preferred the three-volume format because three volumes could be lent to three subscribers | Length was commercially favoured rather than penalised; expansiveness was an asset |
| Sale by the volume, not the reader’s hour | A publisher’s revenue was tied to a physical format | The economic pressure ran towards more pages, not fewer |
| Slow reading, long runs | A novel might occupy its readers for eighteen months | Threads could be left dormant for a month and reinstated by the part’s own recapitulative habits |
The consequences are visible in the books. Vanity Fair (1847–48) and Bleak House (1852–53) were each issued in twenty monthly numbers. Middlemarch (1871–72) appeared in eight parts, a format Eliot and her publisher devised for it. War and Peace began serial publication in the Russian Messenger under a different title before being completed as a book. Anna Karenina was serialised in the same journal between 1875 and 1877, and its final part appeared separately after the editor declined to print Tolstoy’s treatment of the Serbian war — a fact that matters here, because it means the novel’s most disputed structural decision was also a publishing event.
The withdrawal. The three-decker collapsed in the 1890s when the circulating libraries refused to sustain its price, and the single-volume novel became standard. The modernist novel then narrowed the field further, for reasons internal to its ambitions rather than commercial: a form organised around consciousness rather than around action has less use for a second chain of actions. Mrs Dalloway and To the Lighthouse have many lives in them and very few plots.
The present. The contemporary literary novel is conventionally between eighty and a hundred and ten thousand words — roughly a third of Middlemarch. At that length, a subplot occupying twenty per cent of the book is sixteen to twenty-two thousand words, and the arithmetic of the attention tax becomes severe in a way it was not for Dickens. Meanwhile the braided structure survived, and thrives, in the places where length survived: epic fantasy, the multi-generational saga, the historical sequence, and — most vigorously — long-form television, which reinvented the monthly number as the season and rediscovered the A/B/C plot as an industrial method (§10.4).
The inference to draw. The contemporary suspicion of subplots is not a discovery about how narrative works. It is what a two-hundred-year decline in the average length of a novel looks like when it is restated as craft advice. Learn it as a convention of the present market, and read historically enough to know that it is one.
Franco Moretti’s quantitative work — Graphs, Maps, Trees (2005) and the later essays on plot as network — is the natural place to pursue this, because it insists that formal change in the novel is legible as a function of quantity and market rather than only of intention.
3.7 The cognitive strand: what interleaving costs a reader
The claim that braiding has a cost is empirically supported, though the support is indirect and the chapter should not overstate it. Two literatures bear on it.
Situation models and updating. Rolf Zwaan and Gabriel Radvansky’s synthesis (1998) holds that readers construct a situation model of the state of affairs a text describes, tracking a small number of dimensions — protagonist, space, time, causality, intentionality — and incurring a measurable processing cost whenever one of them shifts. A crossing from one plot thread to another typically shifts all five simultaneously: different people, elsewhere, at a different point, pursuing different causes, wanting different things. This is the largest single update a narrative can ask for short of a change of narrator, and it is why Chapter 5 (The Transition) treats thread crossings as the most expensive transitions in the repertoire.
Frame recall. Catherine Emmott’s Narrative Comprehension (1997) supplies the mechanism the writer actually needs. Emmott’s contextual frames are the reader’s running record of who is present in a scene and where; a frame that is left is not deleted but primed — held in a recoverable state — and reactivated when the narrative returns to it. Her account of frame switching and frame recall gives braiding its precise cognitive description: a thread is not re-read on resumption; it is recalled. The reader’s ability to resume depends on the frame being recoverable, and recoverability decays.
Three inferences for the writer, stated as claims about tendency rather than as laws:
- Reinstatement is cheap when the frame is warm and expensive when it is cold. A thread resumed after twenty pages costs a clause of re-anchoring. A thread resumed after two hundred costs a scene of re-establishment — which the reader experiences as the book explaining itself.
- Cost is a function of interval and of distinctiveness together. A thread with a strongly individuated cast, setting and register decays more slowly, because fewer cues are needed to recall it. This is the cognitive justification for the old practical advice to make secondary threads look and sound different.
- The reader’s felt experience of a cold resumption is not confusion but resentment. They do not usually fail to reconstruct the thread; they resent being asked to. This is why the late-arriving and long-dormant thread (§8.5) registers as a pacing complaint rather than a comprehension complaint, and why writers misdiagnose it.
§6.6 converts these into working intervals and a dormancy audit. The chapter states plainly that the intervals there are conventions derived from practice, calibrated by the cognitive literature rather than derived from it. No experiment has established how many pages a novel’s B-plot may lie dormant, and any source that claims otherwise should be distrusted.
3.8 The attention tax, restated — and qualified
The chapter’s first principle can now be stated in full.
The attention tax. Every scene given to a secondary chain is a scene not given to the chain carrying the reader’s primary investment. A subplot therefore does not clear its cost by being well written; it clears its cost by supplying something the main plot structurally cannot supply on its own, and by eventually converting its own kernels into the main plot’s (§3.4).
And now the qualification, which is where the more interesting teaching lies.
The tax metaphor presupposes a single primary investment. It assumes there is a story the reader “came for”, against which everything else is measured. That assumption is accurate for most commercial narrative, for genre fiction of nearly every kind, for the well-made play, for the screenplay, and for the great majority of manuscripts a workshop will see. It is false for the multi-plot novel.
Middlemarch does not have a subplot. It has four plots — Dorothea and Casaubon; Lydgate and Rosamond; Fred Vincy and Mary Garth; Bulstrode — and they are not ranked. The novel’s own account of its method is a web, not a trunk with branches: the narrator, pausing in Chapter 15, says that all the light at her command must be concentrated on this particular web and not dispersed over that tempting range of relevancies called the universe. A web has no main strand. Ask which of the four is “the” plot and the question does not resolve; ask what the relation between them is, and the novel answers at length.
The same is true, with variations, of War and Peace, Bleak House, Vanity Fair and The Brothers Karamazov. In each case, applying the attention tax mechanically produces a confident absurdity: that Eliot should have cut Lydgate, that Tolstoy should have chosen a family, that Dickens should have decided between Chancery and Chesney Wold.
Why, then, teach the tax at all?
Because it is a stress test, not a description. Like the zero-context test in Chapter 9 (The Opening Pages) §3.7, its value lies in its austerity. A thread that survives the tax has a demonstrable function. A thread that fails it may be a co-plot in a multi-plot novel — but the writer must then say so, and accept the obligations that come with it, which are set out at §5.4. In this chapter’s judgement, the commonest structural failure in ambitious manuscripts is not the presence of a second plot. It is a book that has half-declared itself a multi-plot novel and half-retained a hierarchy, so that the reader cannot tell whether the second thread is a co-plot to be weighed or a subplot to be tolerated.
The double position to hold. For a book with one primary investment, the tax is the correct standard and should be applied without mercy. For a book without one, the tax is the wrong instrument, and the right question is Levin’s and Empson’s: what does the relation between these plots produce that neither produces alone?
The rest of the chapter is written to serve both cases. §4 assumes a hierarchy; §5.4 and §6.4 do not.