Does sentence combining improve writing?

Verdict

This topic exists separately from grammar instruction because the evidence forces it. Two independent meta-analyses put the two treatments in the same table with opposite signs — Hillocks 1986 at −0.29 and +0.35, Graham & Perin 2007 at −0.32 and +0.50. Andrews et al. commissioned two separate systematic reviews under the same protocol and reached opposite conclusions. Folding them together would produce an average that describes nothing and would make the archive unable to answer "what should I do instead?", which is the question a person who has just been told grammar does not work actually has.

So: moderate-support, and every word of that is doing work. The direction is consistent across three independent syntheses and one head-to-head randomised comparison. The magnitude is not established, the durability is barely measured, and the age range is narrower than the literature's own titles suggest.

The most important single fact in this topic for a reader with young children: the evidence does not reach them. The EPPI review is titled "…on 5 to 16 year olds' accuracy and quality in written composition." That is the search range. The studies it found ran "predominantly … from fourth grade (9–10 year olds) to tenth grade (15–16 year olds), with the majority clustering in the upper years of primary/elementary schooling and the lower years of secondary schooling." There is no sentence-combining evidence for five-, six- or seven-year-olds. That is a useful answer, not a missing one.

What the evidence shows

Saddler & Graham (2005) is the most informative single study and the reason this is not just a category artefact. Almost every sentence-combining study compares it with business as usual, so the effect could be "structured sentence work beats nothing." Saddler and Graham made the control condition grammar instruction delivered in the identical peer-assisted format, to fourth-graders, for the same time. Same teacher attention, same peer structure, same dose — only the content differs. Sentence combining wins. That is the cleanest demonstration available that the content is doing something.

The Englicious null contains a hint that points here. Wyse et al. found nothing on narrative writing (d = 0.04) but a larger, still non-significant d = 0.14 on sentence generation, and they argue explicitly that whatever signal exists came from "pupils' manipulation of words, phrases and sentences" rather than from the teaching of grammatical terms. That is the sentence-combining hypothesis showing up inside a grammar trial — suggestive, not evidence.

What the review actually claims is narrower than what it gets cited for. Andrews et al. conclude sentence combining improves syntactic maturity — measures like words per T-unit, clauses per sentence. Whether syntactic maturity is the same as better writing is an assumption, and one worth holding loosely: a child can write longer, more embedded sentences that are worse.

Hereditarian-lens assessment

Risk: low, with one honest qualification.

The corpus is experimental and quasi-experimental — children are assigned to sentence combining or to a comparison, so heritable differences do not differ systematically between arms. Saddler & Graham randomise within a single grade cohort with an active control. Nothing in the positive verdict depends on comparing children who happened to receive sentence combining with children who did not.

The qualification is about what a sentence-combining gain means under the archive's premise. Syntactic maturity is downstream of oral language, which is substantially heritable and shows minimal shared-environment influence (Olson 2013 finds shared environment significant only for vocabulary among writing, language and reading measures). Teaching a child to combine sentences gives them a procedure for producing structures they can already comprehend. That is a real skill and worth teaching — but it is floor-raising on a specific technique, not an increase in language capacity, and the delayed-post-test tempering the review reports is exactly what a taught procedure without continued practice looks like.

Saddler & Graham's skilled-versus-less-skilled split should be read the same way as the equivalent split in the grammar topic: descriptive, not a demonstrated aptitude-treatment interaction.

Boundaries & what critics say

• The review's own authors do not think the case is proved. Andrews et al. (2006), summarising both reviews: "in both cases, however, despite over a hundred years of research and debate on the topic, there is insufficient quality of research to prove the case with either approach." The positive verdict here is relative to grammar and relative to nothing — it is not a claim of a well-established effect size.
• No pooled effect size has ever been computed for the EPPI corpus. Eighteen studies, narrative synthesis, no meta-analytic estimate. The only numbers available are Graham & Perin's +0.50 from k = 5 and Hillocks's +0.35 — small k, researcher-scored quality, in a table whose homogeneity test the authors themselves flag.
• No sentence-combining programme has ever cleared an independent-measure standard. Slavin et al.'s Best Evidence synthesis requires measures independent of developers, researchers and teachers; fourteen studies of twelve programmes across Grades 2–12 qualified, and the overall effect on writing was +0.18. Not one was a sentence-combining programme. The honest expectation, applying that discount, is that the true independent-measure effect is a fraction of +0.50.
• The corpus is old and monocultural. 17 of 18 studies from the USA, one from Canada, searches running back to 1900. Andrews et al.'s own research recommendation is to "move beyond the USA into different contexts" and to "undertake some large-scale and longitudinal experimental studies."
• Durability is essentially unmeasured. One study in eighteen ran a delayed post-test. It found gains maintained "albeit less dramatically." Everything else is acquisition-phase — the same weakness the archive records for motor competence.
• Syntactic maturity is not writing quality. Longer, more embedded sentences are the measured outcome in most of this corpus. Conflating that with better writing is the same category slip the archive refuses elsewhere.

Practical guidance

• Use it, from about age 9, as a routine inside writing lessons — not as a separate course. Give two or three short kernel sentences; have children combine them; then have them apply the same move to a sentence in their own draft.
• Pair children up. The best-controlled study used a peer-assisted format at Grade 4, and it is free.
• For 5-to-7-year-olds, do something else. There is no evidence here for that band, the ability to combine sentences with conventional grammar arrives around age seven, and the only robust trial at ages 6–7 found d = 0.14 (ns) on sentence generation. At that age the binding constraints are transcription, spelling and oral language.
• Keep doing it. The one delayed post-test in the literature shows tempering after instruction stops. Treat it as a maintained habit, not a unit to complete.
• Do not pay for a sentence-combining programme. The intervention is a technique, the materials are two sentences on a board, and no branded version has been shown to beat the technique.
• Expect a modest effect. If you are budgeting, budget from Slavin's +0.18 for independently measured writing programmes, not from +0.50.

Open questions

• Nobody has run a modern, adequately powered, independently evaluated RCT of sentence combining. The whole positive case rests on a pre-1990 North American corpus plus one small Grade 4 study. This is the most obviously missing trial in the subject.
• Does it work below age 9? Completely untested. Given that this is where the demand is — the archive's motivating question is about five- to seven-year-olds — it is the highest-value gap in this tranche.
• Does syntactic maturity translate into better writing on an independent measure? Assumed throughout the literature, tested nowhere.
• How much of the Englicious sentence-generation signal is sentence combining? The authors think the manipulation of words and phrases is the active ingredient. A trial that isolated the combining component from the terminology component at ages 6–7 would answer both this and the previous question.