Evidence and Support

A paper is only as convincing as the material a skeptical reader can go check. Everyone agrees with that in the abstract and forgets it in the draft, where confident assertion is cheap and support is work. Evidence is what converts an opinion into a position: the study, the passage, the case, the measurement, the observed detail a reader can weigh independently of how much you want to be right. But evidence never argues on its own. A quotation dropped into a paragraph sits there inert until someone says what it shows, and a number can be perfectly accurate while supporting nothing at all. Both halves matter — choosing support that genuinely fits the claim it is asked to carry, and doing the reasoning that fastens the two together. Most of what follows works from weak paragraphs rather than finished ones, because a broken paragraph shows its machinery.

Section 1What Counts as Evidence

Saying is not showing

Read these two sentences as a stranger would. “The new orientation program has been very successful.” And: “Of the twenty-two nurses who completed the new orientation program in 2024, nineteen were still on the unit twelve months later; under the old program, the twelve-month figure had been eleven of twenty-four.” The first sentence asks you to trust the writer. The second hands you something and lets you decide. That is the whole difference, and it is not a difference of style.

An assertion is a statement offered without support. Assertions are not forbidden — every thesis is one, and every topic sentence is one — but an assertion is a promissory note. Evidence is what redeems it: material outside the writer’s own confidence that a reader can inspect, count, look up, or reread. The test is blunt and useful. Point at a sentence and ask, what here could a reader check without taking my word for it? If nothing, the sentence is a promise, not a payment.

Notice what happens to the vague words when evidence arrives. “Successful” became a retention rate. “Very” disappeared entirely, because it had nothing to do. Adjectives and intensifiers are what writers reach for when they have a conviction and no support; watching your own sentences for very, extremely, clearly, obviously, and devastating is a surprisingly reliable way to find the places where the paper is running on fumes.

The main kinds of evidence, and what each one can carry

Evidence is not one substance. Different kinds answer different questions, and a great deal of weak college writing comes from grabbing whichever kind is easiest to find rather than the kind the claim requires. Six families cover nearly everything you will use.

Empirical data are observations gathered systematically: measurements, survey results, counts, experimental outcomes. Textual evidence is the language of a source itself — a passage from a novel, a clause in a policy, a phrase in an advertisement. Expert testimony is the considered judgment of someone with standing in the field. Cases and examples are particular instances narrated in enough detail to be examined. Anecdotal evidence is individual experience, yours or someone else’s, reported without systematic collection. Analogy and precedent reason from a similar situation to the one at hand: this worked in Denver, this is how the court ruled in an earlier case.

Each family has a ceiling. Empirical data can establish that a pattern exists and cannot establish why anyone should care about it. Textual evidence can show what a document says and cannot show how readers received it. Expert testimony can transfer credibility and cannot substitute for the reasoning the expert used. Analogy can illuminate and can never prove, because the two situations are by definition not identical. Knowing the ceiling keeps you from asking a piece of evidence to do work it was never able to do.

Evidence is audience-relative

There is no such thing as evidence in general. There is only evidence for a particular reader, and readers differ in what they will accept without argument. A citation to a peer-reviewed trial settles the matter for one audience and means nothing to another that distrusts the funding source. A scriptural quotation is decisive within a community that shares the text and inert outside it. A veteran nurse’s clinical intuition carries real weight with colleagues and none with a budget committee.

This is not relativism about truth. It is a practical fact about persuasion, and it changes what you gather. Before you go looking for support, ask what your specific reader already grants. The evidence you need is whatever stands between what they will concede for free and what you want them to accept. Everything else is decoration, and decoration lengthens a paper without strengthening it.

A worked instance. Suppose you are arguing that your hospital should stop rotating night-shift staff onto day shifts every two weeks. Writing for fellow nurses, you can lean on shared experience and short-cycle rotation research, because the audience already believes fatigue is real. Writing for a chief financial officer, shared experience will not move anything; you need overtime costs, turnover replacement costs, and error-related length-of-stay data, because that reader concedes fatigue and disputes only whether fixing it pays. Same position, two entirely different evidence sets.

Section 2Matching Evidence to the Claim

Five kinds of claim ask five different questions

Ancient rhetoricians noticed that arguments stall at predictable points and named them the stases. The modern version, taught in most composition courses with minor variations in the labels, sorts claims into five types. A claim of fact asserts that something is or is not the case. A definitional claim asserts that something belongs in a particular category. A causal claim asserts that one thing produced another. A claim of value asserts that something is good, bad, better, or worse. A claim of policy asserts that something should be done.

The taxonomy is useful for exactly one reason: each type is settled by a different kind of evidence, and mismatches are the single most common structural failure in student support. A causal claim propped up on definitions goes nowhere. A value claim buried under statistics never gets argued, because no quantity of data tells anyone what matters. A policy claim is the greediest of all — it usually needs a fact claim (the problem exists), a causal claim (this is what drives it), a value claim (it is worth fixing), and a feasibility claim (this remedy would work) before it can stand.

Try it on one subject. “Third-shift call volume has risen 40 percent since 2022” is fact, settled by records. “What the dispatch log calls a wellness check is functionally a mental-health call” is definitional, settled by criteria plus cases. “The rise is driven by the county clinic closure” is causal, settled by timing, comparison, and ruled-out alternatives. “Sending patrol officers to these calls is the wrong response” is value, settled by criteria and consequences. “The county should fund a co-responder team” is policy, and it needs all four of the others.

Ask what would settle it before you go looking

The most efficient research habit in this book takes one sentence. Before opening a database, write down: if I found ______, I would have to concede I was wrong. Filling that blank forces you to specify what kind of evidence bears on your claim, which tells you what to search for and, just as importantly, what to stop collecting.

Watch it work. Claim: “The new tutoring center improved outcomes in first-year composition.” What would settle it? Pass rates before and after, ideally with a comparison group of similar students who did not use the center, and some way of handling the fact that students who seek tutoring may already be more motivated. Notice how much that single question just clarified. You now know you need two time periods, a comparison, and an answer to the self-selection problem. A student who skipped this step tends to come back with six articles about tutoring in general and nothing about this center.

The habit also protects you from the most demoralizing experience in research writing: gathering for hours and having nothing usable. Sources are usable when they answer a question you specified in advance. Chapter 8 is where you learn to find and vet those sources; the point here is that searching without a target produces piles rather than support.

When the evidence you found does not fit the claim you made

This happens constantly, and there are only three honest responses: change the claim, find different evidence, or admit the limit out loud. Papers get into trouble when writers choose a fourth option and simply hope the reader will not notice the gap.

Consider a real pattern from student drafts. The claim: “Mandatory overtime causes medication errors.” The evidence found: a survey in which 68 percent of nurses reported feeling less safe during mandated overtime. That evidence is genuine, but it supports a different claim — that nurses perceive reduced safety. Perception and error rate are not the same variable, and a careful reader will say so within a sentence of reading it.

Three repairs, all legitimate. Narrow the claim to what the evidence supports: “A majority of nurses working mandated overtime report reduced confidence in their own safety.” Or go find the evidence the original claim requires: incident reports, error rates by shift length, a study with a comparison group. Or keep the original claim and mark the limit explicitly: “Direct error-rate data by shift length is scarce, so the strongest available support is self-report, which likely understates errors that go unnoticed.” That third move costs a writer nothing. Naming a limitation reads as command of the material, not weakness — and a reader who spots the gap before you do stops trusting everything else.

Section 3Making Evidence Mean Something

The dropped quotation and the orphan statistic

Here is the most common paragraph failure in first-year writing, and it is not a failure of research. The writer states a point, produces a genuinely relevant quotation or number, and then moves to the next point. No sentence in between. The evidence sits in the paragraph like furniture in a hallway.

The reason this fails is that evidence does not carry a single obvious meaning. Any given fact is compatible with several conclusions, which is exactly why arguments about facts happen. When you drop a quotation without comment, you leave the interpretation to the reader — and readers, left alone with evidence, reliably interpret it the way they already believed. The one sentence you skipped was the only sentence doing argumentative work.

What has to go in that slot is analysis: a statement of what the evidence shows and why that supports the claim. Analysis is not summary. If your sentence after the evidence could be replaced by “as you can see above,” you have restated, not analyzed. The reliable test is whether the sentence contains an idea absent from the evidence itself. Restatement recycles the evidence in new words; analysis adds the inference.

The four-move paragraph, and why formulas are scaffolding

You have probably met this shape under an acronym: PEEL, TEAL, ICE, MEAL, claim-evidence-warrant. The initials differ; the architecture does not. A body paragraph normally makes a claim, presents evidence, analyzes it, and states why it matters — then either returns to the governing thesis or hands off to the next paragraph.

Here is the shape filled in, on a subject where the temptation to skip the analysis is strongest. “Claim: The unit’s hand-hygiene campaign fails for reasons that have nothing to do with knowledge. Evidence: The pre-campaign staff survey found that 96 percent of respondents correctly identified the five moments for hand hygiene, and observed compliance in the same month was 51 percent. Analysis: A forty-five-point gap between what staff know and what staff do rules out a knowledge deficit as the driver, which is the deficit the posters were designed to correct. Significance: Any campaign built on reminders is therefore treating the one variable that was never broken, and the money would do more work relocating dispensers into the workflow.” Four sentences, and the paragraph argues.

An honest caveat about formulas, because instructors genuinely disagree here. Some hold that templates like PEEL give inexperienced writers a usable shape and free attention for thinking. Others hold that templates produce mechanical, one-idea-per-paragraph prose that students then have to unlearn. Both observations are correct about different students at different stages. Treat the four moves as a diagnostic rather than a mold: if your paragraph is missing one of them, find out whether that was a choice. Skilled writers distribute the moves across several paragraphs, or embed the analysis inside the sentence introducing the evidence. What they never do is omit the analysis.

Warrants: the assumption that lets evidence count

Between a piece of evidence and the claim it supports there is always an unstated assumption. Philosopher Stephen Toulmin named it the warrant, and it is the hinge on which most real disagreements actually turn. Chapter 7 develops the full model; what you need here is the habit of finding your own warrants and deciding whether your reader shares them.

An example. Evidence: the campus food pantry served 2,400 visits last year. Claim: the college should fund it permanently. The warrant, unstated, is roughly an institution should sustain services its students demonstrably use. Many readers grant that instantly. A trustee focused on mission boundaries might not — that reader could accept every number and still reject the claim, because the assumption connecting them is precisely what they dispute. When a reader agrees with all your facts and rejects your conclusion, you have found an unshared warrant, and no additional facts will help. What helps is arguing for the assumption itself.

To surface a warrant, put the evidence and claim in a sentence joined by because, then ask what has to be true for the joint to hold. “The college should fund the pantry because 2,400 visits occurred” only holds if usage obligates support. Say the assumption aloud in your own draft. If it is uncontroversial, leave it implicit and save the words. If your specific reader would balk, that is the sentence your paragraph was missing.

Section 4Testing Evidence: Enough, Typical, Accurate, Relevant

The STAR audit

A widely taught heuristic in composition, associated with the Writing Arguments tradition, gives evidence four tests under the acronym STAR: sufficiency, typicality, accuracy, relevance. It is memorable and it maps onto the four ways support actually fails, which is why it survives.

Sufficiency asks whether there is enough. One study, one interview, one incident supports a modest claim and collapses under a sweeping one. Typicality — also called representativeness — asks whether your instances stand in fairly for the whole. A vivid case can be entirely accurate and entirely unrepresentative, which is how a single dramatic story steers a policy conversation for a decade. Accuracy asks whether the material is correct and correctly reported, including whether you have preserved the qualifications the original attached. Relevance asks whether the evidence bears on this claim rather than a neighboring one, which is the failure the mandatory-overtime example in Section 2 illustrated.

Run the audit on a draft paragraph rather than on a whole paper. Point at each piece of support and ask the four questions in order. Most drafts have one recurring weakness rather than four scattered ones, and finding your own pattern — you over-generalize, or you always take numbers on trust — is worth more than any single fix.

How much is enough

There is no number, and anyone who gives you one is describing a course policy rather than a principle. Sufficiency is set by three things: the size of the claim, how much resistance the audience brings, and what the discipline conventionally accepts. A claim about one poem needs evidence from that poem. A claim about American poetry in the 1920s needs considerably more, and a reader will notice if it rests on three poems by one poet.

Two useful reference points. James Lind’s 1747 shipboard trial of scurvy treatments allocated twelve sailors to six treatments — two men per group. The pair given oranges and lemons recovered; the other five pairs did not. By modern standards two men per arm is far too few to be conclusive, and Lind himself never presented citrus as a proven cure — nearly fifty years passed before the Royal Navy issued lemon juice as standard. At the other end, a modern trial enrolling thousands can detect an effect so small that it is statistically real and practically trivial. Sufficiency is not a quantity. It is a relationship between the evidence and the size of the claim it is asked to carry.

The practical version for a college paper: match the reach of your sentence to the reach of your support. If you have one well-documented case, write a claim about what that case shows is possible. If you have three studies pointing the same way, you can write about a pattern. If they conflict, the honest claim is about the conflict — and a paper that accurately reports a genuine disagreement in the literature is more sophisticated than one that pretends the question is closed.

Counterevidence and the cherry-pick

Cherry-picking is selecting the evidence that fits and omitting what does not. It is rarely a deliberate lie. Far more often it is what happens when a writer forms a conclusion early and then reads only for confirmation, which is the default setting of the human mind and requires active countermeasures.

The countermeasure is procedural. When you finish gathering, spend a fixed block of time — twenty minutes is enough — searching specifically for the strongest evidence against your position. Not the weakest opposing argument you can find and knock over, which produces the straw man Chapter 7 treats. The strongest. Then decide, in the open, what to do with it.

Counterevidence that you raise yourself does three things at once. It shows you have read widely enough to know it exists. It lets you frame it rather than letting a hostile reader spring it. And it usually improves the claim, because engaging the best objection almost always forces a qualification that makes the thesis more defensible. The paper that acknowledges “two large studies found no effect, both in populations under forty, which is where my claim is weakest” is more trustworthy than the paper that quietly leaves them out and hopes.

Section 5The Two Hardest Kinds: Numbers and Experience

Read the number before you use it

Students hand numbers a level of authority they never extend to sentences. A quotation gets scrutinized; a percentage gets copied. Reverse that instinct. Before a figure enters your draft, it has to survive four questions: percent of what, measured when, in whom, and compared with what.

Percent of what is the denominator, and it is the most frequently missing piece of information in public argument. “Complaints rose 200 percent” means one thing if complaints went from 40 to 120 and something else entirely if they went from 1 to 3. Measured when matters because a 2011 figure about internet use describes a vanished world. In whom matters because a result in one population may not transfer to another. Compared with what matters because a bare number is not evidence of change; it becomes evidence only against a baseline.

Three distinctions will handle most of the trouble. First, relative versus absolute change. A treatment that lowers risk from two in a thousand to one in a thousand cuts relative risk by 50 percent and absolute risk by one tenth of one percentage point. Both are true; the first sells and the second informs. Second, percent versus percentage point. A rate moving from 4 percent to 6 percent has risen two percentage points and 50 percent, and writers who use these interchangeably will eventually publish something false. Third, mean versus median. In any skewed distribution — income, hospital length of stay, wait times — a handful of extreme values drag the mean away from anything typical, which is why the median is usually the honest summary of what an ordinary case looks like.

Correlation, cause, and the third thing

Two things moving together is evidence that they move together. That is all it is on its own. Three other explanations are always live: the causal arrow runs the other way, both are driven by a confounding variable, or the pattern is coincidence in a small sample. Ice cream sales and drowning deaths rise together, and the confound is summer.

This does not mean correlation is worthless — it is usually where causal knowledge begins. It means a correlation supports a causal claim only when you add other things: a plausible mechanism, the right time order, a comparison group, and some effort to rule out the obvious confounds. When you have those, say so; that is what makes the evidence load-bearing. When you do not, write the claim you can support. “Units with the new protocol report fewer falls” is defensible. “The protocol reduced falls” requires the extra work, especially since units that adopt new protocols early are often units already invested in safety.

One further caution about your own drafts. The moment a number appears, prose tends to go passive and vague around it — “it has been shown that rates were impacted.” Name who measured what, in whom, and when. Specificity is not decoration on quantitative evidence; it is the evidence.

Experience as evidence

If you work in a field where practice runs ahead of the published literature — clinical care is the obvious case — you have seen things no article has caught up to. What is that worth in a paper? More than novice writers think and less than they hope, and the honest answer depends on what the experience is asked to do.

Experience is legitimate and often superior evidence for existence claims (this happens), for mechanism made visible (here is how it unfolds), for stakes (here is what it costs a person), and for expertise about practice conditions that outsiders systematically misdescribe. Experience cannot establish frequency, cannot establish cause on its own, and cannot stand in for a population. “In three years on a step-down unit I have never seen the rapid-response criteria applied consistently” is real evidence about implementation variability. It is not evidence about how often criteria are applied nationally, and a reader will draw that line whether or not you do.

The scholarly conversation here is genuinely unsettled, and it is worth knowing that. One tradition in composition studies holds that excluding lived experience from academic argument enforces a narrow and arguably exclusionary standard of proof, one that disadvantages students whose knowledge came from work and community rather than from libraries. Another holds that generalizable claims require generalizable evidence and that the standard is not arbitrary. Both positions have serious defenders. The workable practice, in most college writing, is to use experience where its logic actually fits, mark it clearly as what it is, and pair it with systematic evidence when the claim reaches past the individual case. Written that way, a single well-placed observation can do more for a paragraph than a fourth citation ever would.

Words worth knowing