Software output open in one window and an empty results section in the other? This guide helps graduate nursing students interpret descriptive statistics, t-tests, chi-square, correlation, regression, confidence intervals and effect sizes, and report them in APA 7 without the common misreadings. It explains concepts and formats, and uses no invented data.
If you have statistical software output open in one window and an empty results section in the other, this guide is for you. It is written for master's and doctoral nursing students who must read quantitative studies critically and report statistics in their own papers. It explains the ideas behind the common methods and how to write them up. It does not use invented datasets, because the aim is to help you interpret your own output correctly.
You do not need to derive formulas. You need to know what each statistic tells you, what assumptions it makes, and how to say so in plain words.
Quick answer. Report what you measured (descriptive statistics), how precisely you measured it (confidence intervals), whether the result is likely to be chance (the test and its p value), and how big the finding is (an effect size). Then say what it means for practice, without claiming more than the design allows.
Every test depends on how a variable is measured. Before choosing anything else, classify each variable.
| Level | What it means | Nursing example | Usual summary |
|---|---|---|---|
| Nominal | Categories with no order | Unit type, diagnosis, yes or no | Counts and percentages |
| Ordinal | Ordered categories, unequal gaps | Pain rating categories, Likert item | Median, counts |
| Interval or ratio | Numeric with equal intervals | Age, blood pressure, scale total score | Mean and SD, or median and IQR |
Likert scales are debated. A single Likert item is ordinal, while a total score summed across many items is often treated as interval. State how you treated it and why.
Descriptive statistics summarize your sample. They come first in any results section and often reveal problems before you run a test.
Look at a histogram before deciding how to summarize. A long right tail, as with length of stay, means the mean will exceed the median and the median is the fairer summary. Tell the reader which you chose and why.
A table of participant characteriztics tells readers who was studied and whether the results might apply to their setting. Include the sample size, key demographics and any groups being compared. See APA tables and figures for layout.
A sample is only part of a population, so estimates vary from sample to sample. Inferential statistics quantify how much you can trust what the sample shows.
A p value is the probability of results at least as extreme as yours, if the null hypothesis were true. A small value means the data are unlikely under the null. The conventional threshold of .05 is a convention, not a law of nature.
A 95% confidence interval is a range of plausible values for the true population quantity. If the whole process were repeated many times, about 95% of intervals built this way would contain the true value. A narrow interval means a precise estimate. A wide one means a lot of uncertainty, often because the sample was small.
Confidence intervals are more useful than p values alone because they show both the size and the precision of an effect.
The right test depends on the outcome, the predictor and the design. The table below gives a starting point, and your methodologist has the final word.
| Question | Variables | Common test | Non-parametric alternative |
|---|---|---|---|
| Do two independent groups differ on a numeric outcome? | Numeric outcome, two groups | Independent-samples t-test | Mann-Whitney U |
| Did the same people change between two times? | Numeric outcome, paired | Paired-samples t-test | Wilcoxon signed-rank |
| Do three or more groups differ? | Numeric outcome, 3 or more groups | One-way ANOVA | Kruskal-Wallis |
| Are two categorical variables related? | Two categorical variables | Chi-square test of independence | Fisher's exact test |
| Are two numeric variables related? | Two numeric variables | Pearson correlation | Spearman correlation |
| What predicts a numeric outcome? | Numeric outcome, several predictors | Multiple linear regression | Robust or transformed methods |
| What predicts a yes or no outcome? | Binary outcome, several predictors | Logistic regression | Exact methods for small samples |
A t-test compares two means. It assumes independent observations and roughly normal data, or a sample large enough that the mean behaves well. If variances differ between groups, use a version that does not assume equal variances. When you compare more than two groups, use ANOVA instead of repeated t-tests, which inflates the chance of a false positive. If ANOVA is significant, use post hoc tests to see which groups differ.
The chi-square test asks whether the distribution of one category depends on another, such as whether falls differ by unit type. It compares observed counts with the counts expected if the variables were unrelated. When expected counts are small, a common rule of thumb is fewer than five in some cells, Fisher's exact test is safer.
The correlation coefficient r ranges from negative one to positive one. The sign gives direction and the size gives strength. Pearson's r suits linear relationships between numeric variables, and Spearman's rho suits ordinal or skewed data. Correlation describes association only. It does not show that one variable causes the other.
Regression estimates how the outcome changes with each predictor while holding the others constant. Report the coefficients with confidence intervals, and R squared as the share of variance explained. Check the assumptions: a roughly linear relationship, independent errors, similar spread across the range and approximately normal residuals. Also look for predictors that are highly correlated with each other, which makes coefficients unstable.
When the outcome is binary, logistic regression gives odds ratios. An odds ratio of one means no association. Above one means higher odds of the outcome, and below one means lower odds. If the confidence interval for an odds ratio includes one, the result is not statistically significant at the matching level.
Odds ratios are often misread as risk ratios. When the outcome is common the two can differ a lot, so describe an odds ratio as odds, not as a percentage change in risk.
Statistical significance says an effect is unlikely to be zero. It says nothing about how large or important the effect is. Report an effect size with every main result.
| Measure | Used with | What it tells you | Rough conventions |
|---|---|---|---|
| Cohen's d | Differences between means | Difference in SD units | 0.2 small, 0.5 medium, 0.8 large |
| Pearson's r | Correlations | Strength of association | .1 small, .3 medium, .5 large |
| R squared | Regression | Share of variance explained | Judge by the field |
| Odds ratio | Binary outcomes | Change in odds per unit or group | Interpret with its confidence interval |
| Cramer's V or phi | Chi-square | Strength of association between categories | Judge by the field |
The conventions above are rough guides originally offered for general use. Context matters more. A small effect on a serious outcome in a large population can be important, and a large effect on a trivial measure may not be.
Ask whether the difference is big enough to matter for patients, staff or systems. Where an instrument has a minimal clinically important difference, compare your result with it. A statistically significant change smaller than that threshold may not justify changing practice.
Illustrative example, not a real client.
Problem. A DNP student finished analyzing a pre-post project and wrote in her results section: "The difference was significant, so the program worked."
Tension. Her committee member circled the sentence and asked three questions: how big was the change, how precise was the estimate, and could anything else explain it? She had reported only a p value.
Turn. She rebuilt the paragraph around the four-part sentence in this guide: what was compared, the test with exact p, the effect size with a confidence interval, and a plain-language interpretation. She also added a sentence noting that the design had no comparison group.
Proof. The committee member's next comment was about how the outcome scale related to a clinically meaningful change, which is a discussion-level question rather than a correction of a misreading.
Payoff. She used the same four-part pattern for every result in the chapter, so the section read consistently and took less time to revise.
Send your output, question and rubric, and get a quote for a results write-up or an interpretation check. The price is shown before you pay, and revisions are free for 14 days.
| Misreading | Why it is wrong | Say instead |
|---|---|---|
| "p = .03 means there is a 3% chance the hypothesis is false." | A p value is not the probability that a hypothesis is true. | "If there were no effect, results this extreme would be uncommon." |
| "p = .20 proves there is no difference." | Not significant is not the same as no effect. The study may lack power. | "The data did not provide sufficient evidence of a difference." |
| "The groups are correlated, so A causes B." | Association is not causation, and confounders may explain it. | "A was associated with B." |
| "The result is significant, so it is important." | Large samples make tiny differences significant. | Report effect size and clinical meaning. |
| "The 95% CI means there is a 95% chance the true value is inside." | The interval either contains the value or does not. The 95% refers to the method. | "We are 95% confident, in the sense that this method captures the true value 95% of the time." |
| "I ran twenty tests and one was significant." | Multiple testing raises the chance of false positives. | Pre-specify tests, or adjust for multiple comparisons. |
APA 7 gives clear conventions for statistics, which make results easy to scan. The examples below use placeholders in square brackets so that no data is implied.
| Test | Format template |
|---|---|
| Independent t-test | t([df]) = [value], p = [value], d = [value], 95% CI [lower, upper] |
| Chi-square | χ²([df], N = [n]) = [value], p = [value], Cramer's V = [value] |
| Correlation | r([df]) = [value], p = [value], 95% CI [lower, upper] |
| Regression | b = [value], 95% CI [lower, upper], p = [value]; R² = [value] |
Fill each bracket from your own output, and never round in a way that hides a meaningful difference. For general APA guidance see APA Style. To place these results within the rest of the chapter, use the results and discussion guide.
Suppose you compare two groups and your output gives a p value below .05, a confidence interval for the difference that excludes zero, and a medium effect size. A strong write-up states the finding in words first ("Group A scored higher than Group B"), then gives the numbers, then explains the size in practical terms, and finally notes limits such as a non-random sample. It does not say the intervention "works" unless the design supports that claim.
Every test rests on assumptions, and violating them can make a p value or interval misleading. State in your methods that you checked them, and report what you did if they were not met.
| Test | Key assumptions | How to check | If violated |
|---|---|---|---|
| t-test | Independent observations, roughly normal outcome, similar variances | Histogram or Q-Q plot, Levene's test | Use a version for unequal variances, or Mann-Whitney U |
| ANOVA | Independence, roughly normal residuals, similar variances | Residual plots, Levene's test | Robust ANOVA or Kruskal-Wallis |
| Chi-square | Independent observations, adequate expected counts | Calculate expected counts per cell | Fisher's exact test, or combine categories sensibly |
| Pearson correlation | Linear relationship, no extreme outliers | Scatterplot | Spearman correlation |
| Linear regression | Linearity, independent errors, constant spread, normal residuals, no severe overlap among predictors | Residual plots, variance inflation factors | Transform, use robust methods or simplify the model |
| Logistic regression | Binary outcome, independent observations, enough events for the number of predictors | Count events per predictor | Reduce predictors or use exact methods |
Independence is the assumption students forget most. Repeated measures on the same person, or patients clustered on the same unit, are not independent, and they need methods designed for that structure. Ask your methodologist before you run the standard test.
Two topics regularly appear in committee comments, and both are easier to handle if you plan them early.
Power depends on the effect size you hope to detect, the alpha level, the variability of the outcome and the sample size. Plan the sample size before data collection, and say where your assumed effect size came from, such as prior studies or the smallest difference that would matter clinically.
Report how much data were missing for each key variable, and whether the pattern looked random. Deleting every incomplete case is simple, but it can bias results and shrink the sample. Other approaches, such as multiple imputation, make their own assumptions. Whatever you choose, describe it in the methods and say what it might mean for your findings. Never remove data silently.
For a starting point on planning, the research proposal guide explains where the analysis plan and sample size justification belong.
Reporting both is good practice, and many journals and programs expect it. The interval shows size and precision, and the p value links to the test.
Consider a non-parametric test, a transformation or a method that does not require normality, and consult your methodologist. For large samples, many tests are robust to moderate departures.
Use the mean and SD for roughly symmetrical numeric data. Use the median and IQR for skewed data or data with outliers. Show a histogram if you are unsure.
No. A well-designed study with a non-significant result can still be valuable if you report effect sizes, precision and limitations honestly. Reporting only significant results distorts the literature.
The EQUATOR Network lists reporting guidelines such as CONSORT and STROBE, which describe what to include for each study design.
Good statistical writing is not about impressive numbers. It is about giving the reader enough to judge the finding themselves: who was studied, what was measured, how big the effect was, how sure you are and what you cannot conclude. For choosing the design behind those numbers, read the research methods guide, and for planning the analysis before data collection, the research proposal guide.
If you would like help writing up or interpreting results, get your instant quote. The price is shown before you pay. Delivered work comes with 14 days of free revisions, and refund terms are on the money-back guarantee page. Use any support in line with your institution's academic-integrity rules.