How to interpret a t-test for your action research results
Action research turns a classroom problem into a focused inquiry: a new reading routine, a feedback strategy, a seating arrangement or a behaviour support plan. A t-test can summarise whether an observed mean differs from another mean, but the result is meaningful only when the comparison, measurement and study design support it. For teacher-researchers, the aim is not simply to achieve statistical significance; it is to judge whether the evidence justifies changing practice.
Sound interpretation begins with a precise question and ends with an honest decision about what to do next. Whether your project involves one Brisbane classroom, a small Melbourne school group or students in a Sydney learning support programme, you need to distinguish a change in numbers from evidence that a teaching strategy works. The calculations matter, but so do the assumptions, the size of the effect and the limits of your sample.
Choose the t-test that matches your comparison
A t-test compares sample means to assess a null hypothesis, usually that two population means are equal. The t statistic expresses the observed difference in standard-error units. An independent-samples test compares different groups, such as pupils taught by two methods. A paired test compares matched observations, such as the same pupils before and after an intervention. A one-sample test compares one mean with a fixed reference value.
For a question about whether a teaching strategy changed learning, use a paired test when you measured the same participants twice. A pre-post comparison describes change, but maturation, familiarity with the test or a parallel initiative may also explain it. Without random allocation, a difference between classrooms does not by itself prove that the teaching method caused the difference.
Match the test to the unit of analysis: a pupil should not count as several observations merely because scores were collected on several days. Set whether the hypothesis is two-sided or directional before analysis. Changing the test after seeing results increases false-positive risk. Check admission requirements and your institution’s research guidance for project expectations.
Check assumptions before trusting the output
T-tests assume continuous outcomes and independent observations within each group. For paired data, inspect participants’ change scores rather than separate pre-test and post-test distributions; paired degrees of freedom equal the number of complete pairs minus one, while Welch’s independent test uses an approximation.
With small samples, check for data-entry errors, extreme outliers and severe skew; normality matters most here. An independent-groups test with unequal variances is better handled by Welch’s version than a pooled-variance test. If scores are ordinal or heavily skewed, consider a suitable nonparametric alternative rather than forcing a t-test.
A Brisbane regional placement may yield few complete pairs or many missing sessions. Report the number analysed and avoid conclusions that depend on missing data. If a colleague will gather observations, clear roles in co-teacher collaboration can protect the consistency of the measures.
Read the p-value without misinterpreting it
The p-value is the probability, under the null hypothesis and model assumptions, of a result at least as extreme as the observed one. It is not the probability that the null is true or that your strategy worked. Set an alpha level such as .05 beforehand and report the exact p-value.
A small p-value indicates incompatibility with the null; a large one does not prove there is no effect. A Type I error rejects a true null hypothesis, whereas a Type II error misses a real effect. Small samples and inconsistent measurement increase these risks, so statistical power deserves attention.
In Sydney, a weekly reading probe may fit an action research cycle better than assessment during a NAPLAN window, when schedules and participation can change. Use a one-tailed test only if a directional hypothesis was justified before collection, not because the result later looked convenient.
Use confidence intervals and effect sizes
A 95% confidence interval estimates a range for the mean difference; it does not mean there is a 95% probability that the true value lies in this particular interval. If it includes zero, the result is not significant at the two-sided .05 level, although the estimate may still be informative. An effect size, such as mean change divided by the standard deviation of change scores, shows magnitude relative to variability.
Suppose 18 hypothetical Brisbane pupils improve mean oral reading fluency from 62 to 69 words correct per minute. If the mean paired change is 7 and its standard deviation is 5, then t = 7 ÷ (5 ÷ √18) = 5.94, with 17 degrees of freedom and p < .001. The paired standardised effect is d_z = 1.40, and the 95% confidence interval for the mean change is about 4.52 to 9.48 words.
Keep the calculation visible and interpret the effect in relation to your outcome. Standardised probes depend on consistent materials and a quiet testing space; reviewing campus facilities is one practical way to consider how the learning environment supports reliable teaching and assessment.
Connect statistical significance to classroom change
Statistical significance addresses whether the evidence is compatible with no difference. Practical significance asks whether the change is worthwhile for learners and feasible for the school. In the reading example, seven additional correct words per minute may be educationally meaningful, but the number alone does not establish durability, enjoyment or transfer to unfamiliar texts.
Consider the intervention’s cost, workload and sustainability within Australian school staffing realities. A modest gain that pupils maintain may be more useful than a larger gain that disappears after the novelty ends. Combine test results with observation, pupil feedback and teacher reflection, and record any disagreements between these sources.
In many Australian public schools, achievement reporting uses A–E scales, so an improvement in a reading probe should not be described as an automatic jump in a pupil’s report grade. Explain the measure you used and the context instead. In Melbourne classrooms, including bilingual or mixed-ability groups, compare learner-group patterns where sample size permits before generalising them.
Report your results for busy readers
Provide enough information for another educator to reproduce your analysis. Include each group’s sample size, mean and standard deviation, the mean difference, confidence interval, test type, t statistic, degrees of freedom, exact p-value and effect size. State whether the test was paired or independent and whether you used Welch’s method.
A plain sentence might read: “The matched scores showed an average increase of 7 words correct per minute, 95% CI 4.52 to 9.48, t(17) = 5.94, p < .001, dz = 1.40.” This is clearer than attaching a significance label without a magnitude. Use the notation required by your course or journal rather than copying a template blindly.
Ethical reporting matters as much as numerical accuracy. Follow your university’s ethics process, obtain school approval and parent or guardian consent where required, and explain data storage and withdrawal arrangements. Where the Privacy Act 1988 applies, handle personal information accordingly; research involving Aboriginal and Torres Strait Islander students or communities should follow locally agreed data-sovereignty practices rather than treating cultural knowledge as an extractable data field.
Treat the t-test as one part of action research
A t-test compares means; it does not explain why a result occurred. Other influences may include baseline attainment, teacher experience, attendance, assessment conditions or concurrent support. Include comparison data where feasible and check whether groups were comparable before the intervention, but recognise that baseline similarity does not turn a non-randomised design into a causal experiment.
Triangulation strengthens action research. Combine the test with classroom artefacts, interviews, observation and process notes that show how the change was implemented. If the test and qualitative evidence conflict, investigate measurement and context rather than suppressing the inconvenient source. Negative or mixed findings can guide a more precise next cycle.
Placing your small inquiry within the wider movement in education can help you consider how practitioner-led change contributes to broader educational improvement. Then revise, repeat or scale the strategy according to the evidence and the needs identified in your own setting.
Take your completed analysis and write one plain-English sentence: what changed, by how much and how certain the estimate is. Then add one sentence explaining the educational relevance and one acknowledging the main limitation. Keep the t-test output, the underlying table and the consent or approval details together so the result can be checked, shared and reproduced.
Use that record to decide whether to continue, refine or extend your classroom strategy, and discuss the evidence with colleagues, school leaders and participants. Apply the result now: identify the comparison, verify the assumptions, calculate the effect size and commit to the next evidence-based teaching cycle.