Read every block in the same order
Whatever the analysis, four questions arrive in a fixed sequence. Answer them in that order and you will never report a result the data cannot carry.
- How many cases went in. Check the sample size first, against the number you expected. Missing cases are the most common silent problem in a results chapter.
- Were the conditions met. Assumption tables print alongside the test and they are read before it, not after.
- What did the test find. One row, one comparison, one value to report.
- How large is it. Significance says something is there. An effect size says whether it matters.
The t test block
An independent-samples t test prints two tables. The first gives group means and standard deviations, and it is worth reading on its own, because it tells you the direction of the difference before any test does. A significant result in the opposite direction to your hypothesis is caught here or not at all.
The second table shows two rows for the same comparison. Which row you read depends on the equality-of-variances test printed beside them. Read that first, take the row it points to, and report the value, the degrees of freedom and the significance from that row only. Reporting from both rows is the clearest sign that output was copied rather than read.
The ANOVA block
A one-way ANOVA answers one question: is there a difference somewhere among the groups. The main table gives an F value, two degrees of freedom and a significance value. That is the entire finding, and it is deliberately unspecific.
Which groups differ comes from the post hoc table underneath, where the pairs are compared against each other. Write from those comparisons rather than from the means, because a difference that looks obvious in a bar chart may not survive the pairwise correction. Report the overall test, then the specific pairs, then how large the differences were.
The regression block
Regression prints three tables answering three different questions, and confusing them is the most common error in a quantitative chapter.
- The model summary. R squared says how much of the variation in the outcome the model accounts for. It is not a significance value and it is not the effect of any single predictor.
- The ANOVA table. This tests the model as a whole. A model can be significant here while individual predictors are not, which is a real and reportable result.
- The coefficients table. Each predictor is tested here. The unstandardized coefficient gives the size of the change in the outcome's own units, and the standardized one lets you compare predictors with each other.
- The diagnostics. Printed last, read early. Residual plots and collinearity statistics are what a committee asks about when the rest looks tidy.
Writing the sentence underneath
Every table needs one sentence in words: what was compared, what was found, how large it was, and what that means in the units of the study. Write it immediately, while the output is still on screen. At the end of an analysis a hundred tables look alike, and reconstructing which one produced which claim is slower than the analysis was.
Keep the output itself, alongside the dataset version that produced it. A results chapter gets checked against its tables, and an analysis rerun months later with a slightly different filter will not reproduce the numbers you wrote. Saving the file costs nothing and settles that question permanently.
Two habits keep a chapter defensible. Report exact values rather than rounding toward a threshold, and never let a sentence claim more than the design supports. Where a chapter has stalled in the gap between the output and the words, that gap is exactly what a statistics specialist is for.
Questions
Why does my sample size change between tables?
Because SPSS excludes cases with missing values on the variables used in that particular analysis. A correlation of two variables uses different cases from a regression with five. Read the note under each table and report the number that actually entered each analysis. A chapter quoting one sample size everywhere while its tables disagree gets returned.
Is R squared an effect size?
It is a model-level one. R squared reports how much variation in the outcome the whole model accounts for, which is useful and is not the same as the strength of any single relationship. For that, read the coefficients table. Presenting R squared as if it described one predictor is a common error and an easily caught one.
The overall test is significant but no pair looks different. Now what?
Read the post hoc table rather than the group means. An overall test can detect a spread across groups that no single pairwise comparison survives once it has been corrected. Report that honestly: the analysis found an overall difference and the pairwise comparisons did not localize it. That is a genuine result, and it reads better plainly than forced.
Should I round significance values to the usual thresholds?
Report the exact value the output gives, to three decimal places, and use the very small notation only where the output does. Rounding toward a threshold looks like a decision made after seeing the number. Keep exact values in the tables and reserve interpretation for the sentence underneath, which is where judgment belongs.
Do I paste the software tables into my document?
Follow the style your program requires, which almost always means rebuilt tables rather than pasted images. Output carries variable labels and decimal places that no style guide accepts, so the numbers get retyped into the required format. Keep the original output as a file or an appendix, because a reviewer may ask to see what the software actually produced.
