What Does Paired Mean? How to Run and Read a Paired Samples t-test in SPSS

What does paired mean

In statistics, paired means that two values are matched because they come from the same subject or from two subjects with a clearly defined match. In research data, the most common case is measuring the same group of people at two time points, such as scores before and after a training program. Each person then has two related observations.

A Paired Samples t-test tests whether the means of two measurements differ significantly. SPSS does this by calculating the difference for each pair:

Dᵢ = Xᵢ,after - Xᵢ,before

The software then tests whether the mean of the differences Dᵢ differs from 0. You are therefore testing change within each pair rather than treating the two data columns as two independent groups. This is the key distinction between paired data and the Independent-Samples t-test.

A valid pair needs three basic data conditions. First, both values must belong to the same unit of analysis. Second, each row must preserve the relationship between the two measurements. Third, the tested variable must be numeric and have an appropriate measurement meaning. If a person answered only at the before measurement and has no after measurement, that case is usually not included in a complete paired calculation.

In SPSS, paired is not the name of a column or a type of scale. It describes the relationship between two observations. You may see names such as Pre_Post, Before_After, T1_T2, or Time 1_Time 2, but SPSS will only interpret them correctly when the two variables are placed in the same pair and in the correct order.

The meaning of paired data in quantitative research

Paired data help you answer a question such as: “After the intervention, did the mean score change compared with the score before the intervention?” The intervention could be a course, a policy, a product use, a service version, or another experimental condition. You still need an identifier showing that the before and after values belong to the same person or unit.

For example, you survey 120 students about their confidence before and after taking a course. The Confidence_before and Confidence_after columns must be on the same row for the same student. If you shuffle the row order in one column, SPSS may still run the test, but the paired result loses its meaning because the pairs have been matched incorrectly.

This test is commonly used when one quantitative variable is measured twice. A composite Likert scale or the mean score of several items is often used in business research. You can also read what a scale is if you are unsure whether to use individual items or a representative score for a construct.

A Paired t-test does not identify the cause of a change by itself. If the after score is higher than the before score and the p-value is small, you have evidence of a mean difference in the sample under the stated design. Saying that “the program definitely caused the change” also depends on the research design, control group, measurement period, and other factors.

You should define the research question and the direction of interpretation in advance. If your hypothesis expects the after score to be higher than the before score, keep the variable order consistent and record it in the methods section. This helps you interpret a negative or positive Mean Difference without assuming that a negative sign is always an unfavorable result.

A Paired t-test is usually appropriate for a before-and-after design using the same sample. If you survey two different groups, the data do not automatically become paired even when the two groups have the same sample size. The statistical question and standard-error formula are different in that case.

You can review related material in the Statistics topic to distinguish a paired t-test from other tests before choosing a menu in SPSS.

What counts as acceptable for a paired test

For a Paired Samples t-test, “acceptable” is not one threshold that applies to every table. At minimum, you need to examine the p-value, mean difference, confidence interval, and distributional condition of the difference variable. In many theses, the significance level is set at 5%, which corresponds to rejecting H0 when the p-value is below 0.05. The 0.05 level is a test-design convention, not a claim that every result close to the cutoff has the same practical importance.

The table is a decision guide, not a reason to modify data until the p-value falls below 0.05. If the p-value is greater than or equal to 0.05, write that there is insufficient statistical evidence to conclude that the two means differ. Do not write that H0 has been proven true, because a test provides evidence against H0 or insufficient evidence against it.

A Paired t-test is appropriate when the difference variable is approximately normally distributed, especially with a small sample. With a larger sample, the effect of skewness may be different, but you still need to inspect plots, outliers, and the data-collection process. See outliers in data if a few unusually large differences are pulling the mean in one direction.

Do not use a threshold table to turn a descriptive result into a causal conclusion. For example, a Mean Difference of 0.20 points may be statistically significant but may not have practical importance for service users. Conversely, a practically important change may not produce a p-value below 0.05 if the sample is small or the data are highly dispersed. Report the mean, standard deviation, difference, confidence interval, and p-value together.

If your supervisor asks, “What paired value is acceptable?”, tie the answer to the test design. Compare Sig. (2-tailed) with the selected significance level, usually 0.05, then examine the direction of the difference and the conditions for applying the test. There is no fixed Mean Difference that works for every research topic.

How to read paired results in the output

After you run the test, SPSS usually displays four tables: Paired Samples Statistics, Paired Samples Correlations, Paired Samples Test, and sometimes an earlier diagnostic table if you requested additional statistics. The Correlations table shows whether the two measurements are correlated. It is not the main result for deciding whether the two means differ.

To run the test, go to Analyze > Compare Means > Paired-Samples T Test. In the dialog box, place the first variable in Variable 1 and the second variable in Variable 2, then click OK. Check each variable label before running the analysis. If you place After in Variable 1 and Before in Variable 2, the sign of Mean Difference will be the reverse of the result obtained by placing Before first and After second.

The following is illustrative output, not the result of an actual study. Suppose 80 people have scores for Service_before and Service_after on the same scale.

The Paired Samples Statistics table shows that the after mean is 0.26 points higher than the before mean. This is only the initial description. You cannot conclude that the difference is statistically significant until you read the test table.

Because SPSS calculates Before - After, Mean Difference is -0.260. The negative sign means that the before score is lower than the after score. The confidence interval from -0.370 to -0.150 does not contain 0, and Sig. (2-tailed) is below 0.05. With this illustrative output, there is a statistically significant difference between the two measurements. When reporting the result, state the subtraction order so readers understand the sign of the difference.

If the Paired Samples Correlations table shows Correlation = 0.62 and Sig. = 0.001, this only indicates that the two measurements are correlated in the sample. A high correlation does not mean that the means are equal, and a low correlation does not automatically prove that the means differ. The conclusion about a mean difference must come from Paired Samples Test.

In the Paired Samples Test table, df is usually the number of valid pairs minus 1. If you initially have 80 people but some are missing one of the two measurements, the actual N used in the test may be lower than 80. Compare N in Paired Samples Statistics with the original number of rows in Data View.

What to do when the paired result is not acceptable

If the result does not meet the criterion you defined in advance, work through the following checks in order.

Check how the pairs were matched

Open Data View and check each person’s ID. The before and after variables must be on the same row and must not be misaligned because one column was sorted separately. Also check missing values, codes such as 99 or 999, and cases that contain only one of the two measurements.

If the file was merged from two sources, check the identifiers before copying data. A common error is sorting the before file by name and the after file by ID, then pasting the two columns next to each other. SPSS cannot know which row belongs to which person unless you preserve the order or match the IDs yourself.

Check the variables entered into the test

You may be using the score of one item at the first measurement and the composite score of an entire scale at the second. Standardize the calculation method, coding direction, and reverse-coded items before creating a mean score. If an item is reverse-coded but has not been recoded, the difference may reflect a processing error rather than a real change.

If the two measurements use different scales, such as 1 to 5 at one time and 1 to 7 at the other, Mean Difference will be difficult to interpret directly. Before running the test, check the questionnaire, value labels, and formula used to create the composite variable. Record every change so you can explain it to your supervisor.

Examine the distribution of the difference variable

Create a difference variable with Transform > Compute Variable, such as after - before, then inspect the histogram, Q-Q plot, and outliers. If the difference variable is highly skewed or contains an unusual value, record the reason before excluding a case. Do not delete a person simply because that row makes the p-value larger.

You can go to Analyze > Descriptive Statistics > Explore, place the difference variable in Dependent List, and then select the plots and distribution checks. If there is an outlier, determine whether it is a data-entry error, an unusual respondent, or a valid case. Every decision to exclude data must be connected to data quality, not to whether the result reaches significance.

Consider an alternative test

If the t-test conditions are not appropriate, the Wilcoxon Signed-Rank Test is a commonly considered option for two related measurements when the difference variable does not meet the necessary assumption. State why you selected the alternative and report the median or another description appropriate to the method. Do not run both tests and choose the table with the more favorable result.

In SPSS, go to Analyze > Nonparametric Tests > Related Samples, select the two related variables, and mark Wilcoxon. The menu may differ slightly by SPSS version, but the principle is the same: select the correct two variables in the same pair. Do not describe Wilcoxon as a guaranteed improvement over the Paired t-test. It is more appropriate when the research question, measurement scale, or data distribution calls for it.

If the result is not significant because the p-value is large, keep the result and interpret it accurately. The mean change may be small, the data may be dispersed, the sample may be insufficient, or the change may differ across individuals. Do not switch tests simply to obtain a significant conclusion.

Paired versus Independent-Samples t-test

Paired data consist of two observations linked within each pair. The Independent-Samples t-test compares two independent groups, such as scores from people who attended a course and scores from people who did not. If one student appears in both groups, the groups are no longer independent in the usual sense.

You also need to distinguish paired data from correlation. Correlation measures how two variables change together, while a paired t-test tests the mean of the pairwise differences. A study may report both, but each result answers a different question.

Paired data also differ from moderator and mediator. Moderator and mediator describe a variable’s role in a relational model, while paired describes the matching structure of two measurements. Do not call a “before” or “after” variable a moderator merely because it appears in the same analysis. The material on moderating is more relevant when you are testing whether a variable changes the strength or direction of a relationship.

If you have three or more time points, running several paired t-tests for each pair increases the number of tests and makes the results harder to present. Consider a method that fits the repeated-measures design and your research objective instead of repeating the same test without adjusting how you interpret the results.

A quick check is to write one sentence describing the data before choosing the test. If the sentence is “the same 80 people were measured twice,” a paired test may be appropriate. If it is “80 people are in Group A and 80 different people are in Group B,” an independent test is usually the direction to consider. If it is “the relationship between satisfaction and purchase intention,” you are asking about correlation or a regression model, not paired data.

Common mistakes

The first mistake is using a Paired Samples t-test for two independent groups. Ask whether every value in column A has a corresponding value in column B from the same person. If the answer is no, a paired test is not appropriate.

The second mistake is looking only at Paired Samples Statistics. That table contains the mean, standard deviation, and sample size. The statistical conclusion is in Paired Samples Test, especially t, df, Sig. (2-tailed), and the confidence interval.

The third mistake is interpreting a negative sign as a bad result. The sign of Mean Difference depends on the variable order in Paired Variables. State whether you calculated before minus after or after minus before. A negative sign may simply mean that the first value is lower than the second.

The fourth mistake is reporting Sig. = 0.000. SPSS rounds the displayed value, so write p < 0.001 rather than claiming that p is exactly 0. Finally, do not overlook missing values, outliers, or cases without both measurements.

Another mistake is running the test repeatedly after changing the variable calculation without saving data versions. Keep the syntax or record the variable names, formulas, number of excluded observations, and result from each run. This shows which table is the final result and prevents you from copying old output into Chapter 4.

How to write this in your thesis

You can write: “The Paired Samples t-test showed that the score for [variable name] at [after] was higher/lower than at [before] (M = [value] versus [value]), with a mean difference of [Mean Difference], t([df]) = [t], p [< or =] [p-value]. Therefore, hypothesis [H...] was [accepted/supported or not supported] according to the stated criterion.” Replace every bracketed section with the actual values in your file.

When writing, avoid the word “effect” if your design has only one group measured before and after, without an appropriate control group or other control. You can write that the after-measurement mean was higher than the before-measurement mean and state clearly that this was an observed difference in the sample. This wording matches the output more closely and is safer when the committee asks about causal inference.

Frequently asked questions

What does paired mean in statistics?

Paired data are observations arranged into related pairs, usually two measurements from the same person. A Paired Samples t-test checks whether the mean difference between the two values in each pair differs from 0.

The word paired describes the relationship between observations, not the name of a variable or scale. Check both the data-collection design and the order of rows in the file.

When should you use a Paired Samples t-test?

Use this test when you have exactly two quantitative measurements linked within each pair, such as before and after the same intervention. The two columns must be matched correctly by ID or by the order of the same units.

If the two groups contain different people, consider an Independent-Samples t-test. If the same people are measured at three or more time points, do not automatically run multiple paired t-tests without considering the number of tests.

What value is acceptable for a Paired Samples t-test?

Usually, if Sig. (2-tailed) < 0.05, you conclude that there is a statistically significant mean difference at the 5% level. You should also report the mean, Mean Difference, confidence interval, and assumption checks rather than copying only the p-value.

If the p-value is greater than or equal to 0.05, write that there is insufficient statistical evidence of a mean difference. Do not write that H0 is definitely true, and do not change the data to push the p-value below the threshold.

What is the difference between a Paired t-test and an Independent t-test?

A paired test is for two related measurements within the same pair. An independent test is for two independent groups. Choosing the wrong test can affect the standard error, degrees of freedom, and conclusion.

Equal sample sizes do not make two groups paired. The deciding issue is the relationship between individual observations, not the number of people in each group.

What should you do when the paired test is not statistically significant?

Check the pairing, missing values, coding, difference variable, and outliers. If the data do not fit the t-test, you may consider the Wilcoxon Signed-Rank Test. If the data are valid and the p-value remains large, report that result and avoid changing the data to force statistical significance.

Also examine the size of the difference and the confidence interval. A non-significant result does not make the data worthless. It means that the current sample does not provide sufficient evidence under the selected significance level.

Open your SPSS file now, check the ID and the order of the two variables, then go to Analyze > Compare Means > Paired-Samples T Test. Compare every number in the Paired Samples Test table with your reporting template. If you need to run the analysis on your own .sav or .csv file, see DoThesis M4 data analysis.