Thus far, we compared 3 pairs of exams using 3 t-tests. d = 0.57 (pair 3) - slightly over a medium effect.d = 0.56 (pair 2) - slightly over a medium effect.d = -0.23 (pair 1) - roughly a small effect.However, it's easily computed in Excel as shown below. Sadly, SPSS 27 is the only version that includes it. One way to answer this is computing an effect size measure. Our t-tests show that exam 3 has a lower mean score than the other 2 exams. The same goes for the final test between exams 2 and 3. In a similar vein, the second test (not shown) indicates that the means for exams 1 and 3 do differ statistically significantly, t(18) = 2.46, p = 0.025. The 95% confidence interval includes zero: a zero mean difference is well within the range of likely population outcomes. The mean difference between exams 1 and 2 is not statistically significant at α = 0.05. It should be close to zero if the populations means are equal. The mean is the difference between the sample means. SPSS reports the mean and standard deviation of the difference scores for each pair of variables. The last one - Paired Samples Test- shows the actual test results. SPSS creates 3 output tables when running the test. T-TEST PAIRS=ex1 to ex3 /CRITERIA=CI(.9500) /MISSING=ANALYSIS. *Shorter version below results in exact same output. T-TEST PAIRS=ex1 ex1 ex2 WITH ex2 ex3 ex3 (PAIRED) /CRITERIA=CI(.9500) /MISSING=ANALYSIS. *Syntax pasted from analyze - compare means - paired-samples t-test. We added a shorter alternative to the pasted syntax for which you can bypass the entire dialog. For 3 pairs of variables, you need to do this 3 times.Ĭlicking Paste creates the syntax below. In the dialog below, select each pair of variables and move it to “Paired Variables”. SPSS Paired Samples T-Test DialogsĪnalyze Compare Means Paired Samples T Test If all is good, proceed with the actual tests as shown below. If necessary, set and count missing values for each variable as well. At the very least, run some histograms over the outcome variables and see if these look plausible. We'll do so later on.Īt this point, you should carefully inspect your data. The only way to look into this is actually computing the difference scores between each pair of examns as new variables in our data. Since we've only N = 19 students, we do require the normality assumption. Our exam data probably hold independent observations: each case holds a separate student who didn't interact with the other students while completing the exams. Normality is only needed for small sample sizes, say N < 25 or so. normality: the difference scores must be normally distributed in the population.It therefore requires the same 2 assumptions. Technically, a paired samples t-test is equivalent to a one sample t-test on difference scores. However, this test requires some assumptions so let's look into those first. We'll answer just that by running a paired samples t-test on each pair of exams. SoĪre the sample means different enough to draw this conclusion? However, very different sample means are unlikely and thus suggest that the population means aren't equal after all. So even if the population means are really equal, our sample means may differ a bit. We only have a sample of N = 19 students and sample outcomes tend to differ from population outcomes. Now, we don't have data on the entire student population. Generally, the null hypothesis for a paired samples t-test is that They hold the number of correct answers for each student on all 3 exams. Their data -partly shown below- are in compare-exams.sav. He needs to know if they're equally difficult so he asks his students to complete all 3 exams in random order. SPSS Paired Samples T-Test Tutorial report this ad By Ruben Geert van den Berg under Statistics A-Z & T-TestsĪ paired samples t-test examines if 2 variablesĪre likely to have equal population means.Ī teacher developed 3 exams for the same course.