There is a huge number of tutorials (including videos) online on how  to perform an independent groups t-test with SPSS.  Try a Google search on , for example. 

Because one group is twice as large as the other, homogeneity of variance becomes much more important than it is when the group sizes are equal (or nearly so).  If your variances are heterogeneous, you'll probably want to use the "Equal variances not assumed" test (i.e., the bottom row of the table). 

Strictly speaking you must go back to your experiment because your data are incomplete. Your design is for "paired" samples, and as the word connotes you must have both pre-test and post-test data for each pair. The incompleteness in your data is manifested by SPSS's comment. 

Eddie, how are you getting paired samples from what Calli said?  There is an experimental group and a control group, and one group is twice as large as the other.  That sounds to me like two independent groups.

Bruce, in t-test of paired samples, you will have to make paired observations from  each participant in the exp. group and from each participant in the control group, where you can do a t-test for dependent or paired samples. Then you can compare the difference in the means of the pre-test between the experimental and control groups, and the post-test means between the exp. and control groups using t-test of independent samples. I hope this clarifies. Eddie

Hi

Basically if your data is independent you have to use unpaired t-test for it. Two sampled t-test is also available in MS-Excel you can use it for equal population variances.

I want to thank you for your advice, however from further research i realized that an Independence t-test or a Welch's t-test is appropriate, particularly the Welch's is reported in the independent t-test table as the equal variances not assumed. The Welch's test can also be calculated through the one-way Anova. Here is the reference which supports that Welch's test is robust with both equal variances not assumed and with sample size that differ.

Kohr, R.L & Games, P.A. (1974). Robustness of the analysis of variance, the Welch procedure and a box procedure to heterogeneous variances. Journal of experimental education, 43(1), 61-69.

Eddie, let me restate my question.  I understand the distinction between paired and unpaired (independent groups) t-tests.  I fail to see any evidence in Calli's original question to support the belief that there are paired scores.  The description of the problem is far more consistent (in my opinion) with there being two independent groups.  I was curious about what you see in Calli's description that convinces you there are paired scores.

Yes Bruce. I see your point. It's just that I try to give a "holistic" answer to questions here in RG where many questions are not given in full. But I do agree with your observation. You do have a closer, more consistent perspective of the question than mine. Please accept my upvote to your latest comment.

Calli, you don't need to make your task more complicated than you need to. Welch's test is contained in the SPSS output for an independent samples t-test - the adjusted t-test for unequal variances.

Take Bruce Weaver's sound advise.

And remember that a t-test, or ANOVA, is simply a test of whether the mean of one group is different from the mean of another group. It is not a test of whether all, or even most, of one group score higher than another. And it's certainly not a test of whether any difference is "important". The question, "What is the chance that a randomly-selected member of Group-A scores higher than a randomly-selected member of Group-B?" is completely different issue. For that, you should consider a "Common Language Effect Size" statistic.

For an explanation, see:

McGraw, K. O., and Wong, S. P. (1992). A common language effect size statistic. Psychol. Bull. 111, 361–365. doi: 10.1037/0033-2909.111.2.361

And you can download an excellent spreadsheet to help you make the calculations from:

Lakens, Daniel (2013). Calculating and Reporting Effect Sizes to Facilitate Cumulative Science: A Practical Primer for t-tests and ANOVAs. Frontiers in Psychology 4 DOI=10.3389/fpsyg.2013.00863

http://journal.frontiersin.org/article/10.3389/fpsyg.2013.00863/abstract

http://journal.frontiersin.org/article/10.3389/fpsyg.2013.00863/abstract

I have kind of the same problem.I have conducted a pretest survey with 56 respondents while the post test survey has only 38 respondents.we have a question in both surveys that based on that we can match the pretest and post test participants(a unique code that is used for both pre test and post test).should I match the pre test and post test based on that code then remove the ones who havent answered to the post test?what is the best method for analysis of this pretest post test surveys?what test I can apply for analysis?

What software are you using, Reza Abasi? For a paired t-test, it probably wants a wide file format with two columns (variables) for the pre and post scores. If so, the paired t-test command will automatically omit anyone who has just the pretest score. And the t-test output will (most likely) show the pretest and posttest means for everyone who has both. It may be interesting to also do an unpaired t-test on the pretest scores comparing those who have complete data to those with pretest only. For that one, I would use an "unequal variances" t-test, given the discrepancy in the sample sizes. How to get an unequal variances test depends on what software you are using, but it should be straightforward.

HTH.

Thanks for your reply.

Actually I may use SPSS for analysis,but I haven't used it before.This is part of my master thesis and honestly I haven't used any software for analysis before this.what do you recommend?

And if I understood correctly I don't need to remove the pretest participants data that haven't answered the post test?can you please mention the reference for it,as I need to cite the reference in my thesis?

There are many examples of how to do t-tests in SPSS online. E.g.,

• https://stats.idre.ucla.edu/spss/output/t-test/

For the unpaired t-test I suggested, the only tricky bit might be that you do not currently have a grouping variable to distinguish between those who have complete data and those who are missing the posttest data. You can easily compute a grouping variable, something like the following. I've assumed your posttest score is stored in a variable named posttest. Change that in the code if you have a different variable name.

COMPUTE HasPost = NOT MISSING(posttest).

FORMATS HasPost (F1).

VARIABLE LABELS HasPost "Has posttest data".

VALUE LABELS HasPost 0 "No" 1 "Yes".

Then use HasPost as the grouping variable for an unpaired t-test on the pretest score variable.

HTH.

Thanks for your reply

Here is my question:I have a question in pretest survey and the same question in post test survey for matching each person's pretest and post test survey.As I mentioned earlier now I am going to analyse the data and compare pretest with post test but the problem is that the number of participants to the pretest is not equal to the number of participants for post test survey.So should I remove the ones who have answered the pretest but not the post test survey?I am new to statistics so please mention the reference for it as I need to add it as my reference