2
$\begingroup$

I am trying to analyze the following case study using SPSS. What I have done so far as part of my research is to conduct a case study to measure the affect the introduction of different technology within a case study. To do this I ran a survey pre case study (before change) and another survey post case study (after change). I am now trying to analyse the datasets.

I have imported the datasets into one dataset, with a variable for system to represent which system the results were collected for (0 for system A, 1 for system B). From what I have learnt in stats so far, I need to run a t-test. The case study ran for 7 months, and since the group is more or less the same group (that used both systems and now I want to evaluate), I should run a dependent sample t-test. Am I correct? Both surveys were run anonymously.

It seems to me that dependent sample t-test follows a group of people, and then collects the results of the same people, knowing that person A scored 1 and then scored 2. Since my data is anonymous, I cannot really track this.

$\endgroup$
2
  • $\begingroup$ It's an interesting question, though, what happens to the dependence of a paired sample if you lose data on the pairing... presumably if you are measuring the same people there is some kind of dependence structure in there that does not exist with truly independent samples. $\endgroup$ – Alexis Mar 16 '15 at 16:07
  • $\begingroup$ it's possible to run anonymous studies that retain pairing; they might be worth looking at for future studies $\endgroup$ – Glen_b Jul 25 '15 at 2:33
-2
$\begingroup$

After further research I just learnt that "This situation means that neither a paired t-test nor an independent samples t-test is appropriate. You can't do a paired t-test because this requires actually matching each individuals' pre and post surveys, and you can't do an independent samples t-test because the observations aren't independent.”

$\endgroup$
2
  • 2
    $\begingroup$ You certainly can do an independent samples t-test. An independent samples t-test is always valid for dependent data; you will just have lower power (higher p-values and wider confidence levels) than you would have had if you could do a paired test. $\endgroup$ – Karl Ove Hufthammer Mar 16 '15 at 18:18
  • 1
    $\begingroup$ One further comment: A pre/post study has a very weak design. You might be able to conclude that there was a change from pre to post, but you can in general not conclude that it was because of the intervention. For that, you would need a separate control group, which was not given the intervention. $\endgroup$ – Karl Ove Hufthammer Mar 16 '15 at 18:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.