I am struggling with a data analysis. I have made a survey and got the following answers in the first turn, I put them in a frequency table

  • Don't agree at all: 8
  • Don't agree: 6
  • I don't know: 16
  • Agree: 14
  • Totally agree: 6

In the second turn:

  • Don't agree at all: 8
  • Don't agree: 13
  • I don't know: 9
  • Agree: 12
  • Totally agree: 8 I added weights to the answers: 2 for Don't agree at all, 1 for Don't agree, 0 for Don't know, -1 to agree and -2 for totally agree. I need to know whether their opinion is significantly different or not. I have 50 people who wrote these answers. I know firstly I need to know if the data is Normally distributed or not, then I have to test the variance with the two sample F test and only after this I can turn to the T test. My only problem is, that I have to idea how to solve this in Excel (or SPSS), I have only found examples for data in row and not for frequency tables. Could someone please help me?
  • $\begingroup$ Are the people in the first turn the same as the people in the second turn? If so, can you identify, for any given person, their answer in each turn? $\endgroup$ – Sal Mangiafico Apr 14 at 13:15
  • $\begingroup$ Yes, they are the same people. Unfortunately, no, because they filled it out anonymously. How bad is that? $\endgroup$ – TMoci Apr 14 at 16:04
  • $\begingroup$ I edited by answer below. $\endgroup$ – Sal Mangiafico Apr 14 at 18:50

Since the the same people were sampled in the first turn and in the second turn, the correct approach would be use a test that takes into account the paired or repeated measures of the data. However, since the identity of the participants was not recorded in a way that allows the observations to be paired, that isn't an option. The only approach I know of in this case is to treat the samples (the two turns) as independent, and make note of this in your write-up.

Treating the data is if your observations are independent: For this kind of data, I wouldn't assume or assess if the data are normally distributed. Nor would I use a t-test. A common test for this situation is the Cochran-Armitage test. This isn't available in SPSS, but because you have only two levels of Turn, you can use the linear-by-linear test of trend in SPSS as described in the answer by Hong-Qiu Gu here. Other alternatives: An ideal approach is ordinal regression. Also, if you have an implementation that handles ties, a Mann-Whitney test will give satisfactory results.


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.