0

The following experiment was devised: a group (random) was divided randomly into 2 : Half were give coffee A the others coffee B, all rated it on a 1-5 Likert scale. Results where analyzed using unpaired Student T test.

This was actually given in a test' students were asked' which statistic test is to be used and so on. Now something looks odd or flawed to me in this example and i cant really put my finger on it. Is it my presupposition that there is no "objectively good" coffee and so the correct way is to compare the 2 coffee with a paired T-test? also a single answer is surly unreliable or accurate but is there more to it? aure its simplified but is this misleading? how so/

  • 2
    Generally using t-test for 5-point Likert scale is wrong :) However: what is your question? Sorry but it is unclear what you are asking. – Tim Feb 18 '15 at 20:33
  • 2
    Although the question title asks about "methodology," the question itself seems to focus on interpretation. What really is your question? Could you be more specific? (The allusion to a paired t-test is mysterious because the data you describe have no natural pairing.) – whuber Feb 18 '15 at 20:52
  • 1
    @Tim That is not correct. T-tests can provide valid inference in comparing Likert responses. Look here. – AdamO Feb 18 '15 at 20:56
  • Is it possible this test was simply a comprehension exercise, assessing your critical reading ability? There is nothing inherently wrong with the example you provided, although caveat emptor as you have only paraphrased the nature of the question. – AdamO Feb 18 '15 at 20:58
  • "Something looks odd to me" - so you were not explicitly asked to identify what is wrong here? That would make sense... because there doesn't seem to be anything wrong about the analysis. As @AdamO notes, t-tests can be used for Likert scales, and an unpaired test is certainly appropriate. Maybe your intuition (that something is wrong) is a bit over-sensitive. – Stephan Kolassa Feb 19 '15 at 07:20
  • 1
    To expand: a paired test would be appropriate if each subject had tasted both types of coffee (you could then randomize who gets which type first). If there's only one coffee per tester, it is an unpaired situation, regardless of whether in the end the number of coffee A tests equaled the number of coffee B tests. – cbeleites unhappy with SX Feb 19 '15 at 10:02

0 Answers0