3

Suppose I have a dataset with 1,000 people and they rating 5 shampoos across a number of attributes (cost, smell, texture, how well it cleans).

If I wanted to compare the average of an attribute across brands, would I use an independent t-test? I know that multiple comparisons invite things like a Bonferroni correction, but is it appropriate to use an independent t-test here? I know I should probably do an ANOVA, but really basic, if I wanted to see if the mean for shampoo A on clean was bigger than the mean for shampoo B, would I see an independent or dependent test? And why?

You can think of the dataset as long instead of very wide, with a set of rows for each respondentXbrand combination. Then can I think of the data as multiple independent samples?

Michael Lew
  • 10,995
  • 2
  • 29
  • 47
vashts85
  • 151
  • 1
  • 1
    I'd do a paired t test here. – gammer Feb 01 '17 at 03:33
  • 1
    @gammer you really should explain the rationale for your use of a paired t-test (dependent sample t-test). – Michael Lew Feb 01 '17 at 04:53
  • 1
    @MichaelLew, assuming we're talking about the same samples, the point estimate of the difference would be the same no matter which test you did, and the paired t-test almost always more powerful than the 2-sample t-test – gammer Feb 01 '17 at 05:07
  • 1
    @gammer I agree, and besides being more powerful, the df for paired t test would be correct. – David Lane Feb 01 '17 at 05:26
  • 1
    OP: this thread http://stats.stackexchange.com/questions/38102/paired-versus-unpaired-t-test is about whether to use a paired or unpaired test. The question is good and some (but not all) of the answers there are good. – gammer Feb 02 '17 at 02:43

0 Answers0