PS. Given what I have said above, the issue is not 'higher-tier' and 'lower tier' journals. If the use of a t-test is appropriate, then it should be allowed in any paper; if not, not.
j.arimathea - 11 years ago
Stephan, RR and GuidoB. I agree with you all. The issue is whether the t-test is appropriate for the study involved. In my field, biology [nearly 40 years post-PhD] my feeling is that these days issues worth publishing are likely too complex for a t-test to be useful. But the t-test does have its uses, and the appropriateness must be judged by the context.
j.arimathea - 11 years ago
Stephan, RR and GuidoB. I agree with you all. The issue is whether the t-test is appropriate for the study involved. In my field, biology [nearly 40 years post-PhD] my feeling is that these days issues worth publishing are likely too complex for a t-test to be useful. But the t-test does have its uses, and the appropriateness must be judged by the context.
Stefan - 11 years ago
Of course the t-test is a valid test when all the underlying assumptions are met. However, in real studies these requirements (only 1 comparison, equal variances, normal distribution of the errors) are rarely met. Therefore, the t-test is not appropriate in most studies.
RR - 11 years ago
The appropriateness of various statistical tests all depends on the sample set. Whether a t test can be published in a low-tier or high-tier journal is similarly subject to the relevance of the analysis.
Guido B - 11 years ago
I agree that this is a nonsensical remark. As with any statistical method, its appropriateness depends on your data and research design. If you have conducted a two-group experiment in which you have control over all relevant external factors, and use straightforward measures, a two-sample t-test is fine. If you don't have control over other factors and/or use more complex measures, you would need more sophisticated analytical methods. At least in my field (marketing/management), I think that editors and reviewers at top journals are usually aware of this, and don't demand 'sophisticated' methods irrespective of the data.
CR - 11 years ago
This was an awkward remark indeed. Was there any statistical justification for this remark? I have heard of tests being perfected or proved lowly robust, but t test was always presented as a good option in many cases.
PS. Given what I have said above, the issue is not 'higher-tier' and 'lower tier' journals. If the use of a t-test is appropriate, then it should be allowed in any paper; if not, not.
Stephan, RR and GuidoB. I agree with you all. The issue is whether the t-test is appropriate for the study involved. In my field, biology [nearly 40 years post-PhD] my feeling is that these days issues worth publishing are likely too complex for a t-test to be useful. But the t-test does have its uses, and the appropriateness must be judged by the context.
Stephan, RR and GuidoB. I agree with you all. The issue is whether the t-test is appropriate for the study involved. In my field, biology [nearly 40 years post-PhD] my feeling is that these days issues worth publishing are likely too complex for a t-test to be useful. But the t-test does have its uses, and the appropriateness must be judged by the context.
Of course the t-test is a valid test when all the underlying assumptions are met. However, in real studies these requirements (only 1 comparison, equal variances, normal distribution of the errors) are rarely met. Therefore, the t-test is not appropriate in most studies.
The appropriateness of various statistical tests all depends on the sample set. Whether a t test can be published in a low-tier or high-tier journal is similarly subject to the relevance of the analysis.
I agree that this is a nonsensical remark. As with any statistical method, its appropriateness depends on your data and research design. If you have conducted a two-group experiment in which you have control over all relevant external factors, and use straightforward measures, a two-sample t-test is fine. If you don't have control over other factors and/or use more complex measures, you would need more sophisticated analytical methods. At least in my field (marketing/management), I think that editors and reviewers at top journals are usually aware of this, and don't demand 'sophisticated' methods irrespective of the data.
This was an awkward remark indeed. Was there any statistical justification for this remark? I have heard of tests being perfected or proved lowly robust, but t test was always presented as a good option in many cases.