Assuming you wish to maintain good control over the family-wise alpha, Fisher's LSD should only be used when you have k=3 groups. Only in that case, does it control the family-wise alpha at the per-contrast alpha. David Howell discussed this in his textbook, Statistical Methods for Psychology (many editions available). I believe Thom Baguley also discusses it in his book, Serious Stats. And the following article is another source, if you need citations to convince a supervisor, co-author or reviewer (etc.). HTH.

Article A note on the power of Fisher's least significant difference procedure

PS- Fisher's LSD also assumes reasonably good homogeneity of variance, IIRC.

Assuming you wish to maintain good control over the family-wise alpha, Fisher's LSD should only be used when you have k=3 groups. Only in that case, does it control the family-wise alpha at the per-contrast alpha. David Howell discussed this in his textbook, Statistical Methods for Psychology (many editions available). I believe Thom Baguley also discusses it in his book, Serious Stats. And the following article is another source, if you need citations to convince a supervisor, co-author or reviewer (etc.). HTH.

Article A note on the power of Fisher's least significant difference procedure

PS- Fisher's LSD also assumes reasonably good homogeneity of variance, IIRC.

The Fishers LSD test is basically a set of individual t tests. It is only used as a followup to ANOVA.

•Unlike the Bonferroni, Tukey, Dunnett and Holm methods, Fisher's LSD does not correct for multiple comparisons.

•If you choose to use the Fisher's LSD test, you'll need to account for multiple comparisons when you interpret the data, since the computations themselves do not correct for multiple comparisons.

•The only difference a set of t tests and the Fisher's LSD test, is that t tests compute the pooled SD from only the two groups being compared, while the Fisher's LSD test computes the pooled SD from all the groups (which gains power).

Akash Roy - it is worth clarifying that the Fisher LSD does protect against Type I error. However, it does so only for the complete null hypothesis (mu1 = mu2 ... = mu_k). This protection comes from the omnibus ANOVA F test (which needs to be significant for the Fisher LSD procedure). However, it doesn't protect against partial null hypotheses. It turns out for k = 3 you can have at most one Type error because of the logical connection between hypotheses.

Thank you Bruce Weaver , Akash Roy, and Thom Baguley . I appreciate your views regarding LSD. It's really helpful.