I have a table that uses the Fisher LSD and has significance "in each column". I'm having trouble interpreting it. Is the significance just for that individual column or can they compare to other columns?
I know the superscripts are difficult to read. Here are the superscript (going across here in rows, but going down in the table):
Column 1: a b b
Column 2: c d d
Column 3: e f f
Column 4: g g g
Column 5: h i i
The simple answer is that comparisons as shown in this table are done within each column. For example, there would be no reason to compare mean values of hepatic folate (2nd column) against mean values of hepatic S-adenosylmethione (3rd column) within each treatment group, which is what a cross-column comparison would attempt.
The table does raise a few issues of which you should be aware and that might help you interpret other tables like this. First, the particular test reported, "Fisher's least-significant-difference multiple-comparison test," doesn't do a very good job of dealing with multiple-comparisons, except in the particular case of 3 groups, as happens to be the case in this table.
Second, the analysis of variance underlying this test has an assumption that within-group errors are independent of the magnitudes of the values and of the treatment. That assumption doesn't look so good in this case for some of the columns. Standard errors for control plasma folate are about 10 times those for the folate-deficient groups; standard deviations are better measures of within-group errors, meaning that errors in the 6-member control folate group are about 15 times greater than in the 3-member folate-deficient groups. (And what do you make of the extraordinarily low standard errors in the S-adenosylmethione and S-adenosylhomocysteine values in the 4-week-deficient group?) So the assumptions needed to interpret the significance tests here probably don't hold. In this particular example, the "statistical significance" might have been better evaluated in other ways, but you should be skeptical of almost any statistical test that's based on only 3 individuals in a group.
Third, although presenting data as means $\pm$ SEM is traditional, you should think about how valuable that is for groups with only 3 observations. With so few cases, wouldn't you rather know what the individual observations were? It wouldn't take up much more space, or could be shown graphically.
