On Mon, 01 Sep 2003 03:30:41 GMT, "Gordon Couger"
wrote:

I have no idea what the effect of the discarded data did to the study did.
That's the point.

Well, ask yourself what happens to a data set if you add an outlier
to it: The overall variance in data increases, and the ANOVA may
come out with either too low F-values to allow you to calculate a
figure for least significant difference (LSD) or you may end up with
an inflated figure for LSD. The effect of this may be that you will
reject otherwise significant differences.

Now turn that around: Without the outlier, finding significant
differences will be more likely. In other words, you can reasonably
suspect that those significant difference that are found without the
outlier, might turn into insignificant differences with the outlier
added back in.

However, this study actually found very few significant differences,
although some of those that were found were striking.

The main findings, in short, was that plants in the GE bacteria setup
turned chlorotic and wilted, apparently concomitant with a flush
period of nematode growth during which nematodes reached higher
numbers, and, with fungal feeding nematodes dominating.
Compared to this plants in the non-GE bacteria setups,plants grew
well, nematode numbers did not flush just as much, and bacterial
feeding nematodes remained dominating.

And here's the crunch, there is no way any discarding of outliers can
have 'produced' the finding that plants in the GE bacteria setup
wilted and died, while the plants in the nonGE bacteria setup grew.

So, from a cool minded perspective, you may well have concerns
as a matter of principle as regards handling of outliers,
but you cannot have concerns that this handling has affected
the main findings of the study.

Quite generally, a criticism of a study, on counts that do not
affect its main findings, is insubstantial. Perhaps some would
call such criticism mere nitpicking, I would go that far, because
it may well be educating. However, the point is, you can't 'kill'
a study by flawing it of something that does not change its
conclusions.