Significance without power?

[Bradford Apple]

I am trying to use data from a preliminary experiment to perform power analysis, to be used in planning subsequent experiments.
The xp was analyzed with 1-way ANOVA. Overall significance and a group difference were indicated. However, putting the data into a power analysis indicated that I need a much, much, much larger N to achieve acceptable power, based on the SD's from that xp. So, how can I have statistical significance with extremely low power?

Data:
Control, 15.9 % ± 5.4 (s.e.m.)
Condition1, 17.7% ± 4.9
Condition2, 25.6% ± 5.4
N=8 for each group.

(One-Way ANOVA followed by Dunnett’s post-test, Control vs Condition 1 group q_2,23 = 3.2, P < 0.05 and Control vs. Condtion 2 group, q_2,23 = 2.5, P < 0.05, n = 8 per group

[Lenth, Russell V]

Well, I suppose the problem is with the inputs you used for the power calculations. But there are some serious anomalies in what you do report. Among them is that the degrees of freedom seem wrong. With 3 groups and 8 observations per group, there should be 24-3=21 df. Moreover, the difference between control and cond 1 is clearly not statistically significant, if those really are the SEs of the means. And surely that difference of 1.8 is less, not more, significant than the difference of almost 10 between control and cond 2. There are all kinds of things in there that don't make sense, even before we get to anything that went wrong with the power calculations. I suggest finding a good consultant who can talk this through with you.

Russ

Sent from my iPad

--
You received this message because you are subscribed to the Google Groups "PiFace discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to piface-discuss...@googlegroups.com.
To post to this group, send email to piface-d...@googlegroups.com.
Visit this group at http://groups.google.com/group/piface-discussion.
To view this discussion on the web visit https://groups.google.com/d/msgid/piface-discussion/86101cca-d15a-460d-a2f0-c2786bbd26e8%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.