p values - urgent.. (Sep/20/2006 )

Ok guys,

I am in a tight situation and help me with this and I shall be eternally greatful !!!

Can i dervied a p value to determine significance using duplicate values?? I am comparing control vs treated groups. I use one way annova followed by Dunnet's post-test on Graph pad prism.

Is it right or I need to have three values??

or is my PI harrassing me at the last moment??

---

The above sentence I don't understand.
You get normally a p value of the overall test (in Anova F-test), and for each comparison a p-value. What you show depends on PI, journal, your prefs. I use the overalll p-value in text together with df, F-value, n; the other p in table that shows the differences between comp.
Hope you meant that.

---

ok this is the thing...

I have 3 values of each treatment group from one experiment. I derive the mean and SE.
Similarly, I derive the mean and SE from a duplicate experiement. Now when I have to make a bar graph..I just take the two means and average it and derive the p value.

Yes..I do the p for the all the treatment groups together. However, I need to put the p fpr each comparison on top of its relevant bar (on the graph)

according to my boss I cannot do this since Ihave only technically 2 means which corresponds to 2 values.
is he right??

---

I wouldn't do it also, because you lose information and perhaps differences. I would look if the two identical groups of each of the two experiments are sig. different. If not, you are allowed to pool them to get for each treatment a mean +SE and then make a stat test to get p-value for each treatment comparison.

---

The two means that I get from the duplicate experiemnts are not significantly differnt from each other. In fact, I calculate the SE or SD for the two means and use the averaged mean, SE, n for calculating the p.

For eg.
Exp 1.
Mean of 2-3 values is 83.4 %

Exp 2
mean of 2-3 values is 84.5 %

For p value:
Mean of 83.4 and 84.5
SE of 83.4 and 84.5

I cannot combine the means together if they are not similar, then it would skew my numbers. I unsersand that.
so am i right or wrong.

Thanks a million for your prompt response.

---

I would compare mean from exp. 1 vs. 2 (same treatm) (with chi2-test if you have percentage values or transform and use t-test) > no difference > pool data exp 1 & 2 > 6 values > mean + SE
The other experiments same way, then compare treatments to get p-values., e.g with Anova + Tukey as most common one.
This is the way I would go, if I remember my lessons correct.

---