- Thread starter Sunil Neelam
- Start date

What can you tell from inspecting the data?

Where are the mid points (median or mean) in each sample and how spread out is the data (IQR or st. dev)? What is your sample size? What are you testing: within-subjects treatment effects (in which case your treatment and control wont be independent), or between (in which case your treatment & control should be independent)?

First, test for normality. If you have a stats program that does Anderson-Darling, Ryan-Joiner, or Kolmogorov-Smirnov tests, this is a good way of testing it. Otherwise, you could construct a quantile plot in Excel and visually judge how close the point fall to a 45 degree angle (to do this, complete as below then make a scatter plot with Col A on the x-axis and Col B on the y-axis):

Column A Column B

(i-0.5)/n variables in order from lowest to highest

=(1-0.5)/20 Variable 1 (lowest)

=(2-0.5)/20 Variable 2

=(3-0.5)/20 Variable 3

....

=(20-0.5)/20 Variable 20

If you data appear to be sloping upward by 45 degrees in roughly a straight line then you should be ok to use a paired t-test. If not, then you might have to use a non-parametic equivalent like a Wilcoxon paired signed rank test or a paired sign test.

Make sure you set up your hypotheses correctly too, because this will affect the p-value. So, if you are testing whether Control is not equal to Treatment, then this becomes a two-tailed test. If you are testing if Control > (or<) Treatment, then this becomes a one-tailed test.

Let me be more elaborate.

The experiment is to study the effect of elevated co2 on the performance of 20 different plant germplasm lines.

One set under controlled conditions and the other under treatment (i.e elevated co2)

data such as plant ht, leaf area, leaf dry weight etc about ten such parameters were recorded. Can we do the analysis with the paired t test ? please clarify considering the data follows normal distribution.

Thanks in advance

Then you have a multivariate experiment & I probably would look for ways other than multiple t-tests (or non-parametric equivalent) to test for group differences. If it was me I would be thinking about wehther a multivariate ANOVA test might be more appropriate, possibly followed up by a one-way ANOVA to pin down where the effects were happening. note, you could just go straight to the latter...but once again, make sure you meet the assumptions of the test (and make sure you read what these are first).