choice of t-test

#1
Hello
I have been presented with some data and, being a bit rusty on my statistics am unsure whether I have used the correct t-test. The data shows the results of of farmer trials; 22 plot-pairs, with each plot-pair containing a control and a test.
Based on my understanding, the correct t-test is a paired two sample for means - which I have done.
However, the client has asked me to check other t-tests (i.e. two sample assuming equal or unequal variances).
Based on the F-test, the variances of the two data sets are equal; therefore I could perform a two sample assuming equal variances.
But as I understand it, that type of t-test is not suited to testing pairs of results.
Am I correct?
best wishes,
 
Last edited:
#2
Based on my understanding, the correct t-test is a paired two sample for means - which I have done.
I agree.

But if there is no relation between the pairs then you can as well do a two sample t-test (based on constant variance or a Welch test based on non-constant variance).

But, as an almost philosofical point, if there is a restriction on the randomization, such that the treatment and control was randomized to pairs, then it should be analysed as a pairwise t-test.

(But show the audience what they want to see, the result of the three tests.)
 

Karabiner

TS Contributor
#3
But, as an almost philosofical point, if there is a restriction on the randomization, such that the treatment and control was randomized to pairs, then it should be analysed as a pairwise t-test.
Perhaps the OP could describe in some more detail how the case/control pairs were created?

With kind regards

Karabiner