Paired Differences T-tests for Odd number of Observations Suppose we have $m$ baseline data points and $n$ post-baseline data points where $m < n$. This is for one subject. Is it possible to perform a paired t-test on this data even though $m+n$ is odd? Can we thrown away the data that is not paired and perform the test?
For example, suppose the baseline data is $(1,2,3)$ and the post-baseline data is $(6,7,8,9,10)$ for one subject. Then we can form the pairs $(1,6)$, $(2,7)$ and $(3,8)$. The $9$ and $10$ would be thrown out. I guess if $m$ is not much greater than $n$, then we can still perform a paired t-test without a loss of power? Or perhaps do some data imputation?
 A: Given your comments to help clarify the problem - 
Let's say you measure Bob's weight at baseline 3 times, about a minute apart each. We can call these measures 1a 1b and 1c. Then you do something to him and measure his weight again a month later. Call these measures 2a 2b 2c. You seem to be trying to match 1a with 2a, 1b with 2b, 1c with 2c, and so on. The problem is that there is no reason to match 1a with 2a. In this case, 1a, 1b and 1c are all expected to be the same, except for small differences due to measurement error. It's no more logical to match 1a with 2a than it is to match 1a with 2b or 2c. 
Assuming the error in your measures is very small relative to the scale of the thing you are measuring (i.e. using a digital scale to measure weight in adult humans) the easiest, but not the most "correct", thing to do is just to calculate Bob's mean weight at baseline and follow-up using the respective repeated measures. 
A mixed model (or random effects model, multilevel model, hierarchical linear model) is the most correct thing to do, with your treatment included as a fixed effect. This will properly account for the within and between person variation in your sample. 
