Question

Suppose instead of comparing independent measurements taken from two groups, you used a matched-pairs experiment and one treatment is randomly assigned to each half of the pair. In this case, how should you compute the confidence interval for the difference?

Answer 1

You should use a T distribution to find the critical T value based on the level of confidence. The confidence level is often given to you directly. If not, then look for the significance level alpha and compute C = 1-alpha to get the confidence level. For instance, alpha = 0.05 means C = 1-0.05 = 0.95 = 95% confidence

Use either a table or a calculator to find the critical T value. When you find the critical value, assign it to the variable t.

Next, you'll compute the differences of each pair of values. Form a new column to keep everything organized. Sum everything in this new column to get the sum of the differences, which then you'll divide that by the sample size n to get the mean of the differences. Call this dbar (combination of d and xbar)

After that, you'll need the standard deviation of the differences. I recommend using a calculator to quickly find this. A spreadsheet program is also handy as well. Let sd be the standard deviation of the differences

The confidence interval is in the form (L, U)

L = lower bound

L = dbar - t*sd/sqrt(n)

U = upper bound

U = dbar + t*sd/sqrt(n)