The answer to this question discusses problems associated with calculating P-values for dependent observations. Let's say you have observations from two different groups that are dependent. You consider carrying out a t-test to compare means of the two groups. However, aware that the calculated P-values in a t-test will likely be inaccurate, you calculate means of each group and their 95% confidence intervals. Does calculating confidence intervals avoid problems associated with calculating P-values when observations are dependent?