I am collecting data to look at some correlations and noticed something that I never thought about in-depth during the earlier stage when I had little observations to work with.
Between two of my variables early on I found a significant correlation of around $r=-0.85$ with a small sample size of $N=23$. I know that outliers will skew the point estimate with this little data even if it is deemed to be significant through the $p < 0.05$ approach.
I found that the standard error of correlation can be calculated through $\sqrt \frac{1-r^2}{N-2}$ which in this specific case leads to a standard error of around 0.115.
My question comes when I calculate confidence intervals for r. If I were to create a 95% confidence interval I would have $(-0.85-1.96 \cdot 0.115, -0.85+1.96 \cdot 0.115)$ which evaluates to $(-1.0754, -0.6246)$. As this is a correlation, the range of $r$ can only fall between -1 and 1 so the $-1.0754$ lower range of that 95% confidence interval is not attainable.
In such a case, would the 95% confidence interval just cut off at -1 and become $(-1, -0.6246)$?