Let $p$ be an odd prime. Define $S(p)$ as the sum of all primitive roots modulo $p$ taken from $\left[-\frac{p-1}2,\frac{p-1}2\right]$.
Now here's the strange thing. If the primitive roots were 'random', you'd expect $S(p)$ to be negative about as often as it is positive. However, experiments suggest that, at the very least, $$\lim_{n\to\infty}\frac{\#\{p\le n:S(p)<0\}}{\#\{p\le n:S(p)>0\}}>2.$$ This is extremely counter-intuitive to me.
I don't expect one can compute the limit, or even prove it exists. But is there some intuition behind this phenomenon?
EDIT: If $p\equiv 1\pmod 4$, it is easy to see $S(p)=0$. Now consider $p\equiv 3\pmod 4$. We can go further and split into cases mod $8$. I've computed $S(p)$ for all primes $p<10^5$ with $p\equiv 3\pmod 4$.
The value of $p$ modulo $8$ turns out to matter a whole lot. For $p\equiv 3\pmod 8$, I found $$ \begin{align*} \#\{p\le 10^5:p\equiv 3\pmod 8,S(p)>0\}&=204;\\ \#\{p\le 10^5:p\equiv 3\pmod 8,S(p)=0\}&=10;\\ \#\{p\le 10^5:p\equiv 3\pmod 8,S(p)<0\}&=2196. \end{align*} $$ Meanwhile, the behavior for $p\equiv 7\pmod 8$ is completely different: $$ \begin{align*} \#\{p\le 10^5:p\equiv 7\pmod 8,S(p)>0\}&=1278;\\ \#\{p\le 10^5:p\equiv 7\pmod 8,S(p)=0\}&=39;\\ \#\{p\le 10^5:p\equiv 7\pmod 8,S(p)<0\}&=1083. \end{align*} $$ Refining even further: up to $10^5$ there are $1203$ primes $p$ satisfying $p\equiv 11\pmod {24}$. With the sole exception of the prime $65171$, these primes satisfy $S(p)<0$.