Wait — should I actually decide things this way? (No.) Here's why ⤵
It's a fun gag, but the bars don't mean what you'd want them to mean. Three reasons:
It measures joint probabilities, not the conditional ones you care about.
"I shouldn't have …" results count roughly
P(I did it & I regret it), and "I should have …" results count
P(I didn't & I regret not doing it). But to actually decide, you
want the conditional probabilities — P(regret | I do it) vs
P(regret | I don't). The counts are confounded by how often people do
the thing in the first place, so they can't be compared directly. Try
"run for president": almost nobody does it, so "I shouldn't have run for
president" is vanishingly rare — not because running rarely brings regret, but
because running is itself rare. The bar reflects the base rate, not the regret.
There's a heavy writing-about-it bias. It's really
P(I did it & regret it & wrote about it online). People only
post when they're able and willing to — which skews away from outcomes that are too
embarrassing to admit, or so bad they took away your ability or desire to write about
them at all.
Regret is binary; consequences aren't. A justified decision needs the
expected cost, not just the probability of regret. If regret in one direction
means "wasted five minutes" and in the other means "lost a leg," counting how
often people regret each — without weighing how much — tells you nothing
about what to do.
Selection bias: the two groups aren't the same people. Those who did the thing
and those who chose not to are different populations, often differing in exactly the
traits that drove the choice. So even a clean P(regret | did) vs
P(regret | didn't) wouldn't tell you what to do — the regretful "didn't"
crowd might have regretted doing it too, and vice versa.
You might not be the average person. Even a perfectly measured number is an
average over everyone who wrote about it. Your situation, your values, and your
particular "her" can differ enough that the crowd's tendency simply doesn't apply to
you.