On the other hand, if a variable is accidentally controlled across multiple replication attempts, then it will likely be accidentally controlled outside the lab too.
And this is how we missed that nitrocellulose isn’t actually explosive in a vacuum.
This does still leave one tricky problem: what if we accidentally control some variable?
Another issue is of anti-correlated variables. (Which, if one were being pedantic, could be treated as introducing another variable and accidentally controlling that.)
If I have an event E that requires both A and B, and in my experimental setup A and B cannot occur at the same time, I am going to conclude that E doesn’t happen with high probability. I am not controlling for A accidentally—I do try A—and I am not controlling for B accidentally—I do try B.
This is in some ways worse than the bare “accidentally controlling some variable” because the number of possible combinations grows exponentially with dimension.109 is bad enough. 2109… good luck. Even the bare handshake-pairing (109−1)∗109/2≈5∗1017 is terrible.
And this is how we missed that nitrocellulose isn’t actually explosive in a vacuum.
Another issue is of anti-correlated variables. (Which, if one were being pedantic, could be treated as introducing another variable and accidentally controlling that.)
If I have an event E that requires both A and B, and in my experimental setup A and B cannot occur at the same time, I am going to conclude that E doesn’t happen with high probability. I am not controlling for A accidentally—I do try A—and I am not controlling for B accidentally—I do try B.
This is in some ways worse than the bare “accidentally controlling some variable” because the number of possible combinations grows exponentially with dimension.109 is bad enough. 2109… good luck. Even the bare handshake-pairing (109−1)∗109/2≈5∗1017 is terrible.