Since Duhem-Quine, it has become orthodoxy that scientific hypotheses are not tested in isolation. With regards to God, this means that we need to add on some auxiliary hypotheses to generate some predictions about God. We don’t simply test God; rather, we test the prayer-answering God, the YEC God, etc.
God hypotheses with suitable auxiliary hypotheses are falsifiable, viz., the God that answers prayers. By prayer I mean petitionary prayer, like asking God to cure another person of an illness. I’m not talking about prayer in the sense of meditation. This is old news, but the prayer-answering God has been falsified.
The objector might say that God does answer prayers, but prayer is not testable, because all we have is correlation and not causation; and there may be numerous unknown factors that we’re not accounting for. I don’t think this is a good response because this type of testing is prevalent in the sciences.
Statistician and biologist R. A. Fisher argued that we can infer causal relations from statistical data if there is an appropriate randomization within the experiment. Suppose we want to test a drug’s causal efficacy in curing some disease. We separate a diseased population, randomly, into two equally sized groups, and give the drug to one group. If a sufficiently high percentage of the drugged group is cured then it is reasonable to draw conclusions about the drug’s causal efficacy.
The idea is that in randomizing the two populations any special problems will even out among the two populations. For example, suppose God does not want to heal gentiles during a test for prayer efficacy on cancer patients. And suppose that people prayed for group A and group A is all gentiles. This would mean that we shouldn’t expect any difference in recovery between group A and B. But if we randomize the two populations, it is likely that the gentiles would be evenly distributed into group A and B and we should see group A have higher recovery due to prayer, if prayer works.
The theist may use a skeptical theist response: We don’t know all of God’s reasons. This response won’t work, because randomization would distribute God’s unknown reasons evenly between group A and B. But maybe God doesn’t answer prayers because he doesn’t want to be tested? The problem with this is that there are a lot of prayers that are implicit tests. For example, If I pray to get accepted into a university, I will eventually know whether I got accepted or not.
This is not to say that this type of inference is infallible or without criticism (e.g., maybe for some reason we are unable to randomize). But if you reject this type of testing, you’ll have to reject a lot of the testing in science. The other alternative is to say that there can’t be empirical evidence for or against God.
Okasha, Samir. Causation in Biology.