Category: Method

Why this program deliberately avoids the most capable model available, and what “interpretable” has to mean to be worth anything.

“Social media analysis” is a phrase that should make you suspicious. Here is precisely what this component does and does not do.

Several terms in constant use in this field are not what they appear to be, including one in our own founding material.

We could not find evidence for the usual justification, so here is the one that survives contact with the data.

The AHRQ finding on risk mitigation is stranger than “the evidence is weak,” and it applies to most of the standard toolkit.

All the strongest evidence on tapering harm is observational. That is a real limitation and it is not a reason to ignore it.

When the behavior you are screening for is rare, most of what you flag will be something else. This is arithmetic, not a criticism of anyone’s analytics.

Two things moving at the same time in national data is almost never enough to establish that one caused the other — in either direction.

Prevalence, incidence and persistence answer different questions, and policy written against the wrong one comes out mis-sized.