Common Errors in Statistics (and How to Avoid Them), 4th Edition

FUNDAMENTALS

Few experimenters fail to list number of subjects, doses administered, and dose intervals in their reports. But many fail to provide the details of the associated power for their sample sizes. Feng et al. [2001] found that such careless investigators report a higher proportion of nonsignificant intervention effects, indicating underpowered studies. Your report should include all the estimates used in prescribing your sample sizes, along with the smallest effect of practical interest that you hoped to detect along with the corresponding power to detect that effect.

Too often, inadequate attention is given to describing treatment allocation and the ones who got away. We consider both topics in what follows.

Treatment Allocation¹

Allocation details should be fully described in your reports, including dictated allocation versus allocation discretion, randomization, advance preparation of the allocation sequence, allocation concealment, fixed versus varying allocation proportions, restricted randomization, masking, simultaneous versus sequential randomization, enrollment discretion, and the use of intent to treat (ITT) analyses.

Allocation discretion may be available to the investigator, the patient, both, or neither (dictated allocation). Were investigators permitted to assign treatment based on patient characteristics? Could patients select their own treatment from among a given set of choices?

Was actual (not virtual, quasi-, or pseudo-) randomization employed? Was the ...