CBL-514 Statistics: Confidence Intervals and P Values
Ask how large before asking how significant
CBL-514 papers report statistical comparisons, including P values.[1] A research summary should identify the outcome and effect size before highlighting a significance threshold. The same P value can accompany a small precise difference or a larger but less precise difference.
In the usual frequentist interpretation, a 95% confidence procedure has 95% long-run coverage under its assumptions. It does not assign a 95% probability to the fixed true effect being inside one particular calculated interval. FDA’s E9 guidance discusses estimates, intervals and hypothesis tests together.[2]
A worked interpretation
Hypothetical example, not CBL-514 data: two analyses both estimate a 40 mL difference. One interval extends from 30 to 50 mL; the other extends from 2 to 78 mL. Their central estimates match, but the second leaves much more uncertainty about magnitude.
Neither interval describes the full range of outcomes experienced by individual participants. A confidence interval around a group contrast and a distribution of individual changes are different objects. A chart should identify which it displays.
Likewise, a small P value is not the probability that the treatment works, the probability that a result will replicate or the probability that there are no important biases. It is conditional on a statistical question and assumptions. Measurement problems and selective reporting are not erased by additional decimal places.
What to save from a CBL-514 table
Record the effect estimate with its units, comparator, visit, interval level and analysis set. Keep an accompanying P value in a separate field. If only a significance symbol appears, do not invent an exact value or interval.
Then write one sentence describing what the interval permits you to conclude about magnitude. For instance, note whether it leaves room for both a modest and a substantial difference. Avoid declaring two studies contradictory merely because one crosses a significance threshold and another does not.
Finally, review whether the analysis was prespecified and how many comparisons were tested. A numerical result is strongest when its interpretation remains consistent with the study’s design and reporting context.
Frequently asked questions
Does P below 0.05 prove a large effect?
No. Statistical significance and effect magnitude are separate. Read the estimate, units and uncertainty.
Is an error bar always a confidence interval?
No. It may show a standard deviation, standard error or another quantity. Use the figure legend rather than guessing.
Continue your CBL-514 research
Compare the original study evidence with the documentation for the specific material you are reviewing.
