The study had about 30 participants, with 15 in each arm, so that's not a lot of statistical power to begin with.
Both the low-glucose-label and high-glucose-label arms showed a spike in actual measured blood sugar after ingestion. The difference in blood glucose between the two arms was just about 10%.
The naïve p-value between the two arms was barely significant at 2%, 6%, and 2%, but the p-values were taken from a complex model that probably has p-hacking problems and needs p-value correction.
I would wait for the replication study before believing in this effect. If the effect is real, they only need to double their sample size to get a t-stat of 3 instead of 2, which would give a much more convincing p-value of 0.2% instead of the 2-6% right now.
I applaud the authors for exploring this theory though, and encourage replications to push the p-value down.
Not that my statistics knowledge is great, but I wonder why they don't show fig.4 with SD bars? (Or if the plot represents the model, show the data points the model was fitted to?)
The study had about 30 participants, with 15 in each arm, so that's not a lot of statistical power to begin with.
Both the low-glucose-label and high-glucose-label arms showed a spike in actual measured blood sugar after ingestion. The difference in blood glucose between the two arms was just about 10%.
The naïve p-value between the two arms was barely significant at 2%, 6%, and 2%, but the p-values were taken from a complex model that probably has p-hacking problems and needs p-value correction.
I would wait for the replication study before believing in this effect. If the effect is real, they only need to double their sample size to get a t-stat of 3 instead of 2, which would give a much more convincing p-value of 0.2% instead of the 2-6% right now.
I applaud the authors for exploring this theory though, and encourage replications to push the p-value down.