3 ms·

Here's the article: https://www.pnas.org/content/112/4/1232 https://www.pnas.org/content/112/4/1232 Those numbers aren't errors--they're standard deviations. T

Here's the article: https://www.pnas.org/content/112/4/1232 https://www.pnas.org/content/112/4/1232 Those numbers aren't errors--they're standard deviations. The error would be sqrt(N=12), or about 3.5x, smaller, so 3 and 6 minutes, respectively. They also aren't particularly easy to interpret here because it's a crossover design, so the relevant quantity is the change in each subject's sleep onset (or whatever), which is captured by the mixed-effects model, but not the summary statistic. That said, there are some weird things with the setting: four hours is a lot of reading, and the subjects lived in a fairly dim environment (90 lux) throughout the experiment. Some data (e.g., Chang et al., 2011: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3060589/ https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3060589/) shows that the range of exposure matters, so this effect might be minuscule in real life, where the noontime sun exposes people to thousands of lux.

bildung 7y ago
Perfectly right that standard deviations are not the same es errors! Compared to the average effect size, though, in this case they clearly show there's a very high variation in the observed results. The 4h is really unusually high compared to real-life reading in bed. But compared to the other studies its still quite low! One of the other studies had a whopping 6.5h hours of "bedtime reading".
- mattkrause 7y ago
  Right, but the point of the crossover design and mixed-effects model is that some of that variability is attributable to individual differences: you always fall asleep quickly; I toss and turn for a bit. Imagine you fall asleep in 1 minute normally, but take 11 minutes after light exposure. I take 21 minutes, but 31 after light exposure, and a third person goes from 11 to 21. The standard deviation of each condition is the same size as the effect (10 minutes), but there’s still an effect.

bildung 7y ago

Perfectly right that standard deviations are not the same es errors! Compared to the average effect size, though, in this case they clearly show there's a very high variation in the observed results. The 4h is really unusually high compared to real-life reading in bed. But compared to the other studies its still quite low! One of the other studies had a whopping 6.5h hours of "bedtime reading".

mattkrause 7y ago
Right, but the point of the crossover design and mixed-effects model is that some of that variability is attributable to individual differences: you always fall asleep quickly; I toss and turn for a bit. Imagine you fall asleep in 1 minute normally, but take 11 minutes after light exposure. I take 21 minutes, but 31 after light exposure, and a third person goes from 11 to 21. The standard deviation of each condition is the same size as the effect (10 minutes), but there’s still an effect.