Re:Children born in the pandemic have significantly reduced ...
by jpeloquin on Sunday December 19, 2021 @05:11PM
(#29618967)
> In figure 2, the authors argue that there is a trend. But the graph seems to show the direct measurements of dependent variables with confidence intervals, and they are all non-significant.
From the Figure 2 caption, I'm pretty sure it's showing standard deviations (of the sample) not 95% confidence intervals (of the mean). Two samples may have overlapping ± 1 s.d. intervals despite having significant different means.
Re:Children born in the pandemic have significantly reduced ...
by uniqueuid on Monday December 20, 2021 @02:28AM
(#29622285)
Thanks for catching this, that's even more straight-forward (and arguably less informative, but the best solution would be to have both plots of raw data and marginal effects of course).
In figure 2, the authors argue that there is a trend. But the graph seems to show the direct measurements of dependent variables with confidence intervals, and they are all non-significant. This means that there is no reason to talk about a trend, especially if the difference between significant and non-significant is not significant :)
[edit]: Would really have loved if they showed marginal effects here, removing the influence of control variables.
But the models themselves do something else: Table 3 suggests that covid was used as a binary variable to test for. So they are comparing pre-covid to covid-time and see a significant result there.
Methodologically, it doesn't seem like a bad design, but the sample size is really a bit small to have sufficient power.
Apart from that, the effect size of covid (31) seems really high, but developmental psychology is not my field so I can't say anything about that.