This post is one that has been restored after the hacking. All original comments were lost.
I have no idea what the answer to that question is. And I doubt Gine Roll Skjærvø, Frode Fossøy, and Eivin Røskaft know either. They’re the authors of the peer-reviewed “Solar activity at birth predicted infant survival and women’s fertility in historical Norway“, appearing in the Proceeding of the Royal Society B, a paper that made a splash last week.
From their abstract they say:
Using data on temporal variation in sunspot numbers and individual-based demographic data (N = 8662 births) from Norway between 1676 and 1878, while controlling for maternal effects, socioeconomic status, cohort and ecology, we show that solar activity (total solar irradiance) at birth decreased the probability of survival to adulthood for both men and women. On average, the lifespans of individuals born in a solar maximum period were 5.2 years shorter than those born in a solar minimum period.
Controlling for. Did you see that phrase, or did it swim by unnoticed? It almost invariably means some sort of regression or regression-like model, wee p-values, and mistakes and misinterpretations of causation. Let’s see if that’s the case here.
No expert in Norwegian demographics and economic history I, but I do know the times, they were a’ changin’, and changing rapidly over the late seventeenth to late eighteenth centuries. Did these changes cause lifespans to increase, and for the other things they measured like fertility to shift, over this period? As admitted, I have no idea. I’m guessing Skjærvø and friends don’t know either, because they mention nothing about this subject.
First thing the authors did was to split up years depending on whether the mean annual number of sunspots were higher or lower than some number. “This resulted in 1041 boys and 965 girls born in a solar maximum period ([per-year] mean: 96.9, range: 47.7–154.4) and 3362 boys and 3294 girls born during a solar minimum period (mean: 34.0, range: 0–118.1).”
Since it’s supposed to be solar beams, or whatever, that boogers up bewombed babies or their parents, and solar beams are related to the number of sunspots, why didn’t our team analyze the number of sunspots instead of this curious dichotomization? I have no idea. Onward.
Bring on the models. Generalized linear mixed models, that is. Think of them as fancy frequentist regressions. What is regression? Probably not what you think: see this and this to see what you should think.
“We always included the interaction between solar period (maximum/minimum) and socioeconomic status and the interaction between solar period and population. Non-significant interactions were sequentially excluded from the models,while significant interactions were further explored by rerunning the analyses for each group separately…”
“Significant interactions” means wee p-values. “Socioeconomic status” means for women in 1676, well, I’m betting wasn’t measured all that well.
Their Figure 1 is at the top of this post. Shows the “probability (mean +/- s.e.) of survival to adulthood in relation to solar activity for boys and girls. **p < 0.1 and *p < 0.05. Dark grey bars denote SSmax, light grey bars denote SSmin.”
P-values not so wee after all. The boys were “significant” only by loosening the intolerable loose criterion of 0.05 to twice that. Also, the authors said they ran the analyses “separately for each gender.” Why do that instead of the more usual inclusion of sex in the model? I have no idea. But I do know that when you run separate tests you have to adjust the weeness of the p-values lest they seem too low, an adjustment which they did not do here. In other words, even by the usual hopeless standard of “significance” there’s really nothing to see here.
In another model for the probability of survival to age 20, the coefficient associated with low sunspots numbers was “significant” for girls but not for boys. In another model of fertility, there was no wee p-value for men nor “high status” women, but there was weeness for “low status” women. And so on.
Why these less-than-remarkable signals? The p-values are watery weak, and regression does not causality make. Well, “increased folate degradation in solar maximum periods could result in folate deficiency in pregnant women”. It could at that. Did solar whatsits cause folate degradation in pregnant women? Who knows? And don’t forget “low status” women might have been, but who knows, outside fighting the sun, while “high status” women sheltered. Nowadays that’s opposite, no?
Or how about this? “[S]olar activity could differentially affect the abortion rate of male and female fetuses. If only the most robust male fetuses survive, while all female fetuses survive, male fertility could be less affected.” Speaking as a representative of the more robust sex, I’m hoping this is true.
The rest of the study I’m not so sure about.
1 reply »