SoylentNews is people
SoylentNews
[...] Researchers found children who had a vegetarian diet had similar mean body mass index (BMI), height, iron, vitamin D, and cholesterol levels compared to those who consumed meat. The findings showed evidence that children with a vegetarian diet had almost two-fold higher odds of having underweight, which is defined as below the third percentile for BMI. There was no evidence of an association with overweight or obesity.
Underweight is an indicator of undernutrition, and may be a sign that the quality of the child's diet is not meeting the child's nutritional needs to support normal growth. For children who eat a vegetarian diet, the researchers emphasized access to healthcare providers who can provide growth monitoring, education and guidance to support their growth and nutrition.
[...] A limitation of the study is that researchers did not assess the quality of the vegetarian diets. The researchers note that vegetarian diets come in many forms and the quality of the individual diet may be quite important to growth and nutritional outcomes. The authors say further research is needed to examine the quality of vegetarian diets in childhood, as well as growth and nutrition outcomes among children following a vegan diet, which excludes meat and animal derived products such as dairy, egg, and honey.
Journal Reference:
Laura J. Elliott et al. Vegetarian Diet, Growth, and Nutrition in Early Childhood: A Longitudinal Cohort Study [open] Pediatrics 2022
DOI: 10.1542/peds.2021-052598
(Score: 0) by Anonymous Coward on Thursday May 05 2022, @10:58AM (5 children)
And this is why you have to watch out with putting two statistical conclusions out of context. While they seem contradicting at first (caught me as well),
statistically they could both be valid. I haven't looked into detail, but I expect it to be something in the lines like:
The first statement is about the 9,000 children, where they didn't find any significant differences within the classes on various parameters. (Done by t-test or anova)
The second statement is about the distribution within these classes. They aren't completely the same, but small enought to result in the first statement being true and second statement being observed (example: one class shows 0.1% underweight, while the other shows 0.2% underweight = double the odds, chi^2 test required to test significance).
(Score: 1, Interesting) by Anonymous Coward on Thursday May 05 2022, @12:13PM (4 children)
You can't conclude something is similar or dissimilar based on statistical significance. That requires understanding the practical significance.
In fact the statistical significance only tells you the sample size. If it is stat sig you had big enough sample, if not then it was too small for how variable your measurements are.
(Score: 0) by Anonymous Coward on Thursday May 05 2022, @01:59PM (3 children)
Not sure where you got that from, but as far as I know the whole statistical field is based on the definition below (from wikipedia):
https://en.wikipedia.org/wiki/Statistical_significance [wikipedia.org]
If statistical significance would only mean something in correlation with a sample size, no statistical test would ever produce anything useful. I think what you try to say is that statistical significance tells you more about (unknown) errors, but it doesn't have to get smaller when you get a larger sample size (I know this from experience).
Variabillity within your samples can be a trait of your population. A population with large variabillity will result in a sample with a large variabillity, increasing your sample size won't fix that. Even stronger, on a schientific level it's something you might not want, because you're introducing a bias into your samples if you do this (big no-no).
Tests work often around this variabillity (requiring equal distribution of samples), resulting sometimes in a lower significance. BUT, that's not a shame in itself. It often means you have to tone down your statement, use other tests (e.g. non-parametric tests) AND mention that you saw this great variabillity, to justify your choice in used tests.
(Score: 0) by Anonymous Coward on Thursday May 05 2022, @04:21PM (2 children)
If you measure the heights of a billion people with blood type A vs blood type B and find a statistically significant difference of 0.1 mm, are they similar or dissimilar?
Similar, of course, because 0.1 mm has no practical impact. The statistical significance just shows you collected enough data to find a tiny difference.
Likewise comparing two groups of n=3, will be statistically insignificant even if one is 2 ft taller than the other on average.
(Score: 0) by Anonymous Coward on Thursday May 05 2022, @04:50PM (1 child)
First, I doubt that such experiment would result in a statistically significant difference. This would mean you do a test on the mean (which is 0.1 mm different) and the deviation would be very (extremely) narrow. But let's go with what you say and such thing would be significantly different, then the two classes would not be the same, so dissimilar. It would mean that people with one blood type have a good chance to be, on average, slightly larger.
"Practical impact" has no use in statistics, who decides what's practical? It's not objective.
(Score: 0) by Anonymous Coward on Thursday May 05 2022, @06:24PM
If n = 1 billion even very tiny differences will be significant. Probably orders of magnitude less than 0.1 mm.