Also this week: against pi, statistical populations as useful (or not-so-useful) fictions, weasel words about causality, the public face of your scholarly discipline, confidence intervals vs. vitamin D, and more.
Theoretical physicist Stephen Hawking, perhaps the world’s greatest living scientist, and surely its most famous, passed away this week. He was 76. Diagnosed with motor neurone disease at 21 and told he had two years to live, he went on to predict that black holes could emit radiation, a stunning claim with deep implications for fundamental physics. The prediction was later confirmed. He made many other seminal contributions to physics and cosmology, and wrote the popular science bestseller A Brief History of Time. In his later years, using a wheelchair to get around and speaking with the aid of a computer, he became one of the few scientists instantly recognizable to a wide swath of the general public. He played himself in tv shows and movies, and was the subject of the biopic The Theory of Everything.
Data Colada on how the notion of a statistical population, characterized by fixed but unknown population parameters, is (sometimes) a useful fiction. If you want to estimate the true average height of the students in your statistics class, that’s a perfectly well-defined population from which we can take a random sample just like the textbooks say. But the true average effect of N enrichment on plant species richness doesn’t exist, because there’s no well-defined population (not even an infinite one) of studies of N enrichment on plant species richness, from which we could randomly sample.
An oldie but a goodie: what is the public face of your academic discipline, as represented by what bookstores display on their shelves (or these days, what Amazon displays in response to search queries)? For instance, for “history” in the US the answer is “WW II and the Civil War”. For ecology, the answer presumably is “environmentalism”.
Sticking with Kieran Healy: a very good short tweetstorm on using weasel words to hint at causal claims without actually committing to them. He’s talking about sociology, but how often do you think this is done in ecology?
Still sticking with Kieran Healy: what statistical software would the founding fathers of sociology have used? 🙂 Sample:
What makes a scientific theory “interesting”? A classic discussion. From the social sciences, but it generalizes. (ht Andrew Gelman)
Here’s a great example of a misunderstanding of the confidence interval of a sample mean, made by people with advanced degrees and with potentially-serious public health consequences. Good fodder for an intro stats course; I used it in my intro biostats course this week.
Against pi. (ht Matt Levine)
Am I missing something, or is there a basic statistical error in that article about basic statistical errors? It refers to the intervals calculated as confidence intervals, but I think they are prediction intervals – and they’re referred to as such later on in the article. Confidence intervals don’t tell you anything about where 2.5% of the population will be in relation to a mean or a fitted line.
Yes, the article switches between terms, and it shouldn’t. But the substantive point stands.
Please do not say “confined to a wheelchair.” Wheelchairs are mobility aids– they are freedom, not confinement.
Here’s someone who puts it very well: https://twitter.com/tommysantelli/status/974176034928160768
Jeremy asks the provocative question (do we expect less?) how often ecologists hint at causal claims in their papers. This is certainly an old topic of discussion, but there is fresh paint on it these days because causality is a hot topic, thanks to Judea Pearl’s attempts to codify the necessary requirements for causal claims. I do think social scientists discuss this issue more than ecologists, (and computer scientists are now hot on the topic too). During the past year I have taken three excellent courses on causal statistics, all taught by social scientists, and all through the Statistical Horizons Company. It is interesting to see what topics draw crowds from the human sciences (upcoming list at https://statisticalhorizons.com/seminars/public-seminars).
I notice that you carefully avoided answering my provocative question, Jim. 🙂
Ok, obviously it’s not a question to which I can give a quantitative answer, and I wouldn’t expect anyone to be able to (trying to quantify an answer would be a challenging but interesting exercise…). But I’m still curious about people’s subjective impressions of the literature. The words with which social scientists hint at causal claims without actually committing to them is different than the words with which ecologists do so. Different fields have different typical writing styles. But still: very curious to hear if you think it’s common for ecologists to say things that kind of sound like causal claims but that aren’t actually rigorously supported.
I’m not sure myself. Offhand, I’d say the most common way for ecologists to hint at claims they haven’t actually established is to say their data “suggest” the claim. I think ecologists write that way fairly often. That’s stronger language than saying the data are merely “consistent with” the claim, or “fail to rule out” the claim (though ecologists do often use those phrases too). “Suggest” makes it sound like you’ve almost demonstrated the claim; “consistent with” lacks that connotation. Even though “suggest” doesn’t really *say* anything beyond what “consistent with” says.
Note that I’m sure that if I went back through my old papers I’d find at least one in which I used “suggest” when I should’ve used “consistent with”. We are all sinners. 🙂
The distinctions you use, “suggest” versus “consistent with” seem to be a useful dichotomy. I think it would be a good exercise to look at a sampling of papers to see what methods were used and what language was chosen to present the findings. Rather than try to speculate or go off on some tangent, I will offer to find a slice of time to do so. Do you have a suggested approach? I know you try to quantify answers to questions like this one on a regular basis. In the hope of doing it once instead of too many “go backs”, if we can arrive at a plan, I will pledge a little time to the task.
Wow, thanks for the generous offer Jim! I think this would be an interesting exercise to try to collect a bit of data that improve enough on anecdotal impressions to be worth talking about in a blog post. Shoot me an email and let’s try to come up with a plan. email@example.com
Some of my geologist colleagues like the term “permits”, as in: “the available evidence permits interpretation X”. “allows” is also used in this context.
Both are weaker than “suggests” or “consistent with,” but they still allow the writer to segue into an extended discussion of their pref’d interp at the expense of possibly stronger interps
I like the mental image of the evidence as an indulgent parent, giving a teenager “permission” to take the family car on Saturday night or something. 🙂
“indulgent parent”, ha, I like that, it certainly fits with some folks and their pref’d interps! 😉