if I had just one shot to share an important idea💡here it is. No joke. Look 👀
RATIOS DEMAND LOG SCALE
👇 the coolest non-trivial example when a log Y-axis is required to represent ratios
going from 50 in ~1988 to 25 in ~2001 is as bad as going from 100 in 1970 to 50 in ~1988
Fascinating how the data story is changing – the seeming recent plateau is no longer there. And log *is* the *right* way to represent the nature of the data and tell the story.
RATIOS appear whenever you divide something by something else of a similar class/nature. Ratios are dimensionless. The most common "hidden" case is time series standardisation when some year is taken for 100%, like 1970 in the example above
The thing is: the story (the shape of the line) will change depending on the arbitrary choice of the year taken as 100%. We divide all other years by this standard effectively creating *ratios*
I genuinely believe this is *the* most common #dataviz issue
It also often becomes a problem in statistical analysis – log transformation of the ratios are required before regression/correlation analysis is performed on the data
The realisation of this simple data visualization and analysis rule literally took me ages of slow mind walk: from odds ratios through other explicit ratios to the "hidden" ones like here. Even color bars in the plot legends should be log transformed if they colorcode ratios!
I lost track how many times I've pointed this out 😅
RATIOS DEMAND LOG SCALE
The idea to write a blog post on visualizing ratios correctly lives in my head and to-do list for well over a year now
Let me use this thread to legally bind myself to write it up soon 🔗
• • •
Missing some Tweet in this thread? You can try to
force a refresh
I started Demographic Digest project in my second PhD year (or rather half a year after my modified PhD topic was set). Having struggled through the first feverish reading of the literature of the topic, I realised that I need to radically up the skill of skimming through papers.
Reading deeper and deeper into the specific domain of demographic literature, and not being 100% sure that this was even *my* topic, I also missed a lot a broader view of what was being discussed at the cutting edge of demographic debates.
For the context, @VPrasadMDMPH published this piece statnews.com/2020/04/27/hea…
calling for a more constructive academic discourse of the manifestly wrong IFR estimates by Ioannidis.
So sweet of him.
Let's see if @VPrasadMDMPH was equally mature, responsible, and wise when it was Ioannidis vigorously attacking (in a paper published in the journal where Ioannidis was editor in 2010—19) @GidMK for persistently and convincingly debunking his c19 research
Unlike many statistics and quantities of general use that we tend to see regularly, life expectancy is not observed directly. It's an output of a *mathematical model* called life table.
So, why can't we do without a model?
2/
Consider a seemingly simple task: you want to know how long people live. What can be easier? Let's just see how many years lived those who died recently.
Why not?
Such a metric would be massively driven by population age structure
3/
How can one be satisfied with the "corrected" "finding" that literally claims that missing a couple of months of schooling is the most detrimental thing that can happen to a kid in the remaining life...
but (!) only if it's a US kid; if the kid is from Europe, all fine
Two months since the publication, our critique w/ @GidMK (osf.io/9yqxw) caused the authors to revise the (remember, thoroughly peer-reviewed) paper and publish a correction
THREAD
tl;dr: retraction is still the only way
how it started: how it's going:
Now the paper's page at @JAMANetworkOpen mentions a correction, and 2 comments are published: ours (which is a very short summary of our critique, osf.io/9yqxw) and the response from the authors.
But first, let me sincerely thank you for the open take of the debate – that's an honest and strong move 💪
I have to say, after the first, embarrassing tbh, reactions from the friend circle of the papers' authors I had very little hope for what's happening here.
2/
Involvement of the authors in the open discussion of their results is a cornerstone thing in science. Mistakes are inevitable, and those who can't admit and fix them are hindering the scientific process.