Interesting stuff, I love the range of ideas.
@leoneler - Yeah, it totally can be about jumping to conclusions, and descriptive Epi can allow you to be surprised. I like the quote by Isaac Asimov, who once said, “The most exciting phrase to hear in science, the one that heralds new discoveries, is not ‘Eureka!’ but ‘That’s funny…’” “That’s funny…” is the sound of something catching in your brain. A contradiction, an anomaly, just plain weird: something doesn’t fit what you expected.
@aokhisa I like your last point about standardized processes. I’ll talk about that a bit further on.
@amy.mikhail I wasn’t quite looking at whether it’s useful, more I think it is useful but why do you think that and why do we chose to do it. Your “traditional” perspective is interesting; I’ve never heard of that before. It seems somewhat similar to the idea that you cannot generate a hypothesis and confirm a hypothesis on the same source of information. Is that more what you mean? I think you can use the same cases for both descriptive and inferential analyses. The descriptive is just about getting summary estimates; the inferential is about quantifying uncertainty around those estimates. However, if you were to generate an idea that two groups are different in your descriptive, then you can totally quantify the uncertainty of that idea in “your” data. You just can’t say that those two groups are really different.
I think this relates to the movement in Statistics to stop talking about p-values as evidence and start talking about compatibility. We start to use phrases such as “the data are compatible with this pattern.” Then you’re highlighting that it is just your data, and conclusions should really only happen at a meta level. So scientific changes only start to happen in meta-analyses. But that’s getting a bit off-topic and into the Philosophy of Science. Here’s a paper on compatibility instead of evidence.
@paulablomquist - The devil is indeed in the detail. I like the idea that we’ve agreed, as a community, on key indicators. But how do you think they were arrived at? And did we agree? Or were they somewhat discovered by probability theory?
For instance, why do we calculate R? You could say we calculate R because it tells us how many people, on average, will be infected. Or you could say we calculate R because it is a summary of a distribution, and the most efficient and effective estimate to summarize a distribution in general is a mean, and R is just a mean. So did the Epis decide, or was it kind of already determined? And then, is R, or whatever we have—attack rates and proportions and means—are they the best way?
Now you could also ask: Is R descriptive Epi or inferential? Well, if you have case and contact data, then you can calculate R just by doing a summary of how many people got infected from each case, and that’s just descriptive. But we often don’t have that, so that’s why it sits in inferential usually.
@lnielsen I like how you reference a collective understanding. I think descriptive statistics can be thought of as a common language. I think that provides shortcuts for communication. Interesting that you consider it the “what” and then perhaps the inferential is the “why”? I tend to think of the descriptive as trying to figure out what is going on and how to describe whilst exploring stuff based on possible whys.
@kevin.vanzandvoort Very practical. Yeah, descriptive stuff is often how we spot patterns for stuff that is wrong. I wonder if you consider data cleaning and descriptive to be different things? In clinical trials, cleaning would be separated out as, I guess, descriptive analyses that help you identify issues, and then you sort them. Then descriptive would be the presentation of tables. But I get your point that unless you look at your data, you won’t know if stuff is rubbish, and the way to look at your data is usually descriptive stuff.
@jestrada I’m glad to hear how important you find it. So for you, it helps you really balance when you’ve got loads of information. I think this is a key aspect.
