How valid are the statistics, or rather the conclusions, for some of the (maybe unspoken) claims on some charts? Take the last one, "How many deaths does it take for a disaster in different continents to receive newscoverage?". This suggests that disasters in Europe get reported on much sooner than those in Asia in Africe; which I believe, I'm just wondering to what extent we can conclude that from this data. But it also (maybe only in my perception?) seems to make a moral judgement on that - that that is unfair, biased, discrimination, or just overall morally objectionable. Many comments in this thread seem to focus on that point, or just take it as a given.
The op says for this specific chart 'Note: This data is controlled for several factors, including the number killed and affected, country, year, month, and type of disaster.'. There are no famines in Europe (ok maybe depending on definitions and time period, there were some in the 60's and 70's in former communist countries?), which make the largest number of casualties by far. So how is it 'controlled for' by disaster type? When you look at 'data' for that chart, it doesn't even mention 'disaster type' under 'controlled for'! So what is it?
Furthermore, this chart suggests that every disaster in Europe with at least 1 casualty is reported on. That seems, let's say, rather unlikely. So is this data normalized with Europe as the base? (I'll admit that I only skimmed the actual paper). But that would be weird - what is the point of this graph? What could it show, and what does it show?
Look I'm not saying these people (i.e., the authors of the original paper) deliberately misrepresented anything, or that they set out to prove a point. It just looks more like a 'hey we have a cool idea for what we can do with data, let's make regression models, it'll be fun (and probably at least somewhat informative to others as well)' than some fundamental inquiry into the morality of news reporting. And that's fine - much (dare I say most) published papers have 'only' that level of depth; I've done it myself. But my problem is the big jump from numerical relations between some 'data' and broader conclusions (not by the authors, usually, but by people reporting on it, like the OP).
In this case, the numbers support the intuitions most people would have. I'm not even doubting the results, or the overall trend or bias. What I'm wondering is to what extent we can claim that this sort of analysis actually 'proves' those intuitions. Are so-so analyses, and hand-wavy generalizations made from them, better than having none at all? I used to think 'math is factual and any data that is not completely bogus is better than none at all', but nowadays, I'm not so sure any more.
The op says for this specific chart 'Note: This data is controlled for several factors, including the number killed and affected, country, year, month, and type of disaster.'. There are no famines in Europe (ok maybe depending on definitions and time period, there were some in the 60's and 70's in former communist countries?), which make the largest number of casualties by far. So how is it 'controlled for' by disaster type? When you look at 'data' for that chart, it doesn't even mention 'disaster type' under 'controlled for'! So what is it?
Furthermore, this chart suggests that every disaster in Europe with at least 1 casualty is reported on. That seems, let's say, rather unlikely. So is this data normalized with Europe as the base? (I'll admit that I only skimmed the actual paper). But that would be weird - what is the point of this graph? What could it show, and what does it show?
Look I'm not saying these people (i.e., the authors of the original paper) deliberately misrepresented anything, or that they set out to prove a point. It just looks more like a 'hey we have a cool idea for what we can do with data, let's make regression models, it'll be fun (and probably at least somewhat informative to others as well)' than some fundamental inquiry into the morality of news reporting. And that's fine - much (dare I say most) published papers have 'only' that level of depth; I've done it myself. But my problem is the big jump from numerical relations between some 'data' and broader conclusions (not by the authors, usually, but by people reporting on it, like the OP).
In this case, the numbers support the intuitions most people would have. I'm not even doubting the results, or the overall trend or bias. What I'm wondering is to what extent we can claim that this sort of analysis actually 'proves' those intuitions. Are so-so analyses, and hand-wavy generalizations made from them, better than having none at all? I used to think 'math is factual and any data that is not completely bogus is better than none at all', but nowadays, I'm not so sure any more.