3 ms·
In that case, why bother showing statistics demonstrating correlation unless there's actual evidence to back up some kind of causation? A high percentage of pe
by nailer 17y ago
In that case, why bother showing statistics demonstrating correlation unless there's actual evidence to back up some kind of causation?
A high percentage of people with certain characteristics being fired or having poor bonuses is factually not equivalent to those characteristics being the basis for those people being fired.
There are well established relationships between membership of particular groups and academic performance. It is generally not accepted to say that those people are being discriminated against when marking due to their ethnic groups unless there is specific evidence to prove that such wrongdoing has occurred. In fact, generally accepted reason behind the statistics in one not of discrimination, but of high rates of poverty amongst those groups causing other problems.
- jacquesm 17y agoThe statistics are not the only evidence, they are part of the evidence, and together with that other evidence present the case. It's like being blindfolded and strapped to a chair and having to decided if something is on fire. You don't have any direct evidence. But you can smell smoke, you can feel the heat on your face and you hear a crackling noise. Now none of those by themselves are proof, but together they add up to a preponderance of evidence, and you (the judge) may (rightfully) conclude that there is indeed something on fire. Either that or someone is going to great extremes to trick you. Lawsuits like these are rarely decided on absolutes. Now, the other side can play the same game, they can show partial evidence to the contrary. They can suggest that the noise you hear is made by the wind playing with a plastic bag, the heat is the sun on your face and the smoke smell is because last week there was a fire but there is none today. It wouldn't take a smart judge very long to ask some follow up questions to figure out who has it right and who is fibbing. When it's a really close call or when both sides 'have a point' the judge may order mediation. More likely the parties will then settle because neither will want the case to be decided on such a slim margin because a ruling 'against' could be much more costly than a settlement.
- nailer 17y agoThinking about it, you're right. My statistics brain shouted 'eeek, correlation != causation' but you make a pretty compelling argument with the fire analogy. Using a programming analogy, you could think of a bayesian spam classifier. The word 'viagra' might not on it's own constitute spam, but 'viagra' 'hard p3n1s' and 'meds' scores enough to be safely considered spam. Consider my mind changed.
- jacquesm 17y agoBayes actually deals with evidence so you're spot on. Here is an interesting wikipedia article on that subject: http://en.wikipedia.org/wiki/Evidence_under_Bayes_theorem http://en.wikipedia.org/wiki/Evidence_under_Bayes_theorem