5 ms·
The article only attempts to show there's correlation, and ends up by saying "correlation doesn't imply causation, but maybe there's causation". So basically, w
by bvaldivielso 5y ago
The article only attempts to show there's correlation, and ends up by saying "correlation doesn't imply causation, but maybe there's causation". So basically, we don't know.
I'm not blaming the author, because determining causation is hard, and checking there's correlation between two variables can still be interesting, but the article falls short of answering the question.
- wpietri 5y agoCorrelation is correlated with causation, so I think this sort of exploration of a question is still valuable.
- bvaldivielso 5y agoI agree that it's valuable and interesting, just not a definitive answer to the question
- wpietri 5y agoDid I miss where somebody promised you a definitive answer to the question?
- IshKebab 5y agoThey provide decently good evidence of causation, because of the lack of correlation with CPUs. Correlation doesn't prove causation, but it is evidence for it.
- cinntaile 5y agoHow does the lack of correlation with CPUs verify the GPU correlation?
- beebmam 5y agoCorrelation can be shown to be causation when all factors are both known and controlled for. Controlling for CPU prices is one step towards that goal.
- cinntaile 5y agoThis is a circular argument. Why are CPU prices a confounding factor?
- brokensegue 5y agoone can imagine factors that would cause both CPU and GPU prices to increase together. for example general purpose computing demand or supply-side constraints might cause all chip prices to rise. if those non-crypto factors were confounding (say also happened to correlate with crypto-prices maybe due to some third unknown factor) then we would see CPU and GPU prices rise in response to crypto prices. we do not see that.
- squarefoot 5y ago> for example general purpose computing demand or supply-side constraints might cause all chip prices to rise. This. Those who can't afford anymore powerful video cards now fall back to CPUs with decent integrated graphics, so that the higher than normal demand is therefore affecting their prices too.
- cinntaile 5y agoOk thanks, so it's like saying that we can ignore anything to do with manufacturing?
- blululu 5y agoThey ran a pretty solid analysis of the subject. Not only do they look at correlation, but they regress out the trend line and show a correlation in the noise. Correlation of noise typically does imply a causal connection. They also look at correlation between GPUs that are used for mining and ones that are not and show that the correlation only holds for GPUs used in mining, so the case seems pretty tight. If you are just trying to poke holes in the analysis I would focus more on the possibility of there being other contributing factors that are more significant (i.g. supply chain issues). In any case the author was considerate enough to include the dataset and the python notebook so you can run subsequent analyses on your own. Some quick things worth trying: 1. )Weight both the price of Cryptocurrencies and the price of GPUs for Market Size. 2.) Add in other explanatory variables to try to find the relative contributions of different sources.
- bvaldivielso 5y agoI'm not trying to poke holes in the analysis, and as I said I'm not blaming the author for what they did. I'm just repeating what the author said: > We can see a correlation between GPU prices and the value of crypto currencies, but correlation does not mean causation. A causation is plausible in this case. And then they go on to repeat a very reasonable argument, which I happen to believe to be true, but the kind of argument that everyone was already making before looking at any data. The fact that the correlations do not hold as much for CPUs is good, but still not enough to stablish causation. As you said, there might be some confounding factors that affect crypto and GPUs but not CPUs.