Comment Re:Do not want (Score 1) 579
That's an incredible treatment effect, I will need to see some evidence!
You just simply can't compare raw event numbers when estimating relative risk. Your statement about "twice as deadly" is very likely not true, and certainly not justified from the data you reference. You fail to take into account any sort of denominator when just using the raw events. What if only 27 kids rode school busses each year? What if 2 million did? What if only 43 kids were exposed to H1N1, and they all died? What if everyone was exposed to H1N1, and 43 died? You need to take into account the population, not just events. After all, every(?) child who died last year used toothpaste.
..., and vice-versa.
Dude, go back school.
I think I am going to have a t-shirt made of that...
There has never been a *randomized* control clinical trial showing smoking causes cancer in humans. If you believe smoking causes cancer in humans, then you are somehow making an exception of your own rule. So, do you?
Your points are valid for managing data with SAS. But that's not what a lot of statistics is about of course. Try coding a cubic spline regression algorithm in SAS vs. R and see which you like better. I know that doesn't take away your point about large data. But what happens is that SAS is still taught in many graduate stats departments to manage small (dozens or hundreds of cases) datasets and run regressions on them. This is where using SAS seems pointless to me. Even clinical trial data only number in the thousands for both subjects and variables. This is not "large scale", and free R is perfectly capable on data like these.
I will beg to differ on your "not so bad" conclusion on the SAS language design. R's object-oriented functional model is far superior for designing statistical functions, packages, and systems, including graphical functions. I don't think anyone will deny that. R's evaluation model is based in Scheme semantics, and I think if you're coming to stats from a CompSci background, as many are these days, you're not going to like anything SAS has to offer.
While I agree with your points, the fact is that SAS has such a stranglehold on some industries, specifically the pharma industry, that they haven't had to improve their product much in recent years. I mean, I think in the last few years, the one major feature that their survival models package (proc phreg) got was the ability to include categorical variables with more than 2 categories (i.e., a class statement).
R, which is a GNU project, has taken over completely when it comes to new statistical methods being implemented, and has also taken over everything in graphical research and methods. I think it is only a matter of time before it is the standard, but it will take awhile since there is a lot of money invested in legacy SAS macros and programmers. But they certainly aren't teaching too much SAS at universities these days, it won't be long before students come out knowing a lot more R. R is definitely the future of statistical computing, and SAS is the past. They have recently been trying to concentrate on being a business platform, like an SAP competitor, more than statistical software. I suppose that's a smart move on their part, we'll see how it plays out.
Completely agree with the comments about ugly syntax though, ugh, I would not wish it on anyone.
"I've seen it. It's rubbish." -- Marvin the Paranoid Android