Friday, July 16, 2010

Quote of the Summer

Paraphrase actually:

"Mr. Clement said the medical journal and other critics should trust Statistics Canada."

Also, excellent article from Gardner this morning. This, in particular, made me spit out my Cheerios:

"Then the Earth shook. The change to the census will produce "seriously biased" data, the legendary statistician Ivan Fellegi told this newspaper. It is "indefensible." Coming from a man who spent half a century at Statistics Canada, including 23 years as Chief Statistician, this was rather like Moses returning from the mountain and explaining to the wayward Israelites that, no, you can't worship a golden calf, you idiots."

8 comments:

  1. There seem to be a lot of Liberals in support of the Conservatives on this one: Kinsella, Smart is the New Black, and others. And Statistics Canada themselves are saying this will work. Seems reasonable enough, so why all the fuss?

    ReplyDelete
  2. Well Jaysus, if Warren Kinsella and some guy I've never heard of support the move...

    ReplyDelete
  3. The guy you never heard of, is that the people at Statistics Canada buy any chance ... seems rather reliable to me.

    ReplyDelete
  4. The guy you never heard of, is that the people at Statistics Canada buy any chance ... seems rather reliable to me.

    Link please.

    While you're at it, feel free to prove that 2+2=5. The simple fact of the matter is that it won't work. This isn't up for debate, it's the most basic of sample rules. Nothing will be as reliable as a mandatory survey. To say otherwise is just outing yourself as an idiot.

    ReplyDelete
  5. I've studied statistics and I have never saw anything that said a mandatory survey is more reliable than a voluntary survey. If you know otherwise please provide the link. Indeed, what I did read over and over again was that random surveys were the most reliable. But that's the funny part, because random surveys are voluntary, lol.

    ReplyDelete
  6. But that's the funny part, because random surveys are voluntary, lol.


    Oy vey!

    ReplyDelete
  7. From Gardner's article:

    Del Mastro defended the government's decision. Ivan Fellegi is wrong, Del Mastro said. And he should know. Why, he has taken "a number" of courses on statistics.


    I would like to share a little anecdote: I was taking a graduate level biostatistics course in which my professor pointed out with the great exasperation that a lot of what passes for statistics is garbage today. The reason why is that very rarely do statistics courses delve into the epistemological aspects of statistical modeling and sampling. So, people will proudly set up supposed flawless sampling regimes only to discover that their methodologies are woefully incongruous to the actual question that is being asked . That is the problem with general mathematical illiteracy: 'Random surveys' that are reported are perceived to have higher information value than in reality, especially when some survey samples are non probability.

    Oh, and no, random surveys are not necessarily voluntary. In fact, probability sampling can still yield a bias due to non response. That is why EKOS and other pollsters try very hard to incentivize the people they contact because the more nonresponse bias you get, the more your observed value deviates from the population parameter. There is no debate on this. This is a intrinsic fact in statistical analysis. If the goal is to make population level assessments with representative samples, then mandatory response is, well, mandatory.

    ReplyDelete
  8. That is why EKOS and other pollsters try very hard to incentivize the people they contact because the more nonresponse bias you get, the more your observed value deviates from the population parameter.

    ... and the higher the weight they have to put on created from *drum roll* census numbers.

    ReplyDelete