Tuesday, August 10, 2010

Statistics Lesson For Jane Taber

I don't know if Jane Taber, who sets the agenda at the Globe and Mail, ever took a statistics course in University but I doubt it (if she even attended University). Today a poll was released showing the Conservatives at 34% and the Liberals at 31% with a margin of error of 3.1% to which Jane wrote: "Stephen Harper’s Conservatives and Michael Ignateiff’s Liberals are in a virtual dead heat in a new national opinion poll". I don't think Jane understands probability theory or parameter estimation. When they say 3.1% accuracy 19 times out of 20, they are referring to a 95% confidence interval. That is to say, the Tories are at 34% and Ipsos is 95% certain that the true value (polls estimate the true value) lies between 32.45% and 35.55%. However the parameter estimate is the most probable value in the confidence interval, and the further you get from that 34% towards the outer bands of the interval, the smaller the probability that it is the true value.

Translation: I'm not 95% sure the Tories have a lead, but I am 93% certain that they have a lead. It isn't tied; it is just a rejection of the 95% hypothesis test. Introductory Statistics courses often use the 95% confidence interval, because many tests conducted by science students require a higher level of certainty than opinion polls. In medical research where a microgram can be the difference between life and death in a controlled experiment, you need more scrutiny. Political pollsters don’t require the same level of accuracy as researchers testing a cancer drug.

Also, that 3.1% refers to the entire range of the interval, meaning 1.55% on each side of the parameter estimate; meaning that Ipsos is 95% certain that the lowest possible value of Conservative support is 32.45% and the maximum possible value of Liberal support is 32.55%. That is an overlap of 0.1%. What is the probability that the two parties are tied at 32.5%? The answer, very small, less than 5%.

16 comments:

  1. I thank you for your expertise. My limited understanding of this leads me to conclude that the Conservatives are ahead in the polls - and no amount of slicing and dicing will change the facts on the ground. Poor Jane, always tilting at the windmills. Cheers

    ReplyDelete
  2. o/t
    Remember exStatsCan boss's resignation letter that he made public...
    here's the last couple of sentences:


    '...In closing, I wish the best to my successor. I promise not to comment on how he/she should do the job.
    I do sincerely hope that my successor's professionalism will help run this great organization while defending its reputation.
    Munir A. Sheikh'

    Yah, right!

    Yesterday Shiekh writes an article for the Globe and Mail pleading for a reversal of the decision,
    the G&M is the same paper he said he quit because of their article on page 4,
    today he is on Power & Politics.

    http://www.theglobeandmail.com/news/opinions/we-still-have-time-to-reverse-the-census-decision/article1667194/

    ReplyDelete
  3. A very famous stat.
    "Baseball is 90% mental and the other half is physical" Yogi Berra
    And thats all I got to say about stats.
    billg

    ReplyDelete
  4. I took a course in statistics, failed it in French and finally passed it in English. Only thing that saved me was doing examples over and over again. Also the fact that the male prof starting talking about the fact there are so few Nobel prize winners who are women. So I was damned if I was going to flunk his course!

    Moving right along, I would like to see some statistics on what percentage of media types are liberal and/or leftist.

    ReplyDelete
  5. A public opinion poll such as this one calls 1 in every 30,000 Canadians with a question in order to estimate the total value of all 30 million Canadians. If they estimate a 3% lead with a 3.1% margin of error, then the probability of overlap in the true value (parameter) would be less than 6%. How is it accurate to say they are tied when there is a 94% probability that one party is leading the other?

    A rejection of the 95% hypothesis test (as used in medical research) does not mean that the Conservatives and the Liberals are tied. The Conservatives are still ahead, and of that I am at least 93% certain, if not 94%.

    ReplyDelete
  6. I retired 10 years ago from a responsible position in the medical industry. Whenever I did national polling of medical specialists, depending on the product, I had a statistician review my questions to help me achieve true answers. They needed to be related to the outcome I desired to increase business and to direct my efforts. I also had these same statisticians evaluate the subsequent responses.

    All I am saying is that proper statistical analysis is absolutely necessary to avoid those damned lies we keep hearing about.

    My polling achieved the desired result.

    ReplyDelete
  7. Jane completely ignored todays new poll

    The Canadian Press Harris-Decima survey puts Tory support at 34 per cent, with the Liberals at 28, the NDP at 15, and the Greens at 12.

    http://www.winnipegfreepress.com/breakingnews/tories-rebound-after-mid-summer-slump-hold-six-point-lead-poll--100365904.html

    ReplyDelete
  8. Perfect - you and I understand stats, but all the crap in the media is just that - people who are constantly mis-interpreting stats with NO caveats, which is probably why some of the Stats Can figures and reports should die.

    Ivan Fellegi built his little empire and now it it being shut down and he and his buddies are ticked off, and they are rallying the troops. This is change I can believe in!

    20% of Canadian children living in poverty?? This is insane. This is stats run mad! The people writng reports in Tunney's pasture have never seen a rural town, where you do not need an urban income. They continue to count older children in the basement who are supported by we parents as poor! They report expiration of teacher contracts as job losses.

    It took me a while to get it, but it is true that the power to spout this stuff has to be removed from the rather independent socialist regime of Stats Can., just like Maurice Strong's one world vision of CIDA has to be shut down. No wonder Fellegi (and his mouthpiece) are outraged; but they had no right to construct this oligarchy.. it was just permitted since 1970 by successive government who were not paying attention to what is important.

    ReplyDelete
  9. Hey Iceman,
    This is the type of knowledge that needs to be demonstrated in order to exercise your franchise at the voting booth.
    Well, maybe not a specific knowledge of statistics.
    But, if you had to pass a test in order to qualify yourself as a voter then maybe our society would move in the right direction.

    And I know it opens a can of worms.
    If it was just a few simple questions that would evaluate a voter's awareness of the party platforms and their consequences.

    For instance, 'Jean Chretien has promised to axe the gst if you elect him. Do you realize that by axing it he will either have to incur a deficit or raise other taxes from you to make up for the missing revenue?'

    Keep in mind, as a lifelong conservative, I was relieved that Kim Campbell lost the election. She was a disaster. She sat on the Milgaard file for years as justice minister.

    ReplyDelete
  10. Dance has a point: dumbed-down public "democracy" has to be checked by reason. There was once a rule in Britain and Canada that only property owners could vote. My granny was a Suffragette supporter, so women and the riff raff did get the vote, but who extended the vote to prisoners??? and should thre be some kind of merit test - I would not disagree.

    ReplyDelete
  11. I understand the merits of only allowing informed people to vote, but if you are Canadian and the government controls your life, you should have a voice whether you are a genius or an idiot.

    And L, given that my folks invested $50,000 in my math degree, I should hope that I understand statistics. It just drives me nuts to see a statistically uneducated journalist saying that there is a tie when there is a 93% probability that one party is leading the other.

    ReplyDelete
  12. Perky Tabor owes her "success" to the Liberals who fed her "scoops" when she was elected the first female to lead the Parliamentary Press Gallery. If your listen closely to her analysis on any topic she is as useless as a breast on a nun.

    She and the other "journalists" in Canada still get their cue from the Liberals. Just look at the phony census debate.

    ReplyDelete
  13. Jane will be "toast" when "Fox North" arrives - they will expose all her lies and blatant liberal bias, it will make her look like she looks - washed and worn out!

    ReplyDelete
  14. Not only does she get her cue from the liberals but from her coalition buddies(ndp-bloc) as well who are marxist and socialist that are using the liberal brand to get what they want and she better smile to them or else-what do marxist do to their people if they don't follow orders?

    ReplyDelete