Not quite right

March 27th, 2014 at 12:00 pm by David Farrar

writes in The Letter:

Maybe the Fairfax media is right that will do exceptionally well. In the Herald poll has gone from zero to .8%. As a percentage increase that is an infinite increase. Projected forward at that rate of increase could govern alone. That statement is no sillier than the commentary the Herald has run on its poll. We are not trumpeting ’s spectacular rise because the margin for error in the poll is 3.5%. so might already be on 3%.

That isn’t right. It is a common mistake.

The margin of error normally quoted in a poll is the maximum margin for a result of 50%. It is far less for smaller results such as 0.8%. In fact a 0.8% result for a poll of 1,000 has a margin of error of 0.6% so the 95% confidence range is 0.2% to 1.4%.

Evan at a 99.999999% confidence interval the margin of error for 0.8% is only 1.7%. There is no way at all ACT can be at 3%, just on sampling variation.

Tags: , ,

12 Responses to “Not quite right”

  1. brucehoult (195 comments) says:

    While of course you are correct in essence, and I hate to correct a professional (in both senses) pollster, you are not quite right in detail. The distribution of margin of error at these limits is not symmetrical — as should be obvious to anyone, as it’s not possible for the true population figure to be -0.9%.

    Vote: Thumb up 1 Thumb down 0 You need to be logged in to vote
  2. brucehoult (195 comments) says:

    There is in fact one chance in 837878 of getting 0.8% from a true population of 3% (with 1000 trials), this is 99.9998%, which is considerably more likely than 99.999999%

    The actual 99.999999% confidence interval goes to 3.696%.

    The 95% confidence range goes up to about 1.441%, and 99% confidence range to 1.759%.

    But yes, it’s vanishingly unlikely that the true support is currently 3%.

    Vote: Thumb up 1 Thumb down 0 You need to be logged in to vote
  3. Ed Snack (1,883 comments) says:

    That also assumes that the sampling is in fact accurate and not a biased sample. Does ACT represent a constituency less likely to be surveyed though ?

    Vote: Thumb up 0 Thumb down 0 You need to be logged in to vote
  4. brucehoult (195 comments) says:

    Ed, yes that’s very possible. Is polling still only done to people with landlines? I haven’t had one of those in a decade.

    Vote: Thumb up 1 Thumb down 0 You need to be logged in to vote
  5. mikenmild (11,247 comments) says:

    Ah the old landline question. Typically trotted out years ago by ACT supporters to claim they were under-represented in the polls because they were too busy out and about doing things to be waiting at home for the phone to ring.

    BTW hold the front page – Richard Prebble was wrong about something!

    Vote: Thumb up 0 Thumb down 2 You need to be logged in to vote
  6. berend (1,711 comments) says:

    I think DPF/mikenmild read something in Richard’s sentence he doesn’t say: he is making an assertion that it would be stupid to make silly claims. An example of a silly claim being that ACT could be at 3% because of a margin of error of 3.5%.

    He doesn’t say ACT could be at 3%, and he doesn’t say the appropriate margin of error is 3.5%. He is making a statement about silly claims.

    Vote: Thumb up 1 Thumb down 0 You need to be logged in to vote
  7. Alan (1,087 comments) says:

    What more interesting is changes in VI based on previous voting record but for some reason this is rarely published.

    Vote: Thumb up 0 Thumb down 0 You need to be logged in to vote
  8. Flyingkiwi9 (54 comments) says:

    I understand what DPF is saying but as someone else points out; he’s taking the piss out of the polls

    Vote: Thumb up 1 Thumb down 0 You need to be logged in to vote
  9. Jim Rose (35 comments) says:

    IPredict is much better for the small parties. ACT got 1.3% in 2011. This time around that second seat at 1.4% seems assured.

    IPredict has ACT as a 70% chance in Epsom.

    the conservative party are less of a safe bet now.

    Vote: Thumb up 3 Thumb down 0 You need to be logged in to vote
  10. griffith (1,111 comments) says:

    Will the rise sustained over time?

    Is An act party rise reflected in the range of polls available ?

    Ipredict may be skewed by its self-selection of participants.

    What is ACT latest incarnation?

    How will the party now interpret its principles.

    Vote: Thumb up 0 Thumb down 1 You need to be logged in to vote
  11. alex (304 comments) says:

    I think its more likely ACT are still at effectively 0, and as such will only be returning with whoever the stuffed shirt in Epsom is.

    Vote: Thumb up 1 Thumb down 1 You need to be logged in to vote
  12. Anthony (796 comments) says:

    DPF points out the same sort mistake that stupid reporters quite often make when talking about Acts polling – they say Act is polling at less than the margin of error, implying support for Act may be actually zero. This is quite a stupid course of course for the reasons DPF mentions that a party polling at 0.8% can’t have a margin of error of 3.5%! Any anyway, the fact that a few people said they would vote for Act makes it extremely unlikely their actual support is zero!

    As usual, I look forward to making money on iPredict when Act does get back into parliament. Funny, how I’ve never head any of the smart arsed reporters and left wingers admit they were wrong when Act got back in previously after them saying Act wouldn’t be back! Act has had more final nails in the coffin than I’ve had hot dinners!

    Vote: Thumb up 0 Thumb down 0 You need to be logged in to vote