Question on margin of error

Discussion in 'Politics, Religion, Social Issues' started by Gutwrench, Apr 27, 2016.

  1. Gutwrench Contributor


    Jan 2, 2011
    Long ago I was in a statics class but I was preoccupied and the truth is didn't pay much attention. The only thing I came away from the class is that a poll with a margin of error much > 4-4.5 is considered unreliable.

    can you give me a crash course on poll and margin of error, please
    Thank you in advance
  2. Meister Suspended


    Oct 10, 2013
    Nevermind the margin of error.

    The election polls have proven to be completely false in various elections across europe lately.

    I've worked in statistics before and I can tell you that they are educated guesswork.
  3. Scepticalscribe, Apr 28, 2016
    Last edited: Apr 28, 2016

    Scepticalscribe Contributor


    Jul 29, 2008
    The Far Horizon
    These days, with the increasing importance and influence of social media, a growing tendency with opinion polls, and many surveys, is that a good many of those who have responded to such surveys, or opinion polls, have 'opted-in'. In other words, these days, many of the respondents are self selected, and have chosen to respond to the survey or opinion poll. (This model became very popular as a result of - say - TV shows which sought responses to shows, or quizzes, using premium lines to generate shared profit).

    In practice, this has come to mean that sometimes, the 'sample' responding to an opinion poll may not be representative of anything other than an especially motivated element of the population.

    In previous times, those conducting polls used to contact people, and, if the individuals contacted met the criteria for assembling some sort of representative sample, they could choose whether to respond or not to the survey. Therefore, in practice, the 'older' model of gathering responses for surveys, or opinion polls, was based on an 'opt-out' model, whereby the company selected to conduct the survey was in charge of choosing those to be contacted, and they, in turn, could choose whether or not to participate by responding.

    These issues were discussed in the UK when it became clear that the polls (with one striking exception) had called the (result of the) General Election of last year spectacularly wrongly.

    Mind you, people lie, too, in their responses to surveys or opinion polls. What they say they would like to see happen is not always what they actually end up voting for.
  4. Meister Suspended


    Oct 10, 2013
    All this is true provided they actually bother asking people in the first place ...
  5. aaronvan Suspended


    Dec 21, 2011
    República Cascadia
    Major polls in the United States are still conducted exclusively by land-line which skews the results immensely.
  6. rdowns macrumors Penryn


    Jul 11, 2003
    Pew Research Center will increase the percentage of respondents interviewed on cellphones from 65% to 75% in most of its 2016 telephone surveys. We’re making this change to ensure our survey samples properly represent the now roughly half (47%) of U.S. adults whose only phone is a cellphone.

    Nine-in-ten Americans have a cellphone, and the share of adults who are cellphone-only has steadily increased since 2004, the year the government began tracking the size of this group. To keep pace with the public’s changing habits and lifestyle, we have increased the percentage of respondents interviewed by cellphone nearly every year since 2009.
  7. juanm macrumors 65816


    May 1, 2006
    Fury 161
    Finally something we agree on!
  8. thermodynamic Suspended


    May 3, 2009
    If that's a crime, then we're all guilty at one point and/or another. :D
  9. Robisan macrumors 6502

    Jan 19, 2014
    Here's the nut graph from that link, re-written slightly to make it generic. The example is a poll of 497 people that says candidate X has 25% support with a MOE of 5%:

    A 5 percent MOE in the poll means we can be 95 percent confident that candidate X has somewhere between 20 and 30 percent support among the surveyed population* (25 plus or minus 5 percent). In other words, if we were to conduct this survey many times with different samples of 497 randomly chosen people, 95 out of 100 times the proportion of the survey respondents supporting candidate X, plus or minus the MOE, would contain the actual percentage of support in the surveyed population. We call the range of 20 to 30 percent support the 95 percent confidence interval for this poll.

    (* "surveyed population" simply means the group of people you're polling. E.g. you could be polling "all likely voters" or just "all registered republicans" etc.)​

Share This Page