Filtered dossier

statistics

  • 7 / 7
  • No visual on file
    Statistical Illusions Entry #0446

    Why a lagging statistic invites the wrong correction

    The number describes a situation that has already moved. Acting hard on it applies force in the direction the system has just finished going.

    Novice
    The hidden part #0446

    When the measurement lag matches the system’s response time, a proportional correction becomes an oscillation. Each step is defensible; the sequence is not.

    Open file
  • No visual on file
    Statistical Illusions Entry #0445

    The reason a self-selected panel overstates strong opinions

    Answering costs time, and the people who pay it are those with something to say. Indifference is the hardest state to get into a dataset.

    Master
    The hidden part #0445

    Willingness to answer is itself an attitude, and weighting cannot repair it. Indifference is the state a voluntary survey cannot record.

    Open file
  • No visual on file
    Statistical Illusions Entry #0444

    Why the sample frame decides the result

    Sampling errors shrink with size; frame errors do not. If the list you draw from excludes a group, no amount of sampling from it will find them.

    Adept
    The hidden part #0444

    Size shrinks sampling error and leaves coverage error untouched. A precise estimate of a frame is not an estimate of a population.

    Open file
  • No visual on file
    Statistical Illusions Entry #0443

    The reason a leading indicator stops leading

    It led because of a structure that produced the delay. When the structure changes, the correlation stays in the historical data and leaves the future.

    Novice
    The hidden part #0443

    An indicator leads because of a structure, not because of its history. Fit is measured with data nobody had at the time.

    Open file
  • No visual on file
    Statistical Illusions Entry #0442

    Why a cluster is expected in random data

    Randomness produces clumps. A pattern spread out evenly is the unlikely arrangement, so finding a cluster is not yet finding a cause.

    Adept
    The hidden part #0442

    Independent events arrive in bursts, so a cluster is the expected texture rather than a signal. Evenness is what would need explaining.

    Open file
  • No visual on file
    Statistical Illusions Entry #1127

    Why the winner’s curse inflates first results

    Winner's curse makes the first significant result look stronger than it really is, because extreme estimates are the ones that win.

    Master
    The hidden part #1127

    When many noisy estimates compete, the winner is often the most exaggerated one.

    Open file
  • No visual on file
    Statistical Illusions Entry #1126

    Why Berkson bias bends the sample

    Berkson bias makes two independent traits look related because the sample was filtered by a shared gate, not by reality.

    Adept
    The hidden part #1126

    A filter can create a relationship that the population never had.

    Open file