Trust: Studies in Definition

Prelude: "Would you say that most of the time people try to be helpful, or that they are mostly just looking out for themselves?"

  • A double-barrel question.
    • These equate (or try to test how equatable) two different question forms are.
  • Questions don't directly measure factors of concern, they correlate with them.
    • Enough of them correlating, and you can decide that "here is a Thing".
  • BUT: What form does that correlation take?

Example: g factor

  • g is the observation that there exists a broad variance across "cognitive ability".
  • Other more specific factors correlate more closely with other tasks.
  • Narrower tasks can be observed, and generate the actual data we look at.

One possible structure of conditional dependence that fits the correlations

digraph name {
        "working memory"[style=filled,color=lightblue];
        "spatial intelligence"[style=filled,color=lightblue];
        "EQ"[style=filled,color=lightblue];
        "g"[style=filled,color=lightblue1];
        g -> "working memory" -> "ability to repeat back large numbers"[label="R>0"];
        g -> "spatial intelligence"[label="R>0"];
        "spatial intelligence" -> "ability to solve map problems"[label="R>0"];
        g -> "EQ"[label="R>0"];
        "EQ" -> "ability to recognize emotions on faces"[label="R>0"];
}

Another possible structure

digraph name {
        "working memory"[style=filled,color=lightblue];
        "Sahiti's visual factor"[style=filled,color=lightblue];
        "working memory" -> "ability to repeat back large numbers"[label="R>0"];
        "working memory" -> "ability to solve map problems"[color=red,label="R<0"];
        "Sahiti's visual factor" -> "ability to solve map problems"[label="R>0"];
        "Sahiti's visual factor" -> "ability to recognize emotions on faces"[label="R>0"];
}

Factor analysis gives us too few answers in one iteration.

So the iterative process of science is to

  • look at the data
  • posit two models
    • or five
  • construct a test which differentiates your models
  • find or collect data which fits the parameters of the test
  • rinse and repeat.

Key insight: All definitions and internal workings of a model are contingent on the model.

  • Eg. "Working memory" is only as useful as the model it is defined within. If a better fit comes along, toss it out and redefine everything.

Yamagishi's work is an exercise in redefinition.

  • recap: "Would you say that most of the time people try to be helpful, or that they are mostly just looking out for themselves?"
    • Are the above two question forms likely to be measuring the same thing?
  • Will they be anticorrelated when asked separately?
  • Will the answer to this question correlate with the questions asked separately?
  • What else would you ask in the same questionnaire to correlate this with?

Trust, gullibility, and social intelligence

Threefold thesis statement:

  • The single factor "trust" should be separated out
    • into
      • The Good Faith principle (belief in the goodness of others)
      • prudence/vigilance
    • Trends across time show that Americans' good-faith has held steady, where their prudence has increased
      • Both high-good-faith and low-good-faith people are highly vigilant
  • Good-faith differs significantly between cultures: specifically, American and Japanese
  • "Gullibility", or the likelihood that one will be fooled by a bad actor, should be constructed in terms of prudence, not faith. (This makes its correlation with faith interesting.)

Diagrammatically:

Old model

digraph name {
        "Trust"[style=filled,color=lightblue]
        "Trust" -> "Gullibility"[label="R>0"]
}

New model

digraph name {
        "Good Faith"[style=filled,color=lightblue]
        "Prudence"[style=filled,color=lightblue]
        "Good Faith" -> "Gullibility"[color=purple,label="R=?"]
        "Prudence" -> "Gullibility"[label="R<0"]
}

Distinguishing test for gullibility: Fool me once

How much do people update on negative information about someone's trustworthiness?

Tiger's cave problem (Kakiuchi and Yamagishi, 1997)

  • Iterated prisonner's dilemma variant: extensional. Additional penalty whenever you cooperate and your partner defects.

Ontology is an interesting problem with iterated factor analysis in anthropometry

  • ontology

Author: Nix build user

Validate