Almost nobody takes a trust test out of curiosity. The search happens after something, and the question underneath it is usually about a specific person rather than about the self.
That is the mismatch worth naming first. A questionnaire can describe how readily one person extends trust. It has no access at all to whether the other person deserves it.
22%
of American adults classify as high trusters
41%
as medium
35%
as low
46%
of adults under 30 in the low group, the highest of any age band
Three different things are called trust
The research literature never settled on one construct, and the three traditions measure things that behave differently.
The oldest is dispositional. Rotter built the Interpersonal Trust Scale in 1967 to measure a general expectancy that the word of others can be relied upon, treated as a stable individual difference like any other trait. This is the tradition most online trust quizzes descend from, whether or not they know it.
The second is relational. Rempel, Holmes and Zanna argued in 1985 that trust in a close relationship is not one quantity but three, developing in sequence. Predictability is confidence that a partner's behaviour is consistent, built from observation. Dependability is confidence in the partner's disposition rather than their pattern, which is what gets tested when circumstances change. Faith is the part that runs ahead of the evidence, an expectation about a future that has not happened yet. Their measure weighted faith most heavily, with ten items against three for predictability, which says something about where they thought the action was.
The third is behavioural. Berg, Dickhaut and McCabe built the investment game in 1995, in which one participant may send any part of a real endowment to an anonymous stranger, the amount is tripled, and the stranger decides what to return. Nothing is asked and nothing is self-reported. What people do with actual money under those conditions became the standard behavioural measure, and it correlates with the questionnaires far less tightly than anyone expected.
Which one a quiz can reach
Only the first, with an approximation of the second if the items are written about a named person.
That limit is not a flaw in any particular test. It follows from the method. A self-report instrument asks someone to summarise their own expectations, and expectations about a specific partner are held by the person answering rather than by the partner being described. What comes back is an accurate report of a forecast, and a forecast is not an observation.
The consequence is straightforward and unwelcome. A test cannot say whether a partner is trustworthy. It can say how much evidence this particular person tends to require before extending trust, which is genuinely useful information for interpreting one's own reaction, and useless as a verdict on anyone else.
The finding that reframes the question
There is a large result that bears directly on where to look instead.
Joel and colleagues assembled 43 longitudinal datasets covering 11,196 couples in 2020 and used machine learning to find which self-reported variables actually predict relationship quality. They split the predictors into two families. Relationship-specific variables describe this particular bond, including perceived partner commitment, appreciation, sexual satisfaction and conflict. Individual-difference variables describe the person, including life satisfaction, negative affect, depression and attachment style.
- Relationship-specific reports: 45%
- Individual differences: 21%
Both families matter and one matters more than twice as much. What a person is like in general explains a meaningful slice. What they perceive about this specific relationship explains most of what is explainable.
The implication for a trust score is direct. A dispositional reading is the smaller of the two columns. The larger column is filled with perceptions about one named person, which means the productive question after a betrayal is almost never how trusting am I but what has this person actually done, and how often, and what happened the last time it was raised.
What a score is actually good for
Three uses, none of which is a verdict.
The first is calibration. Someone who learns they sit at the ninth percentile for propensity to trust now knows that their alarm fires earlier than most people's, which does not mean the alarm is wrong. It means the same behaviour from a partner produces a stronger signal in them than it would in someone else, and that is worth knowing before deciding what the signal means.
The second is locating the breach type. Trust does not fail uniformly. Some people are undone by unreliability and untroubled by secrecy, others are the reverse, and knowing which one applies turns a vague sense of wrongness into a sentence that can be said to another person.
The third is the repair question. What restores trust differs by profile, and someone whose sensitivity is to unpredictability needs a sequence of small kept commitments rather than one large apology. The trust patterns test scores 25 items across propensity, betrayal sensitivity and repair tolerance and reports each as a percentile, which is three numbers rather than one and the three do not move together.
| The question being asked | Can a questionnaire answer it |
|---|---|
| How quickly do I extend trust compared to other people | Yes, that is what it measures |
| Which kind of breach lands hardest on me | Yes, if the instrument scores more than one dimension |
| What would rebuild trust for someone like me | Partly, from the same profile |
| Is my partner trustworthy | No, and no instrument can |
| Did something in my past cause this | No, a score has no history in it |
| Should I stay | No, and the framing of the question is the problem |
Reading a low score without catastrophising it
Low trust is not a disorder and the population numbers make that obvious. Pew's national survey found 35 percent of American adults in the low-truster group, rising to 46 percent among adults under 30, which is a very large minority to describe as damaged.
It is also not free. Low dispositional trust has real costs in how much evidence a relationship has to supply before it can settle, and the person paying most of that cost is usually the one holding the disposition. The entry on trust issues covers where the pattern comes from and what actually shifts it, and the specific case of rebuilding after a lie covers the sequence that tends to work.
Nobody searching this phrase wants a percentile. They want to know whether the thing they are feeling is proportionate. A score cannot settle that, but it can say which half of the question belongs to them, and that turns out to be the smaller half.
Sources
- Rotter, J. B. (1967). A new scale for the measurement of interpersonal trust. Journal of Personality, 35(4).
- Rotter, J. B. (1980). Interpersonal trust, trustworthiness, and gullibility. American Psychologist, 35(1).
- Rempel, J. K., Holmes, J. G., Zanna, M. P. (1985). Trust in close relationships. Journal of Personality and Social Psychology, 49(1).
- Berg, J., Dickhaut, J., McCabe, K. (1995). Trust, reciprocity, and social history. Games and Economic Behavior, 10(1).
- Joel, S., Eastwick, P. W., et al. (2020). Machine learning uncovers the most robust self-report predictors of relationship quality across 43 longitudinal couples studies. Proceedings of the National Academy of Sciences, 117(32).
- Pew Research Center (2019). Americans' trust in other Americans: the state of personal trust.