Rules
Part of MBTI sentinels compared: ISTJ, ISFJ, ESTJ, ESFJ
MBTI sentinels traits: why a retest shifts the code
MBTI sentinels traits sit on four cuts through four tallies, so a code can move on a second sitting. Here is the arithmetic, and a retest you can run.
A trait is supposed to be a stable thing about a person. One property of a four-letter code needs no laboratory, no publisher's data, and no argument about what the letters mean: does the same person get the same code twice?
That question organizes most of what can honestly be said about MBTI sentinels traits: the descriptions attached to ISTJ, ISFJ, ESTJ and ESFJ.
The instrument behind those four codes is the Myers-Briggs Type Indicator, published by The Myers-Briggs Company and taught by the Myers & Briggs Foundation. It is administered in named forms: Form M is the standard one, and Form Q adds facet scores beneath each pair. Official administration does not stop at scoring: the person who took the questionnaire also reads the descriptions and confirms which one fits.
Sentinels, by name
The word Sentinel does not come from the instrument. It comes from 16Personalities, a popular test site that sorts the sixteen codes into four roles — Analysts, Diplomats, Sentinels and Explorers — the Sentinels being the four codes that carry both S and J. David Keirsey's earlier grouping calls the same four Guardians. The official type descriptions published by the Myers & Briggs Foundation cover all four without using either label.
What the four share in the model's own mechanics: introverted sensing (Si) sits in the top two slots of the function order, paired with either extraverted thinking (Te) or extraverted feeling (Fe), and extraverted intuition (Ne) sits in the bottom two. The published descriptions read that way:
- ISTJSi first, Te second. Methodical, thorough and dependable about commitments; checks a new plan against precedent and against what has already worked.
- ISFJSi first, Fe second. Attentive to the practical needs of particular people, loyal, quietly dutiful; keeps track of personal details other people let go.
- ESTJTe first, Si second. Organizes and directs, works from explicit standards and procedures, and expects the same reliability back from others.
- ESFJFe first, Si second. Reads the room and keeps a group's arrangements and routines running; works for harmony and belonging.
All four codes carry S and J, so all four descriptions lean on the same two preferences. That is why they read alike at the top and differ mainly in which function leads.
What to take away
- Stability across two sittings is the one property of a personality result you can test yourself. It is also the property a four-letter code handles worst, for reasons that are arithmetic rather than mysterious.
- Each pair is settled by a cut through a continuous tally, so a pair answered near the middle can flip on a second sitting without anything about the person changing.
- Four pairs multiply. Even a code whose individual pairs are quite stable will change as a whole more often than people expect, which is why the best-fit reading exists.
Why stability is the right first question
Two other questions get asked first and both go nowhere fast. Is the model true? Does it measure something real? Those require agreeing on what "real" means and on what the model is claiming, and the argument never ends.
Stability is different. Its operational form: give the same person the same questionnaire twice, some weeks apart, and compare.
That is what repeatability means in any measurement setting. It is the minimum a measure must clear before the harder question of whether it measures what it claims is worth discussing.
A ruler that gives a different length each time is not slightly wrong. It is not yet a ruler.
The arithmetic of four cuts
Work through a hypothetical, chosen purely to show the shape of the problem.
Suppose each of the four pairs, taken on its own, comes back the same on a second sitting nine times out of ten. That would be respectable for a self-report questionnaire. Now ask how often all four come back the same, which is what has to happen for the code to be unchanged.
| Pairs that must agree | Chance all agree, at 90 percent each |
|---|
Two thirds, as a piece of arithmetic rather than a finding: four agreements at nine in ten each multiply to roughly two in three. A third of people would report a different code the second time, and every one of those changes would come back to them as a change of type. Nothing about the person moved. The code is a conjunction of four separate decisions, and conjunctions are fragile.
This is not a scandal about one instrument. It is what happens whenever continuous measurements are cut into categories and the categories are then combined.
The real studies point the same way. The MBTI Manual by Myers and McCaulley (1985) summarizes test-retest research on the instrument, including correlations of continuous scores from ten studies with retest intervals of four to five weeks. Carskadon (1979) reported retest results for each scale on Form G at seven-week intervals for male and female students. Those figures describe the continuous scores pair by pair. The four-letter code is a further step, because it needs every pair to land on the same side of its cut again.
Where the instability concentrates
Not evenly: it clusters in people whose tallies neared a cut.
Someone who answered almost every item on a pair the same way will report that letter again. Someone who split the items nearly evenly is close to a coin flip on that pair, as the notes on continuous scoring set out.
That has a practical consequence for reading trait descriptions of this group. The four sentinel codes are usually described in terms of S and J: attention to the concrete and a preference for settling things.
A person whose S and J tallies were both marginal is reading a description built from two preferences they barely hold. It will feel about half right. The half that misses is the half the description leans on hardest.
What the model does about it
It has an answer, and the answer is the reason the procedure has a third step. After the questionnaire and the explanation, the respondent reads the descriptions and decides which fits. Where the score and the person disagree, the person wins.
Read as measurement design, this is an admission and a repair: the questionnaire generates a hypothesis, not the measurement itself.
Its instability near the cuts is expected, not fatal, and it puts the result where the model says information lives, inside the person. The group overview explains what that means for anyone trying to type someone else.
Free questionnaires drop this step, which is why a changed result from a website feels like a contradiction rather than a prompt, a gap the notes on the free formats go through.
What can honestly be said about the group's traits
Three things, and they are narrower than a profile page.
A description of preferences, in preference grammar. People reporting these codes tend to describe themselves as attending first to what is concrete and present, and as preferring matters settled rather than open. That is a statement about self-description, and it is well supported because it is nearly a restatement of the questionnaire.
The costs alongside the uses. Published descriptions pair the preference for closure with a difficulty in leaving a question open when more information is coming. Popular versions drop the costs, and what is left reads like a strength rather than a preference.
Nothing about ability, and nothing about anyone else. Preference is not skill, and a preference reported by one person says nothing about a person who has not reported it.
Celebrity profiles assign four-letter codes to public figures the same way, from the outside and with no retest behind them. The arithmetic above is the reason to read those codes as a hypothesis about a person rather than as a result about one.
A test you can run this month
Take any questionnaire you like. Write down the four letters and, if it gives them, the four figures. Put the page away for six weeks. Take the same questionnaire again without looking at the first result.
Retest your own code
- Take any questionnaire
- Write down four letters and figures
- Put the page away six weeks
- Retake without looking
- If a letter moved, read both descriptions
- Decide which fits
If the code is the same, you have learned that your tallies are not near the cuts. If a letter moved, look at which one, then read both descriptions for that pair and decide. Either way you end up doing the step that settles a type, which is more than most people who know their code have ever done.
Common questions
Does a changed result mean my personality changed?
Usually not. It more often means one pair sat close to its cut, and small differences in mood or in item wording moved it across.
Are some pairs more stable than others?
Reported stability differs by pair and by sample: the retest studies summarized in the MBTI Manual and Carskadon's Form G study report a separate result for each of the four scales. The pattern to expect is that the pair you answered least consistently is the pair most likely to move.
If the code is unstable, is the model useless?
No. It means the code is a summary rather than a measurement, and that the description reading is doing the real work. Used that way it holds up.
Should I average two results?
Better to read the descriptions for whichever pair moved and choose. Averaging a category is not a defined operation.
Do the sentinel descriptions depend more on some letters than others?
They lean heavily on sensing and judging. A marginal result on either of those makes the group description a poorer fit than the code suggests.
