Holland codes: six types, three letters, and what the letters do not say
John Holland's idea was simple enough to survive seventy years of use. People cluster into six interest types, work environments cluster the same way, and the closer the match between the two, the longer you tend to stay and the better you tend to do. He published the first version of the theory in 1959 and laid it out in full in Making Vocational Choices, whose third edition came out in 1997.
The six types sit on a hexagon. Neighbours share more in common than opposites, which is why a Realistic-Investigative pairing shows up in engineering work and a Realistic-Social pairing rarely does.
The six types
| Type | Pulled toward |
|---|---|
| Realistic (R) | Tools, machines, animals, the outdoors. Work with a visible result. |
| Investigative (I) | Observing, analysing, figuring out how things work. Science and maths. |
| Artistic (A) | Writing, music, design, performance. Loose structures, original output. |
| Social (S) | Teaching, helping, training, treating. People rather than objects. |
| Enterprising (E) | Leading, persuading, selling. Targets, influence, decisions. |
| Conventional (C) | Records, data, procedures. Clear rules and tidy systems. |
How a code gets built
This version has 42 items, seven per type, each rated 0 to 4, so a single type can reach 28. Rank the six totals and the top letters become your code. That is the standard logic across every RIASEC instrument: add up each type, sort, read off the top two or three.
The public tools differ in size rather than in method. The US Department of Labor's O*NET Interest Profiler uses 180 items, 30 per type, with a like / dislike / not sure response and the items mixed rather than grouped, which cuts down on response-style bias. The IIP RIASEC Markers from Liao, Armstrong and Rounds (2008) use 48 items on a five-point scale and are in the public domain.
What the psychometrics look like
The O*NET technical report from 1999 tested the instrument on 1,061 participants. Internal consistency for the six types ran from .93 to .96. In the retest sample the alphas held at .93 to .97 on the first occasion and .91 to .97 on the second. Test-retest correlations came in between .81 and .92, which beat the comparison instrument's .79 to .88.
Convergent validity looked strong: the corresponding scales of O*NET and Interest-Finder correlated highest, with the weakest cross-correlation landing on Enterprising at r = .73.
The structural picture is messier. Fit to the hexagonal circle model gave a Correspondence Index of 0.40 for ONET against 0.60 for Interest-Finder. A principal components analysis found a general factor explaining 43.6% of variance, with Interest-Finder at 49.47%. Prediger's benchmark, from 1982, puts the typical general factor around 41%, so ONET sits a little above it while still below its competitor. Three known wrinkles: Enterprising correlates too strongly with Artistic for a pair that should be separated by one step, correlates too weakly with Social for an adjacent pair, and Realistic and Conventional correlate less than the model predicts.
Does the match actually predict anything
Yes, modestly. Nye and colleagues ran a meta-analysis in 2017 covering 92 studies and tens of thousands of participants: people whose interests line up with their work environment show better performance and stay longer. Bullock-Yowell and Reardon's 2024 open textbook on the hexagon is a decent free overview if you want to go deeper. Modest is the operative word. Interest congruence is one input among several, and it does not carry the weight that the enthusiastic career-advice version of Holland's theory implies.
What a code cannot tell you
Interests are not abilities. Someone with a strong Investigative score may or may not be good at research, and the scale has no way to know. There are no cutoffs anywhere in the instrument, so a low Conventional score is not a deficit. Scores also drift: what you have been exposed to shapes what you report liking, which is why a code taken at 18 and the same code taken at 40 can look quite different.
Treat the three letters as a starting point for exploration rather than a verdict. The hexagon fits the data imperfectly, and the whole model is a simplification of something genuinely complicated. This test is not a diagnosis. It describes interests, and it does not replace a conversation with someone who knows your situation.