Your download will open automatically. If it doesn't, use the Download PDF button.

Career Compass Pro · Institute of Industrial & Organisational Psychology

Technical Manual

Scoring approach, adaptive logic, validity evidence, and key formulas underpinning the four-tier Career Compass Pro assessment suite (Primary, Secondary, Higher Education, Professional).

Edition 2026.1 · Revision date: July 29, 2026

Prepared for chartered psychologists, procurement, and research partners.

Section 1

Constructs & framework

Career Compass Pro operationalises nine peer-reviewed vocational psychology frameworks into a single adaptive engine. Each tier draws from the subset appropriate to its developmental stage.

Framework	Primary construct	Tiers used
Holland RIASEC	Vocational interests (6 types)	All
Super's Life-Career Rainbow	Role salience & life stage	Sec, HE, Prof
Schein Career Anchors	Motivational anchors (8)	HE, Prof
Savickas CAAS	Career adaptability (concern, control, curiosity, confidence)	Sec, HE, Prof
Krumboltz Happenstance	Planned happenstance skills	HE, Prof
Gottfredson Circumscription	Career aspiration boundaries	Primary, Sec
Lent SCCT	Self-efficacy & outcome expectations	All
Big Five (IPIP-NEO short)	Personality facets	HE, Prof
Grit & Growth Mindset	Persistence & mindset	Sec, HE, Prof

Section 2

Scoring approach

2.1 Raw score computation

Each item i contributes to one or more constructs c with a signed loading w_i,c. The raw construct score is:

R_c = Σ_{i ∈ I_c} ( w_i,c · r_i )

where r_i is the participant's response on a 1–5 Likert scale (reverse-keyed items rescaled as 6 − r_i), and I_c is the item set for construct c.

2.2 IRT calibration (2-PL / GRM)

For polytomous items we fit Samejima's Graded Response Model. The probability of response k on item i for a person with latent trait θ is:

P(X_i ≥ k | θ) = 1 / (1 + exp(−a_i (θ − b_i,k)))

Item discrimination a_i is retained where a_i ≥ 0.70; threshold parameters b_i,k are ordered per category. Person parameters θ̂ are estimated by expected a posteriori (EAP) estimation with a N(0, 1) prior.

2.3 Norming to T-scores

Norm groups are stratified by tier × country × age band × gender. Standardised T-scores are:

T_c = 50 + 10 · (θ̂_c − μ_norm) / σ_norm

Percentiles are derived from the empirical norm CDF: P_c = Φ((T_c − 50) / 10) × 100.

2.4 Career Readiness Composite (CRC)

CRC = 0.30·T_adaptability + 0.25·T_{self-efficacy} + 0.20·T_clarity + 0.15·T_exploration + 0.10·T_grit

Weights derived from a multiple regression of the composite onto 12-month career outcome criteria (n = 4,182; R² = .47, p < .001).

2.5 Interest–Career fit (cluster matching)

Career cluster fit F_j for occupation j is computed as the cosine similarity between the participant's RIASEC vector r and the cluster's O*NET-derived profile v_j:

F_j = (r · v_j) / (‖r‖ · ‖v_j‖) · 100

Section 3

Adaptive logic

3.1 Reading & format adaptation

Tier	Reading level (Flesch-Kincaid)	Item format	Coach persona
Primary (8–11)	Grade 3–4	Icon pairs, 3-point smiley scale	Encouraging mentor
Secondary (12–17)	Grade 6–7	5-point Likert, situational vignettes	Peer-like guide
Higher Ed (18–24)	Grade 9–10	5-point Likert, forced-choice pairs	Career strategist
Professional (25+)	Grade 11–12	Likert + reflective prompts	Executive coach

3.2 Item selection (CAT rules)

Within each construct, the next item i* is chosen by maximum Fisher information at the current θ̂ estimate, subject to content-balancing constraints:

i* = argmax_{i ∈ P} I_i(θ̂) = argmax_i a_i² · P_i(θ̂) · (1 − P_i(θ̂))

Termination: SE(θ̂) < 0.30 or 90% of construct blueprint delivered.
Exposure control: Sympson–Hetter randomesque with r-max = 0.20.
Content balancing: enforce minimum items per RIASEC / CAAS / anchor facet.

3.3 Branching & chapter flow

The engine advances participants through developmental chapters. Transition rules use construct-level completion c_done and confidence SE:

advance ⇔ (c_done ≥ 0.90) ∧ (SE_c ≤ 0.35) ∀ c ∈ chapter

3.4 Coach guardrails

Deterministic recommendation core: career, subject and university matches are computed by the scoring engine before any LLM output.
Score-tethered narratives: the LLM receives normed scores plus a curated interpretive context pack; it cannot introduce trait claims that are not in the data.
Non-deterministic language enforced ("your profile suggests…" not "you are…").
Low-confidence / safeguarding-flagged sessions routed to a chartered psychologist.

Section 4

Validity & reliability evidence

4.1 Content validity

Twelve subject-matter experts rated each item on relevance and representativeness. Item-level Content Validity Index (I-CVI) was retained where I-CVI ≥ 0.83; scale-level S-CVI/Ave = 0.91 across the item pool.

4.2 Internal structure (CFA)

CFI ≥ .94 · TLI ≥ .93 · RMSEA ≤ .06 · SRMR ≤ .07

Configural, metric and scalar invariance established across country and gender groups (ΔCFI ≤ .010; ΔRMSEA ≤ .015).

4.3 Reliability

Scale	Cronbach α	McDonald ω	Test–retest (4 wk)
RIASEC (6 scales)	.82 – .89	.84 – .90	.78 – .86
CAAS-24	.88	.90	.83
Career Anchors	.81 – .87	.83 – .88	.79 – .84
Self-efficacy (SCCT)	.91	.92	.85
Grit-S	.83	.85	.80

4.4 Criterion & predictive validity

Concurrent: RIASEC congruence with declared major r = .58 (HE tier, n = 2,140).
Predictive: CRC → 12-month career action completion β = .42, ΔR² = .18 (n = 4,182).
Incremental: CRC over cognitive ability alone ΔR² = .11 for training performance.

4.5 Fairness & DIF

Every item is screened for Differential Item Functioning using Mantel–Haenszel and logistic regression procedures. Items flagged at ETS category B or C are revised or retired before release.

MH χ² = [(Σ (A_k − E(A_k)))² ] / Σ Var(A_k)

Section 5

Norming census

Region	Primary	Secondary	Higher Ed	Professional
West Africa (NG, GH)	1,240	3,110	3,820	2,050
United Kingdom & EU	640	1,880	2,410	1,720
North America	510	1,470	2,090	1,540
Rest of world	380	1,020	1,660	2,260
Total (n = 27,800)	2,770	7,480	9,980	7,570

Section 6

Key references

Holland, J. L. (1997). Making vocational choices (3rd ed.). PAR.
Savickas, M. L., & Porfeli, E. J. (2012). Career Adapt-Abilities Scale. Journal of Vocational Behavior, 80(3), 661–673.
Schein, E. H. (1996). Career anchors revisited. Academy of Management Executive, 10(4), 80–88.
Lent, R. W., Brown, S. D., & Hackett, G. (1994). Social cognitive theory of career development. Journal of Vocational Behavior, 45, 79–122.
Samejima, F. (1969). Estimation of latent ability using a response pattern of graded scores. Psychometrika Monograph, 17.
American Educational Research Association, American Psychological Association, & National Council on Measurement in Education. (2014). Standards for Educational and Psychological Testing.
International Test Commission. (2018). ITC Guidelines for the Use of Tests.

Confidentiality

This manual is intended for chartered psychologists, procurement teams and research partners under NDA. Item parameters and DIF tables are available on request.