You have a questionnaire — newly written, translated from English, or borrowed from another field — and you need evidence that it measures what it claims to measure, in your language and your population. This is the most common reason people come to PEZ, and the area where five instruments now carry my name.

• Item generation and expert review; pilot testing and item analysis
• Exploratory and confirmatory factor analysis; competing factor structures compared
• Reliability across the appropriate coefficients, not just Cronbach’s alpha out of habit
• Convergent, divergent and criterion validity against established measures
• Measurement invariance across gender, age, education level or country before any group is compared
• Translation and cross-cultural adaptation, including back-translation protocols
• Norms and standardisation for Croatian populations

How many, among whom, and how sure can we be? Prevalence work is where a sampling mistake is invisible in the output and fatal in the conclusion. I design the study so the estimate is defensible before anyone starts collecting.

• Sampling frames, stratification and realistic power calculations
• Prevalence and incidence estimates with confidence intervals, weighted where the sample requires it
• Subgroup breakdowns by age, gender, county, school type or sector
• Trend comparison against earlier waves and against national or EU benchmarks
• Public-facing reports for institutions, county authorities and ministries

Knowing that X predicts Y is rarely the interesting part. The question is what carries the effect and for whom it holds — the mechanism and the boundary condition. These models are easy to run and easy to over-claim from; the value is in specifying them honestly.

• Simple, serial and parallel mediation with bootstrapped indirect effects
• Moderation and moderated mediation; conditional process models
• Johnson–Neyman regions of significance rather than arbitrary median splits
• Latent-variable mediation in SEM when measurement error would otherwise bias the estimate
• Honest treatment of causal language: what a cross-sectional mediation model can and cannot support

Which variables actually carry weight once the others are accounted for, how much variance they explain, and whether the same predictors work in different groups. This is the backbone of most applied research questions in education, health and organisational settings.

• Multiple, hierarchical and logistic regression with full assumption diagnostics
• Multilevel models for pupils within classes, staff within institutions, repeated measures within people
• Multivariate multilevel modelling — the subject of my Croatian textbook on the method
• Discriminant analysis and classification; risk-group identification
• Model comparison, cross-validation and honest reporting of overfitting risk

Technical controls fail at the person using them. Since 2015 I have worked with information-security researchers and engineers to measure the behavioural side: what users actually do with passwords, devices and links, how aware they believe they are, and how far apart those two things sit.

• Security-awareness and risky-behaviour measurement using validated instruments — UISAQ and BCISQ, both developed here
• Studies comparing self-assessed risk against measured real behaviour
• Sector-specific assessment: healthcare staff, university and school populations, ICT system users, employees
• Cross-country comparison and translation of security instruments
• Statistical support for engineering and informatics teams publishing in IEEE, MIPRO and applied security journals
• Interpretation for management: which awareness gaps predict which behaviours, and where training changes outcomes

The work that does not fit a standard label: several outcomes moving together, nested and longitudinal designs, latent structures, or a reviewer asking for a method the team has never run.

• Structural equation modelling and path analysis
• MANOVA, MANCOVA and repeated-measures designs
• Cluster analysis and latent profile approaches for typologies
• Longitudinal and cross-lagged models
• Second opinions on an analysis already run, and responses to reviewers’ statistical objections