Welcome to my professional website! As a Professor of Statistics at Qatar University with over 25 years of experience across international institutions, I’ve created this space to share my academic journey and resources with students, colleagues, and fellow researchers.
My expertise spans statistical theory and practical applications using modern computational tools. Whether you’re interested in graphical models, Bayesian methods, public health analytics, biostatistics for hematology and oncology, chronic pain modeling, sensory data analysis, or AI regulation, you’ll find relevant content throughout this site.
Feel free to contact me with questions, collaboration ideas, or feedback on any of the resources shared here.
ORCID ID
Dr. Dhafer Malouche is a Professor of Statistics at Qatar University, with more than 25 years of academic experience across Europe, North Africa, the Middle East, and North America. He earned his Ph.D. in Statistics and Applied Mathematics from Paul Sabatier University and has held faculty positions in Tunisia, Egypt, and Qatar.
His international experience includes appointments as a Fulbright Scholar at Stanford University (2011) and the University of Michigan (2016–2017), as well as long-term research collaborations with Yale University through the MacMillan Center (2014–2019). His work during these periods addressed survey methodology, data quality, governance, and applied statistical modeling in social and health sciences.
Dr. Malouche’s research spans a broad range of statistical and interdisciplinary domains. His methodological work centres on Bayesian networks and graphical models, Bayesian hierarchical modeling for evidence synthesis when only a few studies are available, epidemiological modeling, and reproducible data science workflows in R and Stan. As the statistician on interdisciplinary teams, he works extensively with Qatar Biobank data, leading and co-authoring studies on chronic pain, cardiometabolic risk, autoimmune diseases, hematological abnormalities, and gender disparities in quality of life. His clinical collaborations, in which he designs the analyses and leads the statistical modeling, extend to oncology (pancreatic cancer, T-cell leukemia, diffuse large B-cell lymphoma), neurological disorders (multiple sclerosis), and systematic reviews and meta-analyses of clinical outcomes. Beyond biomedical research, he collaborates on engineering education and educational technology, examining the impact of virtual reality and blended learning on student engagement. He is also engaged in interdisciplinary work at the intersection of statistics, AI regulation, anticipatory law, and finance, applying quantitative foresight methods to emerging legal and economic challenges.
In teaching, Dr. Malouche delivers advanced courses at both undergraduate and doctoral levels, including stochastic processes, actuarial statistics, and advanced biostatistics for PhD students in the biological sciences. He makes extensive use of R, Bayesian methods, and real-world health data, and regularly develops reproducible teaching materials, Shiny applications, and LaTeX-based resources to support applied statistical training.
Beyond academia, Dr. Malouche has served as a statistical consultant for international organizations, including the World Health Organization, contributing to analyses of COVID-19 impacts on mental health and health systems. At Qatar University, he continues to lead interdisciplinary research, supervise graduate students, and contribute to methodological innovation at the intersection of statistics, public health, and data science.
Download my resumé.
Habilitation (Tenure), Problèmes autour de la probabilité et de la statistique, Méthodes et Applications., September 2009
Université de Tunis ElManar Ecole National d'Ingénieurs de Tunis, Tunisia.
Ph.D. in Statistics and Probability, October 1997
Paul Sabatier University, Toulouse, France
Master's Degree (D.E.A.) in Applied Mathematics and Statistics, 1993 - 1994
Paul Sabatier University, Toulouse, France
Bachelor (Maîtrise) in Mathematics, 1989 – 1993
Ecole Normale Supérieure de Bizerte, Tunisia
Authors: Aisha Al-Khinji, Dhafer Malouche
Status: Posted
Abstract: Mesenchymal stem/stromal cell therapies have approached the clinic for two decades, but potency and quality-control problems in cell manufacturing slow their translation, and artificial intelligence is increasingly proposed as a remedy. We mapped the intersection with a bibliometric analysis of 144 documents (95 original articles and 49 reviews, 2009 to 2025) retrieved from PubMed and enriched through OpenAlex and Crossref. Output grew at a compound annual rate of 35.1% and was overwhelmingly preclinical (47.9%); manufacturing and quality-control studies made up 13.2% of the corpus and clinical-stage work only 2.1%. The studies concentrated on cell characterization rather than release-grade potency testing. A reproducible corpus and analysis pipeline accompany the paper.
Authors: Aisha Al-Khinji, Dhafer Malouche
Status: Posted
Abstract: Pregnancy loss reflects sociodemographic, reproductive and metabolic factors acting together, yet these are usually studied one at a time, particularly in Gulf populations. Using logistic regression and Bayesian network structure learning, we analyzed cross-sectional data on 864 ever-married women in the Qatar Biobank, of whom 304 (35.2%) reported at least one pregnancy loss. Maternal age was the dominant correlate: the odds of loss reached 4.44 (95% CI 2.06–9.57) at ages 61 and above relative to women aged 30 or younger, whereas parity lost its association after adjustment for age. In the network, age was the only variable sharing a direct edge with pregnancy loss (bootstrap support 89%). The design is cross-sectional, so all relationships are associational.
Authors: Aisha Al-Khinji, Hadeel Alashwal, Abdullatif Al-Hor, Hissa Al-Kuwari, Farwa Bismi, Nada Maki, Taghreed Abunada, Dhafer Malouche
Status: Published in BMC Medical Research Methodology (2026), under the title “Bayesian and frequentist meta-analysis of GCB-classification heterogeneity in TP53-reporting DLBCL cohorts at small k: an applied case study with reporting guidance”.
Abstract: Meta-analyses of biomarker prevalence in diffuse large B-cell lymphoma often rest on five or fewer cohorts, where heterogeneity statistics destabilize and priors begin to dominate posteriors. Using cohorts that report both TP53 status and cell-of-origin classification as a methodological exemplar (five cohorts, 738 patients), we targeted the probability of germinal-center B-cell-like (GCB) classification with a random-effects frequentist analysis and a hierarchical Bayesian analysis in Stan that share one estimand. Heterogeneity in the GCB proportion was meaningful but imprecisely estimated at k = 5, and the odds ratio for TP53 positivity versus GCB classification, reconstructible in three cohorts, was 0.82 (95% CI 0.53–1.26), consistent with no association. The analysis is exploratory, and future cohort studies should pre-specify joint TP53 and cell-of-origin reporting.
Preprint on Research Square — Published version (DOI)
Authors: Aisha Al-Khinji, Dhafer Malouche
Status: Posted
Abstract: Standardized measurement of population health needs reproducible transformations from raw biobank data to internationally calibrated indicators, a step rarely taken for Middle Eastern cohorts. We built and validated such a framework on 2,400 Qatar Biobank participants, deriving 24 indicators in five domains from published clinical thresholds (EWGSOP2, WHO, ESC/ESH 2018, IDF, GLI-2012, GOLD and CANTAB). Obesity prevalence was 39.5% (95% CI 37.6–41.5) and central obesity 60.6% (58.6–62.5). Against measured blood pressure, self-reported hypertension showed 50.9% sensitivity and 88.9% specificity (κ = 0.28, PABAK = 0.72). The framework recovered the expected sex difference in grip strength (Cohen’s d = 2.41) and supports cross-cohort comparison in Gulf-region settings.
Authors: Chaker Mzoughi, Dhafer Malouche
Status: Posted
Abstract: Agentic AI systems, capable of autonomous goal pursuit, adaptive learning and multi-agent coordination, are being deployed in financial markets faster than the frameworks meant to govern them. We report a systematic scoping review of 30 governance documents published between 2022 and 2025 across seven jurisdictions, coded against a 67-provision instrument and synthesized with a Bayesian Beta–Binomial model. Sixty-one percent of provisions are absent from the majority of the corpus, and transparency is the weakest dimension (estimated coverage 35%, 95% credible interval 30 to 40 percent). We propose a three-axis normative framework (autonomy level by market function by regulatory regime) and illustrate it with a case study of the Qatar Financial Centre.
Authors: Aisha Al Khinji, Dhafer Malouche, Noof Al-Thani, Azza Mustafa, Jazeel Abdulmajeed, Mohamed Al-Kuwari
Status: Posted
Abstract: We estimated the independent associations of dyslipidemia and hypertension with non-alcoholic steatohepatitis (NASH) in a retrospective 1:1 age- and sex-matched case–control study of 894 cases and 894 controls drawn from Primary Health Care Corporation records in Qatar. Conditional logistic regression, which respects the matched design, gave adjusted odds ratios of 1.56 (95% CI 1.25–1.96) for dyslipidemia and 1.39 (95% CI 1.09–1.77) for hypertension. Both were independently associated with NASH, which supports targeted metabolic risk assessment in primary care.
Authors: Aisha Al Khinji, Dhafer Malouche
Status: Published in Journal of Translational Medicine 24, 899 (2026).
Abstract: Chronic pain frequently co-occurs with depressive and anxiety symptoms and doubles or triples the risk of suicidal ideation. Yet the joint pathways linking these conditions remain under-explored in Middle-Eastern settings. We analyzed questionnaire data from 2,363 Qatari adults enrolled in the Qatar Biobank using a data-driven Bayesian Network to map probabilistic interdependencies between chronic pain, mental-health symptoms and demographic factors, with a focus on sequences culminating in suicidal thoughts. The final BN identified self-regret and psychomotor change as the strongest direct predictors of suicidal thoughts, with an early-symptom cascade (sleep problems → anhedonia → depression) amplifying suicide risk. Fatigue linked chronic pain to self-regret, illustrating a physical–psychological bridge.
Preprint on Research Square — Published version (DOI)
Authors: Aisha Al Khinji, Dhafer Malouche, Abdullatif Al-Hor, Hadeel Ashwal, et al.
Status: Posted
Abstract: Autoimmune diseases confer excess cardiovascular disease (CVD) risk, yet comparative profiles across phenotypes in Gulf health systems are unclear. Using routinely collected primary-care records, we analyzed 14,616 adults and derived a composite CVD risk index from a Framingham-based algorithm. Relative to the non-autoimmune reference, adjusted odds of high risk were higher in rheumatoid arthritis (OR 1.20, 95% CI 1.06–1.37), lower in Hashimoto’s thyroiditis (0.49, 0.40–0.60), and not clearly different in SLE or multiple-autoimmune phenotypes. These findings support systematic CVD risk assessment and targeted management in autoimmune populations within Gulf health systems.
Authors: Dhafer Malouche
Year: 2026 — SocArXiv, doi:10.31235/osf.io/y9q5e
Abstract: Prediction markets and polls are usually compared as two forecasts of one event, but under a two-round rule they are not: a first-round poll measures intentions in one assumed line-up, a run-off poll one pairing, and a price on the presidency folds ballot access, the first round and the run-off into one number. Using a Bayesian state-space model of the polls and a model of Kalshi and Polymarket prices, we show that on 31 October 2026 the probability that Marine Le Pen reaches the run-off was 0.96 in the polls and 0.79 in the markets, and the probability that she wins it 0.89 against 0.42. The log ratio of the two splits exactly into ballot access, first round and run-off. The run-off carries 75 percent of the gap (90% interval 70 to 83), ballot access 17 and the first round 8: the markets doubt the run-off, not her first round. The analysis is exploratory; a registered evaluation follows the run-off of 2 May 2027.
Authors: Rim Lahmandi-Ayed, Dhafer Malouche
Year: 2018 — arXiv:1807.08458 [stat.AP]
Abstract: This paper tests whether national R&D investment causally affects schoolchildren’s education performance. OECD PISA scores are merged with the World Bank’s World Development Indicators and Bayesian-network learning is applied to identify a statistical causal relationship between a country’s R&D expenditure and its PISA performance. The effect is long-term: at least about ten years of sustained R&D investment is required before improvements in pupil performance become detectable.
Authors: Hela Jeddi, Dhafer Malouche
Year: 2015 — arXiv:1511.02229 [stat.AP]
Abstract: Wage differentials between males and females in Tunisia are estimated using the Oaxaca–Blinder decomposition applied to the Tunisian Population and Employment Survey (2005). The estimated gender wage gap is approximately 19%, attributable primarily to discrimination (underestimation of female characteristics in the labour market) rather than differences in observable characteristics.
Authors: Dhafer Malouche, Bala Rajaratnam, Benjamin T. Rolfs
Year: 2013 — arXiv:1310.2641 [math.ST]
Abstract: This work examines the formal duality between undirected (concentration) and bidirected (covariance) Gaussian graphical models. Pairwise and global Markov properties are developed under weaker pseudographoid and reverse-pseudographoid rules (in lieu of intersection/composition), and the duality is then used to transport results across the two model classes, including new results on faithfulness implications.
Authors: Dhafer Malouche, Bala Rajaratnam
Year: 2009 — arXiv:0912.2407 [math.PR]
Status: Published as Journal of Probability and Statistics (2011), Article ID 152942.
Abstract: This paper studies covariance graphs (undirected graphs encoding marginal independencies) for multivariate Gaussian distributions, focusing on the case in which the covariance graph is a tree. New conditions are derived under which the covariance graph faithfully represents the marginal-independence structure of a Gaussian random vector.
arXiv:0912.2407 — Published version (DOI)
Authors: Dhafer Malouche
Year: 2007 — arXiv:0705.1613 [math.ST]
Status: Published in Bernoulli 15(4), 1179–1189 (2009).
Abstract: The concentration graph of a random vector is the undirected graph in which absence of an edge encodes full conditional independence given all remaining variables. The full conditional independence structure can be recovered from low-order conditional independence tests; the maximum order required is characterised in terms of minimal separators in the graph.
arXiv:0705.1613 — Published version (DOI)
Authors: Hajer Khlaifi, Dan Istrate, Jacques Demongeot, Jérôme Boudy, Dhafer Malouche
Year: 2018 — HAL:hal-01692422 (JETSAN 2017, Bourges, France)
Status: Extended journal version published in IRBM 39(6), 407–412 (2018).
Abstract: A non-invasive acoustic method is developed for distinguishing swallowing sounds from ambient noise during food intake, in the context of the e-swallhome project for post-stroke dysphagia monitoring. Gaussian Mixture Models trained with the EM algorithm and leave-one-out cross-validation achieve 100% recognition accuracy on the study dataset.
HAL:hal-01692422 — Published version (DOI)
Authors: Khawla Ghribi, Sylvie Sevestre, Zahia Guessoum, Javier Gil-Quijano, Dhafer Malouche, Anis Youssef
Year: 2014 — HAL:cea-01888018 (CISTEM 2014, Tunis, Tunisia)
Status: Published in 2014 IEEE International Conference on Electrical Sciences and Technologies in Maghreb (CISTEM).
Abstract: A taxonomy of multi-agent approaches to managing complex energy systems is proposed, distinguishing reactive (short-term) from anticipative (long-term) management and classifying systems by the level (component, sub-system, system) at which decisions are made.
HAL:cea-01888018 — Published version (DOI)
Authors: Dhafer Malouche, Bala Rajaratnam
Year: 2010 — HAL:inria-00494743 (42èmes Journées de Statistique, SFdS, Marseille)
Abstract: Conference-length presentation of work on covariance graphs that faithfully represent Gaussian marginal independencies when the graph is a tree. Companion to arXiv:0912.2407 and the Journal of Probability and Statistics (2011) paper.
Authors: Rania Bengrich, Dhafer Malouche, Sylvie Sevestre-Ghalila
Year: 2009 — HAL:inria-00386779 (41èmes Journées de Statistique, SFdS, Bordeaux)
Abstract: Graphical-model estimation is coupled with Gaussian Markov random-field modelling for image textures, in order to estimate the neighbourhood graph of a stationary periodic Markov field. The method exploits texture periodicity to recover the underlying spatial-interaction graph.
Authors: Makrem Djebali, Dhafer Malouche, Sylvie Sevestre-Ghalila
Year: 2009 — HAL:inria-00386781 (41èmes Journées de Statistique, SFdS, Bordeaux)
Abstract: Graphical-model variable selection is used to predict osteoporosis status from a clinical dataset with potentially missing observations at prediction time. The selected concentration graph is fitted on complete data, then specialised for prediction patients with missing covariates by re-estimating the model on the available subset of variables.
Authors: Daria Mizza, Tamer El Khouly, Dhafer Malouche, Oluwapelumi Dickson, Ilaria Compagnoni
Year: 2026 — SSRN:6505578
Abstract: Feedback practices in project-based courses may foster intrinsic motivation and informed academic choices, but intervention effects on the psychological needs for autonomy, competence and relatedness have been inconsistent. In a mixed-methods study of an undergraduate digital tools course in architecture across two semesters (N = 160), we examined how feedback supported these needs and whether support related to motivation and program declarations. Procedural autonomy was better supported than substantive autonomy, competence support fell from 91.5% positive for foundational skills to 30.5% for advanced ones, and relatedness was strong with instructors but weak among peers. Perceived need support strongly predicted intrinsic motivation (R² = 0.73), and focus groups identified four communication factors (message consistency, relational tone, interpersonal dynamics and feedback clarity) that explain why measured support diverged from lived experience.
Authors: Dhafer Malouche
Year: 2021 — SSRN:3900615
Abstract: Covidradar24.org is a platform aggregating more than 280 real-time dashboards on COVID-19 morbidity and mortality covering countries, continents and territories. The paper documents the data sources, refresh pipeline and visual-analytic design choices, alongside the reproducible R/Shiny and R Markdown infrastructure used to maintain the dashboards.
Authors: Dhafer Malouche, Habiba Ben Romdhane
Year: 2020 — SSRN:3731481
Abstract: World Bank population and mortality data are combined with the IHME (University of Washington, Seattle) COVID-19 SEIR projections of pandemic deaths for 151 countries to estimate, country by country, the excess crude death rate associated with COVID-19 by 31 December 2020 under each IHME non-pharmaceutical-intervention scenario. Projected new death crude rates, growth rates, and the share of COVID-19 in total annual deaths are reported.
An interactive dashboard for Al-Sharman et al. (2026), Journal of Multidisciplinary Healthcare — a comparative Bayesian network analysis of directional symptom dependencies in multiple sclerosis (MS, n=104) and Parkinson’s disease (PD, n=54). Explore the disease-specific directed networks, hub structure, edge strengths, and bootstrap stability directly in the browser.
An interactive companion dashboard for Al-Khinji et al. (2026), Frontiers in Medicine — a Qatar Biobank study of hematological abnormalities in non-alcoholic steatohepatitis (NASH). The app lets the reader explore the cohort, recompute sex-stratified CBC reference limits, and reproduce the group-comparison results of the paper directly from the browser.
Teaching and research applications, built as worked examples of how data and statistics are used in academic work.
A guided, step-by-step Western blot for the teaching laboratory, from casting the gel to quantifying the bands, with timers, a gel recipe calculator and a sample loading calculator. It is an academic example of the quantitative side of bench work, where recipe volumes and loading amounts come from simple formulas. It uses no dataset, only standard laboratory arithmetic and the values the user types. © 2026 Dhafer Malouche and Aisha Al-Khinji, all rights reserved.
Five calculators for chemists and biologists: a concentration converter, molar mass from a chemical formula, percent solutions, buffer and media recipes, and centrifugation in rpm and ×g. Each answer is shown with the formula and the full working. Formerly ChemCalc, and a companion to CellCalc. It is an academic example of unit-aware quantitative reasoning in the laboratory, and it uses no dataset, only standard formulas and the user’s own numbers. © 2026 Dhafer Malouche and Aisha Al-Khinji, all rights reserved.
A virtual haemocytometer for teaching laboratories. Students place live and dead cells on the counting grid and see viability, cell density and the subculturing volume worked out step by step, together with a check on how evenly the sample was mixed (the coefficient of variation across squares). It is an academic example of how a count becomes an estimate with a quality check. It uses no dataset, only the counts entered by the user. © 2026 Dhafer Malouche and Aisha Al-Khinji, all rights reserved.
A pedagogical calculator for the wet lab: solution mass, solution volume, dilution, serial dilution, cell counting and cell seeding. Rather than returning only a number, every calculator states the formula, substitutes the user’s own values with units carried through each line, walks the arithmetic step by step, and draws a live diagram — a filling flask, dot-density beakers, or a well-plate mixing bar. Automatic unit conversion across M / mM / µM / nM and L / mL / µL, a one-click worked example, and reproducible R, Python and plain-text protocol export.
An offline-first revision app (PWA) for STAT 481 — Multivariate Analysis at Qatar University (Fall 2026). Every card, formula, R snippet and question is extracted automatically from the course LaTeX notes and carries a stable reference back to the file and line it came from, so nothing is paraphrased. Leitner spaced repetition, a searchable formula index, verbatim R output, timed practice exams that unlock only after the paper has been sat, and 56 course figures — all installable, fully offline, with every keystroke of progress kept on the student’s own device.
An offline-first revision app (PWA) for STAT 312 — Stochastic Processes at Qatar University (Fall 2026): a live semester timeline, per-stage revision hubs, spaced-repetition flashcards, an interactive practice-quiz engine built from the Fall 2026 sample quizzes, a searchable formula sheet, and an FAQ. Installable on phones, works fully offline — no backend, no accounts, all progress stored in the browser.
An interactive companion to Al-Sharman et al. (2026), Journal of Multidisciplinary Healthcare, a comparative Bayesian network analysis of directional symptom dependencies in multiple sclerosis (n = 104) and Parkinson’s disease (n = 54). It shows the learned networks, the reported edge strengths and the bootstrap edge stability, and it serves as an example of how to read a Bayesian network without mistaking arrows for causes. It uses only the results printed in the open-access article (CC BY-NC 3.0) and no participant-level data.
An interactive dashboard of the 2026 French municipal elections (rounds of 15 and 22 March 2026). Official Ministry of the Interior results, aggregated from the polling-station level to the 96 metropolitan departments, are mapped with Leaflet and explored by political bloc and party. Built in R (tidyverse, sf) with a static HTML/JS front end deployed on Netlify.
An interactive companion dashboard for Al-Khinji et al. (2026), Frontiers in Medicine — a Qatar Biobank study of hematological abnormalities in non-alcoholic steatohepatitis (NASH). The app lets the reader explore the cohort, recompute sex-stratified CBC reference limits, and reproduce the group-comparison results of the paper directly from the browser.
An interactive teaching tool for Principal Component Analysis: paste or upload a numeric dataset and the app reports the eigen-decomposition of the (correlation or covariance) matrix, the scree plot, the cumulative variance explained, the score and loading matrices, and a coordinated biplot. Companion to StatTables, StatTests, StatRegress, StatCI, StatPower, and StatCorr — runs entirely in the browser, no backend.
An interactive teaching tool that draws many samples from a chosen population, builds a confidence interval from each, and shows visually how often the intervals actually cover the true parameter — curing the 95% confident misconception. Companion to StatTables, StatTests, StatRegress, and StatPower; runs entirely in the browser, no backend.
A teaching-focused, browser-only workbench for linear regression: paste a CSV, fit an OLS model, and read the standard coefficient table, model summary, and residual diagnostics — with a drag-a-point mode for simple regression that updates the fit live. Companion to StatTables and StatTests; everything runs client-side, no data leaves the browser.
A browser-based replacement for the statistical tables at the back of stats textbooks. Pick a distribution ($Z$, $t$, $\chi^{2}$, $F$, Binomial, Poisson, Exponential) and obtain a classical lookup table, a probability calculator with shaded plot, and a step-by-step formula breakdown — all three coordinated, with matching cells highlighted in the table.
A browser-based teaching tool that runs 14 of the most common hypothesis tests — for means, proportions, variances, ANOVA, $\chi^{2}$, and non-parametric problems — with step-by-step formulas, shaded rejection regions, and automatic assumption checks. Sister application to StatTables, designed to replace the test-by-test recipes traditionally given in the appendix of an introductory statistics textbook.