Statistical consulting for theses and research
I support you from study design to interpreting the results: correct analyses, clearly explained, reproducible.
Free introductory call, no obligation.
A well-specified model always starts from the shape of the data.
Tools I work with
RPythonSASGeneralized linear modelsMachine learning
From the initial idea to the response to reviewers, tailored support for students, PhD candidates and researchers.
Dataset cleaning, descriptive statistics, tests and models suited to your study design, with tables and figures ready for the text.
Regression, mixed models, time series, survival analysis, machine learning: I choose the right model and explain why.
Commented R or Python scripts, so you can redo the analysis on your own and answer every question with confidence.
Review of the study design, the assumptions and the methods section, before a reviewer does it.
Additional analyses and clear wording to answer comments on the statistical part of a paper.
One-to-one sessions to truly understand the analyses carried out and present them confidently at your defence.
For research teams
A good model is only useful if someone can use it. I help you put it “into production” on online cloud services, so that it keeps running reliably even when I am no longer involved.
Let's talk about your projectI turn the analysis or model into robust, repeatable code, ready to run automatically.
I make the model available as a web application, dashboard or service that other systems can call. I work on Amazon Web Services (AWS).
I check that the model keeps working over time and train your team so they can manage it independently.
Case studies
Real examples of theses and research, with charts recalculated from the original data. They are presented only in aggregated, anonymised form: no name, centre or client can be recognised. Choose an area or open a case to see the chart, method and limitations of the analysis.
Biostatistics and survival analysis · Case 1
How long it takes for the disease to return after prostate surgery, and which factors bring it forward.
87% → 66% recurrence-free at five years, from low to high risk
Biostatistics and survival analysis · Case 2
Duration of response to first-line therapy in patients with a haematological disease, and which characteristics shorten it.
3.1 vs 2.1 years median duration below and above 65 years of age
Biostatistics and survival analysis · Case 3
Survival after cardiac valve replacement surgery, comparing women and men.
62% of women are alive at five years, against 37% of men
Biostatistics and survival analysis · Case 4
Clinical factors associated with death during hospital stay after emergency aortic surgery.
5,6 odds ratio of death with involvement of the supra-aortic vessels (35% versus 9%)
Psychometrics · Case 5
Validation of an 18-item questionnaire: how many dimensions it measures and how reliable it is.
826 complete questionnaires
Psychometrics · Case 6
How six scales on workplace well-being relate to each other, and how reliable each one is.
0,92 alpha of the work engagement scale, the most reliable
Psychometrics · Case 7
How far gambling among teenagers rises with risk propensity and falls with parental control.
71% of the most risk-prone boys gambled (37% among the most cautious)
General medical studies · Case 8
How the share of patients with positive lymph nodes changes with the extent of surgery and with risk.
5% → 25% patients with positive lymph nodes, from 10 or fewer to 20 or more nodes removed
General medical studies · Case 9
Three circulating markers as predictors of 30-day mortality, in only 57 elderly patients.
57 elderly patients, 16 deaths at 30 days
General medical studies · Case 10
Repeated measures in patients with mild cognitive impairment: depressive symptoms and two cognitive tests, at baseline and after one year.
−1.8 points in depressive symptoms (from 8.2 to 6.4)
General medical studies · Case 11
When cases are few and predictors many: penalised regression to predict the recurrence of a chronic haematoma.
118 patients, of whom 15 with recurrence (13%)
Marketing analytics · Case 12
Predicting which customers of a subscription service are about to leave, to decide whom to contact first.
×4.3 lift: in the highest-risk decile 53% churn, against an average of 12.4%
Marketing analytics · Case 13
Segmenting a market on attitudes and preferences, and seeing how much each group likes the product.
6,6 average liking (1–9 scale) in segment A, 27% of the sample
Marketing analytics · Case 14
Survey of shopping-centre visitors: catchment area and rating of fifteen aspects.
72% of visitors come from less than 100 km away (median 79 km)
Marketing analytics · Case 15
How sales evolve from launch onwards depending on the level of product innovation.
×57 median sales at four years: highly innovative products versus no innovation
Other analyses · Case 16
How claim frequency and cost depend on driver age and no-claims class in an insurance portfolio.
×1.58 claim frequency at ages 18–19, compared with the 45–54 band
Other analyses · Case 17
How far financial statements suffice to predict which companies will default, with a simple index or with a model.
90% default in the lowest-liquidity decile, against 9% in the most liquid
Other analyses · Case 18
Weekly time series of purchase and sale prices, and their seasonality.
0,006 p-value of the cointegration test between purchase and sale price
The first call is free and with no obligation.
You tell me about your goals, data and deadlines. It is free and with no obligation.
You receive a work plan with clear timelines and costs, before we start.
I carry out the analyses and update you at every step, so you can step in early.
Results, code and a clear explanation. I remain available for questions and revisions.
I am Giorgio, Actuarial Data Science Manager. Every day I lead projects that quantify uncertainty using advanced statistical models and AI, and turn it into decisions.
My academic research, carried out in collaboration with other groups, ranges from risk modelling to medical statistics. In theses and papers I bring the same rigour I use at work: explicit hypotheses, justified models, results that can be verified.
I use AI every day, but AI is not accountable for its choices: I am. When a piece of work carries my name, I answer for it to you, to a reviewer or to an examination committee.
No. If you are at the start, I can help you set up the study design and estimate the sample size.
Based on the work required. After the call you receive a written quote, agreed before we begin.
Yes. Your data remains confidential and is used only for your project. Before starting we sign a non-disclosure agreement covering both the collaboration and the data you entrust to me. If the results are ever to be made public, for example in a paper, form and content will always be agreed with you in advance.
Yes, as a working tool. The statistical choices, the verification of results and responsibility for the work delivered remain mine. You remain the author of your thesis or paper and, if needed, I will help you declare the use of tools according to your university's or journal's rules. Client data is uploaded to external tools only with prior consent.
Yes: you receive commented code and an explanation of the results, in person or by video call, so that you can defend them on your own.
Describe your research and your deadline in a few lines. I will reply with an initial assessment and, if needed, we will schedule a call.
Fill in the form: I reply personally, usually within one working day.