Statistical consultations at NRU
This page is meant for the NRU staff (https://nru.dk/) needing assistance with data analysis. You will find information about the statistical support that I provide at NRU, how to get help and what to expect. People outside NRU but working at Rigshospitalet should contact the “statistical advisory service at Rigshospitalet”, see the link at the end of the page.
General recommendations for reporting results of statistical tests and common mistakes are summarized on a separate webpage.
What can I help you with?
-
Group consultation: this is the recommended format for Bachelor and Master students. + students should collect questions, ideally with real data example, and send them to me. + we arrange a meeting where I will discuss one or a few common questions in a 15-30 min, interactive, presentation. The rest of the time is Q & A, leaving time to adress more specialized questions.
Discuss with your supervisor when is the relevant time and who to include. -
Assistance for performing a statistical analysis:
You would help from a statistician to decide upon and carry a statistical analysis to answer a research question. Or you would like help to replicate the statistical methodology from a published paper.
PhD students/new post-docs: invite your supervisor to the first consultation. -
Technical questions:
You have a specific question, it could be a software question, e.g. which package to use in R for fitting a linear mixed model? or a statistical question, e.g. I would like to compare the mean recovery across three treatment groups, do you know a statistical test for doing that?
An example of non-specific question is: What is the best statistical approach to analyze my data? What do you mean by ‘analyze my data’ (compare mean, correlation, variance, …), best with respect to what (type 1 error control, statistical power, unbiasedness, interpretability, …).
Should be related to statistics and not pure data management or visualization question. -
Externalizing the statistical analysis:
You want a statistician to perform the statistical analysis for you. Don’t expect much from me in this case, I will rarely accept to perform the statistical analysis myself (this is at my discretion).
Not for Bachelor/Master/PhD students. -
Planning a study:
You are planning a new study or experiment. You have one or few clearly defined hypotheses that you would like to test but you are wondering how to appropriately design your experiment / study to be able to carry out the tests. This is not my domain of expertise in statistics but I can give you some feedback and help you with simple power calculation.
Make sure to plan sufficient time (> 1 week) for the power calculation as I will generally not be able to help you with “emergency” power calculation.
What can I cannot help you with?
-
“How should I analyze my data?” It is not possible for me to assess the statistical properties of a procedure if I do not precisely understand what hypothesis you want to test nor what quantity you want to estimate.
You should reflect, with your collaborators, to one or several research questions before reaching to me.
For the integrety of subsequent data analysis, you should stay blinded to the data, especially the exposure-outcome relationship. Univariate descriptive statistics can be ok. -
“How is the best statistical approach to answer my research question?” Best with respect to what: type 1 error control, statistical power, bias, interpretability, feasibility, tradition? Those are often conflicting objectives: to achieve optimal type 1 error control one could use permutation tests at the expense of interpretability due to the lack of corresponding confidence intervals. Double robust estimators can be used to minimize bias at the cost of complex statistical procedures.
I can help you with expliciting the pros- and cons- of different statistical procedures. Sometimes there is an obvious better procedure, sometimes several procedures are all reasonnable with different trade-off. You can choose one for the main analysis (typically the most interpretable) and keep the others for sensitivity analyses. -
“Making sure my analysis is correct” or “Can you check my analysis is correct”: those are too broad questions. It would be like going to the doctor and ask him whether he can check that you are heatlhy. Yes with infinite time and ressources the doctor could do many examinations & tests but this is not common practice.
Explicit what your concern is and what you attempted to mitigate it.
How to get help?
- Assistance for performing a statistical analysis/Planning a
study:
You should send me an email (brice.ozenne@nru.dk, please put Consultation in the subject line). The email should include:- a name for your project
- a short description of your study and your questions. By short I mean it can be just a few sentences.
- [optional] additional documents with a sentence explaining at
which part of the document I should look at (especially for articles).
e.g. a display of the design of the study
or the article of a study that you would like to replicate.
Then you can book an appointment (1/2 hour to 1 hour) in my calendar during my consultation hours. I only take one or two consultations like that a week so I may ask you to postpone the meeting if I am too busy.
Note: sending me information by email may seem a loss of time, but it will help me prepare the consultation and so we will be more efficient when we will meet. In some cases it also forces you to write down clearly what you are actually doing / wanting to do and you may yourself find the answers to your questions!
- Technical questions:
You don’t need to book a meeting, you can just show at my office during my consultation hours. Just check in my calendar (in cal) that I don’t already have a meeting. This type of consultation is typically short, i.e. less than 15 minutes. If I am busy, you can also send me an email (brice.ozenne@nru.dk, please put Consultation in the subject line) explaining your problem and I will try to answer within a week.
When can you get help? Consultation hours
I take consultations Tuesday afternoon (13h30-17h30).
If this is completely incompatible with your schedule, for instance because you’re running weekly experiments those days, just ask me and we will figure out a solution.
Authorship
- Assistance for performing a statistical analysis: This help
qualifies me for co-authorship. I expect it to be automatic. If
you/your superviser disagree we should discuss it at the first
meeting. My affiliations are:
- Department of Public Health, Section of Biostatistics, University of Copenhagen, 5 Øster Farimagsgade, 1014 Copenhagen, Denmark
- Neurobiology Research Unit and BrainDrugs, Copenhagen University Hospital, Rigshospitalet, 6-8 Inge Lehmanns Vej, 2100 Copenhagen, Denmark
-
Externalizing the statistical analysis: This help qualifies me for co-authorship. I expect it to be automatic and I will also expect to have a good place (e.g. second, third, second to last).
-
Technical questions: This help does not qualify me for co-authorship.
- Planning a study: This help does not qualify me for co-authorship.
Response time
I usually answer to ‘simple’ questions that I could not solve during the consultation within a week. If you haven’t heard from me within a week, please send me a reminder, I have probably forgotten (sorry!).
If you ask me to review your article you should give me at least two weeks.
What I cannot do is to help you to finish your analysis for tomorrow. Please anticipate a resonable time for the statistical analysis, especially if you have a complicated design or many hypotheses to test.
Software
I can provide you with support using R, e.g. find the appropriate function(s) to perform your analysis, display the results, and understand the outputs from the software.
For other softwares (e.g. SAS, Stata, SPSS, GraphPad Prism, MATLAB, Python, …), I can try to help you to understand the outputs from the software. Note that there are very skilled people using SAS at the section of biostatistics so you should consider using Statistical advisory service at Rigshospitalet (below) if you want to use SAS but you anticipate you will need help in using it. People from the data analysis group at NRU tend to use MATLAB and many psychologists at NRU use SPSS so they can probably help you with these software.
Confidentiality
Please tell me if there is any confidentiality clause relative to your results or your data. By default, I will not share your data with anybody without asking you. However I may discuss the analysis with people at the section of biostatistics (not the results just the method).
What to expect?
- Assistance for performing a statistical analysis
I can help you with:- performing statistical analysis that are suited to your research question and that output interpretable results.
- generating nice figures (if you use the R software).
- writing the “statistical analysis” paragraph from the method section and presenting the results in your manuscript.
- making sure that your analysis is reproducible, i.e. when you get the review 3 months after you are able to reproduce the results you got when you wrote the article (you need to ask specifically for this).
- answering the comments of the reviewers that are related to the statistical analysis.
This typically involves:
- starting from a research question, e.g. is treatment A better than treatment B?
- defining parameter(s) of interest that can be used to answer the research question, e.g. difference in mean outcome under treatment A vs. under treatment B. Often several parameter(s) of interest are possible, e.g. median outcome or frequency of a higher outcome under A vs. B, and which one to choose should mainly depend on the clinical or biological relevance.
- identifying complications like confounding, missing values, …
- deciding on a model to estimate the parameters of interest despite the complications, e.g. a linear model adjusted for age, and expliciting the assumptions, e.g. linear age effect, no other confounder, …
- deciding on an appropriate testing procedure, e.g. Wald test, possible adjustment for multiple comparisons. The definition of parameter of the interest is critical: it is not possible to evaluate the quality of a statistical analysis if we cannot explicit what we want to estimate.
- Technical questions:
I will try my best to indicate the appropriate software/package/function/statistical tool/criteria to use for solving your problem. I will though only try to answer what you asked for. So I will assume that you have given me all the necessary information and if information is missing I will assume that you are in a standard case (e.g. normally distributed outcomes, no missing values, …). In particular, I will (try to) not assess whether what you have been doing seems reasonable to me and I am not responsible for the validity of the statistical analysis.
What I will expect from you
-
to present me the person(s) involved in the project, especially for students it is nice for me to know who is your supervisor. By default I will assume that you are the person coordinating the project. Also tell me if there is another statistician involved because it makes sense that the statisticians agree before telling you what to do.
-
to be responsible for organizing meetings, their frequency, and choosing who should attend the meetings. Importantly, outside the meeting I will never check whether you are making progress or need help. So when you meet technical issues (with the software or what the statistical method is about) it is often a good idea to look for information on Internet and then to reach for help (me or one of your colleagues).
-
to keep track of the decision(s) we have taken during the meeting. When the decisions are related to the manuscript you should also write down the reason(s) for these decisions. It can seem annoying to do but this will help us a lot first and foremost because I will forget what has happened at the last meeting. It will also help me to be consistent over time.
-
to have a somewhat clear and fixed aim. If you have changed your research question between two meetings you should explain it to me. You can ask the same from me if I decide to change the statistical approach.
-
to be open-minded and willing to implement recommandations. You are obviously welcome to disagree with me, suggest classical approaches used in your field, and mention practical constraints (time, programming skills). But you should also have time and willingness to try what I recommend to you, otherwise our discussion is not really worthwhile.
-
before submitting an article with my name on, you need to have had my approval. Send me your article and I will give you some comments. If I mention “major” or “major concerns” or something along those lines this means that we need one more iteration (send article, comments) before I approve it. Note that you’re very welcome to disagree with my comments, then we should just meet to discuss them.
Other places where you can find help for data analysis
The section of biostatistics of the University of Copenhagen offers free statistical advice to faculty, staff, and students at the Faculty of Health and Medical Sciences (SUND) and researchers at Rigshospitalet (including RH Glostrup).
The data analysis group at NRU has a more specialized expertise in medical imaging data processing and analysis.