The LLNL Statistical Consulting Service aims to promote excellence in statistical analysis throughout the laboratory by providing short-term statistical support free-of-charge to LLNL researchers. As part of the Engineering Data Sciences initiative, this consulting service also provides assistance in other areas such as machine learning and decision sciences. Using the service may lead to longer-term analytics support by identifying individuals with relevant expertise.
How to Contact the Statistical Consulting Service
To schedule a consulting session, please send an e-mail to firstname.lastname@example.org with the following information (cut and paste the fields below):
- Name and OUN:
- Project name:
- Project Directorate/Division:
- Sensitivity concerns (classification or export control):
- Brief description of your problem:
Do not include classified or export controlled information in your description.
When to Contact the Statistical Consulting Service
The earlier, the better! In experimental situations, talking with a statistician prior to data collection generally improves the subsequent statistical analysis and prevents costly errors in experimental design. In other data analysis situations, early discussions can lead to more effective modeling without wasted effort. If you have never used the consulting service, please read the section below titled Statistical Consulting Best Practices.
Frequently Asked Questions
Why would someone contact the statistical consulting service?
The service is intended to help researchers who desire expert advice for a short-term statistical problem, but who may not have the resources (or need) to hire a statistician onto their project.
Who can use the consulting service?
Any laboratory employee is eligible for consulting help on their laboratory projects. Note that this includes post-docs working on their institutionally-funded research. The consulting service cannot assist with non-laboratory work, although outside (e.g. university) collaborators may request assistance for projects funded through the LLNL.
How does the service work?
The Statistical Consulting Service provides up to three hours of data analytics support to laboratory programs at no cost to the project. If additional effort is needed, the service attempts to match the program to a consultant with appropriate expertise, who can then be funded by the project itself. Due to the time limitations, the consulting service is better equipped to offer recommendations for analysis strategies rather than data analysis itself. If an expert is required for data analysis, then the project will generally be expected to support this effort. Even if you are not sure if your project meets these criteria, feel free to request a consultation by contacting the consulting service.
Who are the consultants?
The Statistical Consulting Service is staffed by experts in statistics and machine learning, mainly from the Data Analytics Section in the Computational Engineering Division. If you are interested in being included in this service as a consultant, please contact the consulting service.
What are examples of previous consulting projects?
- Educate computational engineers about appropriate analysis strategies and tools for a structural design optimization project. Provide review of final results.
- Recommend experimental design to demonstrate non-inferiority of new DNA based identification system.
- Develop preliminary proposal for uncertainty quantification/stochastic inversion project to address regulatory concerns for federal sponsor.
- Recommend "stopping criterion" for a stochastic simulation to determine when the probability of an undesirable outcome is below a specified threshold.
- Evaluate vendor testing strategies/statistical analysis used to demonstrate that a product meets government requirements.
The members of the consulting service have tremendous breadth of experience in terms of both analysis techniques and application areas, and these examples are intended to be illustrative rather than comprehensive.
Statistical Consulting Best Practices
- Talk to a consultant as early as possible, preferably before data has been collected. Statisticians are trained in experimental design as well as data analysis. The way that data is generated or collected can have an enormous influence on final analysis, and it may be that a small change in experimental procedure can make the final statistical analysis much more straightforward. Consulting a statistician early can also ensure that you are collecting data in an optimal way to answer a particular research question. Talking to a statistician when proposing a project is a good way to check that the proposed goals and experimental resource requirements are in line with each other.
- Be able to describe what your data will look like. Be ready to provide as much information as you can about the data itself. For example, are control variables binary (e.g. on/off) or numeric (e.g. time in seconds)? How many different variables are being measured? Are there potential sources of measurement error/variability in either input or output variables? If you have sample data, please let your consultant know. It may be helpful for them to take a look at it before or after the initial meeting.
- Provide examples of how other people in your field have addressed similar problems. Consultants are available to either explain procedures seen in literature or to suggest modifications if they are needed. Seeing what kind of analysis is considered standard in a field is helpful in that it can indicate what data tends to look like and what pitfalls have been encountered by other researchers in the past. If you have references for similar problems that include statistical analysis, please let your consultant know.
- Make sure that the consultant understands your analysis requirements. This includes ensuring that the consultant understands the scientific problem to be addressed well enough to recommend an appropriate method of analysis. It also means being clear on any limitations of your study, such as experimental restrictions (e.g. research involves human subjects who can only be evaluated once per day) or sample size restrictions (e.g. the budget only allows for 15 specimens).
- Recognize the limitations of statistical analysis. Sometimes the best answer that a consultant can give you is that you are unlikely to be able to accomplish your goals with the proposed study design. For certain tasks, such as identifying small treatment effects or certifying an extremely high level of reliability, large and costly studies are required and meaningful results are unlikely to be found in small samples. However, by consulting a statistician early in the process, such as while writing a proposal, you can ensure that you will have sufficient resources to accomplish your goals.