FAQ : Working with a Statistical Consultant

What services do statistical consultants provide?

On any given project, a statistician may be asked to:
  • Plan data collection, design experiments, and manage data.
  • Perform data cleaning.
  • Account for missing data with an appropriate method.
  • Perform exploratory and descriptive analysis of data.
  • Design statistical models.
  • Produce final results for publication (figures, charts, p-values, etc.)
  • Write methodology section of the research paper.
  • Document his work in a bibliography,

How will communication with a statistical consultant be during collaboration?

We will interact by email, videoconference, and, if needed and possible, during in person meetings. We have worked with multiple online-only collaborations with great success. Honest and transparent communication are essential to any successful scientific project. We will strive to be clear, pertinent, and concise.

At what stage of the project should a statistical consultant be brought on board?

Every single methodological choice impacts the downstream selection of proper statistical techniques needed to answer a set of research questions. Therefore, statisticians are as crucial in the initial phases of a project as in the later stages, when they will deliver the final statistical results. Too often, statisticians are consulted in late stages of a project that already has serious statistical lacunae, preventing publication. We would be happy to help you in later stages if needed, but always remember that the earlier a statistician is involved, the better.

What is data cleaning?

Data are often incorrect, inconsistent or missing. Rigorously handling these data issues is called data cleaning and is crucial for the integrity of the scientific process. Unfortunately, documenting and automating of error-prone computer processes are skills critically lacking in most research laboratories. As a statistician, Justin has built over a decade of experience with advanced data science tools and techniques, in both industry and academia, with big data infrastructures as well as smaller ones (see LinkedIn). We would be glad to help you strive towards high quality data.

What is missing data and how to handle it?

Rarely do we have complete data on hand, even in ideal research settings. Accounting for the statistical implications of missing data is not always trivial. We will write further details on this matter in the future, namely about how we handled a complex missing data problem in a forthcoming research paper. It is always best to consult a statistician if there is any uncertainty related to handling of missing data.

Can a statistician help with exploration of data in search of patterns and insights without being an expert in the relevant field of inquiry?

Multiple fields of inquiry have common underlying mathematical and statistical patterns and structures, namely features related to uncertainty and formulation of causal hypotheses. This is why a statistician can be of great assistance even in the early stages of a project.

How to choose a proper statistical model?

You may be aware that proper statistical models (mixed models, GLMs, SEMs, DAGs, etc.) are often much more powerful tools in research than naïve applications of unsophisticated statistical principles (Mann-Whitney, Kruskal-Wallis, Friedman ANOVA, t-test, Khi-square, etc.). To the untrained, approaching such sophisticated methods can be daunting. We would be glad to help you build such models for your research, while explaining how to do it along the way. If needed, training can also be provided.

Why are non-parametric methods so prevalent?

A common misconception of non-parametric tests (Mann-Whitney, Kruskal-Wallis, Friedman ANOVA, etc.) is that they require less underlying assumptions and are therefore preferable over more sophisticated statistical methods, lest crucial uncertain assumptions (e.g. normality of data, equal variances, etc.) turn out wrong, invalidating our conclusions and damaging our scientific reputation. It is in fact quite the opposite. The additional sophistication of proper statistical methods (e.g. GLM or Linear Mixed Models) is a theoretical price the researcher pays in exchange for added flexibility and decreased uncertainty of conclusions. Making statistical assumptions is a normal part of the scientific process that a statistician can help with. There are multiple techniques of sensitivity analysis that can be used to increase our confidence in otherwise shaky assumptions and plan to discuss them further in a future post.

I am not trained in statistics, and learning these methods feels daunting. What can I do?

If you are interested in learning and growing in statistical sophistication to further your scientific career, we can help. Not only will we provide expert opinion in crucial scientific matters, but we will also strive to make everything clear and accessible to you and your team and can train you and your team in statistics if needed.

How will the final results be delivered?

If you choose to work with us, final results will be delivered ready for publication according to both yours and your journal’s standards. We can use advanced visualization tools in R (and Python) to produce beautiful figures for publication. We will produce tables and figures with the salient data insights for the project, namely p-values estimates of effect sizes, confidence intervals, etc, in formats ready for publication in scientific journals.

What if I am not able to adequately translate a statistician's methodology when I decide to write the final paper?

We will write a methodology section that accurately describes our methodological choices. We are clear, articulate, and concise writers who have extensive experience writing scientific material for peer-reviewed publications.

Will a statistical consultant do research if needed?

We will always strive to do research looking to find, validate, and develop rigorous statistical methods to further your scientific goals. We will thoroughly document any material we used (papers, software packages, books, etc.) in a bibliography ready for publication.