Methods & publishing
Statistical Consulting and Research Data Analysis
Make analytical decisions that fit your research question and the data you actually have. Statistical consulting can support study planning, data preparation, analysis, interpretation and the reporting of results, with a clear record of how each conclusion was reached.
Outlier Science can help define a new analysis, review an existing one or address a specific methodological concern raised during peer review. The starting point is the scientific question and study design. A useful analysis should explain the evidence, its uncertainty and its limits in language the research team can use.
On this page 8 sections
01
Support before, during and after analysis
Before data collection, statistical input can help specify outcomes, comparisons, sampling assumptions and the information needed for the intended analysis. Once data are available, attention shifts to structure, quality, modelling choices and interpretation. At the manuscript stage, a review can examine whether reported methods and results accurately describe the work completed.
These are different engagements. A review of output tables does not include rebuilding the analysis from raw data unless that work is agreed. A written analysis plan does not include collecting observations. By separating planning, execution and review, the scope can focus on the point where additional support will make the most useful contribution.
The service suits individual researchers and teams with a defined empirical question. For a project whose design is still undecided, research design support may be the appropriate starting point. For departments trying to improve staff capability, methods and statistics training offers a knowledge-transfer route.
02
Establish whether the data can answer the question
A dataset needs context. Variable names alone rarely explain how participants were recruited, how outcomes were recorded or why some observations are missing. An assessment can examine the unit of analysis, coding, measurement scales, repeated observations, grouping, eligibility rules and the relationship between the available sample and the population discussed in the paper.
Data preparation should use explicit rules. Potential duplicates, inconsistent values, implausible dates and out-of-range responses need investigation with the research team. Correcting a documented entry error is different from excluding a valid observation because it complicates the result. Where exclusion or recoding is justified, the reason should be recorded and the original source retained according to the agreed data-management arrangements.
Missing data also require explanation. The scope can include a missingness summary, examination of plausible reasons and comparison of suitable approaches. No method can automatically remove all bias caused by missing information. The analysis report should describe the assumptions made and the implications for interpretation.
03
Choose methods and report uncertainty clearly
Method selection depends on the question, design, variable structure and assumptions. A project may require descriptive analysis, comparison of groups, regression, longitudinal methods or another specialist approach. The proposed method must be assessed for the particular problem; a software menu or a technique used in a neighbouring paper is not enough to establish suitability.
The analysis can examine effect estimates, uncertainty intervals, model assumptions and sensitivity to important choices. The American Statistical Association's guidance on p-values explains why statistical significance alone does not establish an effect's magnitude or practical importance. Results should be interpreted with the study context and limitations in view.
If the analysis has several outcomes or many potential comparisons, the plan should explain how those will be handled. Exploratory work can be valuable when it is identified transparently. Searching repeatedly for a preferred result and presenting only the successful attempt would misrepresent the analytical process.
04
Define the analytical outputs you need
A scoped engagement can produce:
Where agreed and technically appropriate, the handover can also include code or syntax, a record of software versions and instructions for reproducing outputs. File formats, software and ownership arrangements should be settled during scoping.
A report should distinguish what was done from what the results mean. The methods account records procedures and decisions; the results communicate estimates and comparisons; interpretation connects those findings with the question. Recommendations for additional work should remain identifiable as recommendations rather than appearing as completed analyses.
Illustrative deliverable: an analysis-plan extract
Imagine a study examining the association between mentoring participation and research confidence among doctoral researchers. A hypothetical analysis-plan extract would define the confidence measure, the comparison of interest, relevant background variables, treatment of missing responses and the primary model. It would also state how uncertainty will be reported and why the findings should be interpreted as associations if the design does not support a causal claim.
A companion table shell could show the planned headings before any results are calculated: analysis population, estimate, uncertainty interval, sample size and explanatory note. The example contains no invented study findings. It illustrates how a deliverable can make the analysis understandable before a table of numbers arrives.
05
A documented workflow from brief to handover
The first step is to agree the question, supplied materials and required output. Where the data are not ready for analysis, a preparation stage can establish the work needed before a reliable quotation for the full analysis is possible. This avoids treating substantial data reconstruction as a minor formatting task.
The analytical plan then records the selected approach and key decisions. Researchers review substantive assumptions, such as the meaning of a variable or whether two groups can reasonably be compared. Questions discovered during analysis should be resolved and documented rather than silently answered on the team's behalf.
The results review is an opportunity to check interpretation, resolve discrepancies with source records and identify legitimate additions. A final handover can bring together the agreed analysis files and explanatory material. Revision boundaries should distinguish correction of the agreed work from new outcomes, a replacement dataset or a substantially different research question.
06
Prepare the materials for a useful consultation
Start with a brief account of the study, the main question, collection procedures and intended output. A protocol, questionnaire, data dictionary and existing analysis plan can provide important context. If the request concerns reviewer criticism, include the relevant comments and the methods or results sections to which they refer.
Before transferring a dataset, agree what is needed and how it can be shared. Participant identifiers are often unnecessary for the analytical task. De-identification also requires attention to indirect identifiers and free-text fields; removing names alone may not make a dataset anonymous. The research team should confirm its authority to share the material and any institutional restrictions.
Our guide to preparing data for statistical consultation provides a practical starting checklist. If data governance, documentation or sharing arrangements need broader work, research data-management support can be scoped separately.
07
What affects cost, timing and feasibility
The work involved depends on the number and structure of datasets, data quality, complexity of the design, specialist methods, expected outputs and amount of interpretation or writing support. A small but poorly documented dataset may require more preparation than a larger, well-structured one. Sample count alone is therefore an incomplete basis for pricing.
Timing depends on access to complete files, responses to data queries and agreement on analytical decisions. A requested deadline can be assessed after an initial review of those requirements. Any plan should include time for researchers to examine the results and confirm that the reporting matches their knowledge of the study.
Consulting cannot create observations that were never collected or guarantee a statistically significant finding. A null, uncertain or unexpected result may be the appropriate result. Substantive authorship, acknowledgements and disclosure should follow the actual contributions and applicable publication rules, as explained on the research integrity page.
08
Statistical consulting questions
Can you analyse data that have already been cleaned?
Yes, but the cleaning decisions and original variable definitions are still relevant. Provide a description of exclusions, recoding and derived measures where available. The scope can include checking those decisions or proceed from a clearly documented, client-approved analysis dataset.
Can you check whether an analysis in my paper is appropriate?
A statistical review can assess alignment between the question, design, methods and interpretation. The depth depends on access to the underlying material. Reading a manuscript can identify reporting problems, while verifying calculations usually requires data, code or sufficiently detailed analytical output.
Will I receive code as well as results?
Code or syntax can be included when it is part of the agreed scope and suitable for the working environment. Specify your required format and access needs before the project begins. A code handover should state dependencies and assumptions so that another authorised researcher can understand its use.
Can you help with a small sample?
A consultation can examine what the sample supports, appropriate summaries and the uncertainty that needs to be communicated. The right response may involve simplifying a model or narrowing the research claim. No analytical technique can make a small or unrepresentative dataset answer every question reliably.
Can additional analyses be added after the first report?
They can be considered, with their purpose and status documented. A change to the agreed question, outcomes or dataset may affect cost and timing. Analyses suggested by the observed results should be reported transparently so readers can distinguish planned work from subsequent exploration.
Is this suitable for a thesis or dissertation?
Methods coaching, analytical advice and institution-permitted support can be scoped around the researcher's learning and responsibilities. Confirm your programme's rules and supervisor expectations before commissioning work. The researcher needs to understand and be able to explain the analysis submitted for assessment.
Tell us your study design, current data stage and the analytical decision you need help with. Discuss your analysis project to establish a suitable scope.