Services / Independent statistical review
Have your analysis checked before someone else checks it
A statistician with no stake in the result goes through analysis you have already done and reports what it will and will not support. Organisations use it before submitting to a journal, before a board acts on an internal figure, and whenever a number has to survive someone looking for the flaw in it.
- Five working days from your documents arriving
- Findings ranked by severity, not a list of nitpicks
- No participant-level data needed at this stage
- One round of follow-up questions included
A qualified statistician with no stake in the outcome examines analysis that has already been carried out and reports what the data does and does not support. The point is independence: a reviewer who had no hand in producing the result has no reason to defend it. Most of what we find is not arithmetic. It is a sample treated as larger than it really is, a comparison that was never pre-specified, or a conclusion that reaches further than the design allows.
What gets checked
Six places results usually fail
These are the questions a statistical reviewer at a journal will ask, worked through before they get the chance.
Unit of analysis
Whether observations were treated as independent when they are not. Repeated measures, patients within wards, pupils within schools. The most common fault we see, and the one that most often invalidates the uncertainty around every result in the paper.
Missing and uneven data
Whether the pattern of missingness and the timing of observations can be treated as unrelated to the outcome, and what the analysis would have to assume for the current findings to hold.
Confounding and comparison
What can honestly be attributed to what, given who is in the sample, how they got there and whether any comparison group exists. Self-selection is checked here.
Effect estimates
Whether the quantities reported are the ones the conclusions actually rest on, and whether their precision is stated correctly rather than optimistically.
Multiplicity
How many comparisons the work is effectively making once exploratory analyses are counted, and what that does to the headline findings.
Reproducibility
Whether each headline result can be reproduced from the methods as documented, what would be needed to reproduce it, and anything that does not add up on the face of the outputs.
Who it is for
When organisations ask for one
- Before submitting an observational study to a peer-reviewed journal
- A reviewer has questioned the statistics and a considered response is needed
- Real-world evidence from an app, registry or wearable, before it goes public
- Analysis inherited from a departed colleague or an outside agency
- A board paper or business case built on an internal analysis
- Evaluation evidence a funder or commissioner will scrutinise
- A figure that has been challenged and needs an impartial opinion
- Due diligence on analysis supplied by a third party
What you get
A report you can act on
Ranked by severity
What needs addressing before submission, what is better disclosed as a stated limitation, and what is already sound. Not an undifferentiated list of comments.
What the data supports
A plain statement of the claims this dataset can carry into print, and the ones it cannot, so you know where the line is before a reviewer draws it for you.
The objections to expect
The specific challenges a reviewer, editor or board member is likely to raise, with what would answer each one.
Scope and price
Two stages, and most stop at the first
Stage one · from £650
The methodological audit, desk-based. Send a methods and analysis summary, a data dictionary and your aggregate outputs. Five working days, fixed price, no participant-level data involved. For most organisations this is the whole engagement.
Stage two · from £950
Reproducing a shortlist of headline results from participant-level data, under a data-sharing agreement. Quoted once the data dictionary has been seen, because the work depends on how the data is structured.
No obligation either way
Stage one stands on its own and often shows that stage two is unnecessary. Every price is agreed in writing before anything starts, and it does not move afterwards.
Governance
How your data is handled
Nothing identifiable at stage one
The first-stage audit works from summaries, a data dictionary and aggregate outputs. No participant-level records are needed and none should be sent.
An agreement comes first
If results are to be reproduced we put a data-sharing agreement in place before anything is transferred. For health research this is special category data and it is treated as such. Our full data governance position is published.
Held only as long as needed
Material is kept for the engagement and deleted within ninety days of completion. We are glad to work to your own agreement or to supply a draft.
Independence
We ask to be acknowledged, not credited
If the work supports a publication we would want to be named in the acknowledgements as independent reviewers, with the scope described accurately, and we would decline co-authorship if it were offered. An author has a stake in the paper being accepted, which is precisely the stake a reviewer must not have. Editors understand the distinction, and an acknowledged independent review carries more weight than an extra name on the byline.
How it works
Three steps, no calls
Send what you have
The methods and analysis summary, the data dictionary, the outputs, and the headline results that matter most if they do not hold.
Approve a fixed quote
Usually within one working day, stating the scope, the deliverable and the date you will have it. Half on acceptance, half on delivery.
Receive the report
Five working days later, with findings ranked by severity and one round of follow-up questions included at no extra cost.
Which do you need
Review, or a statistician on the project?
You have results already
The analysis exists and you want it checked before it goes further. That is this page, from £650.
You have a question, not an answer
Nothing has been analysed yet, or a funder wants a named statistician with a stated role on the bid. That is the independent statistician service, from £950.
You are not sure which
Describe the situation in the quote form and we will tell you which one fits, including when the answer is neither.
Common questions
Answers before you ask
What is an independent statistical review?
A qualified statistician with no stake in the outcome examines analysis that has already been carried out and reports what the data does and does not support. It is used before submitting to a journal, before a board acts on an internal analysis, and where a result has to withstand external scrutiny.
How much does a statistical review cost?
The first-stage desk review is fixed at £650 and takes five working days from the documents arriving. Reproducing selected results from participant-level data is quoted separately, usually between £950 and £1,450 depending on how the data is structured. Every price is agreed in writing before any work starts.
Do you need our raw data?
Not for the first stage. A methods and analysis summary, a data dictionary and your aggregate outputs are enough to audit the methodology. Participant-level data is only needed if you want results reproduced, and we would not accept it without a data-sharing agreement in place first.
What do we receive?
A written report with every finding ranked by severity: what needs addressing before submission, what is better disclosed as a limitation, and what is already sound. It closes with a plain statement of what the dataset can and cannot support as a claim, and the objections a reviewer or editor is likely to raise. One round of follow-up questions is included.
Will you be named as an author on our paper?
No, and we would decline if asked. Co-authorship removes the independence that makes the review worth having. We are glad to be acknowledged as independent reviewers, with the scope of the review described accurately, which carries more weight with an editor.
What is the difference between this and an independent statistician?
A review checks work that already exists. An independent statistician is engaged to design or carry out the work, and can be named on a funding bid with a stated role. If you have results and want them checked, you want a review. If you have a question and no analysis yet, you want a statistician.
How long does it take?
Five working days from the documents arriving. We quote within one working day of your enquiry, and no call is required at any stage.
Keep reading
Related
Clustered data and the unit of analysis
The fault we find most often, and how to avoid it before anyone reviews you.
Effective sample size calculator
Find out what your dataset is really worth in two minutes, free.
Independent statistician
For projects that need a named statistician on the bid or the analysis designed from scratch.
Send what you have. Get a fixed quote.
Describe the analysis you want checked and what it has to survive. A statistician replies with a price and a date, usually within one working day.