- Home
- /
- By your needs
- /
- Design a study
A study is won or lost before the first data point is collected.
Later analysis cannot fully make up for a poorly framed design. This is the statistical groundwork that comes first: framing the question properly, sizing the sample, choosing how to measure and anticipating the difficult designs before you start.
Most of the problems that surface when the data are analysed do not originate in the analysis: they originate in the design. An insufficient sample size, an instrument that does not measure what it is believed to measure, a selection bias that was already present in the sampling. When these flaws are detected once the data have been collected, they often cannot be fully corrected and have to be accepted as limitations of the study.
That earlier phase decides whether the study will be able to answer its question. It is not paperwork for the research ethics committee: it determines how far the conclusions can be supported and which limitations will come with the study. Statistical work done well at this stage can prevent rework and strengthen the validity of everything that follows.
It covers three strands, depending on where you are: framing the study and its protocol, deciding how to measure and whom to sample, and resolving designs that fall outside the standard (interventions, small samples, Bayesian approaches). You can start with whichever applies to you.
This is for you if…
You are applying for funding or for approval from a committee and need to justify the sample size and the design with statistical judgement, not with a rough guess.
Your research question is clear, but you don't know which design answers it best, or how many subjects you need to detect the effect you are looking for.
You are going to measure something that cannot be measured directly (quality of life, risk perception, adherence) and you need a validated instrument, not an improvised scale.
Your study is out of the ordinary: a sample that is going to be small, an intervention to evaluate, a context where the classical frequentist approach falls short.
How we can help
Frame the study and its protocol
The backbone of the study: which design answers your question, what sample size you need to detect the expected effect, which analysis is planned in advance. The statistical plan that a committee or a funder expects to see, written with the rigour it demands and traceable down to each decision.
View service →The statistical design of academic and pragmatic trials: randomisation, power calculation, endpoint definition and analysis plan. Our ground is the research trial, not formal regulatory submission to the EMA or the FDA, which is a separate field.
View service →Define how to measure and sample
When what you want to measure cannot be measured with a device (perception, attitude, quality of life), the instrument has to be validated before you can trust its results. Reliability, item analysis, internal consistency, factor structure: gathering evidence that the scale measures the intended construct reliably and validly.
View service →The sampling design that makes your survey represent the population you are interested in, not just the people who responded. Stratified, cluster and design-weighted sampling: an appropriate sampling design makes it possible to obtain defensible population estimates, provided that selection, data collection and weighting are carried out correctly.
View service →Resolve special designs, interventions or small samples
When the sample is limited and there is defensible prior information, or when you want to express the results as direct probabilities about the parameters. The Bayesian approach is the right tool for specific scenarios where the classical frequentist approach falls short, and it is proposed when the problem calls for it, not by default.
View service →The statistical planning and analysis of studies that evaluate a nutrition intervention: how to structure the comparison, define the variables, control for confounders and estimate the effect. The statistical work of the study; the intervention is defined and delivered by the research team.
View service →Not sure which of these fits your situation?
Tell us what stage you are at and we will tell you what work is needed, or whether none is. No commitment, and to the point.
Let's talk about your studyHow we have approached it in other projects
Before you take the step
Do I need to have the statistical design decided before getting in touch, or can I come with just the research question?
Coming with the research question is enough; deciding the design is part of the work. This is for anyone who is clear about what they want to find out but not about which design answers it best or how many subjects they need. You don't need to arrive with the type of study, the measurement scale or the analysis method already chosen: those decisions are made from the question, the context and the real constraints of the project (time, access to the population, budget). What does help is to bring the question as tightly defined as possible and an idea of the population and the outcome of interest. If the question is still vague, narrowing it down is the first step of the design, not a prerequisite.
Is it too early to get in touch if I haven't collected any data yet?
Quite the opposite: this is exactly the right moment for this work. Designing a study is the statistical phase before the first data point is collected, and most of the limitations that appear at the analysis stage (an insufficient sample, an instrument that does not measure what it is believed to measure, a bias already present in the sampling) cannot be corrected afterwards: they are carried forward as limits of the study. Getting in touch before collecting data makes it possible to size the sample, choose how to measure and anticipate the difficult designs while those decisions can still be made, which strengthens the validity of everything that follows. If what you need now is to analyse a study that has already closed, the place to start is **I have data and don't know where to start**; designing a new phase, a follow-up or a validation does belong here, even if you already have earlier data.
My study needs several things at once (sizing the sample, validating a scale, an unusual design). Are they handled separately or as a single piece of work?
As a single piece of design work, even if it combines several services. The statistical groundwork is organised into three strands (framing the study and its protocol, deciding how to measure and whom to sample, and resolving designs that fall outside the standard), and a single project often touches more than one. Sizing the sample, validating a measurement instrument and choosing the approach for a small-sample design are decisions that constrain one another: the sample size depends on how things are measured, and the planned analysis method depends on the design. That is why they are planned coherently, not as separate engagements that later fail to fit together. You don't have to commission all three strands: you start with the one that matches where the study is.
Another team has already prepared a design or a protocol. Can you review it instead of starting from scratch?
Yes: reviewing a design that has already been drafted is as valid a starting point as starting from scratch. It fits when another team has prepared the protocol, the sample size calculation or the analysis plan and you want an independent statistical view before applying for funding, going to a committee or starting data collection. The review assesses whether the design answers the question, whether the sample size is sufficient to detect the effect you are looking for, whether the measurement instrument is validated and whether the assumptions hold. It is not about redoing the other team's work, but about checking where the design can be strengthened while it can still be changed. If what you already have is an analysis carried out on collected data, that review is the methodological second opinion, within **I have data and don't know where to start**.
What if I know my sample is going to be small, or that my study is out of the ordinary?
A study with a necessarily small sample or an unusual design is not a lost cause: it is precisely one of the strands this work covers. It arises when the available population is limited (a rare disease, a very specific subgroup), when there is an intervention to evaluate, or when the classical frequentist approach falls short and there is defensible prior information to draw on. In those scenarios, the design is planned with the tools suited to the problem, including the Bayesian approach when the case calls for it, not by default. What we don't do is force a standard design onto a problem that is not standard. The honest limit: with very few observations, a good design limits the damage and clarifies what can be claimed, but it does not create statistical power that the sample cannot provide.
I'm applying for funding or for approval from a committee. What do you need from me to justify the design?
To justify the design to a committee or a funder, what is needed above all is a clear research question and the context of the study: target population, outcome of interest, access constraints and an idea of the effect considered relevant to detect. This applies to anyone who has to defend the sample size and the design in writing with statistical judgement, not with a rough guess. With that, the work produces the statistical plan these bodies expect to see: the sample size calculation with its reasoning, the justified design and the analysis planned in advance, traceable down to each decision. You don't need to bring it worked out (working it out is the job), but the sharper the question and the relevant effect, the more solid the justification will be.
Designing well is the beginning. Once the study is under way and the data start coming in, the natural next step is to analyse them: see how we can help if you already have the data and don't know where to start. And if you are still deciding whether your question needs this groundwork, let's talk before you begin.
Shall we talk?
Tell us what you are working on and we will give you a straight answer on how we can help.