Overview
Many quantitative problems that surface at the analysis stage are really design problems: a survey that cannot support the causal claim in the title, a comparison group that differs from the treatment group in obvious ways, a key variable measured with a single untested item, or a sample too small to detect a realistic effect. None of these can be fixed with statistics afterwards.
We work through your design from question to analysis plan. That means clarifying what kind of claim you want to make (descriptive, associational or causal), selecting the design family that can support it, defining variables and how each will be measured, identifying threats to internal, external, construct and statistical conclusion validity, and matching the sampling strategy, sample size and statistical test to your hypotheses.
The output is a design you can explain and defend: in your proposal, your ethics application and your methodology chapter. If you are pre-registering, we help you turn it into a pre-registration on OSF or AsPredicted.
Who it’s for and the problems it solves
This is for you if you are…
- Undergraduate and master's students planning a quantitative dissertation
- Doctoral researchers preparing a proposal or confirmation review
- Researchers in education, health, psychology and business designing interventions or evaluations
- Anyone whose ethics committee or supervisor questioned the design or sample size
Common problems we help with
- Your title says 'the effect of' but your plan is a one-off survey
- You cannot randomize, and do not know which quasi-experimental design is credible
- You are not sure how to measure a construct or whether to use an existing scale
- Your supervisor asked about confounders, control variables or validity threats
- You need a sample-size justification for your proposal or ethics form
What’s included — and what isn’t
Included
- Clarifying the research question and the type of claim it requires, with testable hypotheses where appropriate
- Choosing a design family and specific design, with a written justification and alternatives considered
- Experimental designs: between-subjects, within-subjects, factorial, pretest–posttest control group, cluster-randomized and crossover designs
- Quasi-experimental designs: nonequivalent control group, interrupted time series, regression discontinuity and difference-in-differences logic
- Correlational and predictive designs, including cross-sectional versus longitudinal (panel or cohort) options
- Survey designs: cross-sectional and repeated surveys, mode choice, and how questionnaire design links to your constructs
- Operationalization: defining independent, dependent, mediating, moderating and control variables and choosing validated instruments
- A validity threats table with the mitigation for each threat (randomization, blinding, matching, covariates, attention checks)
- Sampling strategy and a sample-size or power justification
- A pre-specified analysis plan linking each hypothesis to a statistical test
Not included
- Writing your proposal or methodology chapter for you to submit as your own work
- Designing studies intended to produce a predetermined result
- Formal ethics approval, which only your institution's committee can give
- Legal or regulatory advice on clinical trials; we support design, and regulatory requirements must be confirmed with your sponsor or authority
What you receive
- Design brief
- A concise document stating the question, the design chosen and why, the alternatives considered and the claims the design will and will not support.
- Variables and measures table
- Each variable with its conceptual definition, operational definition, instrument, level of measurement and source.
- Validity threats table
- Plausible threats to internal, external, construct and statistical conclusion validity with the mitigation planned for each.
- Sample-size justification
- The calculation or simulation behind your target sample, with the assumptions stated so a reviewer can check them.
- Analysis plan
- Hypothesis-by-hypothesis mapping to statistical tests, handling of missing data and planned robustness checks; convertible into a pre-registration.
How it works
Question and claim
We start from what you want to be able to say at the end and the constraints you face (access, time, ethics, budget).
Design options
We set out two or three feasible designs with their strengths and weaknesses and agree one with you.
Measures and threats
We define variables, select instruments and plan how to handle the main validity threats.
Sample and analysis
We justify the sample size and write the analysis plan, then review the complete design with you.
Design families and when they fit
- Randomized experiments
- The strongest design for causal claims when you can control assignment. Trade-offs include artificial settings, attrition and ethical limits on withholding treatment.
- Quasi-experiments
- Used when randomization is impossible. Credibility depends on a good comparison group, pre-intervention data and explicit handling of selection bias.
- Correlational designs
- Suited to describing relationships and prediction. They cannot establish causation, and conclusions must be worded accordingly; longitudinal data helps with temporal order.
- Survey research
- Efficient for describing populations and testing models of relationships between constructs. Quality depends on sampling frame, response rate and measurement validity.
- Validity frameworks
- We use the Shadish, Cook and Campbell typology of internal, external, construct and statistical conclusion validity to structure threats and mitigations.
- Reporting guidelines
- CONSORT for randomized trials, TREND for non-randomized evaluations and STROBE for observational studies, used at the design stage so you collect what you will later need to report.
- Pre-registration
- OSF Registries and AsPredicted templates for confirmatory studies, clinical trial registries where required.
Matching the claim to the design
The words in your research question commit you to a design. 'Is there a relationship between' can be answered by a correlational survey. 'Does X improve Y' implies a causal claim that needs a comparison, ideally randomized, and control over timing. 'What proportion of' asks for a descriptive design with a representative sample. We check your wording against your plan and adjust one or the other.
- Describe → descriptive survey with a defined sampling frame
- Associate or predict → correlational or cross-sectional design
- Explain a cause → randomized or strong quasi-experimental design
- Track change → longitudinal panel or repeated measures
Trade-offs we will make explicit
Every design trades something. Within-subjects designs need fewer participants but risk carryover and practice effects. Cluster randomization matches how schools or clinics work but needs larger samples because of intraclass correlation. Online surveys reach people cheaply but make it harder to know who responded. We set these trade-offs out so you can choose knowingly and write a candid limitations section.
Planning the sample and the analysis together
Sample size follows from the design and the analysis, not the other way round. A 2×2 factorial ANOVA, a multiple regression with eight predictors and a multilevel model with pupils nested in classes all need different calculations. For anything beyond simple cases we use our dedicated sampling and power analysis support, so the numbers in your proposal hold up.
What the work can look like
Illustrative example — fictional data
Illustrative example: peer tutoring and first-year statistics performance
A fictional master's student wanted to test whether a peer tutoring program improves first-year statistics results. Students choose to attend, so random assignment is impossible. Details are invented.
Design chosen
A nonequivalent comparison group design with a pretest: attendees versus non-attendees in the same course, with prior mathematics grade and a baseline statistics anxiety score measured in week one. Self-selection was identified as the main threat to internal validity.
Mitigation and analysis
The analysis plan specified ANCOVA adjusting for baseline scores, with propensity score weighting as a robustness check. Conclusions were worded as 'associated with higher scores after adjusting for prior attainment', not as proof that tutoring caused the difference.
Illustrative example with invented details; not a real client project.
What you’ll need to provide
- Your research topic, question or draft hypotheses
- Proposal guidelines, supervisor feedback or ethics committee comments
- Practical constraints: access to participants, setting, time frame and whether randomization is possible
- Any instruments or scales you are considering
- Relevant prior studies, especially ones reporting effect sizes
You don’t need everything ready to ask for a quote — tell us what you have and we’ll say what’s missing.
What affects price and turnaround
Price depends on
- Complexity of the design (number of groups, factors, time points or levels)
- Whether instrument selection or adaptation is needed
- Depth of the sample-size work (formula-based versus simulation)
- Whether a pre-registration or trial registration draft is included
- Number of review rounds with you and your supervisor
Turnaround depends on
- How settled your research question is at the start
- Availability of prior effect sizes or pilot data
- Time needed for your supervisor or team to agree design choices
Every project is quoted individually. How we price our work.
Questions people ask before starting
Can I make causal claims from a survey?
Generally not strong ones. Cross-sectional surveys show associations. Longitudinal data, natural experiments or causal inference methods can strengthen the argument, but we help you word claims so they match what the design can support.
Should I use an existing scale or write my own questions?
Use a validated scale where one fits your construct and population, as it gives you evidence of reliability and validity. If you adapt or translate one, plan for pilot testing and possibly factor analysis.
My supervisor wants a different design. What happens then?
We explain the trade-offs of both options in writing so you can discuss them. Your supervisor's decision prevails, and we help you implement and justify the design you agree on.
Do you also run the analysis later?
We can, through our data analysis services, and planning both together avoids mismatches. You can also take the analysis plan and run it yourself.
Is this suitable for clinical trials?
We can help with design logic, outcome selection and sample size for trials. Regulatory, sponsorship and registration obligations must be confirmed with your institution and the relevant authority.