Statistics that support your claim.
Study design, sample size, hypothesis testing, regression and survival analysis for medical research. Work in SPSS, R or STATA; output is publication-ready with documented assumptions.
What a statistical analysis involves
Statistical analysis sits between the data collection and the claim you want to make. Done well, it tells you whether the data actually supports the claim, and to what degree; done poorly, it produces a p-value that no reviewer takes seriously. The gap is usually not in the choice of test but in the things that happen before and after - checking assumptions, handling missing data, adjusting for multiple comparisons, documenting the analysis so someone else can reproduce it.
We work with clinical researchers, postgraduates and institutional teams on analyses that range from a single-hypothesis t-test through to multi-variable survival models and systematic-review meta-analyses. Every analysis starts with a statistical analysis plan (SAP) agreed before we touch the data. Every model is checked for assumption violations. Every output is traceable back to a script, not to a button-click.
Our statisticians work in SPSS, R and STATA, with smaller jobs sometimes in JASP or Jamovi for teaching contexts. The software choice is driven by what the project needs, what the journal expects, and what your team can maintain afterwards.
Analyses we handle
- Descriptive statistics and data cleaning (missingness, outliers, transformations)
- Parametric tests: t-test, paired t-test, ANOVA, ANCOVA, repeated-measures ANOVA
- Non-parametric tests: Mann-Whitney U, Wilcoxon signed-rank, Kruskal-Wallis, Friedman
- Categorical data: Chi-square, Fisher's exact, McNemar, Cochran's Q
- Correlation and regression: Pearson, Spearman, linear regression, logistic regression, multinomial, ordinal
- Time-to-event analysis: Kaplan-Meier, log-rank test, Cox proportional hazards
- Diagnostic accuracy: sensitivity, specificity, PPV, NPV, AUC / ROC analysis
- Sample size and power calculation for planned studies (G*Power, PS, pwr package)
- Mixed-effects and hierarchical models for clustered or longitudinal data
- Meta-analysis (fixed and random effects, heterogeneity, subgroup, meta-regression)
- Propensity-score matching and IPTW for observational comparative studies
- Missing-data strategies: complete-case, multiple imputation (MICE, mi)
SPSS, R or STATA.
Different jobs, different tools. We pick by study design and the reproducibility your supervisor or journal expects, not by vendor preference.
| SPSS | R | STATA | |
|---|---|---|---|
| When we reach for it | Standard clinical tests under supervisor-driven workflows | Reproducible pipelines, meta-analysis, mixed-effects models | Epidemiology, survival, longitudinal or panel data |
| Reproducibility | Syntax files; menu-driven history can be lost | Full .R script plus RMarkdown report | Full .do file logging every command |
| Licence | Commercial | Free, open-source | Commercial |
How we work on an analysis
- 01
Design consult and SAP
We start with a 30-minute consult on the study design, primary and secondary outcomes, hypotheses and any planned subgroup analyses. Together we produce a statistical analysis plan (SAP) - which tests, which covariates, what counts as a primary vs secondary finding, how multiple testing will be handled. The SAP is agreed before we see the data.
- 02
Sample size and power
For studies still at the planning stage, we calculate the required sample size under defensible effect-size assumptions, typically using G*Power, PS or R's pwr package. The power calculation is documented with inputs (effect size, alpha, beta, allocation ratio) so it reproduces under audit.
- 03
Data preparation
We receive the dataset (SPSS .sav, STATA .dta, CSV or Excel), run through a data dictionary, handle missingness per the SAP, apply recodes and derived variables, and generate a cleaned working dataset. The cleaning is all in script, so every transformation is reproducible.
- 04
Model fitting and assumption checks
Each planned model is fitted; assumptions are checked (normality where relevant, homoscedasticity, linearity, proportional hazards, independence of observations). Where assumptions are violated, we apply the correct remedy - transformation, robust standard errors, a non-parametric alternative - rather than reporting the nominal test and ignoring the violation.
- 05
Reporting and replication
Outputs are written up in publication-ready tables (following CONSORT or STROBE structure), figures with proper confidence intervals, and a methods paragraph that names the software version, each test, every covariate and every sensitivity analysis. Scripts (.do, .R, .syntax) are shared so your team can re-run the analysis on updated data.
What you get
- Statistical analysis plan agreed before any analysis begins
- Sample size and power calculation document (for new studies)
- Cleaned dataset with data dictionary and transformations logged in script
- All planned analyses run with assumption checks documented
- Publication-ready tables (Table 1 baseline, main results, subgroups) in journal format
- Figures: Kaplan-Meier curves, forest plots, ROC curves, with 95% confidence intervals
- Methods paragraph citing software version, package versions and every test applied
- Reproducible scripts (.R, .do, .sps) for your team to run on updated or extended data
Standards we follow
CONSORT 2025 (replacing CONSORT 2010 in April 2025) specifies how statistical methods, sample size justification, and analysis results must be reported in randomised trials across a 30-item checklist.
Items 12, 13, 14 and 16 cover statistical methods for observational studies.
International regulatory framework for trial statistics - used in industry and increasingly in academic trials.
Transparent reporting of prediction model studies (development and validation).
Royal Statistical Society guidance on analysis reproducibility and pre-registration.
Common questions about statistical analysis
Should I ask for SPSS, R or STATA?
SPSS is the standard in Indian medical postgraduate work - most university statistics departments teach it, IBM licences are common, and supervisors expect SPSS output. R is the standard in methodological and epidemiology work where reproducibility matters; almost all meta-analysis, survival modelling and advanced statistics in medical journals is done in R. STATA sits between the two, dominant in health economics and trials. For a thesis we default to SPSS unless your supervisor prefers otherwise; for a systematic review meta-analysis we default to R. The output is identical - the software just differs in what your team can maintain.
What sample size do I need for my study?
It depends on the primary outcome, the effect size you expect (or consider clinically meaningful), alpha (typically 0.05, two-sided), desired power (0.80 is a floor; 0.90 is better), and the design - two-arm, repeated measures, survival, cluster-randomised. Underpowered studies are the single biggest cause of wasted postgraduate research. We run the calculation in G*Power or R's pwr package, document every input, and often recommend a 10-15% buffer for dropout. If your design has multiple primary outcomes or subgroup-specific hypotheses, the calculation has to account for that.
I have a lot of missing data. What should I do?
Missing data is one of the most mis-handled parts of medical statistics. Complete-case analysis (dropping anyone with a missing value) is only defensible if missingness is demonstrably random and rare; otherwise it biases results. For most thesis-scale datasets with under 10% missingness per variable we recommend multiple imputation using chained equations (MICE in R, mi in STATA), with sensitivity analysis under different assumptions. For structured missingness (e.g., loss-to-follow-up in trials) we document the pattern and apply the strategy CONSORT expects.
How do you handle multiple testing?
Every additional test inflates the probability of at least one false positive. For pre-specified primary and secondary outcomes, we apply a correction (Bonferroni, Holm, or false-discovery rate per Benjamini-Hochberg) based on how many hypotheses are formally tested. For exploratory subgroup analyses we report them as exploratory, do not claim statistical significance, and note in the methods and limitations that they are hypothesis-generating. Dropping the correction because it 'makes the results weaker' is the fastest way to get a methods-section critique from a reviewer.
Can your analysis be re-run if I get new data later?
Yes - every analysis is written as a script (.R, .do, .sps syntax), not as point-and-click output. You keep the script. When data arrives later, your team or we re-run the script against the new dataset and regenerate the tables and figures automatically. This is one of the clearest signals of a rigorous analysis; it is also the main reason journals increasingly ask for analysis code at submission.
You might also need
Medical Thesis & Dissertation
We support MD, MS, DNB and DM candidates through every stage of the thesis - synopsis for the ethics committee, methodology design, data analysis, chapter drafts, viva preparation.
Read more § WritingManuscript & Article Writing
Original research articles, case reports, reviews, editorials and letters, written for the journal you are targeting.
Read more § WritingLiterature Review
Systematic, scoping and narrative reviews for medical research.
Read moreWhat we don't do.
No ghost-authorship
Your name, your voice, your work. We do not put our names on your thesis or paper.
No publication guarantees
Peer review is not ours to promise. We make the work stronger; the journal decides.
No fixed turnaround promises
Good research takes the time it takes. We scope honestly, not optimistically.
No shortcuts
No AI paraphrasing to game similarity checks. No plagiarism. No sentence-swapping.
Talk to us about your project.
Send over an abstract, a chapter or a rough outline of what you need. We reply within one working day with a clear scope and a fixed quote, no obligation to proceed.