Overview
Descriptive statistics tell the reader who and what you studied; inferential tests tell them whether differences or associations are likely to hold beyond your sample. Both are easy to get subtly wrong. A mean reported for a heavily skewed variable, an independent t-test run on paired data, or a string of uncorrected comparisons can each undermine an otherwise careful study.
We start from your research questions and the measurement level of each variable (nominal, ordinal, interval or ratio), then choose tests on that basis rather than on what the software menu offers. For every test we check its assumptions, decide what to do when they fail — a robust alternative, a transformation or a non-parametric test — and write that decision down so you can justify it.
You receive annotated syntax, clean output and tables formatted to your style guide, together with a plain-language explanation of each result. The aim is that you can explain every number in your results chapter without looking at our notes.
Who it’s for and the problems it solves
This is for you if you are…
- Undergraduate and master's students writing a quantitative dissertation for the first time
- Doctoral researchers whose supervisor has asked for effect sizes, assumption checks or corrections for multiple comparisons
- Health, education and psychology researchers comparing groups or conditions in experimental and survey studies
- Professionals analyzing evaluation or audit data who need defensible, clearly reported tests
Common problems we help with
- You are unsure whether to use a t-test, ANOVA, Mann–Whitney U or Kruskal–Wallis for your data
- Levene's or Shapiro–Wilk tests came back significant and you do not know what to do next
- Your ANOVA is significant but you cannot tell which groups differ or which post hoc test to use
- A chi-square table has several cells with expected counts below 5
- Your results report only p-values and your supervisor keeps asking about effect sizes
What’s included — and what isn’t
Included
- Mapping each research question or hypothesis to an appropriate test, with a short rationale
- Descriptive statistics suited to each variable: means and SDs, medians and IQRs, frequencies and percentages, cross-tabulations
- Independent, paired and one-sample t-tests, including Welch's correction when variances differ
- One-way, factorial and repeated-measures ANOVA, ANCOVA and MANOVA, with planned contrasts or post hoc tests
- Chi-square tests of independence and goodness of fit, Fisher's exact test and McNemar's test
- Non-parametric alternatives: Mann–Whitney U, Wilcoxon signed-rank, Kruskal–Wallis, Friedman and Spearman's rho
- Assumption checks (normality, homogeneity of variance, sphericity, homogeneity of regression slopes) with documented decisions
- Effect sizes with confidence intervals: Cohen's d or Hedges' g, eta-squared or omega-squared, Cramér's V, rank-biserial r
- Multiple-comparison control (Bonferroni, Holm, Tukey HSD, Games–Howell, false discovery rate) where needed
- APA 7-style tables and figures, plus annotated SPSS syntax, R or Stata code
Not included
- Running tests until one is significant, or selecting tests after seeing results without disclosing it
- Writing your results or discussion chapter for submission — we explain results so you can write them in your own words
- Inventing, deleting or altering cases to change an outcome; any outlier handling is rule-based and reported
- Promising that a result will be significant or that an examiner will accept a particular test
What you receive
- Test selection table
- Each research question listed with its variables, measurement levels, chosen test, fallback test and the reason for the choice.
- Descriptive profile
- Sample characteristics and variable summaries in APA-style tables, with notes on skewness, missing data and outliers.
- Inferential results pack
- Test statistics, degrees of freedom, exact p-values, effect sizes with 95% confidence intervals and post hoc results, in tables ready to adapt.
- Assumption log
- What was checked, how (tests and plots), what was found and what was done about it, so the analysis can be defended in a viva or review.
- Syntax or script
- Commented SPSS syntax, R script or Stata do-file that reproduces every table from the raw data file.
How it works
Share questions and data
You send your research questions or hypotheses, the dataset and the questionnaire or codebook. We flag mismatches between questions and data early.
Agree the test plan
We propose a test for each question, the alpha level, how multiple comparisons will be handled and what happens if assumptions fail. You approve it before analysis.
Screen and describe
We check coding, missing values and outliers, then produce descriptive statistics and assumption diagnostics.
Run and report
We run the agreed tests, compute effect sizes and intervals, and build the tables and figures.
Explain and clarify
A walkthrough of the output, followed by a round of questions so you can write up confidently.
Tests we commonly run and how we choose between them
- Two-group comparisons
- Independent or paired t-tests (Welch's by default for unequal variances), with Mann–Whitney U or Wilcoxon signed-rank when the outcome is ordinal or badly non-normal in small samples.
- Three or more groups
- One-way or factorial ANOVA with Tukey, Games–Howell or planned contrasts; Welch's ANOVA when variances differ; Kruskal–Wallis with Dunn's tests as a rank-based option.
- Repeated measures
- Repeated-measures ANOVA with Mauchly's test and Greenhouse–Geisser correction, Friedman's test, or a mixed model when there are missing time points.
- Covariates and multiple outcomes
- ANCOVA with a check of homogeneity of regression slopes; MANOVA with Box's M, Pillai's trace and follow-up univariate tests or discriminant analysis.
- Categorical associations
- Pearson chi-square, Fisher's exact test for sparse tables, McNemar for paired proportions, with Cramér's V or odds ratios as effect sizes.
- Correlation
- Pearson's r or Spearman's rho with confidence intervals and scatterplots, and partial correlations where a control variable matters.
Assumptions: checked, not just reported
Every parametric test carries assumptions — independence of observations, approximate normality of residuals or group distributions, equal variances, and for repeated measures, sphericity. We check them with plots (Q–Q plots, boxplots, residual histograms) as well as formal tests, because in large samples Shapiro–Wilk flags trivial departures and in small samples it misses serious ones.
When an assumption fails we choose among defensible options: Welch-type corrections, bootstrapped confidence intervals, a transformation that still makes the result interpretable, or a rank-based test. We explain what each option changes about the question being answered — a Mann–Whitney U test does not compare means, for example — so your write-up describes the result accurately.
Beyond the p-value
A p-value only tells you how surprising the data would be if there were no effect. Reviewers and examiners increasingly expect the size of the effect and the precision of the estimate too. We report standardized effect sizes with 95% confidence intervals and, where your field uses them, raw mean differences in the original units, which are often easier for readers to judge.
We also help you interpret non-significant results honestly: absence of evidence is not evidence of absence, and a wide confidence interval usually signals an underpowered comparison rather than proof that groups are the same. If equivalence is what you want to claim, we can discuss equivalence testing (TOST) instead.
- Cohen's d / Hedges' g for mean differences
- Partial eta-squared or omega-squared for ANOVA designs
- Cramér's V, phi or odds ratios for contingency tables
- Rank-biserial correlation for Mann–Whitney and Wilcoxon tests
Multiple testing and other traps
Running many tests inflates the chance of a false positive. If you compare ten outcomes at alpha = .05, one spurious significant result is not surprising. We agree in advance which comparisons are confirmatory, apply Holm, Bonferroni or false-discovery-rate control where appropriate, and label exploratory results as exploratory.
Group-comparison tests describe association, not causation, unless the groups were formed by random assignment. For survey or observational data we help you word findings as differences between groups rather than effects of group membership, and suggest when a regression with controls would answer your question better.
What the work can look like
Illustrative example — fictional data
Sample excerpt: comparing three teaching formats
A master's student compared exam scores across in-person, hybrid and online sections of the same course (n = 186). All figures are invented to show how results are reported.
Assumption check
Levene's test indicated unequal variances, F(2, 183) = 4.12, p = .018, and group sizes were unbalanced (74, 61, 51). Welch's ANOVA was therefore used, with Games–Howell post hoc comparisons.
Main result
Scores differed by format, Welch's F(2, 118.4) = 6.35, p = .002, ω² = .05, 95% CI [.01, .11]. Hybrid students scored higher than online students, mean difference = 6.8 points, 95% CI [2.1, 11.5], Hedges' g = 0.52.
How to word it
Students were not randomly assigned to formats, so the write-up describes a difference between sections and discusses self-selection as a limitation, rather than claiming the hybrid format caused higher scores.
Illustrative example with invented data. Your own tables follow your variables, sample and required style guide.
What you’ll need to provide
- Research questions or hypotheses, ideally numbered
- Your dataset (SPSS .sav, Excel, CSV or Stata .dta), anonymized
- The questionnaire, codebook or description of how each variable was measured
- Any template, style guide or supervisor feedback that affects reporting
- Your preferred software and whether you want to learn to run the tests yourself
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
- Number of research questions and tests to be run
- Complexity of the design (factorial, repeated measures, covariates, multivariate outcomes)
- How much data cleaning and recoding is needed before analysis
- Whether you need a delivered analysis, a guided session or a review of your own output
- Depth of reporting support: tables only, or tables with a detailed interpretation walkthrough
Turnaround depends on
- Whether the dataset is clean and coded consistently when you send it
- How quickly the test plan can be agreed with you (and your supervisor, if needed)
- Your submission date and the number of tests involved
Every project is quoted individually. How we price our work.
Questions people ask before starting
My data are not normally distributed. Do I have to use non-parametric tests?
Not necessarily. With moderate or large samples, t-tests and ANOVA are fairly robust to non-normality, and Welch-type corrections or bootstrapping often handle the problem better than switching tests. Non-parametric tests answer a slightly different question, so we choose based on your data and what you want to conclude, and explain the choice.
Which post hoc test should I use after ANOVA?
It depends on whether variances are equal, whether group sizes differ and whether you planned specific comparisons. Tukey HSD suits equal variances; Games–Howell suits unequal variances; planned contrasts are more powerful when you had specific hypotheses before seeing the data.
Can you just check the analysis I have already run in SPSS?
Yes. Send your output and syntax (or a description of the menu choices) together with the data. We check test choice, assumptions, reporting and effect sizes, and list any corrections with reasons.
Will you write my results chapter?
No. We provide the analysis, tables and a clear explanation of each result, and we can review your draft for statistical accuracy. The chapter itself is your work, which also prepares you for questions from your examiners.
I only have a small sample. Can I still run these tests?
Often yes, but with care. Small samples reduce power and make assumption checks unreliable, so we may recommend exact tests, bootstrapped intervals or a more descriptive framing. We will tell you plainly what your sample can and cannot support.
Is my data kept confidential?
We ask for anonymized data, use it only for your project and do not share it. If your ethics approval sets specific storage or deletion rules, tell us and we will follow them.