tools / statistics_in_your_browser
Statistics tools: paste, click, understand.
Paste your data, understand the result: t-test, power analysis, group comparison with boxplot & ANOVA and linear regression – computed with the same procedures as R, right in your browser. Your data are never transmitted anywhere. Results can be copied in APA format at a click.
tool_01 / t_test
t-test: do two groups differ?
Example: one department received training, one did not – do the satisfaction scores really differ, or could the difference be chance? That is exactly what the t-test answers.
How do I enter my data?
Copy one column of values per group from Excel or LimeSurvey and paste it into the field – one value per line. Comma, semicolon or space separation works too, and both decimal points and German decimal commas (e.g. ‘4,2’) are recognised.
Independent or paired – which do I choose?
Independent groups (Welch): different people in group A and group B (e.g. trained vs. not trained). The groups may differ in size.
Paired: the same people were measured twice (e.g. before/after). Line 1 in both fields must belong to the same person.
Good to know: A t-test assumes roughly normally distributed values and is sensitive to outliers; with small or skewed samples it is only a rough indication. These tools do not replace statistical consulting: they do not check, for instance, whether the method even fits your question. When in doubt ask me – a short introductory call is free of charge.
Result in APA format:
Computation: Welch two-sample t-test or paired t-test, two-sided, as in t.test() in R; effect size Cohen's d; the error bars in the chart show 95% confidence intervals.
tool_02 / power_analysis
Power analysis: how many participants do I need?
Every study starts with the sample size question: too few participants and even a real effect goes undetected; too many and you waste time and budget (and possibly run into ethics requirements). This tool computes the sample size needed for a t-test.
What is the effect size d – and which one do I choose?
Cohen's d describes how large the expected difference is relative to the spread. Cohen's rules of thumb: d = 0.2 small, d = 0.5 medium, d = 0.8 large. Without prior knowledge, d = 0.5 is a common but optimistic assumption – when in doubt, plan for a smaller effect.
What do α and power mean?
α (significance level): the accepted risk of “finding” a difference that does not exist at all – 5 % (0.05) is usual.
Power: the chance of actually detecting an effect that is really there – 80 % (0.80) is usual, and 90 % is becoming more common in research.
The curve shows how steeply the required total sample grows as the expected effect shrinks – the green dot marks your input. Computation analogous to pwr.t.test() in R (two-sided); for other designs (ANOVA, correlations, repeated measures) get in touch.
Summary to copy:
Good to know: The required sample depends heavily on the assumed effect size – estimate it too optimistically and the study is quickly underpowered. These tools do not replace statistical consulting: they do not check, for instance, whether the method even fits your question. When in doubt ask me – a short introductory call is free of charge.
Further test procedures, other effect size measures, and exact rather than approximated sample sizes. Free of charge from the Universität Düsseldorf.
Download G*Power →tool_03 / describe_groups
Describe & compare groups (boxplot + ANOVA)
You have measurements from several groups – sites, departments or cohorts, say – and want to see and compare them? Enter up to six groups here. You get descriptives, a boxplot with every individual data point, and from two groups onwards an automatic statistical comparison (one-way ANOVA).
How do I read a boxplot?
The box covers the middle 50 % of the values: its lower edge is the lower quartile (25 % of the values lie below it), its upper edge the upper quartile (75 % below). The thick line is the median – half of all values lie above it, half below. The whiskers reach to the last value within 1.5 times the box height; anything beyond that counts as an outlier. Each dot is a real observation. The black diamond marks the mean, the line through it ± 1 standard deviation.
Median and mean far apart? Then the distribution is skewed – usually a few large values are pulling the mean upwards.
What does the ANOVA do?
The one-way analysis of variance (ANOVA) tests whether the means of two or more groups differ by more than chance would explain. With exactly two groups it is equivalent to the t-test. It only tells you that there are differences – which groups differ is shown by the boxplot, or by a follow-up pairwise comparison.
Good to know: The ANOVA only tells you that groups differ somewhere – which ones is settled by a post-hoc test, and the group variances should be similar. These tools do not replace statistical consulting: they do not check, for instance, whether the method even fits your question. When in doubt ask me – a short introductory call is free of charge.
Result in APA format:
Computation: quartiles as in R (type 7), whiskers up to 1.5 × IQR; one-way ANOVA with η² as the effect size.
tool_04 / linear_regression
Linear regression: are two measures related?
Example: does confidence in using a tool grow with the number of training hours? Paste matching pairs of values – line 1 on the left belongs with line 1 on the right. You get the regression line in a scatter plot, the equation, and a plain-language interpretation.
What do b, R² and p mean?
b (slope): how much y changes on average when x increases by 1. R²: what proportion of the differences in y is “explained” by x (0–100 %). p: whether the relationship goes beyond chance. And always bear in mind: a relationship is not a cause – ice cream sales and sunburn are related too.
Good to know: Regression assumes a linear relationship – and a relationship is never proof of cause and effect. These tools do not replace statistical consulting: they do not check, for instance, whether the method even fits your question. When in doubt ask me – a short introductory call is free of charge.
Result in APA format:
Computation: simple linear regression (least squares) as in lm(y ~ x) in R, two-sided test of the slope.
more_than_tools
Your analysis is more complex?
These tools cover the simplest cases. For anything beyond that – from study design through choosing the right method to the automated result report – I am here as a researcher and as an Excel and R Markdown trainer.
Send an enquiry