Free tool, data and statistics
Statistical test chooser
Tell it about your variables and groups, and it names a test.
Answer at most five short questions and the chooser names the statistical test for your study, plus a fallback test for when assumptions fail.
Many scholars come here when objectives are fixed and the data sheet is half filled. The guide has only said “do the appropriate test”, and SPSS is showing forty menus. You still run the test in SPSS yourself. The chooser tells which menu to open and why, in a line you can adapt for Chapter 3.
Not sure the answer fits your study? WhatsApp us or call, and we’ll look at it with you.
How the tool works
Inside, the chooser is a small decision tree only. You see one question, you answer it, and based on that answer it decides which question to show you next.
- What do you want to find out? There are four aims to pick from. You may want to compare groups, or see if two variables are related. Some scholars want to predict an outcome, and some want to check one group against a known value.
- What kind of outcome do you have? It may be continuous (scores, amounts, a Likert scale total), it may be categories, or just yes and no.
- Different people, or the same people measured again? This one comes only when groups are compared on a continuous outcome.
- One grouping factor or two? Comes only if the groups are different people.
- Two groups or time points, or three or more?
So five questions at the most, and many times only two. The answers end in one of ten tests. Two groups of different people with a continuous outcome give the independent samples t-test, Welch’s version. Welch’s version doesn’t need equal variances in the two groups. Even if your variances are equal, you lose almost nothing by using it. Three or more groups lead to a one-way ANOVA, with post hoc tests after it. The same people measured three or more times get a repeated measures ANOVA. With two grouping factors, for example gender and district, you get a two-way ANOVA.
If the outcome is a category and you want to compare or relate, the chooser gives chi-square test of independence, and if the expected counts are small, Fisher’s exact test is the fallback. For predicting a continuous outcome it suggests linear regression. A yes-or-no outcome to predict goes to binary logistic regression.
With every result you will see a second test written, the fallback. You use that one if your data fails the assumptions. The Mann-Whitney U test is the fallback for the t-test. For the paired t-test it is the Wilcoxon signed-rank test, and the Kruskal-Wallis test is there for the one-way ANOVA. After you know your test, open our power analysis calculator to see how many respondents that test needs.
Questions scholars ask
Why does it pick Welch’s t-test and not the usual one?
The usual t-test expects the same variance in both groups. Welch’s version does not need that, and if the variances are equal anyway it loses very little power. For this reason Delacre, Lakens and Leys (2017) recommend Welch’s version as the default. In SPSS, the Independent Samples T Test output already have a row called “Equal variances not assumed”, and that row is Welch’s version.
My data isn’t normal. Do I switch to the fallback test straight away?
Not immediately. If you have 30 or more people in every group, moderate skew is usually not a big problem for t-test and ANOVA. First see the histogram and the skewness value, then decide. If the skew is very bad or the groups are very small, go for the fallback test and explain about the reason in Chapter 3.
I have two groups and a Likert scale total. Is that continuous?
Mostly it is treated as continuous. When you take the total or mean of five or more Likert items, it behaves close to a continuous score, and most theses analyse it with t-test or ANOVA. For single items analysed one by one, the fallback tests are safer.
Can I run several t-tests in place of one ANOVA?
No. Every extra t-test you run pushes up the chance of a false positive. If you have three groups, do the one-way ANOVA first. After that run post hoc tests like Tukey’s, because they control for the number of comparisons.
Is my answer recorded?
No. The chooser works inside your browser itself. Our server only logs that some test was suggested, and which test it was is not sent to us.
Stuck on a chapter? Talk it through with us.
Call or WhatsApp, say where you are in your thesis, and we’ll tell you plainly what kind of help makes sense, if any.