Guide

Analyze Your Thesis Data in JASP: Start to Finish

You have a data file, a deadline, and a supervisor who said "just run the analysis." This guide walks the whole path in JASP (free, point-and-click, and built to produce APA-style output), from opening your CSV to the finished results sentence. Every step links to the StatsCapybara lesson that explains the why, and there's a real dataset to follow along with.

Why JASP

JASP is free, open-source statistics software developed at the University of Amsterdam. For a thesis it has three real advantages: the menus are organized by analysis rather than by 40 years of accumulated dialog boxes, the output updates live as you tick options (so exploring costs nothing), and the tables it produces are already close to APA format. If your department requires SPSS instead, the same journey, with the same potholes, is covered in the companion guide From SPSS Output to APA Results.

Follow along for real: this guide uses sleep-experiment.csv from our practice datasets: 80 volunteers randomly assigned to a normal night's sleep or a night restricted to four hours, then scored on attention lapses. Download it and every number below will appear on your own screen.

Step 0: Start tidy, keep the raw file sacred

JASP can only be as good as the spreadsheet you feed it. Two rules before you even open the software:

  • One row per observation, one column per variable, one value per cell. No merged cells, no color-coding-as-data, no two variables squeezed into one column. If your file breaks these rules, fix it first. The Tidy Data lesson shows every classic sin and its repair, and if your export needs real surgery, work through Clean Your Survey Data, Step by Step before returning here.
  • Never edit the raw export. Copy it, clean the copy, and keep a note of what you changed. Your future self — and anyone auditing your thesis — needs the original to exist. The Cleaning Workflow lesson turns this into a habit.

While you're at it, make sure every cryptic column name has a line in a codebook: what it measures, its units, and how missing values are marked. Blank cells are the safest missing-value marker; numeric codes like -99 will silently poison your means.

Step 1: Import, then check the variable types

In JASP: ☰ menu → Open → Computer, and pick your .csv (JASP also reads SPSS .sav and Excel files). The data appear in a spreadsheet view.

Now the step everyone skips: check the little icon in each column header. JASP guessed each variable's measurement level, and it guesses wrong exactly when it matters: a condition coded 1/2 will be treated as a number, group averages and all. Click the icon to fix it: the ruler means scale (continuous), the ordered bars mean ordinal, the overlapping circles mean nominal. If the distinction feels fuzzy, five minutes with Types of Data will save you a semester of confusion.

For sleep-experiment.csv: condition should be nominal, errors scale, and participant_id is just an ID, so mark it nominal and nothing will ever average it.

Step 2: Descriptives first, always

Before any test: Descriptives in the ribbon, drag your outcome into Variables, and (for a group design) drag the grouping variable into Split. Under Plots, tick Distribution plots and Boxplots.

You're looking for three things:

  • Sanity. Is n what it should be? Are the minimum and maximum possible values for your measure? An age of 511 or a reaction time of 3 ms is a data-entry error, not a finding; the validation lesson shows how these creep in.
  • Shape. Roughly symmetric, or badly skewed? One hump or two? This preview tells you whether the assumption checks in Step 3 will be drama or formality.
  • Outliers. Points far beyond the boxplot whiskers deserve investigation — investigation, not silent deletion. Outliers: Detect, Investigate, Decide gives you the triage rules.

For the sleep data you should see: normal sleep M = 20.07, SD = 5.92; restricted sleep M = 22.77, SD = 6.02, with 40 people per group and no impossible values. (Want the same numbers without opening JASP? Paste a column into our descriptives calculator.)

Step 3: Check the assumptions

Every classical test makes assumptions, and JASP will check the big ones inside the analysis itself. For a t-test, tick Assumption checks to get normality tests and Levene's test for equal variances in the same output panel.

Two honest rules for reading them, both argued properly in Assumptions & When They Break:

  • Plots beat p-values. With large samples, normality tests flag harmless wobbles; with small samples, they miss disasters. Look at the Q-Q plot: points hugging the line are what "normal enough" looks like.
  • Have the fallback ready. Unequal variances? Use the Welch row, which many statisticians simply always do. Hopeless skew with a small sample? The non-parametric alternatives trade a little power for a lot of safety.

For the sleep data, the two groups' spreads are nearly identical — Levene's F(1, 78) = 0.10, p = .758 — so equal variances is comfortable here.

Step 4: Confirm you're running the right test

You probably planned this before collecting data (if not, no judgment here, but read why planning first matters). The choice comes down to your outcome type, your group structure, and whether the same people were measured more than once. Two of our tools do the deciding with you: the quick Which test should I use? chooser, and Plan My Analysis, which turns your answers into a printable plan with a sample-size estimate. The long-form reasoning lives in Choosing Statistics for Your Dissertation.

Our example is the simplest case: one continuous outcome (errors), two independent groups (condition): an independent-samples t-test.

Step 5: Run it

In JASP: T-Tests → Independent Samples T-Test. Drag errors into Dependent Variables and condition into Grouping Variable. Under Tests, tick Welch alongside Student; under Additional Statistics, tick Descriptives and Effect size (Cohen's d); tick the Assumption checks. The output appears instantly:

Step 6: Read the output like an examiner

  • t and df: the test statistic and its degrees of freedom (here 78 = 40 + 40 − 2). The df going into your write-up must match your n; examiners check.
  • p: two-tailed by default, which is almost always what you want. p = 0.047 means: if sleep restriction truly did nothing, data this extreme would turn up about 5% of the time. It is not the probability the effect is real — Hypothesis-Testing Logic untangles that knot.
  • Cohen's d: the size of the effect in standard-deviation units, and the number your discussion section actually cares about. d = 0.45 is a medium-ish effect; Effect Size & Power explains the benchmarks and their limits, and the effect-size converter translates d into overlap you can picture.

Step 7: Write it up

The pattern is: what was compared, the descriptives, then the statistics, in the past tense, with no interpretation yet (that belongs in the Discussion, says From Output to Results Section):

Participants in the restricted-sleep condition made more attention lapses (M = 22.77, SD = 6.02) than participants who slept normally (M = 20.07, SD = 5.92), t(78) = 2.02, p = .047, d = 0.45.

Mind the typography: Latin statistical symbols (t, p, d, M, SD) are italic, p gets no leading zero, and p is reported exactly — never as p = .000. The full rules live in Reporting Statistics in APA Style, and the APA results formatter will build (and sanity-check) any such sentence from your raw numbers.

When things don't go to plan

  • p = .06. Not "a trend toward significance." Report the exact p, the effect size, and its confidence interval, and interpret honestly; Writing About Non-Significant Results shows exactly how, and why a non-significant thesis is not a failed thesis.
  • Assumptions failed. Unequal variances → Welch row (already in your output). Serious non-normality with small n → the matching non-parametric test (for two independent groups, Mann–Whitney; it's one tick away in the same JASP panel).
  • An outlier is driving everything. Run the analysis with and without it, report both, and justify whatever you decide; the outlier triage lesson gives the decision rules. Never delete silently.
  • You're tempted to try "just one more" analysis. Every extra look inflates your false-positive rate. The Replication Crisis shows where that road ends; label exploratory analyses as exploratory and your integrity survives contact with the data.

The whole path, one breath: tidy file → import → fix variable types → descriptives and plots → assumption checks → the planned test → effect size → APA sentence. Eight small steps. No single one is hard; the trick is doing them in order.

Keep going