Practice Datasets

Eighteen finishing totals from the 2019 Nathan's Famous Fourth of July hot-dog eating contest, in place order. It is the smallest file here and the only one that is not simulated. Its size suits working by hand: eighteen numbers you can sort on paper, a clear right skew, a mode that actually exists, and one competitor so far ahead of the field that every outlier rule flags him. Source: Major League Eating's published results for the men's final, 4 July 2019.

Try these

1. Work out the mean, the median and the mode by hand, then say which way the batch is skewed (§1.3 Describing Data).
2. Build the five-number summary the textbook way and apply the 1.5 × IQR rule. Then open the descriptives calculator with these eighteen values already in the box. Flipping its Quartiles switch gives you the textbook, SPSS and R/Python answers to the same question, and they are not the same answer.
3. Compute the sum of squares, the variance and the standard deviation the long way, with a column of deviations. Eighteen rows is small enough that this is a ten-minute job and not a punishment.
4. Drop the winner and redo the mean and the median. Which one barely moves?

Eighty volunteers drew their night at random: either a normal one, or one cut to four hours. The next morning each of them sat a 30-minute vigilance task. The outcome is the number of attention lapses (errors). Did sleep restriction hurt performance, and by how much?

Try these

1. Run an independent-samples t-test comparing errors across the two conditions.
2. Report and interpret Cohen's d: the study was built around a medium effect (check it with the effect-size converter).
3. Before trusting the p-value, check the assumptions (normality, equal variances).

Three groups of 35 students revised for the same exam using one of three techniques (rereading, flashcards, or practice testing), then sat the exam (scored 0–100). Which method works best? The spread also differs by group, so this doubles as practice in checking equal variances.

Try these

1. Run a one-way ANOVA across the three methods.
2. Check equal variances (Levene's test). The SDs are deliberately unequal.
3. Follow up with post-hoc tests (e.g. Tukey) to see which methods actually differ.

Sixty people reported their average daily screen time and rated their sleep quality (0–10). There is a negative relationship in the data, but one person (row 60) reports 11.5 hours of screen time and excellent sleep, far off to the right of everyone else. The dataset shows what one influential point can do.

Try these

1. Plot it, then compute Pearson's r. Always look before you trust a coefficient.
2. Fit a simple regression and inspect leverage & influence (Cook's distance) for row 60.
3. Recompute r without that point, then decide what to do with it, and report both.

A classic 2×2 memory experiment: participants encoded a word list either shallowly or deeply, then were tested in the same room or a different one (25 per cell). The number of words recalled shows an interaction: context matters far more when encoding was deep.

Try these

1. Run a two-way factorial ANOVA. You want the encoding × context interaction.
2. Draw the interaction plot of the four cell means (non-parallel lines = interaction).
3. Put the interaction into words: the effect of context depends on encoding depth.

An 8-item wellbeing questionnaire (each item 1–5). Two items are negatively worded and must be reverse-coded before scoring: q3 and q6. A few cells are blank (real surveys have missing data). Practice reverse-coding, building a composite, and checking the scale's reliability.

Try these

1. Reverse-code q3 and q6 (new = 6 − old), then compute a composite (mean of the 8 items).
2. Assess the scale with Cronbach's α.
3. Decide how to handle the missing values (listwise vs pairwise) and note how many cases each keeps.

A deliberately dirty clinic intake sheet, built for the Data course. It hides duplicates, impossible values, a unit mix-up, inconsistent category spellings, and three different date formats. Your job is to find every problem before you analyze anything, and to never edit the raw file, only a cleaning script.

Try these

1. Find and document every data-quality problem. How many can you spot before reading the solution?
2. Build a reproducible cleaning workflow: de-duplicate, standardize sex, fix the units, parse the dates.
3. Write the validation rules (allowed ranges, allowed categories) that would have caught each error at entry.

Sixty athletes were measured on a strength lift (1RM, kg) at weeks 0, 6, and 12, split between a training program and a control group. The data is in wide format (one row per athlete, one column per timepoint), perfect for practicing reshaping and repeated-measures analysis.

Try these

1. Pivot wide → long so each row is one athlete-at-one-time (R/JASP usually want long).
2. Run a repeated-measures / mixed ANOVA with time (within) and group (between).
3. Interpret the group × time interaction: does the program gain more than the control?

Records for 160 graduate applicants: undergraduate GPA, a GRE-style test score, whether they had research experience, and the binary outcome: admitted or not. A binary outcome means logistic regression, not linear.

Try these

1. Fit a logistic regression: admitted ~ gpa + gre_score + research.
2. Interpret the odds ratios (convert them from the coefficients; the effect-size tool handles OR too).
3. As a warm-up, compare admit rates by research with a chi-square test on the 2×2 table.

A hundred and fifty students rated their mood on arrival at campus, having walked or cycled, taken public transport, or driven. Three conditions and a continuous outcome make it an ANOVA, but students chose their own commute mode, and the two extra columns show why that matters. Walkers live 2 km from campus and drivers live 10 km away. This is the file for practicing an analysis with control variables, and for watching a group difference shrink by nearly half when you add them.

Try these

1. Run a one-way ANOVA of mood on mode, then check equal variances with Levene's test.
2. Follow up with the planned contrast comparing active travel against the average of the two motorized modes, coefficients [1, −½, −½].
3. Refit it as a regression with dummy coding, confirm R² equals η², then add both covariates and read the coefficients as partial (Multiple Regression).
4. Check VIF and the residual plot, and explain the change in the mode coefficients from the covariate imbalance in step 1.