Scientific Analysis - Category Home
Scientific Analysis - Topics
Handling missing values without quietly inventing them
Data cleaning decisions that quietly change your conclusions
My analysis ran fine on my laptop and broke on a co-author's machine
What metadata does a dataset need before anyone else can reanalyze it?
Differential privacy and the trouble with exact counts in shared datasets
Genetic evidence vs epidemiological evidence in the same causal claim
Metadata standards: why your dataset is useless without them
Subgroup analyses that look convincing and mean nothing
Interpreting an effect size that is statistically significant but tiny
Reusing someone else's dataset only to find the column labels do not match the paper
Two meta-analyses of the same question reaching opposite conclusions
Same data and same method but a different software version flips significance
Reading a forest plot without fooling yourself
Observational data vs a randomized trial on the same exposure
How much of a manuscript's statistics can a non-statistician referee check?
Software choices that change your numbers: Excel vs R vs Python
Can reviewers really judge statistics without the raw data?
Reporting an analysis so someone else can redo it
Surrogate endpoints that looked good on paper and failed in patients
What a chi-square test means when half the cells are empty
Why fixed-effects and random-effects meta-analyses disagree on the same trials
Why a percentage change hides the size of the base it came from
What evidence does a regulator need before a drug goes to market?
Small multiples vs animation for showing change over time
Bayesian confirmation runs into the old evidence problem the moment you try it on history
Analyzing survey data with weights so the sample represents the population
What is lost when a continuous variable is split at the median
How do you handle a result that is statistically significant but scientifically trivial?
Why counting people rather than events changes the answer
Cost-benefit analysis as evidence: what gets counted and what gets ignored
Choosing between a line chart and a scatter plot for time-ordered measurements
Exploratory analysis versus confirmatory testing on the same dataset