What is statistics actually for?

This page exists because most stats courses start with a formula and expect you to figure out why later. The why comes first here.

The one-sentence answer

Statistics is the science of being honest about what you don’t know.

If you flip a coin three times and it comes up heads three times, two people can give you very different answers about whether the coin is fair:

  • A storyteller says “all heads, must be rigged”.

  • A statistician says “yes, all heads — but I’d see this 1 in 8 times with a fair coin, so I am not yet sure”.

The statistician’s answer is less exciting and more useful. Statistics is the discipline of keeping the second answer instead of the first one when the data is too thin to support a confident claim.

The three things every analysis must do

Whether you’re testing a drug or comparing two web designs or estimating the population of fish in a lake, the work is the same three steps:

  1. State the question precisely. “Does the drug work?” is too vague. “Is the 30-day mortality rate of patients given drug A different from those given drug B, by more than 1 percentage point?” — that’s a question.

  2. Pick an estimator. An estimator is a recipe that turns data into a single number that’s supposed to answer the question. For the drug question, the estimator might be “the difference in mean mortality between the two groups, weighted by inverse probability of treatment to handle that the groups weren’t randomly assigned”.

  3. Quantify how wrong it might be. This is the standard error, the confidence interval, the p-value. Every honest answer ends in plus or minus something. An answer without that “plus or minus” is a guess wearing a hat.

MORIE has functions for each of those three steps. The fn/ tree (10,000+ small files, each one a single formula) is the estimator vocabulary; the modules in morie.causal, morie.effects, morie.inference are the recipes that combine them.

The four kinds of question this software is good at

  1. Did X cause Y? Causal inference. IPW, propensity scores, double machine learning, instrumental variables. See Causal Inference once you’re past this track.

  2. Are these two groups actually different? Hypothesis testing. t-tests, chi-square, robust alternatives when the data is messy. MORIE Inference Engine.

  3. What’s the population value? Estimation. Means, proportions, regression coefficients, with proper confidence intervals.

  4. Where is X higher than usual? Spatial / temporal pattern detection. Moran’s I, kriging, change-point detection.

You don’t have to read all four. Pick the one that matches the question you walked in with.

The four kinds of question this software is NOT good at

To save you time:

  • Making pretty charts. Use matplotlib, ggplot2, or D3. We produce numbers, not graphics. The numbers feed your charts.

  • Cleaning a dataset. We assume the dataset is already in the shape you want. Use pandas / dplyr for the wrangling step.

  • Predicting the future from a tweet. Forecasting needs specific time-series tooling. We have some (morie.fn TimeSeries category), but Prophet / statsforecast / GluonTS are more focused if forecasting is your only goal.

  • Replacing a stats degree. If the question is hard, the answer is usually nuanced — a tool that gives you “the answer” with no caveats is lying to you.