Statistical consulting

Designing the study and running the analysis for a question you have to answer, with a plain account of how sure the result is.

Write a brief Book a meeting

The parts of a study

Most engagements use a handful of the parts below. Which ones depend on the question and the decision it feeds.

Framing the question: Turning a vague ask into a question that data can answer, with the decision it feeds stated first.
Sample size: Working out how many observations are needed to tell a real effect from noise before any data is collected.
Experiment design: Setting up a test so that the comparison is fair and the result can be read without guessing.
Causal inference: Estimating the effect of a change when a clean experiment was not or could not be run.
Model choice: Picking a method that matches the question, and preferring a simpler one when it answers as well.
Margins of error: Every estimate is given with its margin of error, worked out from the data.
Checking assumptions: Testing whether the method's assumptions hold on your data, and saying where they do not.

Designing the study before reading the result

Many analyses start once the data is already in hand, and by then the important choices have been made for you. The sample is whatever happened to be collected, and nobody wrote down in advance what a convincing result would look like. An analysis built on that footing can still produce a number, but the number carries assumptions that were never checked.

The order that holds up runs the other way. Amos works out what decision hangs on the question, then settles how the data has to be gathered to answer it. He fixes how many observations are needed and which comparison will show the effect through the noise. Where a controlled test is possible, it is designed so the groups differ in a single way. Where one is not, the effect is estimated from the variation a business already produces, and the weaknesses of that route are set out in the report.

Amos writes the analysis code himself, so the person who chose the method is also the one who reports its limits.

How to start

The brief

Write down the question and what you will do once it is answered. Say what data you hold or can collect, and by when you need the answer.

Before collection

Where possible the study is designed before any data is gathered, while the sample size and the comparison can still be chosen.

Who does the work

Amos Weiskopf studied geography at the Hebrew University of Jerusalem and computer science at the University of York, and he writes the code his analysis rests on.

Questions buyers ask

What is statistical consulting?

It is help designing a study and analysing the data for a business or research question, from deciding the sample size to estimating an effect where no experiment was run.

How much does statistical consulting cost?

It depends on the method the question calls for and how much data work it involves. The price is agreed after the brief.

Do I need an experiment, or can you work from data I already have?

Both are possible. A designed experiment gives the cleanest answer. Where one cannot be run, the effect can often be estimated from variation your business already produces, with the limits of that route stated.

How do you decide the sample size?

It comes from how large an effect would matter to your decision and how much the data varies. Amos works this out before collection starts, so you know whether the planned study can tell a real effect from chance.

What happens if the data cannot answer the question?

The report says so plainly, whether the problem is a sample that is too small or two explanations the data cannot separate.

Who runs the analysis?

Amos does, from the design of the study to the code that runs it.

Engagements by written brief.

Work begins from a documented problem statement, not a sales call.

Write a brief Or book a meeting