STATISTICAL CONSULTATION FOR YOUR RESEARCH PROJECT

Dan Tandberg, MD

Professor Emeritus

Department of Emergency Medicine

UNM School of Medicine

http://www.nmia.com/~tandberg

ALWAYS USE A STATISTICAL CONSULTANT FOR ANY RESEARCH PROJECT!

They will help you with design and variable measurement issues that will increase the chances that you end up with an analyzable data set.

There are many inherent subjectivities in Frequentist statistical analysis [model selection, data distribution, curve fitting algorithm, multiple comparison procedures, etc.]. You need someone outside of your project workgroup to make sure that these choices are sensible.

HOW SHOULD YOU USE A STATISTICAL CONSULTANT?

Involve them early [during the design phase]

Use them for what they're good at

[see next few sections]

PLANNING & DESIGN OF EXPERIMENTS AND STUDIES

Study design

Subject selection

Exclusion criteria

Blinding

Reduction of bias

Selection of controls

Survey instrument design

Variable coding

Human/animal subjects issues

SAMPLE SIZE ESTIMATION

How much data do you need to answer your question(s)?

How much variability should you expect?

How big of a difference do you wish to be able to detect?

What are the acceptable Type I and Type II error rates?

PREPARATION FOR DATA COLLECTION

Collecting and recording your data

Computer analysis issues

Error avoidance

Data file security

PREPARATION OF DATA FOR ANALYSIS

Transfer to electronic media

Text, database, or spreadsheet format

Translate to statistical package formats

Error checking

Missing value coding

STATISTICAL ANALYSIS OF DATA*

Counts

Proportions

Comparisons

Exploratory data analysis

Confidence intervals

Hypothesis tests

Multivariate methods

PRESENTATION OF RESULTS

How should you present your findings?

Tables?

Plots?

Text?

WRITING MANUSCRIPTS

Approaches to writing up your project

[best order=tables & graphs, methods, results, intro, discussion, refs, title, abstract]

Efficient handling of references [‘endnotes’ in Microsoft word]

Selection of an appropriate journal

"Writers' block"

Similar published paper

REVIEW OF MANUSCRIPTS

You only get one good "read" from each individual you give your paper to.

WHO CAN HELP STUDENTS WITH STATISTICAL SUPPORT AT UNMSOM?

If your mentor does not have a departmental statistical consultant, these people can often be of help to students.

Betty Skipper, PhD

Cameron Crandall, MD PhD

Chris Stidley, PhD

Clifford Qualls, PhD

Dan Tandberg, MD

Ed Bedrick, PhD

Kristine Tollestrup, PhD

Ron Schrader, PhD

DON’T

Try to do it without outside help

Expect you mentor to know it all

Experiment on people without their consent

Lose your data

Show up with a ‘box’ of data you’ve already collected

Collect the wrong amount of data

Wait ‘til the night before the deadline