I'm not a statistics person so having the right book can be just as important as teaching the right concepts.

My two favorite stats books are: 
1.  What is a p-value anyway? 34 stories to help you actually understand statistics (Andrew Vickers)
2.  The essential guide to effect sizes: Statistical power, meta-analysis, and the interpretation of research results (Paul Ellis)

Both are written in a clear style, use good examples, and they even manage to add humor to a generally dry subject.  I can highly recommend both and neither will break the bank--both are available through Amazon for less than $70, i.e. less than $35 for each.

Helena

Helena M. VonVille, MLS, MPH
Library Director
University of Texas School of Public Health Library Houston, TX  77030
[log in to unmask]
713.500.9131 (office)
713.500.9125 (fax)
Need help? Contact us using Ask a Librarian
Or text us: 713-965-6968



-----Original Message-----
From: Evidence based health (EBH) [mailto:[log in to unmask]] On Behalf Of Steve Simon, P.Mean Consulting
Sent: Friday, June 14, 2013 12:00 PM
To: [log in to unmask]
Subject: Re: EBM and Statistics

On 6/14/2013 6:28 AM, Weyant, Robert J wrote:

> I am looking for input on what would be considered a "minimum set"
> of statistical concepts to be taught in an intro Evidence-based Health 
> Care Course to first professional health care students (physicians, 
> nurses, dentists, etc.).  My experience has been that we
> teach "too much" stats, and turn off people.   So I am curious what
> others consider to be essential statistical knowledge  for competency 
> in EBHC.

You could argue that all of the concepts associated with the critical appraisal step are statistical in nature, but let's exclude those concepts that are associated with the design of an study (randomization, blinding, surrogate outcomes, etc.) and focus on concepts associated with data analysis.

My list of essential statistical concepts were described in Chapter 6 (What do all these numbers mean?) of my book about Evidence Based Medicine (Statistical Evidence in Medical Trials). Here's the material that I covered:
--> Samples and populations
--> Type I and II errors
--> Confidence intervals
--> P-values
--> Odds ratio, relative risk (and NNT)
--> Correlation
--> Survival curves
--> Prevalence and incidence

With the exception of samples and populations, these are listed roughly in priority order.

If I were writing the book again, I might add a bit about linear regression to the material on correlation. I was told to leave out diagnostic tests, so I did not include sensitivity, specificity, likelihood ratios, etc.

I also have a chapter on Systematic Overviews/Meta-analysis, and I introduce three other statistical concepts in that chapter:
--> the Forest Plot, and
--> the Funnel Plot.
--> Cochran's Q (and I-squared),

Again these are listed in priority. If you had to understand one concept in meta-analysis, it would be how to read and interpret a forest plot.

I agree with the comment of Dr. Ogston that we want to create consumers of statistics. The turn-off is not Statistics, per se, but the belief among many Statistics teachers that you have to know enough to be able to produce these statistics yourself.

Also, there is a strong aversion in your audience to formulas. It shouldn't be that way. Formulas are our friends and allow us to express in a single line what otherwise might take hundreds of words to describe. But it is what it is. Most people hate formulas. The only blessing for me is that all the aversion to formulas means that I can charge slightly more than the minimum wage for my consulting projects.

It is possible to teach Statistics without formulas. The formulas become very important for producers of statistics, but consumers don't need the formulas.

Here are topics that I would stay away from:
--> t-tests
--> ANOVA
--> Non-parametric tests
--> the normal distribution
--> probability concepts

These are all excellent topics, but tend to be heavy on formulas and are often of more interest to those who want to run their own studies and produce their own statistics.

If this list of essentials that I gave above is too long, I'd pare it back to just confidence intervals and p-values (which implies at least some discussion of Type I/II errors). If you don't understand confidence intervals and p-values (and appreciate all the misinterpretations that are common for these concepts) then you can't really read the primary studies effectively. The next most important thing would be odds ratios (and relative risk and NNT).

Steve Simon, [log in to unmask], Standard Disclaimer.
Sign up for the Monthly Mean, the newsletter that dares to call itself average at www.pmean.com/news