Statistical analyses in the real world
This presentation is the property of its rightful owner.
Sponsored Links
1 / 36

Statistical analyses in the real world PowerPoint PPT Presentation


  • 46 Views
  • Uploaded on
  • Presentation posted in: General

Statistical analyses in the real world. Paul Williams Lawrence Berkeley National Laboratory. Statistical significance. Theory: The probably of observing something by chance Practice: The accepted threshold P<0.05 is publishable P>0.05 is not publishable

Download Presentation

Statistical analyses in the real world

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

Presentation Transcript


Statistical analyses in the real world

Statistical analyses in the real world

Paul Williams

Lawrence Berkeley National Laboratory


Statistical significance

Statistical significance

  • Theory: The probably of observing something by chance

  • Practice: The accepted threshold

    • P<0.05 is publishable

    • P>0.05 is not publishable

    • The promise to produce P<0.05 is a primary consideration for NIH funding


Choice of an appropriate statistical test

Choice of an appropriate statistical test

  • Familiarity to the reader, reviewer, granting agency

  • Succinctness, in consideration of journal space, simplicity in presenting results

  • Statistical tests promise the world, which we can then promised to funding agencies, journals, etc.


Classical regression model

Classical regression model


1 the real world may befuddle the description of even the simplest relationships

1. The real world may befuddle the description of even the simplest relationships


Relationship of average bmi to weekly distance walked in women

Relationship of average BMI to weekly distance walked in women

kg/m2

Williams PT.Med Sci Sports Exerc. 2005 371893-901.


Relationship of percentiles of the bmi distribution to weekly distance walked in women

Relationship of percentiles of the BMI distribution to weekly distance walked in women

kg/m2

Williams PT.Med Sci Sports Exerc. 2005 371893-901.


Convex regression curve

Convex regression curve


Regression slope by percentile

Regression slope by percentile

Slope (kg/m2 per km/wk)

Williams PT.Med Sci Sports Exerc. 2005 371893-901.


Decline in bmi per km wk run or walked

Decline in BMI per km/wk run or walked

Slope (∆BMI per 1km difference in

weekly distance)

Williams PT.Med Sci Sports Exerc. 2005 371893-901.


Correspondence between bmis of female runners and walkers

Correspondence between BMIs of female runners and walkers


Q q plot of female runners and walkers bmi distribution

Q-Q plot of female runners’ and walkers’ BMI distribution

Walkers’ BMI distribution


Estimated change in bmi per one km wk difference in running distance

Estimated change in BMI per one km/wk difference in running distance

Slope (∆BMI per 1km difference in

weekly distance)


Published applications

Published applications:

“… the effects of physical activity, alcohol, and weight

reduction on HDL-C levels may be, to a large extent,

dependent on the initial level with the greatest

improvement achieved in subjects with high HDL

and the least improvement in those having low

HDL-C levels.”

Williams PT.The relationships of vigorous exercise,

alcohol, and adiposity to low and high high-density

lipoprotein-cholesterol levels.Metabolism.

2004 Jun;53(6):700-9.


Published applications1

Published applications:

“We speculate that the reported greater increases in

triglycerides per unit of adiposity in whites than blacks,

in men than women, and in low-density lipoprotein

(LDL) pattern B than A are all consistent with the

relationships we observe.”

Williams PT. Relationship of adiposity to the population

distribution of plasma triglyceride concentrations in

vigorously active men and women. Atherosclerosis.

2004 Jun;174(2):363-71.


Published applications2

Published applications:

“We speculate that the reported greater increases in

triglycerides per unit of adiposity in whites than blacks,

in men than women, and in low-density lipoprotein

(LDL) pattern B than A are all consistent with the

relationships we observe.”

Williams PT. Relationship of adiposity to the population

distribution of plasma triglyceride concentrations in

vigorously active men and women. Atherosclerosis.

2004 Jun;174(2):363-71.


Published applications3

Published applications:

“These results are consistent with the hypothesis that

running promotes the greatest weight loss specifically

in those individuals who have the most to gain from

losing weight.”

Williams PT. Vigorous exercise and the population

distribution of body weight.Int J Obes Relat Metab

Disord. 2004 Jan;28 (1):120-8


2 the complexities of the real world may negate most statistical analyses

2. The complexities of the real world may negate most statistical analyses


Classical regression model1

Classical regression model


Statistical analyses in the real world

After assigning significance, the second most important contribution of statistics to research scientist is adjustment

  • For example, walkers may be leaner than nonwalkers but is it because they eat better.

  • Statistical adjustment is usually a sufficient argument for journals, funding agencies, etc.


Classical statistical adjustment

Classical statistical adjustment

Triglycerides (mmol/L)


Alternative statistical adjustment

Alternative statistical adjustment

% reduction

Williams PT.Metabolism

. 2004;53:700-9.


3 analysis of change data

3. Analysis of change data

DBMI=a +bDdistance


D bmi a bd distance

DBMI=a +bDdistance

Cross-sectional

relationship

BMI

Distance


Translating change data into a relationship

Translating change data into a relationship

Doesn’t correspond

To:

DBMI=a +

bDdistance

gDdistance2

Cross-sectional

relationship

BMI

The amount of change depends upon the starting and ending distance


Annual change in men s bmi by reported running distance

Annual change in men’s BMI by reported running distance


Exposure model relating d bmi to change in running distance

Exposure model relating DBMI to change in running distance

Estimated DBMI due to a “dj-cj” km/wk change

in running distance

Williams PT, Wood PD. Int J Obes (Lond). 2006 30:543-51


Exposure model to weight change by d running

Exposure model to weight change by Drunning


4 measurement error

4. Measurement error


Original interpretation of acls

Original interpretation of ACLS

Blair SN, et al. JAMA. 1995;273:1093-8.


Variables measured with error

Variables measured with error

Second fitness measurement (Treadmill duration)

Williams PT. Med Sci Sports Exerc. 2003;35:736-40.


Our interpretation of acls

Our interpretation of ACLS

Fitness measured

Fitness measured

Williams PT. Med Sci Sports Exerc. 2003;35:736-40.


Measurement error model

Measurement error model


Simulation of measurement error

Simulation of measurement error


Simulation results

Simulation results


Simulation versus reported results

Simulation versus reported results

Williams PT. Med Sci Sports Exerc. 2003;35:736-40.


  • Login