/
How to Lie With Statistics How to Lie With Statistics

How to Lie With Statistics - PowerPoint Presentation

natalia-silvester
natalia-silvester . @natalia-silvester
Follow
351 views
Uploaded On 2018-09-26

How to Lie With Statistics - PPT Presentation

Im so confident The universe is not one of clockwork Everything we measure has uncertainty How can we confidently draw conclusions and make decisions without sacrificing our intelligence Quantify Your Uncertainty ID: 680806

confidence study lancet data study confidence data lancet deaths vitamin common error cold degree interval group survey scope adjusted

Share:

Link:

Embed:

Download Presentation from below link

Download Presentation The PPT/PDF document "How to Lie With Statistics" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

Slide1

How to Lie With StatisticsSlide2
Slide3

I'm so confident!

The universe is not one of clockwork

Everything we measure has uncertainty

How can we confidently draw conclusions and make decisions without sacrificing our intelligence?Slide4

Quantify Your Uncertainty!

Family of intervals to handle this

Confidence

Prediction

Comprised of

Point estimator

Degree of Confidence

Uncertainty

Rely on

Underlying distribution (assumption)Data collectionSlide5

Formula

 

 

Z*,t*

Margin of Error

 Slide6

What could go wrong?

Poor data collection

"I talked to my 10 closest friends – Romney has got the election in the BAG!"

Wrong assumptions

Excessive uncertainty

Unrealistic expectations

Degree of confidence is too high/low

Wrong type of interval

Prediction versus ConfidenceSlide7

Degree of Confidence

What are we confident in?

The Process

If we collect data from the same population, over and over and over again, we can expect our interval to contain the true (correct) value x% of the

time where x is our

degree of confidence

(confidence level)

"The feeling or belief that

one can rely on someone

or something"

-- What I got

Googling

"What is Confidence"Slide8

Degree of Confidence

What is confidence NOT?

A probability of any single interval containing the true value

The % of all population values contained in the interval

A product of your self-esteemSlide9

Lancet Survey

October 29, 2004

1

The Lancet

published a study on the effects of the US invasion of Iraq, specifically on the number of additional civilian deaths due to the invasion

The data estimated 98,000 excess deaths

They were 95% confident that the true number of excess deaths was between 8,000 and 194,000

1

Les, R.,

Lafta

, R., Garfield, R.,

Khudhairi

, J., & Burnham, G. (2004).

"Mortality before and after the 2003 invasion of Iraq

: Cluster sample survey. The Lancet, 1857-1864.Slide10

Lancet Survey

October

11, 2006

2

The Lancet

published a follow-up study

The

data now

estimated 654,965 excess deathsThey were 95% confident that the true number of excess deaths was between 392,979

and

942,636

Claimed that they asked for death certificates 87% of the time and that 92% of the time they asked they were shown a certificate (80% of reported deaths had a certificate to verify the death)

Iraqi Ministry of Health only issued ~50,000 death certificates for post-war violent deaths

2 Burnham, G.,

Lafta, R., Doocy, S., & Roberts, L. (2006). Mortality after the 2003 invasion of Iraq

: a cross-sectional cluster sample survey. The Lancet, Retrieved from http://www.brussellstribunal.org/pdf/lancet111006.pdfSlide11

Statistical Ties

Des Moines Register Iowa Poll Results Published 9/29/2012

Obama 49%

Romney 45%

Other 4%

Undecided 2%

Margin of Error ±3.8%

Who is winning?

Register claims Obama has a 4% lead

Statistically inconclusiveThe poll results give an ~84% chance that Obama is really aheadSlide12

Multiplicity of ErrorSlide13

Multiplicity of Error

Dr. Elisabeth

Targ

made headlines when she identified an effect of remote faith healing

The study results were published in the peer-reviewed journal

The Western Journal of Medicine

under the title "A randomized double-blind study of the effect of distant healing in a population with advanced AIDS. Report of a small scale study"

Results from the double blind study showed that members of the control group spent 600% more days in the hospital and contracted 300% as many AIDS-related illnesses

Considered the most rigorous, scientifically valid experiment to demonstrate the effect of remote faith healingSlide14

Multiplicity of Error

What was learned until after the fact:

Study was designed to look for differences in mortality rates

Triple-drug Anti-Retroviral therapy became common practice one month into the six month

study

Only one person died – insufficient data to make conclusions

Adjusted the scope to look at HIV physical symptoms and Quality of Life

Also inconclusive

Adjusted scope again to look at Mood State scores

Control group performed better than treatment group

Adjusted scope again to look at Hospital Stays and Doctor Visits

Treatment group performed statistically significantly better

Adjusted scope to include at 23 illnesses associated with AIDS

Data hadn't been collected as part of the study design, so blindness was broken and data was aggregated from hospital charts

Sharpshooter fallacySpray bullets down range, then draw your target around a clusterSlide15

Multiplicity of Error

Keeping ourselves honest

Bonferroni

correction

Don't use

α

, use

where m is the number of comparisons

Conservative

Holm-

Bonferroni

correction

Compare the smallest p-value to

If it is significant, continue testing

Otherwise, stop

Compare the next smallest p-value to

Continue comparing the p-values until one is not significant, then stopLess conservativeOthers that are implemented in statistics packages (Minitab, JMP, etc…)Tukey's HSDDunnet'sAnd many more (you can't get a Ph.D. in statistics without inventing a correction to name after yourself)

 Slide16

Biggest Lie Ever Told

Vitamin C prevents or helps cure the Common Cold

Linus Pauling (an incredibly bright, talented scientist) championed the need for

megadoses

of Vitamin C to prevent or cure the Common Cold (and cancer) with the publishing of his book

Vitamin C and the Common Cold

in 1970

Had a whole institute setup to further study the benefits of

megadosing

Vitamin CAt least 30 experiments have been performed to test the ability of Vitamin C to prevent or cure the Common Cold in large groups of people (dates of experiments go from 1967-2001)

Not a single one has found a link between Vitamin C and the

Common Cold