info on diet questionnaire

[Ruhs, Barbara]

I hope this helps!
Barb Ruhs, MS, RD
-----Original Message-----
From:	Teresa Fung [SMTP:tfung at hsph.harvard.edu]
Sent:	Wednesday, June 10, 1998 4:41 PM
To:	Ruhs, Barbara
Cc:	'dietetics'
Subject:	Re: FW: info on  diet questionnaire

HI Barbara,

The "bootstrapping" or "bootstrap" method that I know is to get an idea
if
any outliers has a big impact on the statistical analysis. However, that
doesn't mean that there isn't another "bootstrap" method to estimate
confidence interval.

The bootstrap method that I remember is:
Given a sample size of N, you can find out a bunch of statistical facts
like mean,etc. However, if there are outliers, they can really skew the
mean (or other statistical estimates). So one can create artificial
samples of size N using the orginal
sample that one has and see if the new mean (or other statistical
estimates) is similar to the original one. This is done by randomly
choose
N subjects from the orginal
sample (also has a sample size of N), but with REPLACEMENT. Therefore,
the
same subject can be in this artificial sample more than once.
Statistical
analysis is done again on this artifical sample(s). If there is big
changes in the new mean (or any other statistical estimates), there is
likely to be at least an outlier.

For example: you want to find out mean blood pressure in town X. You
randomly choose 5 people and hope that their mean will give you an
estimate of the mean blood pressure of town X (OK this is really overly
simplified). These 5 people are a,b,c,d,e and you calculate their mean
value. Now to do bootstrap. YOu have to create a new artifical sample of
size 5. YOu randomly draw a person out of your original 5, say c got
chosen. You then record the
blood pressure of person c, then put person c back to the group and
randomly draw again until you get 5 blood pressure values. Now, note
that
the second time you draw, there is a 20% chance that you will get person
c
again. If you do, do reuse person c. Keep repeating replacing drawn
subjects and keep drawing until you get to 5 blood pressure value. Take
the mean and see if it is quite different. YOu can do this to create
many
artificial samples and give you an idea of outliers. If there is no
outliers, the mean from the artificial sample(s) should be very close to
the orginial sample mean.


Sincerely,
Teresa Fung
MS,RD
Doctoral Student
Department of Epidemiology and Nutrition
Harvard School of Public Health




> -----Original Message-----

> From:	cinasalc at impsat1.com.ar [SMTP:cinasalc at impsat1.com.ar]

> Sent:	Wednesday, June 10, 1998 2:08 PM

> To:	Public Health Nutrition Discussion and Information Group

> Subject:	info on  diet questionnaire

> 

> 1)  does anyone know of any papers where you can find questionnaires

> on 

> diet and nutrition to investigate food habits? 

> 

> 2)  I have read a study where a procedure called "bootstrapping" to

> estimate

> 99% confidence interval is mentioned. Could anybody explain me what 

> "bootstrapping" means?

> 

> Thank you so much 

> Analia Fuchs, MD, MPh

> cinasalc at impsat1.com.ar