sampling & estimation

27
Sampling & Estimation

Upload: conley

Post on 29-Jan-2016

61 views

Category:

Documents


0 download

DESCRIPTION

Sampling & Estimation. Normal Distribution. Normal Sample. Binomial Distribution. Estimation. Sampling. Sampling of the Mean. The more observations the better!. Surprice!!!!!. Sampling of the Variance. Sampling of the proportion. How accurate are these estimates?. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Sampling & Estimation

Sampling & Estimation

Page 2: Sampling & Estimation

Normal Distribution

Page 3: Sampling & Estimation

Normal Sample

Page 4: Sampling & Estimation

Binomial Distribution

Page 5: Sampling & Estimation

Estimation

Page 6: Sampling & Estimation
Page 7: Sampling & Estimation

Sampling

Page 8: Sampling & Estimation

Sampling of the Mean

Page 9: Sampling & Estimation

The more observations the better!

Surprice!!!!!

Page 10: Sampling & Estimation

Sampling of the Variance

Page 11: Sampling & Estimation

Sampling of the proportion

Page 12: Sampling & Estimation

How accurate are these estimates?

Can we use that to report the uncertainty

in a clever way?

Page 13: Sampling & Estimation

Rule of

A random variable is very seldom more than two standard deviations away from the expected value.

A random variable is very seldom more than two standard deviations away from the expected value.

Page 14: Sampling & Estimation

… Ehh, we don’t know that one!

Page 15: Sampling & Estimation

Confidence Interval for the Mean when the variance is not know

Page 16: Sampling & Estimation

Confidence intervals for the variance

It looks like …..

Page 17: Sampling & Estimation

A 95% approximate interval for a proportion

Assume normality

BUT WHAT IF THIS INTERVAL

CONTAINS 0 OR 1?This would be possible if n is small, if p is nearly zero or if p is nearly one.

Page 18: Sampling & Estimation

Log-Transformation

Believe me!Assume normality

Use the expontial transformation, and write

But what if the interval contains

one?

This could happen if n is relatively small and p is nearly one.

Page 19: Sampling & Estimation

Logit-transformation

and it also looks like log(1-p), for p approx one.

Looks like the

log-transformation, for p small

To go the other way

Page 20: Sampling & Estimation

The function logit(p) The function expit(p)

Page 21: Sampling & Estimation

Logit-transformation

Assume normality

To get a 95% CI for p, we use the expit-transformation

Now we are happy!

Page 22: Sampling & Estimation

Why didn’t I just tell you about the logit-transformation in the first place?Because, when comparing proportions (risks), you may consider

To get 95% CI here, you’ll need all three approaches.

Page 23: Sampling & Estimation

How to calculate CI’s in SPSS

• It is easy (sort of) in the case of normally distributed variables

• More or less impossible in case of binomial (Use Excel)

Page 24: Sampling & Estimation

Assume we have a dataset with a variable called: Alcohol

Hmmmm

Page 25: Sampling & Estimation

Choose

• Analyze

• General Linear Model

• Univariate

Choose

• Analyze

• General Linear Model

• Univariate

Page 26: Sampling & Estimation

• Drag the variable Alcohol into Dependent Variable

• Click Options

• Choose Parameter estimates

• Drag the variable Alcohol into Dependent Variable

• Click Options

• Choose Parameter estimates

Page 27: Sampling & Estimation

… And now we get