Section 1.3 - Data collection May 20, 2013 (1040)

advertisement
Section 1.3 - Data collection
Summer 2013 - Math 1040
May 20, 2013
(1040)
M 1040 1.3
May 20, 2013
1/8
Roadmap
Data collection.
Experimental design.
Sampling techniques.
(1040)
M 1040 1.3
May 20, 2013
2/8
Observational Study vs Experiment
Observational studies gather data without influencing the response. The
data are direct observations and measurements.
Example: A finance and risk engineering professor analyzed songs on the
Billboard Hot 100 found songs with a high “beat variance” are popular
when the stock market is calm and songs with steadier beats are hot when
the market is volatile. (All Things Considered, January 22, 2009) This is
the theory of Phil Maymin from New York University from an
observational study.
(1040)
M 1040 1.3
May 20, 2013
3/8
Surveys and simulations
In addition, data can be collected using a survey. The challenge is to
measure a characteristic of a population (people in this case) without
influencing the results.
If situations are impractical, unethical, or too dangerous to administer on
an actual population, a simulation can be used instead. Simulations use a
mathematical, statistical, or physical model to reproduce the conditions of
a situation or process.
(1040)
M 1040 1.3
May 20, 2013
4/8
Experimental design
Experiments gather data by administering different levels of treatments
to samples from a population, and responses to the treatments are
observed. Control groups, in which no treatments are applied, may be
used as a baseline.
A well-designed experimental has control over influential factors, are
randomized, and are repeatable.
Influential factors include confounding variables, placebo effects, and
possile reactivity effects such as the Hawthorne effect.
(1040)
M 1040 1.3
May 20, 2013
5/8
Bias
We wish to minimize bias that influence our measurements when otherwise
they would not have an effect. Biased results can occur:
When potential lurking/confounding variables are not identified.
When the subjects know which experimental group they belong to.
When the sample is not randomized.
When survey questions impose biases.
(1040)
M 1040 1.3
May 20, 2013
6/8
Randomization, that is, randomly assigning subjects to different
treatment groups, together with large sample sizes, improve the validity of
experimental results. Blocks, or groups of subjects with similar traits, will
also improve results.
Replication is the repetition of an experiment under similar or the same
conditions.
Example: In the late 1980’s Fleischmann and Jones announced, upon
pressure of The University of Utah, that they have achieved discovery of
cold fusion. Physicists in other countries failed at replication of the
experiment.
(1040)
M 1040 1.3
May 20, 2013
7/8
Assignements
Assignment:
1
Read pages 16 through 22.
2
The recommended exercises. Monday: Exercises 1-16
Vocabulary:
Observational study, experiment, survey, simulation, confounding variables,
randomization, replication. Bias.
Understand:
Identify the best type of data collection for a study, the design of an
experiment, and possible biases that may occur.
(1040)
M 1040 1.3
May 20, 2013
8/8
Download