# BIOL 243 Lecture Notes - Lecture 2: Sampling Error, Cohort Study, Confounding

60 views6 pages

25 Aug 2016

School

Department

Course

Professor

BIOL 243

Week 2

Sampling & sampling error

Ideal sampling process:

Select a group of units that are a good representation for the population

Can be broken down to 4 components

1. Units must have a known and non-zero probability of being included in your sample

2. Unbiased, should not just be one sample, must include the true population

3. Independent, When selecting one unit to be in your sample, should not influence other

units being selected

4. Each possible sample has equal chance of being selected

Sampling error (Sampling variation)

It is NOT a mistake

A natural variation from one random sample to another

Designing observational studies

Main goal is to characterize an existing population

Characterize existing populations

Find correlations

Have confounding variables

Have multiple study designs

Draw backs are correlative, but not causal

oCorrelation shows the trends, but causation is rarely definable with this method

Confounding variables, are variables we haven’t observed, but are likely driving

relations to those we have observed

Retrospective vs prospective

Retrospective Prospective

Looks back in time

More likely to find correlation

Look at all who have a disease

Look for commonalities for those that

have a disease that are not common in

those who are not diseased

Can be done for a snap shot in time

Prone to confounding errors

Studies are short

Looks forward in time

More likely to find causation

Start with an initial group of people who

represent the population well

Look for individuals who disease as they

age

Find a relation between disease and age

Studies take longer

find more resources at oneclass.com

find more resources at oneclass.com

Cross-sectional surveys

1. Simple random survey

2. Stratified survey

3. Cluster survey

Try to characterize something in an entire population

For correlations within the population

4. Case-control study

5. Cohort study

Look at subparts of a population and look for correlation (People who have disease vs those

that don’t

1) Simple random survey

a. Ideal sampling design for taking a survey for a snapshot in time

a.i. Example: finding peoples income

b. Avoids problems such as bias and independence

2) Stratified survey

a. Division of a population into separate groups called Strata

b. Strata is a character used to separate sampling units

c. Income differences in a city, strata would be neighbourhoods

d. Each strata is then sampled

3) Cluster survey

a. Division of population into groups called Clusters

b. Clusters are then random chosen for sampling

4) Case-control study

a. Do not characterize entire population, only subsets

b. Useful in rare events

c. Look for traits common in one group compared to another that cause that rare event

find more resources at oneclass.com

find more resources at oneclass.com