Data & Significance Testing (Ch. 10 & 13)
Tuesday, January 29, 20131:00 PM
Chapters 10 & 13 without the sections on effect sizes
RESULTS: FROM SAMPLES TO POPULATION
Whether significant or not:
1) Look at data (frequency distribution, etc.)
2) Idea of measuring effect size
3) Carrying out the power of your test, experience, or analyses
- If psychologists need statistics or analyses of results, it’s because we never have
access to population but we’re being asked to describe it well.
- We need to go from symbols.
- Circular problem, we want to choose good samples which describe the
population, but we need to know what the population is. Important to have
many samples to describe a population well. Each sample that we test is just an
estimate, approximation of what the population is.
- Choose your sample well
o In an ideal world, your sample is your population.
- We do stats to make sure we have a certain degree of certainly to be concluding
something that is correct. Want to make sure we draw the right conclusion and if
the chance of not drawing the right conclusion is more than 5%, we don’t take
A good sample is higher in stability and less in bias
- Bias- central tendency (average)
- Bias sample must represent population
- Stability- variability
- Need to care about both central tendency of sample and how the data are
- Idea sample is always a sample where the central tendency is that of the
population (problem, we don’t know this), and strong and reliable= not too much
variation in the date.
- Good sample= representative of the population and not too variable.
If you know your population has distributed itself with more variability (ie.
stereotyping), in order to have a sample that is stable, you will need more
- If you know your population has a lot of variability, your sample needs to be large
in order to be stable.
- If not a lot of variability, you will need less subjects.
- Phenomena that is strong and not prone to a lot of variability- a good sample
that would represent this population well can be a small one, because you could
afford to have less subjects with less variability.
Therefore, when wondering how many subjects are needed for a study, consider
the size of the relationship between your variables: what you measure and what
you vary. Ie. Alzheimer’s between patients, not a lot of variability. Consider
phenomena of what you are studying. Perception science = not a lot of subjects
needed. The correlationship you measure (the phenomena) is strong.
Illustrations: frequency distribution
- Frequency polygon
- Stem-and-leaf display** (affords most detail in the data, keeps exactly all raw
data as it is, doesn’t cluster)
**Never statistically analyze your data before you know how they look, how they
- Statistical analyses compare central tendency in light of variability
- Mean (average)
- Median (50 percentile)
- Standard deviation
Important to always present central tendency with variability. Never central
tendency alone, sometimes variability alone is ok