PSYC20006 Lecture Notes - Lecture 1: Null Hypothesis, Repeated Measures Design, Statistical Hypothesis Testing
1 Jul 2018
School
Department
Course
Professor
PSYC20006 Biological Psychology
WEEKS 1 - 5: STATISTICS & IMAGING METHODS
LECTURE 3 – 4 (W2): Statistical Hypothesis Testing
Statistical Hypothesis Testing
Use it because of limitations, e.g: small sample size; group not representative
of population; measurement error; random factors; luck/chance
Need to compare results to probability distribution representing chance:
oHow likely that result found by chance or if it’s a real difference
oSome extreme outcomes highly unlikely
oIt’s just SD
Construct theoretical test distribution for hypothesis that everything is due to
chance
oFind how unlikely empirical result because of chance distribution (H0
distribution) - reject if highly unlikely
oDistribution looks different depending on degrees of freedom (df)
df = # free variables given we know that the average = 0
df larger with more people tested, better estimate
Test Distribution: t-distribution
Contains: expected mean; standard error of the mean (SM)
Don't know population sd (or variance)
t-value distribution varies with df
obroader if lower; normal if larger
df calculated from sample size (n)
Experimental designs
Between-groups designs (independent-measures design)
o2 groups, values come from different people
oAdvantages
Measurements are truly independent
No concern about learning effects from repeated exposure
oDisadvantages
People in different groups might be different in various ways: IQ, motivation
etc. need large sample size or counterbalance factors that might influence results
Can’t study behaviour over time
Within group designs (repeated-measures design)
o1 group, values for both experimental conditions from same people
oAdvantage
No differences in baseline, personality, IQ, motivation etc.
Can study changes in behaviour over time
Usually can test less people
oDisadvantages
Measurements not independent need to calculate variance differently
People know treatment after 1st condition, can’t be naïve in second round
must counterbalance order of conditions to avoid unwanted order effects
1
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Weeks 1 - 5: statistics & imaging methods. Lecture 3 4 (w2): statistical hypothesis testing. Use it because of limitations, e. g: small sample size; group not representative of population; measurement error; random factors; luck/chance o o o o chance. Need to compare results to probability distribution representing chance: How likely that result found by chance or if it"s a real difference. Construct theoretical test distribution for hypothesis that everything is due to. Find how unlikely empirical result because of chance distribution (h0 distribution) - reject if highly unlikely o. Distribution looks different depending on degrees of freedom (df) df = # free variables given we know that the average = 0 df larger with more people tested, better estimate. Contains: expected mean; standard error of the mean (sm) Don"t know population sd (or variance) t-value distribution varies with df broader if lower; normal if larger df calculated from sample size (n) No concern about learning effects from repeated exposure.
